Index: head/sys/net/altq/altq_cbq.c =================================================================== --- head/sys/net/altq/altq_cbq.c (revision 343994) +++ head/sys/net/altq/altq_cbq.c (revision 343995) @@ -1,565 +1,564 @@ /*- * Copyright (c) Sun Microsystems, Inc. 1993-1998 All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the SMCC Technology * Development Group at Sun Microsystems, Inc. * * 4. The name of the Sun Microsystems, Inc nor may not be used to endorse or * promote products derived from this software without specific prior * written permission. * * SUN MICROSYSTEMS DOES NOT CLAIM MERCHANTABILITY OF THIS SOFTWARE OR THE * SUITABILITY OF THIS SOFTWARE FOR ANY PARTICULAR PURPOSE. The software is * provided "as is" without express or implied warranty of any kind. * * These notices must be retained in any copies of any part of this software. * * $KAME: altq_cbq.c,v 1.19 2003/09/17 14:23:25 kjc Exp $ * $FreeBSD$ */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef ALTQ_CBQ /* cbq is enabled by ALTQ_CBQ option in opt_altq.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * Forward Declarations. */ static int cbq_class_destroy(cbq_state_t *, struct rm_class *); static struct rm_class *clh_to_clp(cbq_state_t *, u_int32_t); static int cbq_clear_interface(cbq_state_t *); static int cbq_request(struct ifaltq *, int, void *); static int cbq_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); static struct mbuf *cbq_dequeue(struct ifaltq *, int); static void cbqrestart(struct ifaltq *); static void get_class_stats(class_stats_t *, struct rm_class *); static void cbq_purge(cbq_state_t *); /* * int * cbq_class_destroy(cbq_mod_state_t *, struct rm_class *) - This * function destroys a given traffic class. Before destroying * the class, all traffic for that class is released. */ static int cbq_class_destroy(cbq_state_t *cbqp, struct rm_class *cl) { int i; /* delete the class */ rmc_delete_class(&cbqp->ifnp, cl); /* * free the class handle */ for (i = 0; i < CBQ_MAX_CLASSES; i++) if (cbqp->cbq_class_tbl[i] == cl) cbqp->cbq_class_tbl[i] = NULL; if (cl == cbqp->ifnp.root_) cbqp->ifnp.root_ = NULL; if (cl == cbqp->ifnp.default_) cbqp->ifnp.default_ = NULL; return (0); } /* convert class handle to class pointer */ static struct rm_class * clh_to_clp(cbq_state_t *cbqp, u_int32_t chandle) { int i; struct rm_class *cl; if (chandle == 0) return (NULL); /* * first, try optimistically the slot matching the lower bits of * the handle. if it fails, do the linear table search. */ i = chandle % CBQ_MAX_CLASSES; if ((cl = cbqp->cbq_class_tbl[i]) != NULL && cl->stats_.handle == chandle) return (cl); for (i = 0; i < CBQ_MAX_CLASSES; i++) if ((cl = cbqp->cbq_class_tbl[i]) != NULL && cl->stats_.handle == chandle) return (cl); return (NULL); } static int cbq_clear_interface(cbq_state_t *cbqp) { int again, i; struct rm_class *cl; #ifdef ALTQ3_CLFIER_COMPAT /* free the filters for this interface */ acc_discard_filters(&cbqp->cbq_classifier, NULL, 1); #endif /* clear out the classes now */ do { again = 0; for (i = 0; i < CBQ_MAX_CLASSES; i++) { if ((cl = cbqp->cbq_class_tbl[i]) != NULL) { if (is_a_parent_class(cl)) again++; else { cbq_class_destroy(cbqp, cl); cbqp->cbq_class_tbl[i] = NULL; if (cl == cbqp->ifnp.root_) cbqp->ifnp.root_ = NULL; if (cl == cbqp->ifnp.default_) cbqp->ifnp.default_ = NULL; } } } } while (again); return (0); } static int cbq_request(struct ifaltq *ifq, int req, void *arg) { cbq_state_t *cbqp = (cbq_state_t *)ifq->altq_disc; IFQ_LOCK_ASSERT(ifq); switch (req) { case ALTRQ_PURGE: cbq_purge(cbqp); break; } return (0); } /* copy the stats info in rm_class to class_states_t */ static void get_class_stats(class_stats_t *statsp, struct rm_class *cl) { statsp->xmit_cnt = cl->stats_.xmit_cnt; statsp->drop_cnt = cl->stats_.drop_cnt; statsp->over = cl->stats_.over; statsp->borrows = cl->stats_.borrows; statsp->overactions = cl->stats_.overactions; statsp->delays = cl->stats_.delays; statsp->depth = cl->depth_; statsp->priority = cl->pri_; statsp->maxidle = cl->maxidle_; statsp->minidle = cl->minidle_; statsp->offtime = cl->offtime_; statsp->qmax = qlimit(cl->q_); statsp->ns_per_byte = cl->ns_per_byte_; statsp->wrr_allot = cl->w_allotment_; statsp->qcnt = qlen(cl->q_); statsp->avgidle = cl->avgidle_; statsp->qtype = qtype(cl->q_); #ifdef ALTQ_RED if (q_is_red(cl->q_)) red_getstats(cl->red_, &statsp->red[0]); #endif #ifdef ALTQ_RIO if (q_is_rio(cl->q_)) rio_getstats((rio_t *)cl->red_, &statsp->red[0]); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->q_)) codel_getstats(cl->codel_, &statsp->codel); #endif } int cbq_pfattach(struct pf_altq *a) { struct ifnet *ifp; int s, error; if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) return (EINVAL); s = splnet(); error = altq_attach(&ifp->if_snd, ALTQT_CBQ, a->altq_disc, cbq_enqueue, cbq_dequeue, cbq_request, NULL, NULL); splx(s); return (error); } int -cbq_add_altq(struct pf_altq *a) +cbq_add_altq(struct ifnet *ifp, struct pf_altq *a) { cbq_state_t *cbqp; - struct ifnet *ifp; - if ((ifp = ifunit(a->ifname)) == NULL) + if (ifp == NULL) return (EINVAL); if (!ALTQ_IS_READY(&ifp->if_snd)) return (ENODEV); /* allocate and initialize cbq_state_t */ cbqp = malloc(sizeof(cbq_state_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (cbqp == NULL) return (ENOMEM); CALLOUT_INIT(&cbqp->cbq_callout); cbqp->cbq_qlen = 0; cbqp->ifnp.ifq_ = &ifp->if_snd; /* keep the ifq */ /* keep the state in pf_altq */ a->altq_disc = cbqp; return (0); } int cbq_remove_altq(struct pf_altq *a) { cbq_state_t *cbqp; if ((cbqp = a->altq_disc) == NULL) return (EINVAL); a->altq_disc = NULL; cbq_clear_interface(cbqp); if (cbqp->ifnp.default_) cbq_class_destroy(cbqp, cbqp->ifnp.default_); if (cbqp->ifnp.root_) cbq_class_destroy(cbqp, cbqp->ifnp.root_); /* deallocate cbq_state_t */ free(cbqp, M_DEVBUF); return (0); } int cbq_add_queue(struct pf_altq *a) { struct rm_class *borrow, *parent; cbq_state_t *cbqp; struct rm_class *cl; struct cbq_opts *opts; int i; if ((cbqp = a->altq_disc) == NULL) return (EINVAL); if (a->qid == 0) return (EINVAL); /* * find a free slot in the class table. if the slot matching * the lower bits of qid is free, use this slot. otherwise, * use the first free slot. */ i = a->qid % CBQ_MAX_CLASSES; if (cbqp->cbq_class_tbl[i] != NULL) { for (i = 0; i < CBQ_MAX_CLASSES; i++) if (cbqp->cbq_class_tbl[i] == NULL) break; if (i == CBQ_MAX_CLASSES) return (EINVAL); } opts = &a->pq_u.cbq_opts; /* check parameters */ if (a->priority >= CBQ_MAXPRI) return (EINVAL); /* Get pointers to parent and borrow classes. */ parent = clh_to_clp(cbqp, a->parent_qid); if (opts->flags & CBQCLF_BORROW) borrow = parent; else borrow = NULL; /* * A class must borrow from it's parent or it can not * borrow at all. Hence, borrow can be null. */ if (parent == NULL && (opts->flags & CBQCLF_ROOTCLASS) == 0) { printf("cbq_add_queue: no parent class!\n"); return (EINVAL); } if ((borrow != parent) && (borrow != NULL)) { printf("cbq_add_class: borrow class != parent\n"); return (EINVAL); } /* * check parameters */ switch (opts->flags & CBQCLF_CLASSMASK) { case CBQCLF_ROOTCLASS: if (parent != NULL) return (EINVAL); if (cbqp->ifnp.root_) return (EINVAL); break; case CBQCLF_DEFCLASS: if (cbqp->ifnp.default_) return (EINVAL); break; case 0: if (a->qid == 0) return (EINVAL); break; default: /* more than two flags bits set */ return (EINVAL); } /* * create a class. if this is a root class, initialize the * interface. */ if ((opts->flags & CBQCLF_CLASSMASK) == CBQCLF_ROOTCLASS) { rmc_init(cbqp->ifnp.ifq_, &cbqp->ifnp, opts->ns_per_byte, cbqrestart, a->qlimit, RM_MAXQUEUED, opts->maxidle, opts->minidle, opts->offtime, opts->flags); cl = cbqp->ifnp.root_; } else { cl = rmc_newclass(a->priority, &cbqp->ifnp, opts->ns_per_byte, rmc_delay_action, a->qlimit, parent, borrow, opts->maxidle, opts->minidle, opts->offtime, opts->pktsize, opts->flags); } if (cl == NULL) return (ENOMEM); /* return handle to user space. */ cl->stats_.handle = a->qid; cl->stats_.depth = cl->depth_; /* save the allocated class */ cbqp->cbq_class_tbl[i] = cl; if ((opts->flags & CBQCLF_CLASSMASK) == CBQCLF_DEFCLASS) cbqp->ifnp.default_ = cl; return (0); } int cbq_remove_queue(struct pf_altq *a) { struct rm_class *cl; cbq_state_t *cbqp; int i; if ((cbqp = a->altq_disc) == NULL) return (EINVAL); if ((cl = clh_to_clp(cbqp, a->qid)) == NULL) return (EINVAL); /* if we are a parent class, then return an error. */ if (is_a_parent_class(cl)) return (EINVAL); /* delete the class */ rmc_delete_class(&cbqp->ifnp, cl); /* * free the class handle */ for (i = 0; i < CBQ_MAX_CLASSES; i++) if (cbqp->cbq_class_tbl[i] == cl) { cbqp->cbq_class_tbl[i] = NULL; if (cl == cbqp->ifnp.root_) cbqp->ifnp.root_ = NULL; if (cl == cbqp->ifnp.default_) cbqp->ifnp.default_ = NULL; break; } return (0); } int cbq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { cbq_state_t *cbqp; struct rm_class *cl; class_stats_t stats; int error = 0; if ((cbqp = altq_lookup(a->ifname, ALTQT_CBQ)) == NULL) return (EBADF); if ((cl = clh_to_clp(cbqp, a->qid)) == NULL) return (EINVAL); if (*nbytes < sizeof(stats)) return (EINVAL); get_class_stats(&stats, cl); if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0) return (error); *nbytes = sizeof(stats); return (0); } /* * int * cbq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pattr) * - Queue data packets. * * cbq_enqueue is set to ifp->if_altqenqueue and called by an upper * layer (e.g. ether_output). cbq_enqueue queues the given packet * to the cbq, then invokes the driver's start routine. * * Assumptions: called in splimp * Returns: 0 if the queueing is successful. * ENOBUFS if a packet dropping occurred as a result of * the queueing. */ static int cbq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) { cbq_state_t *cbqp = (cbq_state_t *)ifq->altq_disc; struct rm_class *cl; struct pf_mtag *t; int len; IFQ_LOCK_ASSERT(ifq); /* grab class set by classifier */ if ((m->m_flags & M_PKTHDR) == 0) { /* should not happen */ printf("altq: packet for %s does not have pkthdr\n", ifq->altq_ifp->if_xname); m_freem(m); return (ENOBUFS); } cl = NULL; if ((t = pf_find_mtag(m)) != NULL) cl = clh_to_clp(cbqp, t->qid); if (cl == NULL) { cl = cbqp->ifnp.default_; if (cl == NULL) { m_freem(m); return (ENOBUFS); } } cl->pktattr_ = NULL; len = m_pktlen(m); if (rmc_queue_packet(cl, m) != 0) { /* drop occurred. some mbuf was freed in rmc_queue_packet. */ PKTCNTR_ADD(&cl->stats_.drop_cnt, len); return (ENOBUFS); } /* successfully queued. */ ++cbqp->cbq_qlen; IFQ_INC_LEN(ifq); return (0); } static struct mbuf * cbq_dequeue(struct ifaltq *ifq, int op) { cbq_state_t *cbqp = (cbq_state_t *)ifq->altq_disc; struct mbuf *m; IFQ_LOCK_ASSERT(ifq); m = rmc_dequeue_next(&cbqp->ifnp, op); if (m && op == ALTDQ_REMOVE) { --cbqp->cbq_qlen; /* decrement # of packets in cbq */ IFQ_DEC_LEN(ifq); /* Update the class. */ rmc_update_class_util(&cbqp->ifnp); } return (m); } /* * void * cbqrestart(queue_t *) - Restart sending of data. * called from rmc_restart in splimp via timeout after waking up * a suspended class. * Returns: NONE */ static void cbqrestart(struct ifaltq *ifq) { cbq_state_t *cbqp; struct ifnet *ifp; IFQ_LOCK_ASSERT(ifq); if (!ALTQ_IS_ENABLED(ifq)) /* cbq must have been detached */ return; if ((cbqp = (cbq_state_t *)ifq->altq_disc) == NULL) /* should not happen */ return; ifp = ifq->altq_ifp; if (ifp->if_start && cbqp->cbq_qlen > 0 && (ifp->if_drv_flags & IFF_DRV_OACTIVE) == 0) { IFQ_UNLOCK(ifq); (*ifp->if_start)(ifp); IFQ_LOCK(ifq); } } static void cbq_purge(cbq_state_t *cbqp) { struct rm_class *cl; int i; for (i = 0; i < CBQ_MAX_CLASSES; i++) if ((cl = cbqp->cbq_class_tbl[i]) != NULL) rmc_dropall(cl); if (ALTQ_IS_ENABLED(cbqp->ifnp.ifq_)) cbqp->ifnp.ifq_->ifq_len = 0; } #endif /* ALTQ_CBQ */ Index: head/sys/net/altq/altq_codel.c =================================================================== --- head/sys/net/altq/altq_codel.c (revision 343994) +++ head/sys/net/altq/altq_codel.c (revision 343995) @@ -1,477 +1,476 @@ /* * CoDel - The Controlled-Delay Active Queue Management algorithm * * Copyright (C) 2013 Ermal Luçi * Copyright (C) 2011-2012 Kathleen Nichols * Copyright (C) 2011-2012 Van Jacobson * Copyright (C) 2012 Michael D. Taht * Copyright (C) 2012 Eric Dumazet * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. The names of the authors may not be used to endorse or promote products * derived from this software without specific prior written permission. * * Alternatively, provided that this notice is retained in full, this * software may be distributed under the terms of the GNU General * Public License ("GPL") version 2, in which case the provisions of the * GPL apply INSTEAD OF those given above. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. * * $FreeBSD$ */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef ALTQ_CODEL /* CoDel is enabled by ALTQ_CODEL option in opt_altq.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include static int codel_should_drop(struct codel *, class_queue_t *, struct mbuf *, u_int64_t); static void codel_Newton_step(struct codel_vars *); static u_int64_t codel_control_law(u_int64_t t, u_int64_t, u_int32_t); #define codel_time_after(a, b) ((int64_t)(a) - (int64_t)(b) > 0) #define codel_time_after_eq(a, b) ((int64_t)(a) - (int64_t)(b) >= 0) #define codel_time_before(a, b) ((int64_t)(a) - (int64_t)(b) < 0) #define codel_time_before_eq(a, b) ((int64_t)(a) - (int64_t)(b) <= 0) static int codel_request(struct ifaltq *, int, void *); static int codel_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); static struct mbuf *codel_dequeue(struct ifaltq *, int); int codel_pfattach(struct pf_altq *a) { struct ifnet *ifp; if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) return (EINVAL); return (altq_attach(&ifp->if_snd, ALTQT_CODEL, a->altq_disc, codel_enqueue, codel_dequeue, codel_request, NULL, NULL)); } int -codel_add_altq(struct pf_altq *a) +codel_add_altq(struct ifnet *ifp, struct pf_altq *a) { struct codel_if *cif; - struct ifnet *ifp; struct codel_opts *opts; - if ((ifp = ifunit(a->ifname)) == NULL) + if (ifp == NULL) return (EINVAL); if (!ALTQ_IS_READY(&ifp->if_snd)) return (ENODEV); opts = &a->pq_u.codel_opts; cif = malloc(sizeof(struct codel_if), M_DEVBUF, M_NOWAIT | M_ZERO); if (cif == NULL) return (ENOMEM); cif->cif_bandwidth = a->ifbandwidth; cif->cif_ifq = &ifp->if_snd; cif->cl_q = malloc(sizeof(class_queue_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (cif->cl_q == NULL) { free(cif, M_DEVBUF); return (ENOMEM); } if (a->qlimit == 0) a->qlimit = 50; /* use default. */ qlimit(cif->cl_q) = a->qlimit; qtype(cif->cl_q) = Q_CODEL; qlen(cif->cl_q) = 0; qsize(cif->cl_q) = 0; if (opts->target == 0) opts->target = 5; if (opts->interval == 0) opts->interval = 100; cif->codel.params.target = machclk_freq * opts->target / 1000; cif->codel.params.interval = machclk_freq * opts->interval / 1000; cif->codel.params.ecn = opts->ecn; cif->codel.stats.maxpacket = 256; cif->cl_stats.qlength = qlen(cif->cl_q); cif->cl_stats.qlimit = qlimit(cif->cl_q); /* keep the state in pf_altq */ a->altq_disc = cif; return (0); } int codel_remove_altq(struct pf_altq *a) { struct codel_if *cif; if ((cif = a->altq_disc) == NULL) return (EINVAL); a->altq_disc = NULL; if (cif->cl_q) free(cif->cl_q, M_DEVBUF); free(cif, M_DEVBUF); return (0); } int codel_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { struct codel_if *cif; struct codel_ifstats stats; int error = 0; if ((cif = altq_lookup(a->ifname, ALTQT_CODEL)) == NULL) return (EBADF); if (*nbytes < sizeof(stats)) return (EINVAL); stats = cif->cl_stats; stats.stats = cif->codel.stats; if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0) return (error); *nbytes = sizeof(stats); return (0); } static int codel_request(struct ifaltq *ifq, int req, void *arg) { struct codel_if *cif = (struct codel_if *)ifq->altq_disc; struct mbuf *m; IFQ_LOCK_ASSERT(ifq); switch (req) { case ALTRQ_PURGE: if (!ALTQ_IS_ENABLED(cif->cif_ifq)) break; if (qempty(cif->cl_q)) break; while ((m = _getq(cif->cl_q)) != NULL) { PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m)); m_freem(m); IFQ_DEC_LEN(cif->cif_ifq); } cif->cif_ifq->ifq_len = 0; break; } return (0); } static int codel_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) { struct codel_if *cif = (struct codel_if *) ifq->altq_disc; IFQ_LOCK_ASSERT(ifq); /* grab class set by classifier */ if ((m->m_flags & M_PKTHDR) == 0) { /* should not happen */ printf("altq: packet for %s does not have pkthdr\n", ifq->altq_ifp->if_xname); m_freem(m); PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m)); return (ENOBUFS); } if (codel_addq(&cif->codel, cif->cl_q, m)) { PKTCNTR_ADD(&cif->cl_stats.cl_dropcnt, m_pktlen(m)); return (ENOBUFS); } IFQ_INC_LEN(ifq); return (0); } static struct mbuf * codel_dequeue(struct ifaltq *ifq, int op) { struct codel_if *cif = (struct codel_if *)ifq->altq_disc; struct mbuf *m; IFQ_LOCK_ASSERT(ifq); if (IFQ_IS_EMPTY(ifq)) return (NULL); if (op == ALTDQ_POLL) return (qhead(cif->cl_q)); m = codel_getq(&cif->codel, cif->cl_q); if (m != NULL) { IFQ_DEC_LEN(ifq); PKTCNTR_ADD(&cif->cl_stats.cl_xmitcnt, m_pktlen(m)); return (m); } return (NULL); } struct codel * codel_alloc(int target, int interval, int ecn) { struct codel *c; c = malloc(sizeof(*c), M_DEVBUF, M_NOWAIT | M_ZERO); if (c != NULL) { c->params.target = machclk_freq * target / 1000; c->params.interval = machclk_freq * interval / 1000; c->params.ecn = ecn; c->stats.maxpacket = 256; } return (c); } void codel_destroy(struct codel *c) { free(c, M_DEVBUF); } #define MTAG_CODEL 1438031249 int codel_addq(struct codel *c, class_queue_t *q, struct mbuf *m) { struct m_tag *mtag; uint64_t *enqueue_time; if (qlen(q) < qlimit(q)) { mtag = m_tag_locate(m, MTAG_CODEL, 0, NULL); if (mtag == NULL) mtag = m_tag_alloc(MTAG_CODEL, 0, sizeof(uint64_t), M_NOWAIT); if (mtag == NULL) { m_freem(m); return (-1); } enqueue_time = (uint64_t *)(mtag + 1); *enqueue_time = read_machclk(); m_tag_prepend(m, mtag); _addq(q, m); return (0); } c->drop_overlimit++; m_freem(m); return (-1); } static int codel_should_drop(struct codel *c, class_queue_t *q, struct mbuf *m, u_int64_t now) { struct m_tag *mtag; uint64_t *enqueue_time; if (m == NULL) { c->vars.first_above_time = 0; return (0); } mtag = m_tag_locate(m, MTAG_CODEL, 0, NULL); if (mtag == NULL) { /* Only one warning per second. */ if (ppsratecheck(&c->last_log, &c->last_pps, 1)) printf("%s: could not found the packet mtag!\n", __func__); c->vars.first_above_time = 0; return (0); } enqueue_time = (uint64_t *)(mtag + 1); c->vars.ldelay = now - *enqueue_time; c->stats.maxpacket = MAX(c->stats.maxpacket, m_pktlen(m)); if (codel_time_before(c->vars.ldelay, c->params.target) || qsize(q) <= c->stats.maxpacket) { /* went below - stay below for at least interval */ c->vars.first_above_time = 0; return (0); } if (c->vars.first_above_time == 0) { /* just went above from below. If we stay above * for at least interval we'll say it's ok to drop */ c->vars.first_above_time = now + c->params.interval; return (0); } if (codel_time_after(now, c->vars.first_above_time)) return (1); return (0); } /* * Run a Newton method step: * new_invsqrt = (invsqrt / 2) * (3 - count * invsqrt^2) * * Here, invsqrt is a fixed point number (< 1.0), 32bit mantissa, aka Q0.32 */ static void codel_Newton_step(struct codel_vars *vars) { uint32_t invsqrt, invsqrt2; uint64_t val; /* sizeof_in_bits(rec_inv_sqrt) */ #define REC_INV_SQRT_BITS (8 * sizeof(u_int16_t)) /* needed shift to get a Q0.32 number from rec_inv_sqrt */ #define REC_INV_SQRT_SHIFT (32 - REC_INV_SQRT_BITS) invsqrt = ((u_int32_t)vars->rec_inv_sqrt) << REC_INV_SQRT_SHIFT; invsqrt2 = ((u_int64_t)invsqrt * invsqrt) >> 32; val = (3LL << 32) - ((u_int64_t)vars->count * invsqrt2); val >>= 2; /* avoid overflow in following multiply */ val = (val * invsqrt) >> (32 - 2 + 1); vars->rec_inv_sqrt = val >> REC_INV_SQRT_SHIFT; } static u_int64_t codel_control_law(u_int64_t t, u_int64_t interval, u_int32_t rec_inv_sqrt) { return (t + (u_int32_t)(((u_int64_t)interval * (rec_inv_sqrt << REC_INV_SQRT_SHIFT)) >> 32)); } struct mbuf * codel_getq(struct codel *c, class_queue_t *q) { struct mbuf *m; u_int64_t now; int drop; if ((m = _getq(q)) == NULL) { c->vars.dropping = 0; return (m); } now = read_machclk(); drop = codel_should_drop(c, q, m, now); if (c->vars.dropping) { if (!drop) { /* sojourn time below target - leave dropping state */ c->vars.dropping = 0; } else if (codel_time_after_eq(now, c->vars.drop_next)) { /* It's time for the next drop. Drop the current * packet and dequeue the next. The dequeue might * take us out of dropping state. * If not, schedule the next drop. * A large backlog might result in drop rates so high * that the next drop should happen now, * hence the while loop. */ while (c->vars.dropping && codel_time_after_eq(now, c->vars.drop_next)) { c->vars.count++; /* don't care of possible wrap * since there is no more * divide */ codel_Newton_step(&c->vars); /* TODO ECN */ PKTCNTR_ADD(&c->stats.drop_cnt, m_pktlen(m)); m_freem(m); m = _getq(q); if (!codel_should_drop(c, q, m, now)) /* leave dropping state */ c->vars.dropping = 0; else /* and schedule the next drop */ c->vars.drop_next = codel_control_law(c->vars.drop_next, c->params.interval, c->vars.rec_inv_sqrt); } } } else if (drop) { /* TODO ECN */ PKTCNTR_ADD(&c->stats.drop_cnt, m_pktlen(m)); m_freem(m); m = _getq(q); drop = codel_should_drop(c, q, m, now); c->vars.dropping = 1; /* if min went above target close to when we last went below it * assume that the drop rate that controlled the queue on the * last cycle is a good starting point to control it now. */ if (codel_time_before(now - c->vars.drop_next, 16 * c->params.interval)) { c->vars.count = (c->vars.count - c->vars.lastcount) | 1; /* we dont care if rec_inv_sqrt approximation * is not very precise : * Next Newton steps will correct it quadratically. */ codel_Newton_step(&c->vars); } else { c->vars.count = 1; c->vars.rec_inv_sqrt = ~0U >> REC_INV_SQRT_SHIFT; } c->vars.lastcount = c->vars.count; c->vars.drop_next = codel_control_law(now, c->params.interval, c->vars.rec_inv_sqrt); } return (m); } void codel_getstats(struct codel *c, struct codel_stats *s) { *s = c->stats; } #endif /* ALTQ_CODEL */ Index: head/sys/net/altq/altq_fairq.c =================================================================== --- head/sys/net/altq/altq_fairq.c (revision 343994) +++ head/sys/net/altq/altq_fairq.c (revision 343995) @@ -1,909 +1,908 @@ /* * Copyright (c) 2008 The DragonFly Project. All rights reserved. * * This code is derived from software contributed to The DragonFly Project * by Matthew Dillon * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in * the documentation and/or other materials provided with the * distribution. * 3. Neither the name of The DragonFly Project nor the names of its * contributors may be used to endorse or promote products derived * from this software without specific, prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $DragonFly: src/sys/net/altq/altq_fairq.c,v 1.1 2008/04/06 18:58:15 dillon Exp $ * $FreeBSD$ */ /* * Matt: I gutted altq_priq.c and used it as a skeleton on which to build * fairq. The fairq algorithm is completely different then priq, of course, * but because I used priq's skeleton I believe I should include priq's * copyright. * * Copyright (C) 2000-2003 * Sony Computer Science Laboratories Inc. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * FAIRQ - take traffic classified by keep state (hashed into * mbuf->m_pkthdr.altq_state_hash) and bucketize it. Fairly extract * the first packet from each bucket in a round-robin fashion. * * TODO - better overall qlimit support (right now it is per-bucket). * - NOTE: red etc is per bucket, not overall. * - better service curve support. * * EXAMPLE: * * altq on em0 fairq bandwidth 650Kb queue { std, bulk } * queue std priority 3 bandwidth 400Kb \ * fairq (buckets 64, default, hogs 1Kb) qlimit 50 * queue bulk priority 2 bandwidth 100Kb \ * fairq (buckets 64, hogs 1Kb) qlimit 50 * * pass out on em0 from any to any keep state queue std * pass out on em0 inet proto tcp ..... port ... keep state queue bulk */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef ALTQ_FAIRQ /* fairq is enabled in the kernel conf */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * function prototypes */ static int fairq_clear_interface(struct fairq_if *); static int fairq_request(struct ifaltq *, int, void *); static void fairq_purge(struct fairq_if *); static struct fairq_class *fairq_class_create(struct fairq_if *, int, int, u_int, struct fairq_opts *, int); static int fairq_class_destroy(struct fairq_class *); static int fairq_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); static struct mbuf *fairq_dequeue(struct ifaltq *, int); static int fairq_addq(struct fairq_class *, struct mbuf *, u_int32_t); static struct mbuf *fairq_getq(struct fairq_class *, uint64_t); static struct mbuf *fairq_pollq(struct fairq_class *, uint64_t, int *); static fairq_bucket_t *fairq_selectq(struct fairq_class *, int); static void fairq_purgeq(struct fairq_class *); static void get_class_stats(struct fairq_classstats *, struct fairq_class *); static struct fairq_class *clh_to_clp(struct fairq_if *, uint32_t); int fairq_pfattach(struct pf_altq *a) { struct ifnet *ifp; int error; if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) return (EINVAL); error = altq_attach(&ifp->if_snd, ALTQT_FAIRQ, a->altq_disc, fairq_enqueue, fairq_dequeue, fairq_request, NULL, NULL); return (error); } int -fairq_add_altq(struct pf_altq *a) +fairq_add_altq(struct ifnet *ifp, struct pf_altq *a) { struct fairq_if *pif; - struct ifnet *ifp; - if ((ifp = ifunit(a->ifname)) == NULL) + if (ifp == NULL) return (EINVAL); if (!ALTQ_IS_READY(&ifp->if_snd)) return (ENODEV); pif = malloc(sizeof(struct fairq_if), M_DEVBUF, M_WAITOK | M_ZERO); pif->pif_bandwidth = a->ifbandwidth; pif->pif_maxpri = -1; pif->pif_ifq = &ifp->if_snd; /* keep the state in pf_altq */ a->altq_disc = pif; return (0); } int fairq_remove_altq(struct pf_altq *a) { struct fairq_if *pif; if ((pif = a->altq_disc) == NULL) return (EINVAL); a->altq_disc = NULL; fairq_clear_interface(pif); free(pif, M_DEVBUF); return (0); } int fairq_add_queue(struct pf_altq *a) { struct fairq_if *pif; struct fairq_class *cl; if ((pif = a->altq_disc) == NULL) return (EINVAL); /* check parameters */ if (a->priority >= FAIRQ_MAXPRI) return (EINVAL); if (a->qid == 0) return (EINVAL); if (pif->pif_classes[a->priority] != NULL) return (EBUSY); if (clh_to_clp(pif, a->qid) != NULL) return (EBUSY); cl = fairq_class_create(pif, a->priority, a->qlimit, a->bandwidth, &a->pq_u.fairq_opts, a->qid); if (cl == NULL) return (ENOMEM); return (0); } int fairq_remove_queue(struct pf_altq *a) { struct fairq_if *pif; struct fairq_class *cl; if ((pif = a->altq_disc) == NULL) return (EINVAL); if ((cl = clh_to_clp(pif, a->qid)) == NULL) return (EINVAL); return (fairq_class_destroy(cl)); } int fairq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { struct fairq_if *pif; struct fairq_class *cl; struct fairq_classstats stats; int error = 0; if ((pif = altq_lookup(a->ifname, ALTQT_FAIRQ)) == NULL) return (EBADF); if ((cl = clh_to_clp(pif, a->qid)) == NULL) return (EINVAL); if (*nbytes < sizeof(stats)) return (EINVAL); get_class_stats(&stats, cl); if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0) return (error); *nbytes = sizeof(stats); return (0); } /* * bring the interface back to the initial state by discarding * all the filters and classes. */ static int fairq_clear_interface(struct fairq_if *pif) { struct fairq_class *cl; int pri; /* clear out the classes */ for (pri = 0; pri <= pif->pif_maxpri; pri++) { if ((cl = pif->pif_classes[pri]) != NULL) fairq_class_destroy(cl); } return (0); } static int fairq_request(struct ifaltq *ifq, int req, void *arg) { struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; IFQ_LOCK_ASSERT(ifq); switch (req) { case ALTRQ_PURGE: fairq_purge(pif); break; } return (0); } /* discard all the queued packets on the interface */ static void fairq_purge(struct fairq_if *pif) { struct fairq_class *cl; int pri; for (pri = 0; pri <= pif->pif_maxpri; pri++) { if ((cl = pif->pif_classes[pri]) != NULL && cl->cl_head) fairq_purgeq(cl); } if (ALTQ_IS_ENABLED(pif->pif_ifq)) pif->pif_ifq->ifq_len = 0; } static struct fairq_class * fairq_class_create(struct fairq_if *pif, int pri, int qlimit, u_int bandwidth, struct fairq_opts *opts, int qid) { struct fairq_class *cl; int flags = opts->flags; u_int nbuckets = opts->nbuckets; int i; #ifndef ALTQ_RED if (flags & FARF_RED) { #ifdef ALTQ_DEBUG printf("fairq_class_create: RED not configured for FAIRQ!\n"); #endif return (NULL); } #endif #ifndef ALTQ_CODEL if (flags & FARF_CODEL) { #ifdef ALTQ_DEBUG printf("fairq_class_create: CODEL not configured for FAIRQ!\n"); #endif return (NULL); } #endif if (nbuckets == 0) nbuckets = 256; if (nbuckets > FAIRQ_MAX_BUCKETS) nbuckets = FAIRQ_MAX_BUCKETS; /* enforce power-of-2 size */ while ((nbuckets ^ (nbuckets - 1)) != ((nbuckets << 1) - 1)) ++nbuckets; if ((cl = pif->pif_classes[pri]) != NULL) { /* modify the class instead of creating a new one */ IFQ_LOCK(cl->cl_pif->pif_ifq); if (cl->cl_head) fairq_purgeq(cl); IFQ_UNLOCK(cl->cl_pif->pif_ifq); #ifdef ALTQ_RIO if (cl->cl_qtype == Q_RIO) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (cl->cl_qtype == Q_RED) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (cl->cl_qtype == Q_CODEL) codel_destroy(cl->cl_codel); #endif } else { cl = malloc(sizeof(struct fairq_class), M_DEVBUF, M_WAITOK | M_ZERO); cl->cl_nbuckets = nbuckets; cl->cl_nbucket_mask = nbuckets - 1; cl->cl_buckets = malloc( sizeof(struct fairq_bucket) * cl->cl_nbuckets, M_DEVBUF, M_WAITOK | M_ZERO); cl->cl_head = NULL; } pif->pif_classes[pri] = cl; if (flags & FARF_DEFAULTCLASS) pif->pif_default = cl; if (qlimit == 0) qlimit = 50; /* use default */ cl->cl_qlimit = qlimit; for (i = 0; i < cl->cl_nbuckets; ++i) { qlimit(&cl->cl_buckets[i].queue) = qlimit; } cl->cl_bandwidth = bandwidth / 8; cl->cl_qtype = Q_DROPTAIL; cl->cl_flags = flags & FARF_USERFLAGS; cl->cl_pri = pri; if (pri > pif->pif_maxpri) pif->pif_maxpri = pri; cl->cl_pif = pif; cl->cl_handle = qid; cl->cl_hogs_m1 = opts->hogs_m1 / 8; cl->cl_lssc_m1 = opts->lssc_m1 / 8; /* NOT YET USED */ #ifdef ALTQ_RED if (flags & (FARF_RED|FARF_RIO)) { int red_flags, red_pkttime; red_flags = 0; if (flags & FARF_ECN) red_flags |= REDF_ECN; #ifdef ALTQ_RIO if (flags & FARF_CLEARDSCP) red_flags |= RIOF_CLEARDSCP; #endif if (pif->pif_bandwidth < 8) red_pkttime = 1000 * 1000 * 1000; /* 1 sec */ else red_pkttime = (int64_t)pif->pif_ifq->altq_ifp->if_mtu * 1000 * 1000 * 1000 / (pif->pif_bandwidth / 8); #ifdef ALTQ_RIO if (flags & FARF_RIO) { cl->cl_red = (red_t *)rio_alloc(0, NULL, red_flags, red_pkttime); if (cl->cl_red != NULL) cl->cl_qtype = Q_RIO; } else #endif if (flags & FARF_RED) { cl->cl_red = red_alloc(0, 0, cl->cl_qlimit * 10/100, cl->cl_qlimit * 30/100, red_flags, red_pkttime); if (cl->cl_red != NULL) cl->cl_qtype = Q_RED; } } #endif /* ALTQ_RED */ #ifdef ALTQ_CODEL if (flags & FARF_CODEL) { cl->cl_codel = codel_alloc(5, 100, 0); if (cl->cl_codel != NULL) cl->cl_qtype = Q_CODEL; } #endif return (cl); } static int fairq_class_destroy(struct fairq_class *cl) { struct fairq_if *pif; int pri; IFQ_LOCK(cl->cl_pif->pif_ifq); if (cl->cl_head) fairq_purgeq(cl); pif = cl->cl_pif; pif->pif_classes[cl->cl_pri] = NULL; if (pif->pif_poll_cache == cl) pif->pif_poll_cache = NULL; if (pif->pif_maxpri == cl->cl_pri) { for (pri = cl->cl_pri; pri >= 0; pri--) if (pif->pif_classes[pri] != NULL) { pif->pif_maxpri = pri; break; } if (pri < 0) pif->pif_maxpri = -1; } IFQ_UNLOCK(cl->cl_pif->pif_ifq); if (cl->cl_red != NULL) { #ifdef ALTQ_RIO if (cl->cl_qtype == Q_RIO) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (cl->cl_qtype == Q_RED) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (cl->cl_qtype == Q_CODEL) codel_destroy(cl->cl_codel); #endif } free(cl->cl_buckets, M_DEVBUF); free(cl, M_DEVBUF); return (0); } /* * fairq_enqueue is an enqueue function to be registered to * (*altq_enqueue) in struct ifaltq. */ static int fairq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) { struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; struct fairq_class *cl = NULL; /* Make compiler happy */ struct pf_mtag *t; u_int32_t qid_hash = 0; int len; IFQ_LOCK_ASSERT(ifq); /* grab class set by classifier */ if ((m->m_flags & M_PKTHDR) == 0) { /* should not happen */ printf("altq: packet for %s does not have pkthdr\n", ifq->altq_ifp->if_xname); m_freem(m); return (ENOBUFS); } if ((t = pf_find_mtag(m)) != NULL) { cl = clh_to_clp(pif, t->qid); qid_hash = t->qid_hash; } if (cl == NULL) { cl = pif->pif_default; if (cl == NULL) { m_freem(m); return (ENOBUFS); } } cl->cl_flags |= FARF_HAS_PACKETS; cl->cl_pktattr = NULL; len = m_pktlen(m); if (fairq_addq(cl, m, qid_hash) != 0) { /* drop occurred. mbuf was freed in fairq_addq. */ PKTCNTR_ADD(&cl->cl_dropcnt, len); return (ENOBUFS); } IFQ_INC_LEN(ifq); return (0); } /* * fairq_dequeue is a dequeue function to be registered to * (*altq_dequeue) in struct ifaltq. * * note: ALTDQ_POLL returns the next packet without removing the packet * from the queue. ALTDQ_REMOVE is a normal dequeue operation. * ALTDQ_REMOVE must return the same packet if called immediately * after ALTDQ_POLL. */ static struct mbuf * fairq_dequeue(struct ifaltq *ifq, int op) { struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; struct fairq_class *cl; struct fairq_class *best_cl; struct mbuf *best_m; struct mbuf *m = NULL; uint64_t cur_time = read_machclk(); int pri; int hit_limit; IFQ_LOCK_ASSERT(ifq); if (IFQ_IS_EMPTY(ifq)) { return (NULL); } if (pif->pif_poll_cache && op == ALTDQ_REMOVE) { best_cl = pif->pif_poll_cache; m = fairq_getq(best_cl, cur_time); pif->pif_poll_cache = NULL; if (m) { IFQ_DEC_LEN(ifq); PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m)); return (m); } } else { best_cl = NULL; best_m = NULL; for (pri = pif->pif_maxpri; pri >= 0; pri--) { if ((cl = pif->pif_classes[pri]) == NULL) continue; if ((cl->cl_flags & FARF_HAS_PACKETS) == 0) continue; m = fairq_pollq(cl, cur_time, &hit_limit); if (m == NULL) { cl->cl_flags &= ~FARF_HAS_PACKETS; continue; } /* * Only override the best choice if we are under * the BW limit. */ if (hit_limit == 0 || best_cl == NULL) { best_cl = cl; best_m = m; } /* * Remember the highest priority mbuf in case we * do not find any lower priority mbufs. */ if (hit_limit) continue; break; } if (op == ALTDQ_POLL) { pif->pif_poll_cache = best_cl; m = best_m; } else if (best_cl) { m = fairq_getq(best_cl, cur_time); if (m != NULL) { IFQ_DEC_LEN(ifq); PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m)); } } return (m); } return (NULL); } static int fairq_addq(struct fairq_class *cl, struct mbuf *m, u_int32_t bucketid) { fairq_bucket_t *b; u_int hindex; uint64_t bw; /* * If the packet doesn't have any keep state put it on the end of * our queue. XXX this can result in out of order delivery. */ if (bucketid == 0) { if (cl->cl_head) b = cl->cl_head->prev; else b = &cl->cl_buckets[0]; } else { hindex = bucketid & cl->cl_nbucket_mask; b = &cl->cl_buckets[hindex]; } /* * Add the bucket to the end of the circular list of active buckets. * * As a special case we add the bucket to the beginning of the list * instead of the end if it was not previously on the list and if * its traffic is less then the hog level. */ if (b->in_use == 0) { b->in_use = 1; if (cl->cl_head == NULL) { cl->cl_head = b; b->next = b; b->prev = b; } else { b->next = cl->cl_head; b->prev = cl->cl_head->prev; b->prev->next = b; b->next->prev = b; if (b->bw_delta && cl->cl_hogs_m1) { bw = b->bw_bytes * machclk_freq / b->bw_delta; if (bw < cl->cl_hogs_m1) cl->cl_head = b; } } } #ifdef ALTQ_RIO if (cl->cl_qtype == Q_RIO) return rio_addq((rio_t *)cl->cl_red, &b->queue, m, cl->cl_pktattr); #endif #ifdef ALTQ_RED if (cl->cl_qtype == Q_RED) return red_addq(cl->cl_red, &b->queue, m, cl->cl_pktattr); #endif #ifdef ALTQ_CODEL if (cl->cl_qtype == Q_CODEL) return codel_addq(cl->cl_codel, &b->queue, m); #endif if (qlen(&b->queue) >= qlimit(&b->queue)) { m_freem(m); return (-1); } if (cl->cl_flags & FARF_CLEARDSCP) write_dsfield(m, cl->cl_pktattr, 0); _addq(&b->queue, m); return (0); } static struct mbuf * fairq_getq(struct fairq_class *cl, uint64_t cur_time) { fairq_bucket_t *b; struct mbuf *m; b = fairq_selectq(cl, 0); if (b == NULL) m = NULL; #ifdef ALTQ_RIO else if (cl->cl_qtype == Q_RIO) m = rio_getq((rio_t *)cl->cl_red, &b->queue); #endif #ifdef ALTQ_RED else if (cl->cl_qtype == Q_RED) m = red_getq(cl->cl_red, &b->queue); #endif #ifdef ALTQ_CODEL else if (cl->cl_qtype == Q_CODEL) m = codel_getq(cl->cl_codel, &b->queue); #endif else m = _getq(&b->queue); /* * Calculate the BW change */ if (m != NULL) { uint64_t delta; /* * Per-class bandwidth calculation */ delta = (cur_time - cl->cl_last_time); if (delta > machclk_freq * 8) delta = machclk_freq * 8; cl->cl_bw_delta += delta; cl->cl_bw_bytes += m->m_pkthdr.len; cl->cl_last_time = cur_time; cl->cl_bw_delta -= cl->cl_bw_delta >> 3; cl->cl_bw_bytes -= cl->cl_bw_bytes >> 3; /* * Per-bucket bandwidth calculation */ delta = (cur_time - b->last_time); if (delta > machclk_freq * 8) delta = machclk_freq * 8; b->bw_delta += delta; b->bw_bytes += m->m_pkthdr.len; b->last_time = cur_time; b->bw_delta -= b->bw_delta >> 3; b->bw_bytes -= b->bw_bytes >> 3; } return(m); } /* * Figure out what the next packet would be if there were no limits. If * this class hits its bandwidth limit *hit_limit is set to no-zero, otherwise * it is set to 0. A non-NULL mbuf is returned either way. */ static struct mbuf * fairq_pollq(struct fairq_class *cl, uint64_t cur_time, int *hit_limit) { fairq_bucket_t *b; struct mbuf *m; uint64_t delta; uint64_t bw; *hit_limit = 0; b = fairq_selectq(cl, 1); if (b == NULL) return(NULL); m = qhead(&b->queue); /* * Did this packet exceed the class bandwidth? Calculate the * bandwidth component of the packet. * * - Calculate bytes per second */ delta = cur_time - cl->cl_last_time; if (delta > machclk_freq * 8) delta = machclk_freq * 8; cl->cl_bw_delta += delta; cl->cl_last_time = cur_time; if (cl->cl_bw_delta) { bw = cl->cl_bw_bytes * machclk_freq / cl->cl_bw_delta; if (bw > cl->cl_bandwidth) *hit_limit = 1; #ifdef ALTQ_DEBUG printf("BW %6ju relative to %6u %d queue %p\n", (uintmax_t)bw, cl->cl_bandwidth, *hit_limit, b); #endif } return(m); } /* * Locate the next queue we want to pull a packet out of. This code * is also responsible for removing empty buckets from the circular list. */ static fairq_bucket_t * fairq_selectq(struct fairq_class *cl, int ispoll) { fairq_bucket_t *b; uint64_t bw; if (ispoll == 0 && cl->cl_polled) { b = cl->cl_polled; cl->cl_polled = NULL; return(b); } while ((b = cl->cl_head) != NULL) { /* * Remove empty queues from consideration */ if (qempty(&b->queue)) { b->in_use = 0; cl->cl_head = b->next; if (cl->cl_head == b) { cl->cl_head = NULL; } else { b->next->prev = b->prev; b->prev->next = b->next; } continue; } /* * Advance the round robin. Queues with bandwidths less * then the hog bandwidth are allowed to burst. */ if (cl->cl_hogs_m1 == 0) { cl->cl_head = b->next; } else if (b->bw_delta) { bw = b->bw_bytes * machclk_freq / b->bw_delta; if (bw >= cl->cl_hogs_m1) { cl->cl_head = b->next; } /* * XXX TODO - */ } /* * Return bucket b. */ break; } if (ispoll) cl->cl_polled = b; return(b); } static void fairq_purgeq(struct fairq_class *cl) { fairq_bucket_t *b; struct mbuf *m; while ((b = fairq_selectq(cl, 0)) != NULL) { while ((m = _getq(&b->queue)) != NULL) { PKTCNTR_ADD(&cl->cl_dropcnt, m_pktlen(m)); m_freem(m); } ASSERT(qlen(&b->queue) == 0); } } static void get_class_stats(struct fairq_classstats *sp, struct fairq_class *cl) { fairq_bucket_t *b; sp->class_handle = cl->cl_handle; sp->qlimit = cl->cl_qlimit; sp->xmit_cnt = cl->cl_xmitcnt; sp->drop_cnt = cl->cl_dropcnt; sp->qtype = cl->cl_qtype; sp->qlength = 0; if (cl->cl_head) { b = cl->cl_head; do { sp->qlength += qlen(&b->queue); b = b->next; } while (b != cl->cl_head); } #ifdef ALTQ_RED if (cl->cl_qtype == Q_RED) red_getstats(cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_RIO if (cl->cl_qtype == Q_RIO) rio_getstats((rio_t *)cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_CODEL if (cl->cl_qtype == Q_CODEL) codel_getstats(cl->cl_codel, &sp->codel); #endif } /* convert a class handle to the corresponding class pointer */ static struct fairq_class * clh_to_clp(struct fairq_if *pif, uint32_t chandle) { struct fairq_class *cl; int idx; if (chandle == 0) return (NULL); for (idx = pif->pif_maxpri; idx >= 0; idx--) if ((cl = pif->pif_classes[idx]) != NULL && cl->cl_handle == chandle) return (cl); return (NULL); } #endif /* ALTQ_FAIRQ */ Index: head/sys/net/altq/altq_hfsc.c =================================================================== --- head/sys/net/altq/altq_hfsc.c (revision 343994) +++ head/sys/net/altq/altq_hfsc.c (revision 343995) @@ -1,1751 +1,1746 @@ /*- * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved. * * Permission to use, copy, modify, and distribute this software and * its documentation is hereby granted (including for commercial or * for-profit use), provided that both the copyright notice and this * permission notice appear in all copies of the software, derivative * works, or modified versions, and any portions thereof. * * THIS SOFTWARE IS EXPERIMENTAL AND IS KNOWN TO HAVE BUGS, SOME OF * WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON PROVIDES THIS * SOFTWARE IN ITS ``AS IS'' CONDITION, AND ANY EXPRESS OR IMPLIED * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. * * Carnegie Mellon encourages (but does not require) users of this * software to return any improvements or extensions that they make, * and to grant Carnegie Mellon the rights to redistribute these * changes without encumbrance. * * $KAME: altq_hfsc.c,v 1.24 2003/12/05 05:40:46 kjc Exp $ * $FreeBSD$ */ /* * H-FSC is described in Proceedings of SIGCOMM'97, * "A Hierarchical Fair Service Curve Algorithm for Link-Sharing, * Real-Time and Priority Service" * by Ion Stoica, Hui Zhang, and T. S. Eugene Ng. * * Oleg Cherevko added the upperlimit for link-sharing. * when a class has an upperlimit, the fit-time is computed from the * upperlimit service curve. the link-sharing scheduler does not schedule * a class whose fit-time exceeds the current time. */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef ALTQ_HFSC /* hfsc is enabled by ALTQ_HFSC option in opt_altq.h */ #include #include #include #include #include #include #include #if 1 /* ALTQ3_COMPAT */ #include #include #include #endif /* ALTQ3_COMPAT */ #include #include #include #include #include #include #include #include /* * function prototypes */ static int hfsc_clear_interface(struct hfsc_if *); static int hfsc_request(struct ifaltq *, int, void *); static void hfsc_purge(struct hfsc_if *); static struct hfsc_class *hfsc_class_create(struct hfsc_if *, struct service_curve *, struct service_curve *, struct service_curve *, struct hfsc_class *, int, int, int); static int hfsc_class_destroy(struct hfsc_class *); static struct hfsc_class *hfsc_nextclass(struct hfsc_class *); static int hfsc_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); static struct mbuf *hfsc_dequeue(struct ifaltq *, int); static int hfsc_addq(struct hfsc_class *, struct mbuf *); static struct mbuf *hfsc_getq(struct hfsc_class *); static struct mbuf *hfsc_pollq(struct hfsc_class *); static void hfsc_purgeq(struct hfsc_class *); static void update_cfmin(struct hfsc_class *); static void set_active(struct hfsc_class *, int); static void set_passive(struct hfsc_class *); static void init_ed(struct hfsc_class *, int); static void update_ed(struct hfsc_class *, int); static void update_d(struct hfsc_class *, int); static void init_vf(struct hfsc_class *, int); static void update_vf(struct hfsc_class *, int, u_int64_t); static void ellist_insert(struct hfsc_class *); static void ellist_remove(struct hfsc_class *); static void ellist_update(struct hfsc_class *); struct hfsc_class *hfsc_get_mindl(struct hfsc_if *, u_int64_t); static void actlist_insert(struct hfsc_class *); static void actlist_remove(struct hfsc_class *); static void actlist_update(struct hfsc_class *); static struct hfsc_class *actlist_firstfit(struct hfsc_class *, u_int64_t); static __inline u_int64_t seg_x2y(u_int64_t, u_int64_t); static __inline u_int64_t seg_y2x(u_int64_t, u_int64_t); static __inline u_int64_t m2sm(u_int64_t); static __inline u_int64_t m2ism(u_int64_t); static __inline u_int64_t d2dx(u_int); static u_int64_t sm2m(u_int64_t); static u_int dx2d(u_int64_t); static void sc2isc(struct service_curve *, struct internal_sc *); static void rtsc_init(struct runtime_sc *, struct internal_sc *, u_int64_t, u_int64_t); static u_int64_t rtsc_y2x(struct runtime_sc *, u_int64_t); static u_int64_t rtsc_x2y(struct runtime_sc *, u_int64_t); static void rtsc_min(struct runtime_sc *, struct internal_sc *, u_int64_t, u_int64_t); static void get_class_stats_v0(struct hfsc_classstats_v0 *, struct hfsc_class *); static void get_class_stats_v1(struct hfsc_classstats_v1 *, struct hfsc_class *); static struct hfsc_class *clh_to_clp(struct hfsc_if *, u_int32_t); /* * macros */ #define is_a_parent_class(cl) ((cl)->cl_children != NULL) #define HT_INFINITY 0xffffffffffffffffULL /* infinite time value */ int hfsc_pfattach(struct pf_altq *a) { struct ifnet *ifp; int s, error; if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) return (EINVAL); s = splnet(); error = altq_attach(&ifp->if_snd, ALTQT_HFSC, a->altq_disc, hfsc_enqueue, hfsc_dequeue, hfsc_request, NULL, NULL); splx(s); return (error); } int -hfsc_add_altq(struct pf_altq *a) +hfsc_add_altq(struct ifnet *ifp, struct pf_altq *a) { struct hfsc_if *hif; - struct ifnet *ifp; - if ((ifp = ifunit(a->ifname)) == NULL) + if (ifp == NULL) return (EINVAL); if (!ALTQ_IS_READY(&ifp->if_snd)) return (ENODEV); hif = malloc(sizeof(struct hfsc_if), M_DEVBUF, M_NOWAIT | M_ZERO); if (hif == NULL) return (ENOMEM); TAILQ_INIT(&hif->hif_eligible); hif->hif_ifq = &ifp->if_snd; /* keep the state in pf_altq */ a->altq_disc = hif; return (0); } int hfsc_remove_altq(struct pf_altq *a) { struct hfsc_if *hif; if ((hif = a->altq_disc) == NULL) return (EINVAL); a->altq_disc = NULL; (void)hfsc_clear_interface(hif); (void)hfsc_class_destroy(hif->hif_rootclass); free(hif, M_DEVBUF); return (0); } int hfsc_add_queue(struct pf_altq *a) { struct hfsc_if *hif; struct hfsc_class *cl, *parent; struct hfsc_opts_v1 *opts; struct service_curve rtsc, lssc, ulsc; if ((hif = a->altq_disc) == NULL) return (EINVAL); opts = &a->pq_u.hfsc_opts; if (a->parent_qid == HFSC_NULLCLASS_HANDLE && hif->hif_rootclass == NULL) parent = NULL; else if ((parent = clh_to_clp(hif, a->parent_qid)) == NULL) return (EINVAL); if (a->qid == 0) return (EINVAL); if (clh_to_clp(hif, a->qid) != NULL) return (EBUSY); rtsc.m1 = opts->rtsc_m1; rtsc.d = opts->rtsc_d; rtsc.m2 = opts->rtsc_m2; lssc.m1 = opts->lssc_m1; lssc.d = opts->lssc_d; lssc.m2 = opts->lssc_m2; ulsc.m1 = opts->ulsc_m1; ulsc.d = opts->ulsc_d; ulsc.m2 = opts->ulsc_m2; cl = hfsc_class_create(hif, &rtsc, &lssc, &ulsc, parent, a->qlimit, opts->flags, a->qid); if (cl == NULL) return (ENOMEM); return (0); } int hfsc_remove_queue(struct pf_altq *a) { struct hfsc_if *hif; struct hfsc_class *cl; if ((hif = a->altq_disc) == NULL) return (EINVAL); if ((cl = clh_to_clp(hif, a->qid)) == NULL) return (EINVAL); return (hfsc_class_destroy(cl)); } int hfsc_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { struct hfsc_if *hif; struct hfsc_class *cl; union { struct hfsc_classstats_v0 v0; struct hfsc_classstats_v1 v1; } stats; size_t stats_size; int error = 0; if ((hif = altq_lookup(a->ifname, ALTQT_HFSC)) == NULL) return (EBADF); if ((cl = clh_to_clp(hif, a->qid)) == NULL) return (EINVAL); if (version > HFSC_STATS_VERSION) return (EINVAL); memset(&stats, 0, sizeof(stats)); switch (version) { case 0: get_class_stats_v0(&stats.v0, cl); stats_size = sizeof(struct hfsc_classstats_v0); break; case 1: get_class_stats_v1(&stats.v1, cl); stats_size = sizeof(struct hfsc_classstats_v1); break; } if (*nbytes < stats_size) return (EINVAL); if ((error = copyout((caddr_t)&stats, ubuf, stats_size)) != 0) return (error); *nbytes = stats_size; return (0); } /* * bring the interface back to the initial state by discarding * all the filters and classes except the root class. */ static int hfsc_clear_interface(struct hfsc_if *hif) { struct hfsc_class *cl; /* clear out the classes */ while (hif->hif_rootclass != NULL && (cl = hif->hif_rootclass->cl_children) != NULL) { /* * remove the first leaf class found in the hierarchy * then start over */ for (; cl != NULL; cl = hfsc_nextclass(cl)) { if (!is_a_parent_class(cl)) { (void)hfsc_class_destroy(cl); break; } } } return (0); } static int hfsc_request(struct ifaltq *ifq, int req, void *arg) { struct hfsc_if *hif = (struct hfsc_if *)ifq->altq_disc; IFQ_LOCK_ASSERT(ifq); switch (req) { case ALTRQ_PURGE: hfsc_purge(hif); break; } return (0); } /* discard all the queued packets on the interface */ static void hfsc_purge(struct hfsc_if *hif) { struct hfsc_class *cl; for (cl = hif->hif_rootclass; cl != NULL; cl = hfsc_nextclass(cl)) if (!qempty(cl->cl_q)) hfsc_purgeq(cl); if (ALTQ_IS_ENABLED(hif->hif_ifq)) hif->hif_ifq->ifq_len = 0; } struct hfsc_class * hfsc_class_create(struct hfsc_if *hif, struct service_curve *rsc, struct service_curve *fsc, struct service_curve *usc, struct hfsc_class *parent, int qlimit, int flags, int qid) { struct hfsc_class *cl, *p; int i, s; if (hif->hif_classes >= HFSC_MAX_CLASSES) return (NULL); #ifndef ALTQ_RED if (flags & HFCF_RED) { #ifdef ALTQ_DEBUG printf("hfsc_class_create: RED not configured for HFSC!\n"); #endif return (NULL); } #endif #ifndef ALTQ_CODEL if (flags & HFCF_CODEL) { #ifdef ALTQ_DEBUG printf("hfsc_class_create: CODEL not configured for HFSC!\n"); #endif return (NULL); } #endif cl = malloc(sizeof(struct hfsc_class), M_DEVBUF, M_NOWAIT | M_ZERO); if (cl == NULL) return (NULL); cl->cl_q = malloc(sizeof(class_queue_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (cl->cl_q == NULL) goto err_ret; TAILQ_INIT(&cl->cl_actc); if (qlimit == 0) qlimit = 50; /* use default */ qlimit(cl->cl_q) = qlimit; qtype(cl->cl_q) = Q_DROPTAIL; qlen(cl->cl_q) = 0; qsize(cl->cl_q) = 0; cl->cl_flags = flags; #ifdef ALTQ_RED if (flags & (HFCF_RED|HFCF_RIO)) { int red_flags, red_pkttime; u_int m2; m2 = 0; if (rsc != NULL && rsc->m2 > m2) m2 = rsc->m2; if (fsc != NULL && fsc->m2 > m2) m2 = fsc->m2; if (usc != NULL && usc->m2 > m2) m2 = usc->m2; red_flags = 0; if (flags & HFCF_ECN) red_flags |= REDF_ECN; #ifdef ALTQ_RIO if (flags & HFCF_CLEARDSCP) red_flags |= RIOF_CLEARDSCP; #endif if (m2 < 8) red_pkttime = 1000 * 1000 * 1000; /* 1 sec */ else red_pkttime = (int64_t)hif->hif_ifq->altq_ifp->if_mtu * 1000 * 1000 * 1000 / (m2 / 8); if (flags & HFCF_RED) { cl->cl_red = red_alloc(0, 0, qlimit(cl->cl_q) * 10/100, qlimit(cl->cl_q) * 30/100, red_flags, red_pkttime); if (cl->cl_red != NULL) qtype(cl->cl_q) = Q_RED; } #ifdef ALTQ_RIO else { cl->cl_red = (red_t *)rio_alloc(0, NULL, red_flags, red_pkttime); if (cl->cl_red != NULL) qtype(cl->cl_q) = Q_RIO; } #endif } #endif /* ALTQ_RED */ #ifdef ALTQ_CODEL if (flags & HFCF_CODEL) { cl->cl_codel = codel_alloc(5, 100, 0); if (cl->cl_codel != NULL) qtype(cl->cl_q) = Q_CODEL; } #endif if (rsc != NULL && (rsc->m1 != 0 || rsc->m2 != 0)) { cl->cl_rsc = malloc(sizeof(struct internal_sc), M_DEVBUF, M_NOWAIT); if (cl->cl_rsc == NULL) goto err_ret; sc2isc(rsc, cl->cl_rsc); rtsc_init(&cl->cl_deadline, cl->cl_rsc, 0, 0); rtsc_init(&cl->cl_eligible, cl->cl_rsc, 0, 0); } if (fsc != NULL && (fsc->m1 != 0 || fsc->m2 != 0)) { cl->cl_fsc = malloc(sizeof(struct internal_sc), M_DEVBUF, M_NOWAIT); if (cl->cl_fsc == NULL) goto err_ret; sc2isc(fsc, cl->cl_fsc); rtsc_init(&cl->cl_virtual, cl->cl_fsc, 0, 0); } if (usc != NULL && (usc->m1 != 0 || usc->m2 != 0)) { cl->cl_usc = malloc(sizeof(struct internal_sc), M_DEVBUF, M_NOWAIT); if (cl->cl_usc == NULL) goto err_ret; sc2isc(usc, cl->cl_usc); rtsc_init(&cl->cl_ulimit, cl->cl_usc, 0, 0); } cl->cl_id = hif->hif_classid++; cl->cl_handle = qid; cl->cl_hif = hif; cl->cl_parent = parent; s = splnet(); IFQ_LOCK(hif->hif_ifq); hif->hif_classes++; /* * find a free slot in the class table. if the slot matching * the lower bits of qid is free, use this slot. otherwise, * use the first free slot. */ i = qid % HFSC_MAX_CLASSES; if (hif->hif_class_tbl[i] == NULL) hif->hif_class_tbl[i] = cl; else { for (i = 0; i < HFSC_MAX_CLASSES; i++) if (hif->hif_class_tbl[i] == NULL) { hif->hif_class_tbl[i] = cl; break; } if (i == HFSC_MAX_CLASSES) { IFQ_UNLOCK(hif->hif_ifq); splx(s); goto err_ret; } } + cl->cl_slot = i; if (flags & HFCF_DEFAULTCLASS) hif->hif_defaultclass = cl; if (parent == NULL) { /* this is root class */ hif->hif_rootclass = cl; } else { /* add this class to the children list of the parent */ if ((p = parent->cl_children) == NULL) parent->cl_children = cl; else { while (p->cl_siblings != NULL) p = p->cl_siblings; p->cl_siblings = cl; } } IFQ_UNLOCK(hif->hif_ifq); splx(s); return (cl); err_ret: if (cl->cl_red != NULL) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_destroy(cl->cl_codel); #endif } if (cl->cl_fsc != NULL) free(cl->cl_fsc, M_DEVBUF); if (cl->cl_rsc != NULL) free(cl->cl_rsc, M_DEVBUF); if (cl->cl_usc != NULL) free(cl->cl_usc, M_DEVBUF); if (cl->cl_q != NULL) free(cl->cl_q, M_DEVBUF); free(cl, M_DEVBUF); return (NULL); } static int hfsc_class_destroy(struct hfsc_class *cl) { - int i, s; + int s; if (cl == NULL) return (0); if (is_a_parent_class(cl)) return (EBUSY); s = splnet(); IFQ_LOCK(cl->cl_hif->hif_ifq); if (!qempty(cl->cl_q)) hfsc_purgeq(cl); if (cl->cl_parent == NULL) { /* this is root class */ } else { struct hfsc_class *p = cl->cl_parent->cl_children; if (p == cl) cl->cl_parent->cl_children = cl->cl_siblings; else do { if (p->cl_siblings == cl) { p->cl_siblings = cl->cl_siblings; break; } } while ((p = p->cl_siblings) != NULL); ASSERT(p != NULL); } - for (i = 0; i < HFSC_MAX_CLASSES; i++) - if (cl->cl_hif->hif_class_tbl[i] == cl) { - cl->cl_hif->hif_class_tbl[i] = NULL; - break; - } - + cl->cl_hif->hif_class_tbl[cl->cl_slot] = NULL; cl->cl_hif->hif_classes--; IFQ_UNLOCK(cl->cl_hif->hif_ifq); splx(s); if (cl->cl_red != NULL) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_destroy(cl->cl_codel); #endif } IFQ_LOCK(cl->cl_hif->hif_ifq); if (cl == cl->cl_hif->hif_rootclass) cl->cl_hif->hif_rootclass = NULL; if (cl == cl->cl_hif->hif_defaultclass) cl->cl_hif->hif_defaultclass = NULL; IFQ_UNLOCK(cl->cl_hif->hif_ifq); if (cl->cl_usc != NULL) free(cl->cl_usc, M_DEVBUF); if (cl->cl_fsc != NULL) free(cl->cl_fsc, M_DEVBUF); if (cl->cl_rsc != NULL) free(cl->cl_rsc, M_DEVBUF); free(cl->cl_q, M_DEVBUF); free(cl, M_DEVBUF); return (0); } /* * hfsc_nextclass returns the next class in the tree. * usage: * for (cl = hif->hif_rootclass; cl != NULL; cl = hfsc_nextclass(cl)) * do_something; */ static struct hfsc_class * hfsc_nextclass(struct hfsc_class *cl) { if (cl->cl_children != NULL) cl = cl->cl_children; else if (cl->cl_siblings != NULL) cl = cl->cl_siblings; else { while ((cl = cl->cl_parent) != NULL) if (cl->cl_siblings) { cl = cl->cl_siblings; break; } } return (cl); } /* * hfsc_enqueue is an enqueue function to be registered to * (*altq_enqueue) in struct ifaltq. */ static int hfsc_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) { struct hfsc_if *hif = (struct hfsc_if *)ifq->altq_disc; struct hfsc_class *cl; struct pf_mtag *t; int len; IFQ_LOCK_ASSERT(ifq); /* grab class set by classifier */ if ((m->m_flags & M_PKTHDR) == 0) { /* should not happen */ printf("altq: packet for %s does not have pkthdr\n", ifq->altq_ifp->if_xname); m_freem(m); return (ENOBUFS); } cl = NULL; if ((t = pf_find_mtag(m)) != NULL) cl = clh_to_clp(hif, t->qid); if (cl == NULL || is_a_parent_class(cl)) { cl = hif->hif_defaultclass; if (cl == NULL) { m_freem(m); return (ENOBUFS); } } cl->cl_pktattr = NULL; len = m_pktlen(m); if (hfsc_addq(cl, m) != 0) { /* drop occurred. mbuf was freed in hfsc_addq. */ PKTCNTR_ADD(&cl->cl_stats.drop_cnt, len); return (ENOBUFS); } IFQ_INC_LEN(ifq); cl->cl_hif->hif_packets++; /* successfully queued. */ if (qlen(cl->cl_q) == 1) set_active(cl, m_pktlen(m)); return (0); } /* * hfsc_dequeue is a dequeue function to be registered to * (*altq_dequeue) in struct ifaltq. * * note: ALTDQ_POLL returns the next packet without removing the packet * from the queue. ALTDQ_REMOVE is a normal dequeue operation. * ALTDQ_REMOVE must return the same packet if called immediately * after ALTDQ_POLL. */ static struct mbuf * hfsc_dequeue(struct ifaltq *ifq, int op) { struct hfsc_if *hif = (struct hfsc_if *)ifq->altq_disc; struct hfsc_class *cl; struct mbuf *m; int len, next_len; int realtime = 0; u_int64_t cur_time; IFQ_LOCK_ASSERT(ifq); if (hif->hif_packets == 0) /* no packet in the tree */ return (NULL); cur_time = read_machclk(); if (op == ALTDQ_REMOVE && hif->hif_pollcache != NULL) { cl = hif->hif_pollcache; hif->hif_pollcache = NULL; /* check if the class was scheduled by real-time criteria */ if (cl->cl_rsc != NULL) realtime = (cl->cl_e <= cur_time); } else { /* * if there are eligible classes, use real-time criteria. * find the class with the minimum deadline among * the eligible classes. */ if ((cl = hfsc_get_mindl(hif, cur_time)) != NULL) { realtime = 1; } else { #ifdef ALTQ_DEBUG int fits = 0; #endif /* * use link-sharing criteria * get the class with the minimum vt in the hierarchy */ cl = hif->hif_rootclass; while (is_a_parent_class(cl)) { cl = actlist_firstfit(cl, cur_time); if (cl == NULL) { #ifdef ALTQ_DEBUG if (fits > 0) printf("%d fit but none found\n",fits); #endif return (NULL); } /* * update parent's cl_cvtmin. * don't update if the new vt is smaller. */ if (cl->cl_parent->cl_cvtmin < cl->cl_vt) cl->cl_parent->cl_cvtmin = cl->cl_vt; #ifdef ALTQ_DEBUG fits++; #endif } } if (op == ALTDQ_POLL) { hif->hif_pollcache = cl; m = hfsc_pollq(cl); return (m); } } m = hfsc_getq(cl); if (m == NULL) panic("hfsc_dequeue:"); len = m_pktlen(m); cl->cl_hif->hif_packets--; IFQ_DEC_LEN(ifq); PKTCNTR_ADD(&cl->cl_stats.xmit_cnt, len); update_vf(cl, len, cur_time); if (realtime) cl->cl_cumul += len; if (!qempty(cl->cl_q)) { if (cl->cl_rsc != NULL) { /* update ed */ next_len = m_pktlen(qhead(cl->cl_q)); if (realtime) update_ed(cl, next_len); else update_d(cl, next_len); } } else { /* the class becomes passive */ set_passive(cl); } return (m); } static int hfsc_addq(struct hfsc_class *cl, struct mbuf *m) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) return rio_addq((rio_t *)cl->cl_red, cl->cl_q, m, cl->cl_pktattr); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) return red_addq(cl->cl_red, cl->cl_q, m, cl->cl_pktattr); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) return codel_addq(cl->cl_codel, cl->cl_q, m); #endif if (qlen(cl->cl_q) >= qlimit(cl->cl_q)) { m_freem(m); return (-1); } if (cl->cl_flags & HFCF_CLEARDSCP) write_dsfield(m, cl->cl_pktattr, 0); _addq(cl->cl_q, m); return (0); } static struct mbuf * hfsc_getq(struct hfsc_class *cl) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) return rio_getq((rio_t *)cl->cl_red, cl->cl_q); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) return red_getq(cl->cl_red, cl->cl_q); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) return codel_getq(cl->cl_codel, cl->cl_q); #endif return _getq(cl->cl_q); } static struct mbuf * hfsc_pollq(struct hfsc_class *cl) { return qhead(cl->cl_q); } static void hfsc_purgeq(struct hfsc_class *cl) { struct mbuf *m; if (qempty(cl->cl_q)) return; while ((m = _getq(cl->cl_q)) != NULL) { PKTCNTR_ADD(&cl->cl_stats.drop_cnt, m_pktlen(m)); m_freem(m); cl->cl_hif->hif_packets--; IFQ_DEC_LEN(cl->cl_hif->hif_ifq); } ASSERT(qlen(cl->cl_q) == 0); update_vf(cl, 0, 0); /* remove cl from the actlist */ set_passive(cl); } static void set_active(struct hfsc_class *cl, int len) { if (cl->cl_rsc != NULL) init_ed(cl, len); if (cl->cl_fsc != NULL) init_vf(cl, len); cl->cl_stats.period++; } static void set_passive(struct hfsc_class *cl) { if (cl->cl_rsc != NULL) ellist_remove(cl); /* * actlist is now handled in update_vf() so that update_vf(cl, 0, 0) * needs to be called explicitly to remove a class from actlist */ } static void init_ed(struct hfsc_class *cl, int next_len) { u_int64_t cur_time; cur_time = read_machclk(); /* update the deadline curve */ rtsc_min(&cl->cl_deadline, cl->cl_rsc, cur_time, cl->cl_cumul); /* * update the eligible curve. * for concave, it is equal to the deadline curve. * for convex, it is a linear curve with slope m2. */ cl->cl_eligible = cl->cl_deadline; if (cl->cl_rsc->sm1 <= cl->cl_rsc->sm2) { cl->cl_eligible.dx = 0; cl->cl_eligible.dy = 0; } /* compute e and d */ cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul); cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len); ellist_insert(cl); } static void update_ed(struct hfsc_class *cl, int next_len) { cl->cl_e = rtsc_y2x(&cl->cl_eligible, cl->cl_cumul); cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len); ellist_update(cl); } static void update_d(struct hfsc_class *cl, int next_len) { cl->cl_d = rtsc_y2x(&cl->cl_deadline, cl->cl_cumul + next_len); } static void init_vf(struct hfsc_class *cl, int len) { struct hfsc_class *max_cl, *p; u_int64_t vt, f, cur_time; int go_active; cur_time = 0; go_active = 1; for ( ; cl->cl_parent != NULL; cl = cl->cl_parent) { if (go_active && cl->cl_nactive++ == 0) go_active = 1; else go_active = 0; if (go_active) { max_cl = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead); if (max_cl != NULL) { /* * set vt to the average of the min and max * classes. if the parent's period didn't * change, don't decrease vt of the class. */ vt = max_cl->cl_vt; if (cl->cl_parent->cl_cvtmin != 0) vt = (cl->cl_parent->cl_cvtmin + vt)/2; if (cl->cl_parent->cl_vtperiod != cl->cl_parentperiod || vt > cl->cl_vt) cl->cl_vt = vt; } else { /* * first child for a new parent backlog period. * add parent's cvtmax to vtoff of children * to make a new vt (vtoff + vt) larger than * the vt in the last period for all children. */ vt = cl->cl_parent->cl_cvtmax; for (p = cl->cl_parent->cl_children; p != NULL; p = p->cl_siblings) p->cl_vtoff += vt; cl->cl_vt = 0; cl->cl_parent->cl_cvtmax = 0; cl->cl_parent->cl_cvtmin = 0; } cl->cl_initvt = cl->cl_vt; /* update the virtual curve */ vt = cl->cl_vt + cl->cl_vtoff; rtsc_min(&cl->cl_virtual, cl->cl_fsc, vt, cl->cl_total); if (cl->cl_virtual.x == vt) { cl->cl_virtual.x -= cl->cl_vtoff; cl->cl_vtoff = 0; } cl->cl_vtadj = 0; cl->cl_vtperiod++; /* increment vt period */ cl->cl_parentperiod = cl->cl_parent->cl_vtperiod; if (cl->cl_parent->cl_nactive == 0) cl->cl_parentperiod++; cl->cl_f = 0; actlist_insert(cl); if (cl->cl_usc != NULL) { /* class has upper limit curve */ if (cur_time == 0) cur_time = read_machclk(); /* update the ulimit curve */ rtsc_min(&cl->cl_ulimit, cl->cl_usc, cur_time, cl->cl_total); /* compute myf */ cl->cl_myf = rtsc_y2x(&cl->cl_ulimit, cl->cl_total); cl->cl_myfadj = 0; } } if (cl->cl_myf > cl->cl_cfmin) f = cl->cl_myf; else f = cl->cl_cfmin; if (f != cl->cl_f) { cl->cl_f = f; update_cfmin(cl->cl_parent); } } } static void update_vf(struct hfsc_class *cl, int len, u_int64_t cur_time) { u_int64_t f, myf_bound, delta; int go_passive; go_passive = qempty(cl->cl_q); for (; cl->cl_parent != NULL; cl = cl->cl_parent) { cl->cl_total += len; if (cl->cl_fsc == NULL || cl->cl_nactive == 0) continue; if (go_passive && --cl->cl_nactive == 0) go_passive = 1; else go_passive = 0; if (go_passive) { /* no more active child, going passive */ /* update cvtmax of the parent class */ if (cl->cl_vt > cl->cl_parent->cl_cvtmax) cl->cl_parent->cl_cvtmax = cl->cl_vt; /* remove this class from the vt list */ actlist_remove(cl); update_cfmin(cl->cl_parent); continue; } /* * update vt and f */ cl->cl_vt = rtsc_y2x(&cl->cl_virtual, cl->cl_total) - cl->cl_vtoff + cl->cl_vtadj; /* * if vt of the class is smaller than cvtmin, * the class was skipped in the past due to non-fit. * if so, we need to adjust vtadj. */ if (cl->cl_vt < cl->cl_parent->cl_cvtmin) { cl->cl_vtadj += cl->cl_parent->cl_cvtmin - cl->cl_vt; cl->cl_vt = cl->cl_parent->cl_cvtmin; } /* update the vt list */ actlist_update(cl); if (cl->cl_usc != NULL) { cl->cl_myf = cl->cl_myfadj + rtsc_y2x(&cl->cl_ulimit, cl->cl_total); /* * if myf lags behind by more than one clock tick * from the current time, adjust myfadj to prevent * a rate-limited class from going greedy. * in a steady state under rate-limiting, myf * fluctuates within one clock tick. */ myf_bound = cur_time - machclk_per_tick; if (cl->cl_myf < myf_bound) { delta = cur_time - cl->cl_myf; cl->cl_myfadj += delta; cl->cl_myf += delta; } } /* cl_f is max(cl_myf, cl_cfmin) */ if (cl->cl_myf > cl->cl_cfmin) f = cl->cl_myf; else f = cl->cl_cfmin; if (f != cl->cl_f) { cl->cl_f = f; update_cfmin(cl->cl_parent); } } } static void update_cfmin(struct hfsc_class *cl) { struct hfsc_class *p; u_int64_t cfmin; if (TAILQ_EMPTY(&cl->cl_actc)) { cl->cl_cfmin = 0; return; } cfmin = HT_INFINITY; TAILQ_FOREACH(p, &cl->cl_actc, cl_actlist) { if (p->cl_f == 0) { cl->cl_cfmin = 0; return; } if (p->cl_f < cfmin) cfmin = p->cl_f; } cl->cl_cfmin = cfmin; } /* * TAILQ based ellist and actlist implementation * (ion wanted to make a calendar queue based implementation) */ /* * eligible list holds backlogged classes being sorted by their eligible times. * there is one eligible list per interface. */ static void ellist_insert(struct hfsc_class *cl) { struct hfsc_if *hif = cl->cl_hif; struct hfsc_class *p; /* check the last entry first */ if ((p = TAILQ_LAST(&hif->hif_eligible, elighead)) == NULL || p->cl_e <= cl->cl_e) { TAILQ_INSERT_TAIL(&hif->hif_eligible, cl, cl_ellist); return; } TAILQ_FOREACH(p, &hif->hif_eligible, cl_ellist) { if (cl->cl_e < p->cl_e) { TAILQ_INSERT_BEFORE(p, cl, cl_ellist); return; } } ASSERT(0); /* should not reach here */ } static void ellist_remove(struct hfsc_class *cl) { struct hfsc_if *hif = cl->cl_hif; TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist); } static void ellist_update(struct hfsc_class *cl) { struct hfsc_if *hif = cl->cl_hif; struct hfsc_class *p, *last; /* * the eligible time of a class increases monotonically. * if the next entry has a larger eligible time, nothing to do. */ p = TAILQ_NEXT(cl, cl_ellist); if (p == NULL || cl->cl_e <= p->cl_e) return; /* check the last entry */ last = TAILQ_LAST(&hif->hif_eligible, elighead); ASSERT(last != NULL); if (last->cl_e <= cl->cl_e) { TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist); TAILQ_INSERT_TAIL(&hif->hif_eligible, cl, cl_ellist); return; } /* * the new position must be between the next entry * and the last entry */ while ((p = TAILQ_NEXT(p, cl_ellist)) != NULL) { if (cl->cl_e < p->cl_e) { TAILQ_REMOVE(&hif->hif_eligible, cl, cl_ellist); TAILQ_INSERT_BEFORE(p, cl, cl_ellist); return; } } ASSERT(0); /* should not reach here */ } /* find the class with the minimum deadline among the eligible classes */ struct hfsc_class * hfsc_get_mindl(struct hfsc_if *hif, u_int64_t cur_time) { struct hfsc_class *p, *cl = NULL; TAILQ_FOREACH(p, &hif->hif_eligible, cl_ellist) { if (p->cl_e > cur_time) break; if (cl == NULL || p->cl_d < cl->cl_d) cl = p; } return (cl); } /* * active children list holds backlogged child classes being sorted * by their virtual time. * each intermediate class has one active children list. */ static void actlist_insert(struct hfsc_class *cl) { struct hfsc_class *p; /* check the last entry first */ if ((p = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead)) == NULL || p->cl_vt <= cl->cl_vt) { TAILQ_INSERT_TAIL(&cl->cl_parent->cl_actc, cl, cl_actlist); return; } TAILQ_FOREACH(p, &cl->cl_parent->cl_actc, cl_actlist) { if (cl->cl_vt < p->cl_vt) { TAILQ_INSERT_BEFORE(p, cl, cl_actlist); return; } } ASSERT(0); /* should not reach here */ } static void actlist_remove(struct hfsc_class *cl) { TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist); } static void actlist_update(struct hfsc_class *cl) { struct hfsc_class *p, *last; /* * the virtual time of a class increases monotonically during its * backlogged period. * if the next entry has a larger virtual time, nothing to do. */ p = TAILQ_NEXT(cl, cl_actlist); if (p == NULL || cl->cl_vt < p->cl_vt) return; /* check the last entry */ last = TAILQ_LAST(&cl->cl_parent->cl_actc, acthead); ASSERT(last != NULL); if (last->cl_vt <= cl->cl_vt) { TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist); TAILQ_INSERT_TAIL(&cl->cl_parent->cl_actc, cl, cl_actlist); return; } /* * the new position must be between the next entry * and the last entry */ while ((p = TAILQ_NEXT(p, cl_actlist)) != NULL) { if (cl->cl_vt < p->cl_vt) { TAILQ_REMOVE(&cl->cl_parent->cl_actc, cl, cl_actlist); TAILQ_INSERT_BEFORE(p, cl, cl_actlist); return; } } ASSERT(0); /* should not reach here */ } static struct hfsc_class * actlist_firstfit(struct hfsc_class *cl, u_int64_t cur_time) { struct hfsc_class *p; TAILQ_FOREACH(p, &cl->cl_actc, cl_actlist) { if (p->cl_f <= cur_time) return (p); } return (NULL); } /* * service curve support functions * * external service curve parameters * m: bits/sec * d: msec * internal service curve parameters * sm: (bytes/machclk tick) << SM_SHIFT * ism: (machclk ticks/byte) << ISM_SHIFT * dx: machclk ticks * * SM_SHIFT and ISM_SHIFT are scaled in order to keep effective digits. we * should be able to handle 100K-100Gbps linkspeed with 256 MHz machclk * frequency and at least 3 effective digits in decimal. * */ #define SM_SHIFT 24 #define ISM_SHIFT 14 #define SM_MASK ((1LL << SM_SHIFT) - 1) #define ISM_MASK ((1LL << ISM_SHIFT) - 1) static __inline u_int64_t seg_x2y(u_int64_t x, u_int64_t sm) { u_int64_t y; /* * compute * y = x * sm >> SM_SHIFT * but divide it for the upper and lower bits to avoid overflow */ y = (x >> SM_SHIFT) * sm + (((x & SM_MASK) * sm) >> SM_SHIFT); return (y); } static __inline u_int64_t seg_y2x(u_int64_t y, u_int64_t ism) { u_int64_t x; if (y == 0) x = 0; else if (ism == HT_INFINITY) x = HT_INFINITY; else { x = (y >> ISM_SHIFT) * ism + (((y & ISM_MASK) * ism) >> ISM_SHIFT); } return (x); } static __inline u_int64_t m2sm(u_int64_t m) { u_int64_t sm; sm = (m << SM_SHIFT) / 8 / machclk_freq; return (sm); } static __inline u_int64_t m2ism(u_int64_t m) { u_int64_t ism; if (m == 0) ism = HT_INFINITY; else ism = ((u_int64_t)machclk_freq << ISM_SHIFT) * 8 / m; return (ism); } static __inline u_int64_t d2dx(u_int d) { u_int64_t dx; dx = ((u_int64_t)d * machclk_freq) / 1000; return (dx); } static u_int64_t sm2m(u_int64_t sm) { u_int64_t m; m = (sm * 8 * machclk_freq) >> SM_SHIFT; return (m); } static u_int dx2d(u_int64_t dx) { u_int64_t d; d = dx * 1000 / machclk_freq; return ((u_int)d); } static void sc2isc(struct service_curve *sc, struct internal_sc *isc) { isc->sm1 = m2sm(sc->m1); isc->ism1 = m2ism(sc->m1); isc->dx = d2dx(sc->d); isc->dy = seg_x2y(isc->dx, isc->sm1); isc->sm2 = m2sm(sc->m2); isc->ism2 = m2ism(sc->m2); } /* * initialize the runtime service curve with the given internal * service curve starting at (x, y). */ static void rtsc_init(struct runtime_sc *rtsc, struct internal_sc * isc, u_int64_t x, u_int64_t y) { rtsc->x = x; rtsc->y = y; rtsc->sm1 = isc->sm1; rtsc->ism1 = isc->ism1; rtsc->dx = isc->dx; rtsc->dy = isc->dy; rtsc->sm2 = isc->sm2; rtsc->ism2 = isc->ism2; } /* * calculate the y-projection of the runtime service curve by the * given x-projection value */ static u_int64_t rtsc_y2x(struct runtime_sc *rtsc, u_int64_t y) { u_int64_t x; if (y < rtsc->y) x = rtsc->x; else if (y <= rtsc->y + rtsc->dy) { /* x belongs to the 1st segment */ if (rtsc->dy == 0) x = rtsc->x + rtsc->dx; else x = rtsc->x + seg_y2x(y - rtsc->y, rtsc->ism1); } else { /* x belongs to the 2nd segment */ x = rtsc->x + rtsc->dx + seg_y2x(y - rtsc->y - rtsc->dy, rtsc->ism2); } return (x); } static u_int64_t rtsc_x2y(struct runtime_sc *rtsc, u_int64_t x) { u_int64_t y; if (x <= rtsc->x) y = rtsc->y; else if (x <= rtsc->x + rtsc->dx) /* y belongs to the 1st segment */ y = rtsc->y + seg_x2y(x - rtsc->x, rtsc->sm1); else /* y belongs to the 2nd segment */ y = rtsc->y + rtsc->dy + seg_x2y(x - rtsc->x - rtsc->dx, rtsc->sm2); return (y); } /* * update the runtime service curve by taking the minimum of the current * runtime service curve and the service curve starting at (x, y). */ static void rtsc_min(struct runtime_sc *rtsc, struct internal_sc *isc, u_int64_t x, u_int64_t y) { u_int64_t y1, y2, dx, dy; if (isc->sm1 <= isc->sm2) { /* service curve is convex */ y1 = rtsc_x2y(rtsc, x); if (y1 < y) /* the current rtsc is smaller */ return; rtsc->x = x; rtsc->y = y; return; } /* * service curve is concave * compute the two y values of the current rtsc * y1: at x * y2: at (x + dx) */ y1 = rtsc_x2y(rtsc, x); if (y1 <= y) { /* rtsc is below isc, no change to rtsc */ return; } y2 = rtsc_x2y(rtsc, x + isc->dx); if (y2 >= y + isc->dy) { /* rtsc is above isc, replace rtsc by isc */ rtsc->x = x; rtsc->y = y; rtsc->dx = isc->dx; rtsc->dy = isc->dy; return; } /* * the two curves intersect * compute the offsets (dx, dy) using the reverse * function of seg_x2y() * seg_x2y(dx, sm1) == seg_x2y(dx, sm2) + (y1 - y) */ dx = ((y1 - y) << SM_SHIFT) / (isc->sm1 - isc->sm2); /* * check if (x, y1) belongs to the 1st segment of rtsc. * if so, add the offset. */ if (rtsc->x + rtsc->dx > x) dx += rtsc->x + rtsc->dx - x; dy = seg_x2y(dx, isc->sm1); rtsc->x = x; rtsc->y = y; rtsc->dx = dx; rtsc->dy = dy; return; } static void get_class_stats_v0(struct hfsc_classstats_v0 *sp, struct hfsc_class *cl) { sp->class_id = cl->cl_id; sp->class_handle = cl->cl_handle; #define SATU32(x) (u_int32_t)uqmin((x), UINT_MAX) if (cl->cl_rsc != NULL) { sp->rsc.m1 = SATU32(sm2m(cl->cl_rsc->sm1)); sp->rsc.d = dx2d(cl->cl_rsc->dx); sp->rsc.m2 = SATU32(sm2m(cl->cl_rsc->sm2)); } else { sp->rsc.m1 = 0; sp->rsc.d = 0; sp->rsc.m2 = 0; } if (cl->cl_fsc != NULL) { sp->fsc.m1 = SATU32(sm2m(cl->cl_fsc->sm1)); sp->fsc.d = dx2d(cl->cl_fsc->dx); sp->fsc.m2 = SATU32(sm2m(cl->cl_fsc->sm2)); } else { sp->fsc.m1 = 0; sp->fsc.d = 0; sp->fsc.m2 = 0; } if (cl->cl_usc != NULL) { sp->usc.m1 = SATU32(sm2m(cl->cl_usc->sm1)); sp->usc.d = dx2d(cl->cl_usc->dx); sp->usc.m2 = SATU32(sm2m(cl->cl_usc->sm2)); } else { sp->usc.m1 = 0; sp->usc.d = 0; sp->usc.m2 = 0; } #undef SATU32 sp->total = cl->cl_total; sp->cumul = cl->cl_cumul; sp->d = cl->cl_d; sp->e = cl->cl_e; sp->vt = cl->cl_vt; sp->f = cl->cl_f; sp->initvt = cl->cl_initvt; sp->vtperiod = cl->cl_vtperiod; sp->parentperiod = cl->cl_parentperiod; sp->nactive = cl->cl_nactive; sp->vtoff = cl->cl_vtoff; sp->cvtmax = cl->cl_cvtmax; sp->myf = cl->cl_myf; sp->cfmin = cl->cl_cfmin; sp->cvtmin = cl->cl_cvtmin; sp->myfadj = cl->cl_myfadj; sp->vtadj = cl->cl_vtadj; sp->cur_time = read_machclk(); sp->machclk_freq = machclk_freq; sp->qlength = qlen(cl->cl_q); sp->qlimit = qlimit(cl->cl_q); sp->xmit_cnt = cl->cl_stats.xmit_cnt; sp->drop_cnt = cl->cl_stats.drop_cnt; sp->period = cl->cl_stats.period; sp->qtype = qtype(cl->cl_q); #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_getstats(cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_getstats((rio_t *)cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_getstats(cl->cl_codel, &sp->codel); #endif } static void get_class_stats_v1(struct hfsc_classstats_v1 *sp, struct hfsc_class *cl) { sp->class_id = cl->cl_id; sp->class_handle = cl->cl_handle; if (cl->cl_rsc != NULL) { sp->rsc.m1 = sm2m(cl->cl_rsc->sm1); sp->rsc.d = dx2d(cl->cl_rsc->dx); sp->rsc.m2 = sm2m(cl->cl_rsc->sm2); } else { sp->rsc.m1 = 0; sp->rsc.d = 0; sp->rsc.m2 = 0; } if (cl->cl_fsc != NULL) { sp->fsc.m1 = sm2m(cl->cl_fsc->sm1); sp->fsc.d = dx2d(cl->cl_fsc->dx); sp->fsc.m2 = sm2m(cl->cl_fsc->sm2); } else { sp->fsc.m1 = 0; sp->fsc.d = 0; sp->fsc.m2 = 0; } if (cl->cl_usc != NULL) { sp->usc.m1 = sm2m(cl->cl_usc->sm1); sp->usc.d = dx2d(cl->cl_usc->dx); sp->usc.m2 = sm2m(cl->cl_usc->sm2); } else { sp->usc.m1 = 0; sp->usc.d = 0; sp->usc.m2 = 0; } sp->total = cl->cl_total; sp->cumul = cl->cl_cumul; sp->d = cl->cl_d; sp->e = cl->cl_e; sp->vt = cl->cl_vt; sp->f = cl->cl_f; sp->initvt = cl->cl_initvt; sp->vtperiod = cl->cl_vtperiod; sp->parentperiod = cl->cl_parentperiod; sp->nactive = cl->cl_nactive; sp->vtoff = cl->cl_vtoff; sp->cvtmax = cl->cl_cvtmax; sp->myf = cl->cl_myf; sp->cfmin = cl->cl_cfmin; sp->cvtmin = cl->cl_cvtmin; sp->myfadj = cl->cl_myfadj; sp->vtadj = cl->cl_vtadj; sp->cur_time = read_machclk(); sp->machclk_freq = machclk_freq; sp->qlength = qlen(cl->cl_q); sp->qlimit = qlimit(cl->cl_q); sp->xmit_cnt = cl->cl_stats.xmit_cnt; sp->drop_cnt = cl->cl_stats.drop_cnt; sp->period = cl->cl_stats.period; sp->qtype = qtype(cl->cl_q); #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_getstats(cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_getstats((rio_t *)cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_getstats(cl->cl_codel, &sp->codel); #endif } /* convert a class handle to the corresponding class pointer */ static struct hfsc_class * clh_to_clp(struct hfsc_if *hif, u_int32_t chandle) { int i; struct hfsc_class *cl; if (chandle == 0) return (NULL); /* * first, try optimistically the slot matching the lower bits of * the handle. if it fails, do the linear table search. */ i = chandle % HFSC_MAX_CLASSES; if ((cl = hif->hif_class_tbl[i]) != NULL && cl->cl_handle == chandle) return (cl); for (i = 0; i < HFSC_MAX_CLASSES; i++) if ((cl = hif->hif_class_tbl[i]) != NULL && cl->cl_handle == chandle) return (cl); return (NULL); } #endif /* ALTQ_HFSC */ Index: head/sys/net/altq/altq_hfsc.h =================================================================== --- head/sys/net/altq/altq_hfsc.h (revision 343994) +++ head/sys/net/altq/altq_hfsc.h (revision 343995) @@ -1,338 +1,339 @@ /*- * Copyright (c) 1997-1999 Carnegie Mellon University. All Rights Reserved. * * Permission to use, copy, modify, and distribute this software and * its documentation is hereby granted (including for commercial or * for-profit use), provided that both the copyright notice and this * permission notice appear in all copies of the software, derivative * works, or modified versions, and any portions thereof. * * THIS SOFTWARE IS EXPERIMENTAL AND IS KNOWN TO HAVE BUGS, SOME OF * WHICH MAY HAVE SERIOUS CONSEQUENCES. CARNEGIE MELLON PROVIDES THIS * SOFTWARE IN ITS ``AS IS'' CONDITION, AND ANY EXPRESS OR IMPLIED * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH * DAMAGE. * * Carnegie Mellon encourages (but does not require) users of this * software to return any improvements or extensions that they make, * and to grant Carnegie Mellon the rights to redistribute these * changes without encumbrance. * * $KAME: altq_hfsc.h,v 1.12 2003/12/05 05:40:46 kjc Exp $ * $FreeBSD$ */ #ifndef _ALTQ_ALTQ_HFSC_H_ #define _ALTQ_ALTQ_HFSC_H_ #include #include #include #include #include #ifdef __cplusplus extern "C" { #endif struct service_curve_v0 { u_int m1; /* slope of the first segment in bits/sec */ u_int d; /* the x-projection of the first segment in msec */ u_int m2; /* slope of the second segment in bits/sec */ }; struct service_curve_v1 { u_int64_t m1; /* slope of the first segment in bits/sec */ u_int d; /* the x-projection of the first segment in msec */ u_int64_t m2; /* slope of the second segment in bits/sec */ }; /* Latest version of struct service_curve_vX */ #define HFSC_SERVICE_CURVE_VERSION 1 /* special class handles */ #define HFSC_NULLCLASS_HANDLE 0 #define HFSC_MAX_CLASSES 64 /* hfsc class flags */ #define HFCF_RED 0x0001 /* use RED */ #define HFCF_ECN 0x0002 /* use RED/ECN */ #define HFCF_RIO 0x0004 /* use RIO */ #define HFCF_CODEL 0x0008 /* use CoDel */ #define HFCF_CLEARDSCP 0x0010 /* clear diffserv codepoint */ #define HFCF_DEFAULTCLASS 0x1000 /* default class */ /* service curve types */ #define HFSC_REALTIMESC 1 #define HFSC_LINKSHARINGSC 2 #define HFSC_UPPERLIMITSC 4 #define HFSC_DEFAULTSC (HFSC_REALTIMESC|HFSC_LINKSHARINGSC) struct hfsc_classstats_v0 { u_int class_id; u_int32_t class_handle; struct service_curve_v0 rsc; struct service_curve_v0 fsc; struct service_curve_v0 usc; /* upper limit service curve */ u_int64_t total; /* total work in bytes */ u_int64_t cumul; /* cumulative work in bytes done by real-time criteria */ u_int64_t d; /* deadline */ u_int64_t e; /* eligible time */ u_int64_t vt; /* virtual time */ u_int64_t f; /* fit time for upper-limit */ /* info helpful for debugging */ u_int64_t initvt; /* init virtual time */ u_int64_t vtoff; /* cl_vt_ipoff */ u_int64_t cvtmax; /* cl_maxvt */ u_int64_t myf; /* cl_myf */ u_int64_t cfmin; /* cl_mincf */ u_int64_t cvtmin; /* cl_mincvt */ u_int64_t myfadj; /* cl_myfadj */ u_int64_t vtadj; /* cl_vtadj */ u_int64_t cur_time; u_int32_t machclk_freq; u_int qlength; u_int qlimit; struct pktcntr xmit_cnt; struct pktcntr drop_cnt; u_int period; u_int vtperiod; /* vt period sequence no */ u_int parentperiod; /* parent's vt period seqno */ int nactive; /* number of active children */ /* codel, red and rio related info */ int qtype; struct redstats red[3]; struct codel_stats codel; }; struct hfsc_classstats_v1 { u_int class_id; u_int32_t class_handle; struct service_curve_v1 rsc; struct service_curve_v1 fsc; struct service_curve_v1 usc; /* upper limit service curve */ u_int64_t total; /* total work in bytes */ u_int64_t cumul; /* cumulative work in bytes done by real-time criteria */ u_int64_t d; /* deadline */ u_int64_t e; /* eligible time */ u_int64_t vt; /* virtual time */ u_int64_t f; /* fit time for upper-limit */ /* info helpful for debugging */ u_int64_t initvt; /* init virtual time */ u_int64_t vtoff; /* cl_vt_ipoff */ u_int64_t cvtmax; /* cl_maxvt */ u_int64_t myf; /* cl_myf */ u_int64_t cfmin; /* cl_mincf */ u_int64_t cvtmin; /* cl_mincvt */ u_int64_t myfadj; /* cl_myfadj */ u_int64_t vtadj; /* cl_vtadj */ u_int64_t cur_time; u_int32_t machclk_freq; u_int qlength; u_int qlimit; struct pktcntr xmit_cnt; struct pktcntr drop_cnt; u_int period; u_int vtperiod; /* vt period sequence no */ u_int parentperiod; /* parent's vt period seqno */ int nactive; /* number of active children */ /* codel, red and rio related info */ int qtype; struct redstats red[3]; struct codel_stats codel; }; /* * HFSC_STATS_VERSION is defined in altq.h to work around issues stemming * from mixing of public-API and internal bits in each scheduler-specific * header. */ #ifdef _KERNEL /* * kernel internal service curve representation * coordinates are given by 64 bit unsigned integers. * x-axis: unit is clock count. for the intel x86 architecture, * the raw Pentium TSC (Timestamp Counter) value is used. * virtual time is also calculated in this time scale. * y-axis: unit is byte. * * the service curve parameters are converted to the internal * representation. * the slope values are scaled to avoid overflow. * the inverse slope values as well as the y-projection of the 1st * segment are kept in order to avoid 64-bit divide operations * that are expensive on 32-bit architectures. * * note: Intel Pentium TSC never wraps around in several thousands of years. * x-axis doesn't wrap around for 1089 years with 1GHz clock. * y-axis doesn't wrap around for 4358 years with 1Gbps bandwidth. */ /* kernel internal representation of a service curve */ struct internal_sc { u_int64_t sm1; /* scaled slope of the 1st segment */ u_int64_t ism1; /* scaled inverse-slope of the 1st segment */ u_int64_t dx; /* the x-projection of the 1st segment */ u_int64_t dy; /* the y-projection of the 1st segment */ u_int64_t sm2; /* scaled slope of the 2nd segment */ u_int64_t ism2; /* scaled inverse-slope of the 2nd segment */ }; /* runtime service curve */ struct runtime_sc { u_int64_t x; /* current starting position on x-axis */ u_int64_t y; /* current starting position on x-axis */ u_int64_t sm1; /* scaled slope of the 1st segment */ u_int64_t ism1; /* scaled inverse-slope of the 1st segment */ u_int64_t dx; /* the x-projection of the 1st segment */ u_int64_t dy; /* the y-projection of the 1st segment */ u_int64_t sm2; /* scaled slope of the 2nd segment */ u_int64_t ism2; /* scaled inverse-slope of the 2nd segment */ }; struct hfsc_class { u_int cl_id; /* class id (just for debug) */ + u_int cl_slot; /* slot in hif class table */ u_int32_t cl_handle; /* class handle */ struct hfsc_if *cl_hif; /* back pointer to struct hfsc_if */ int cl_flags; /* misc flags */ struct hfsc_class *cl_parent; /* parent class */ struct hfsc_class *cl_siblings; /* sibling classes */ struct hfsc_class *cl_children; /* child classes */ class_queue_t *cl_q; /* class queue structure */ union { struct red *cl_red; /* RED state */ struct codel *cl_codel; /* CoDel state */ } cl_aqm; #define cl_red cl_aqm.cl_red #define cl_codel cl_aqm.cl_codel struct altq_pktattr *cl_pktattr; /* saved header used by ECN */ u_int64_t cl_total; /* total work in bytes */ u_int64_t cl_cumul; /* cumulative work in bytes done by real-time criteria */ u_int64_t cl_d; /* deadline */ u_int64_t cl_e; /* eligible time */ u_int64_t cl_vt; /* virtual time */ u_int64_t cl_f; /* time when this class will fit for link-sharing, max(myf, cfmin) */ u_int64_t cl_myf; /* my fit-time (as calculated from this class's own upperlimit curve) */ u_int64_t cl_myfadj; /* my fit-time adjustment (to cancel history dependence) */ u_int64_t cl_cfmin; /* earliest children's fit-time (used with cl_myf to obtain cl_f) */ u_int64_t cl_cvtmin; /* minimal virtual time among the children fit for link-sharing (monotonic within a period) */ u_int64_t cl_vtadj; /* intra-period cumulative vt adjustment */ u_int64_t cl_vtoff; /* inter-period cumulative vt offset */ u_int64_t cl_cvtmax; /* max child's vt in the last period */ u_int64_t cl_initvt; /* init virtual time (for debugging) */ struct internal_sc *cl_rsc; /* internal real-time service curve */ struct internal_sc *cl_fsc; /* internal fair service curve */ struct internal_sc *cl_usc; /* internal upperlimit service curve */ struct runtime_sc cl_deadline; /* deadline curve */ struct runtime_sc cl_eligible; /* eligible curve */ struct runtime_sc cl_virtual; /* virtual curve */ struct runtime_sc cl_ulimit; /* upperlimit curve */ u_int cl_vtperiod; /* vt period sequence no */ u_int cl_parentperiod; /* parent's vt period seqno */ int cl_nactive; /* number of active children */ TAILQ_HEAD(acthead, hfsc_class) cl_actc; /* active children list */ TAILQ_ENTRY(hfsc_class) cl_actlist; /* active children list entry */ TAILQ_ENTRY(hfsc_class) cl_ellist; /* eligible list entry */ struct { struct pktcntr xmit_cnt; struct pktcntr drop_cnt; u_int period; } cl_stats; }; /* * hfsc interface state */ struct hfsc_if { struct hfsc_if *hif_next; /* interface state list */ struct ifaltq *hif_ifq; /* backpointer to ifaltq */ struct hfsc_class *hif_rootclass; /* root class */ struct hfsc_class *hif_defaultclass; /* default class */ struct hfsc_class *hif_class_tbl[HFSC_MAX_CLASSES]; struct hfsc_class *hif_pollcache; /* cache for poll operation */ u_int hif_classes; /* # of classes in the tree */ u_int hif_packets; /* # of packets in the tree */ u_int hif_classid; /* class id sequence number */ TAILQ_HEAD(elighead, hfsc_class) hif_eligible; /* eligible list */ #ifdef ALTQ3_CLFIER_COMPAT struct acc_classifier hif_classifier; #endif }; /* * Kernel code always wants the latest version - avoid a bunch of renames in * the code to the current latest versioned name. */ #define service_curve __CONCAT(service_curve_v, HFSC_SERVICE_CURVE_VERSION) #else /* _KERNEL */ #ifdef PFIOC_USE_LATEST /* * Maintaining in-tree consumers of the ioctl interface is easier when that * code can be written in terms old names that refer to the latest interface * version as that reduces the required changes in the consumers to those * that are functionally necessary to accommodate a new interface version. */ #define hfsc_classstats __CONCAT(hfsc_classstats_v, HFSC_STATS_VERSION) #define service_curve __CONCAT(service_curve_v, HFSC_SERVICE_CURVE_VERSION) #else /* * When building out-of-tree code that is written for the old interface, * such as may exist in ports for example, resolve the old struct tags to * the v0 versions. */ #define hfsc_classstats __CONCAT(hfsc_classstats_v, 0) #define service_curve __CONCAT(service_curve_v, 0) #endif /* PFIOC_USE_LATEST */ #endif /* _KERNEL */ #ifdef __cplusplus } #endif #endif /* _ALTQ_ALTQ_HFSC_H_ */ Index: head/sys/net/altq/altq_priq.c =================================================================== --- head/sys/net/altq/altq_priq.c (revision 343994) +++ head/sys/net/altq/altq_priq.c (revision 343995) @@ -1,645 +1,644 @@ /*- * Copyright (C) 2000-2003 * Sony Computer Science Laboratories Inc. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: altq_priq.c,v 1.11 2003/09/17 14:23:25 kjc Exp $ * $FreeBSD$ */ /* * priority queue */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #ifdef ALTQ_PRIQ /* priq is enabled by ALTQ_PRIQ option in opt_altq.h */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * function prototypes */ static int priq_clear_interface(struct priq_if *); static int priq_request(struct ifaltq *, int, void *); static void priq_purge(struct priq_if *); static struct priq_class *priq_class_create(struct priq_if *, int, int, int, int); static int priq_class_destroy(struct priq_class *); static int priq_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); static struct mbuf *priq_dequeue(struct ifaltq *, int); static int priq_addq(struct priq_class *, struct mbuf *); static struct mbuf *priq_getq(struct priq_class *); static struct mbuf *priq_pollq(struct priq_class *); static void priq_purgeq(struct priq_class *); static void get_class_stats(struct priq_classstats *, struct priq_class *); static struct priq_class *clh_to_clp(struct priq_if *, u_int32_t); int priq_pfattach(struct pf_altq *a) { struct ifnet *ifp; int s, error; if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) return (EINVAL); s = splnet(); error = altq_attach(&ifp->if_snd, ALTQT_PRIQ, a->altq_disc, priq_enqueue, priq_dequeue, priq_request, NULL, NULL); splx(s); return (error); } int -priq_add_altq(struct pf_altq *a) +priq_add_altq(struct ifnet * ifp, struct pf_altq *a) { struct priq_if *pif; - struct ifnet *ifp; - if ((ifp = ifunit(a->ifname)) == NULL) + if (ifp == NULL) return (EINVAL); if (!ALTQ_IS_READY(&ifp->if_snd)) return (ENODEV); pif = malloc(sizeof(struct priq_if), M_DEVBUF, M_NOWAIT | M_ZERO); if (pif == NULL) return (ENOMEM); pif->pif_bandwidth = a->ifbandwidth; pif->pif_maxpri = -1; pif->pif_ifq = &ifp->if_snd; /* keep the state in pf_altq */ a->altq_disc = pif; return (0); } int priq_remove_altq(struct pf_altq *a) { struct priq_if *pif; if ((pif = a->altq_disc) == NULL) return (EINVAL); a->altq_disc = NULL; (void)priq_clear_interface(pif); free(pif, M_DEVBUF); return (0); } int priq_add_queue(struct pf_altq *a) { struct priq_if *pif; struct priq_class *cl; if ((pif = a->altq_disc) == NULL) return (EINVAL); /* check parameters */ if (a->priority >= PRIQ_MAXPRI) return (EINVAL); if (a->qid == 0) return (EINVAL); if (pif->pif_classes[a->priority] != NULL) return (EBUSY); if (clh_to_clp(pif, a->qid) != NULL) return (EBUSY); cl = priq_class_create(pif, a->priority, a->qlimit, a->pq_u.priq_opts.flags, a->qid); if (cl == NULL) return (ENOMEM); return (0); } int priq_remove_queue(struct pf_altq *a) { struct priq_if *pif; struct priq_class *cl; if ((pif = a->altq_disc) == NULL) return (EINVAL); if ((cl = clh_to_clp(pif, a->qid)) == NULL) return (EINVAL); return (priq_class_destroy(cl)); } int priq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { struct priq_if *pif; struct priq_class *cl; struct priq_classstats stats; int error = 0; if ((pif = altq_lookup(a->ifname, ALTQT_PRIQ)) == NULL) return (EBADF); if ((cl = clh_to_clp(pif, a->qid)) == NULL) return (EINVAL); if (*nbytes < sizeof(stats)) return (EINVAL); get_class_stats(&stats, cl); if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0) return (error); *nbytes = sizeof(stats); return (0); } /* * bring the interface back to the initial state by discarding * all the filters and classes. */ static int priq_clear_interface(struct priq_if *pif) { struct priq_class *cl; int pri; #ifdef ALTQ3_CLFIER_COMPAT /* free the filters for this interface */ acc_discard_filters(&pif->pif_classifier, NULL, 1); #endif /* clear out the classes */ for (pri = 0; pri <= pif->pif_maxpri; pri++) if ((cl = pif->pif_classes[pri]) != NULL) priq_class_destroy(cl); return (0); } static int priq_request(struct ifaltq *ifq, int req, void *arg) { struct priq_if *pif = (struct priq_if *)ifq->altq_disc; IFQ_LOCK_ASSERT(ifq); switch (req) { case ALTRQ_PURGE: priq_purge(pif); break; } return (0); } /* discard all the queued packets on the interface */ static void priq_purge(struct priq_if *pif) { struct priq_class *cl; int pri; for (pri = 0; pri <= pif->pif_maxpri; pri++) { if ((cl = pif->pif_classes[pri]) != NULL && !qempty(cl->cl_q)) priq_purgeq(cl); } if (ALTQ_IS_ENABLED(pif->pif_ifq)) pif->pif_ifq->ifq_len = 0; } static struct priq_class * priq_class_create(struct priq_if *pif, int pri, int qlimit, int flags, int qid) { struct priq_class *cl; int s; #ifndef ALTQ_RED if (flags & PRCF_RED) { #ifdef ALTQ_DEBUG printf("priq_class_create: RED not configured for PRIQ!\n"); #endif return (NULL); } #endif #ifndef ALTQ_CODEL if (flags & PRCF_CODEL) { #ifdef ALTQ_DEBUG printf("priq_class_create: CODEL not configured for PRIQ!\n"); #endif return (NULL); } #endif if ((cl = pif->pif_classes[pri]) != NULL) { /* modify the class instead of creating a new one */ s = splnet(); IFQ_LOCK(cl->cl_pif->pif_ifq); if (!qempty(cl->cl_q)) priq_purgeq(cl); IFQ_UNLOCK(cl->cl_pif->pif_ifq); splx(s); #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_destroy(cl->cl_codel); #endif } else { cl = malloc(sizeof(struct priq_class), M_DEVBUF, M_NOWAIT | M_ZERO); if (cl == NULL) return (NULL); cl->cl_q = malloc(sizeof(class_queue_t), M_DEVBUF, M_NOWAIT | M_ZERO); if (cl->cl_q == NULL) goto err_ret; } pif->pif_classes[pri] = cl; if (flags & PRCF_DEFAULTCLASS) pif->pif_default = cl; if (qlimit == 0) qlimit = 50; /* use default */ qlimit(cl->cl_q) = qlimit; qtype(cl->cl_q) = Q_DROPTAIL; qlen(cl->cl_q) = 0; qsize(cl->cl_q) = 0; cl->cl_flags = flags; cl->cl_pri = pri; if (pri > pif->pif_maxpri) pif->pif_maxpri = pri; cl->cl_pif = pif; cl->cl_handle = qid; #ifdef ALTQ_RED if (flags & (PRCF_RED|PRCF_RIO)) { int red_flags, red_pkttime; red_flags = 0; if (flags & PRCF_ECN) red_flags |= REDF_ECN; #ifdef ALTQ_RIO if (flags & PRCF_CLEARDSCP) red_flags |= RIOF_CLEARDSCP; #endif if (pif->pif_bandwidth < 8) red_pkttime = 1000 * 1000 * 1000; /* 1 sec */ else red_pkttime = (int64_t)pif->pif_ifq->altq_ifp->if_mtu * 1000 * 1000 * 1000 / (pif->pif_bandwidth / 8); #ifdef ALTQ_RIO if (flags & PRCF_RIO) { cl->cl_red = (red_t *)rio_alloc(0, NULL, red_flags, red_pkttime); if (cl->cl_red == NULL) goto err_ret; qtype(cl->cl_q) = Q_RIO; } else #endif if (flags & PRCF_RED) { cl->cl_red = red_alloc(0, 0, qlimit(cl->cl_q) * 10/100, qlimit(cl->cl_q) * 30/100, red_flags, red_pkttime); if (cl->cl_red == NULL) goto err_ret; qtype(cl->cl_q) = Q_RED; } } #endif /* ALTQ_RED */ #ifdef ALTQ_CODEL if (flags & PRCF_CODEL) { cl->cl_codel = codel_alloc(5, 100, 0); if (cl->cl_codel != NULL) qtype(cl->cl_q) = Q_CODEL; } #endif return (cl); err_ret: if (cl->cl_red != NULL) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_destroy(cl->cl_codel); #endif } if (cl->cl_q != NULL) free(cl->cl_q, M_DEVBUF); free(cl, M_DEVBUF); return (NULL); } static int priq_class_destroy(struct priq_class *cl) { struct priq_if *pif; int s, pri; s = splnet(); IFQ_LOCK(cl->cl_pif->pif_ifq); #ifdef ALTQ3_CLFIER_COMPAT /* delete filters referencing to this class */ acc_discard_filters(&cl->cl_pif->pif_classifier, cl, 0); #endif if (!qempty(cl->cl_q)) priq_purgeq(cl); pif = cl->cl_pif; pif->pif_classes[cl->cl_pri] = NULL; if (pif->pif_maxpri == cl->cl_pri) { for (pri = cl->cl_pri; pri >= 0; pri--) if (pif->pif_classes[pri] != NULL) { pif->pif_maxpri = pri; break; } if (pri < 0) pif->pif_maxpri = -1; } IFQ_UNLOCK(cl->cl_pif->pif_ifq); splx(s); if (cl->cl_red != NULL) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_destroy((rio_t *)cl->cl_red); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_destroy(cl->cl_red); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_destroy(cl->cl_codel); #endif } free(cl->cl_q, M_DEVBUF); free(cl, M_DEVBUF); return (0); } /* * priq_enqueue is an enqueue function to be registered to * (*altq_enqueue) in struct ifaltq. */ static int priq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) { struct priq_if *pif = (struct priq_if *)ifq->altq_disc; struct priq_class *cl; struct pf_mtag *t; int len; IFQ_LOCK_ASSERT(ifq); /* grab class set by classifier */ if ((m->m_flags & M_PKTHDR) == 0) { /* should not happen */ printf("altq: packet for %s does not have pkthdr\n", ifq->altq_ifp->if_xname); m_freem(m); return (ENOBUFS); } cl = NULL; if ((t = pf_find_mtag(m)) != NULL) cl = clh_to_clp(pif, t->qid); if (cl == NULL) { cl = pif->pif_default; if (cl == NULL) { m_freem(m); return (ENOBUFS); } } cl->cl_pktattr = NULL; len = m_pktlen(m); if (priq_addq(cl, m) != 0) { /* drop occurred. mbuf was freed in priq_addq. */ PKTCNTR_ADD(&cl->cl_dropcnt, len); return (ENOBUFS); } IFQ_INC_LEN(ifq); /* successfully queued. */ return (0); } /* * priq_dequeue is a dequeue function to be registered to * (*altq_dequeue) in struct ifaltq. * * note: ALTDQ_POLL returns the next packet without removing the packet * from the queue. ALTDQ_REMOVE is a normal dequeue operation. * ALTDQ_REMOVE must return the same packet if called immediately * after ALTDQ_POLL. */ static struct mbuf * priq_dequeue(struct ifaltq *ifq, int op) { struct priq_if *pif = (struct priq_if *)ifq->altq_disc; struct priq_class *cl; struct mbuf *m; int pri; IFQ_LOCK_ASSERT(ifq); if (IFQ_IS_EMPTY(ifq)) /* no packet in the queue */ return (NULL); for (pri = pif->pif_maxpri; pri >= 0; pri--) { if ((cl = pif->pif_classes[pri]) != NULL && !qempty(cl->cl_q)) { if (op == ALTDQ_POLL) return (priq_pollq(cl)); m = priq_getq(cl); if (m != NULL) { IFQ_DEC_LEN(ifq); if (qempty(cl->cl_q)) cl->cl_period++; PKTCNTR_ADD(&cl->cl_xmitcnt, m_pktlen(m)); } return (m); } } return (NULL); } static int priq_addq(struct priq_class *cl, struct mbuf *m) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) return rio_addq((rio_t *)cl->cl_red, cl->cl_q, m, cl->cl_pktattr); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) return red_addq(cl->cl_red, cl->cl_q, m, cl->cl_pktattr); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) return codel_addq(cl->cl_codel, cl->cl_q, m); #endif if (qlen(cl->cl_q) >= qlimit(cl->cl_q)) { m_freem(m); return (-1); } if (cl->cl_flags & PRCF_CLEARDSCP) write_dsfield(m, cl->cl_pktattr, 0); _addq(cl->cl_q, m); return (0); } static struct mbuf * priq_getq(struct priq_class *cl) { #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) return rio_getq((rio_t *)cl->cl_red, cl->cl_q); #endif #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) return red_getq(cl->cl_red, cl->cl_q); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) return codel_getq(cl->cl_codel, cl->cl_q); #endif return _getq(cl->cl_q); } static struct mbuf * priq_pollq(cl) struct priq_class *cl; { return qhead(cl->cl_q); } static void priq_purgeq(struct priq_class *cl) { struct mbuf *m; if (qempty(cl->cl_q)) return; while ((m = _getq(cl->cl_q)) != NULL) { PKTCNTR_ADD(&cl->cl_dropcnt, m_pktlen(m)); m_freem(m); } ASSERT(qlen(cl->cl_q) == 0); } static void get_class_stats(struct priq_classstats *sp, struct priq_class *cl) { sp->class_handle = cl->cl_handle; sp->qlength = qlen(cl->cl_q); sp->qlimit = qlimit(cl->cl_q); sp->period = cl->cl_period; sp->xmitcnt = cl->cl_xmitcnt; sp->dropcnt = cl->cl_dropcnt; sp->qtype = qtype(cl->cl_q); #ifdef ALTQ_RED if (q_is_red(cl->cl_q)) red_getstats(cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_RIO if (q_is_rio(cl->cl_q)) rio_getstats((rio_t *)cl->cl_red, &sp->red[0]); #endif #ifdef ALTQ_CODEL if (q_is_codel(cl->cl_q)) codel_getstats(cl->cl_codel, &sp->codel); #endif } /* convert a class handle to the corresponding class pointer */ static struct priq_class * clh_to_clp(struct priq_if *pif, u_int32_t chandle) { struct priq_class *cl; int idx; if (chandle == 0) return (NULL); for (idx = pif->pif_maxpri; idx >= 0; idx--) if ((cl = pif->pif_classes[idx]) != NULL && cl->cl_handle == chandle) return (cl); return (NULL); } #endif /* ALTQ_PRIQ */ Index: head/sys/net/altq/altq_subr.c =================================================================== --- head/sys/net/altq/altq_subr.c (revision 343994) +++ head/sys/net/altq/altq_subr.c (revision 343995) @@ -1,1936 +1,1936 @@ /*- * Copyright (C) 1997-2003 * Sony Computer Science Laboratories Inc. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: altq_subr.c,v 1.21 2003/11/06 06:32:53 kjc Exp $ * $FreeBSD$ */ #include "opt_altq.h" #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif #include #include #include #include #include /* machine dependent clock related includes */ #include #include #include #include #if defined(__amd64__) || defined(__i386__) #include /* for pentium tsc */ #include /* for CPUID_TSC */ #include /* for cpu_feature */ #endif /* __amd64 || __i386__ */ /* * internal function prototypes */ static void tbr_timeout(void *); int (*altq_input)(struct mbuf *, int) = NULL; static struct mbuf *tbr_dequeue(struct ifaltq *, int); static int tbr_timer = 0; /* token bucket regulator timer */ #if !defined(__FreeBSD__) || (__FreeBSD_version < 600000) static struct callout tbr_callout = CALLOUT_INITIALIZER; #else static struct callout tbr_callout; #endif #ifdef ALTQ3_CLFIER_COMPAT static int extract_ports4(struct mbuf *, struct ip *, struct flowinfo_in *); #ifdef INET6 static int extract_ports6(struct mbuf *, struct ip6_hdr *, struct flowinfo_in6 *); #endif static int apply_filter4(u_int32_t, struct flow_filter *, struct flowinfo_in *); static int apply_ppfilter4(u_int32_t, struct flow_filter *, struct flowinfo_in *); #ifdef INET6 static int apply_filter6(u_int32_t, struct flow_filter6 *, struct flowinfo_in6 *); #endif static int apply_tosfilter4(u_int32_t, struct flow_filter *, struct flowinfo_in *); static u_long get_filt_handle(struct acc_classifier *, int); static struct acc_filter *filth_to_filtp(struct acc_classifier *, u_long); static u_int32_t filt2fibmask(struct flow_filter *); static void ip4f_cache(struct ip *, struct flowinfo_in *); static int ip4f_lookup(struct ip *, struct flowinfo_in *); static int ip4f_init(void); static struct ip4_frag *ip4f_alloc(void); static void ip4f_free(struct ip4_frag *); #endif /* ALTQ3_CLFIER_COMPAT */ /* * alternate queueing support routines */ /* look up the queue state by the interface name and the queueing type. */ void * altq_lookup(name, type) char *name; int type; { struct ifnet *ifp; if ((ifp = ifunit(name)) != NULL) { /* read if_snd unlocked */ if (type != ALTQT_NONE && ifp->if_snd.altq_type == type) return (ifp->if_snd.altq_disc); } return NULL; } int altq_attach(ifq, type, discipline, enqueue, dequeue, request, clfier, classify) struct ifaltq *ifq; int type; void *discipline; int (*enqueue)(struct ifaltq *, struct mbuf *, struct altq_pktattr *); struct mbuf *(*dequeue)(struct ifaltq *, int); int (*request)(struct ifaltq *, int, void *); void *clfier; void *(*classify)(void *, struct mbuf *, int); { IFQ_LOCK(ifq); if (!ALTQ_IS_READY(ifq)) { IFQ_UNLOCK(ifq); return ENXIO; } ifq->altq_type = type; ifq->altq_disc = discipline; ifq->altq_enqueue = enqueue; ifq->altq_dequeue = dequeue; ifq->altq_request = request; ifq->altq_clfier = clfier; ifq->altq_classify = classify; ifq->altq_flags &= (ALTQF_CANTCHANGE|ALTQF_ENABLED); IFQ_UNLOCK(ifq); return 0; } int altq_detach(ifq) struct ifaltq *ifq; { IFQ_LOCK(ifq); if (!ALTQ_IS_READY(ifq)) { IFQ_UNLOCK(ifq); return ENXIO; } if (ALTQ_IS_ENABLED(ifq)) { IFQ_UNLOCK(ifq); return EBUSY; } if (!ALTQ_IS_ATTACHED(ifq)) { IFQ_UNLOCK(ifq); return (0); } ifq->altq_type = ALTQT_NONE; ifq->altq_disc = NULL; ifq->altq_enqueue = NULL; ifq->altq_dequeue = NULL; ifq->altq_request = NULL; ifq->altq_clfier = NULL; ifq->altq_classify = NULL; ifq->altq_flags &= ALTQF_CANTCHANGE; IFQ_UNLOCK(ifq); return 0; } int altq_enable(ifq) struct ifaltq *ifq; { int s; IFQ_LOCK(ifq); if (!ALTQ_IS_READY(ifq)) { IFQ_UNLOCK(ifq); return ENXIO; } if (ALTQ_IS_ENABLED(ifq)) { IFQ_UNLOCK(ifq); return 0; } s = splnet(); IFQ_PURGE_NOLOCK(ifq); ASSERT(ifq->ifq_len == 0); ifq->ifq_drv_maxlen = 0; /* disable bulk dequeue */ ifq->altq_flags |= ALTQF_ENABLED; if (ifq->altq_clfier != NULL) ifq->altq_flags |= ALTQF_CLASSIFY; splx(s); IFQ_UNLOCK(ifq); return 0; } int altq_disable(ifq) struct ifaltq *ifq; { int s; IFQ_LOCK(ifq); if (!ALTQ_IS_ENABLED(ifq)) { IFQ_UNLOCK(ifq); return 0; } s = splnet(); IFQ_PURGE_NOLOCK(ifq); ASSERT(ifq->ifq_len == 0); ifq->altq_flags &= ~(ALTQF_ENABLED|ALTQF_CLASSIFY); splx(s); IFQ_UNLOCK(ifq); return 0; } #ifdef ALTQ_DEBUG void altq_assert(file, line, failedexpr) const char *file, *failedexpr; int line; { (void)printf("altq assertion \"%s\" failed: file \"%s\", line %d\n", failedexpr, file, line); panic("altq assertion"); /* NOTREACHED */ } #endif /* * internal representation of token bucket parameters * rate: (byte_per_unittime << TBR_SHIFT) / machclk_freq * (((bits_per_sec) / 8) << TBR_SHIFT) / machclk_freq * depth: byte << TBR_SHIFT * */ #define TBR_SHIFT 29 #define TBR_SCALE(x) ((int64_t)(x) << TBR_SHIFT) #define TBR_UNSCALE(x) ((x) >> TBR_SHIFT) static struct mbuf * tbr_dequeue(ifq, op) struct ifaltq *ifq; int op; { struct tb_regulator *tbr; struct mbuf *m; int64_t interval; u_int64_t now; IFQ_LOCK_ASSERT(ifq); tbr = ifq->altq_tbr; if (op == ALTDQ_REMOVE && tbr->tbr_lastop == ALTDQ_POLL) { /* if this is a remove after poll, bypass tbr check */ } else { /* update token only when it is negative */ if (tbr->tbr_token <= 0) { now = read_machclk(); interval = now - tbr->tbr_last; if (interval >= tbr->tbr_filluptime) tbr->tbr_token = tbr->tbr_depth; else { tbr->tbr_token += interval * tbr->tbr_rate; if (tbr->tbr_token > tbr->tbr_depth) tbr->tbr_token = tbr->tbr_depth; } tbr->tbr_last = now; } /* if token is still negative, don't allow dequeue */ if (tbr->tbr_token <= 0) return (NULL); } if (ALTQ_IS_ENABLED(ifq)) m = (*ifq->altq_dequeue)(ifq, op); else { if (op == ALTDQ_POLL) _IF_POLL(ifq, m); else _IF_DEQUEUE(ifq, m); } if (m != NULL && op == ALTDQ_REMOVE) tbr->tbr_token -= TBR_SCALE(m_pktlen(m)); tbr->tbr_lastop = op; return (m); } /* * set a token bucket regulator. * if the specified rate is zero, the token bucket regulator is deleted. */ int tbr_set(ifq, profile) struct ifaltq *ifq; struct tb_profile *profile; { struct tb_regulator *tbr, *otbr; if (tbr_dequeue_ptr == NULL) tbr_dequeue_ptr = tbr_dequeue; if (machclk_freq == 0) init_machclk(); if (machclk_freq == 0) { printf("tbr_set: no cpu clock available!\n"); return (ENXIO); } IFQ_LOCK(ifq); if (profile->rate == 0) { /* delete this tbr */ if ((tbr = ifq->altq_tbr) == NULL) { IFQ_UNLOCK(ifq); return (ENOENT); } ifq->altq_tbr = NULL; free(tbr, M_DEVBUF); IFQ_UNLOCK(ifq); return (0); } tbr = malloc(sizeof(struct tb_regulator), M_DEVBUF, M_NOWAIT | M_ZERO); if (tbr == NULL) { IFQ_UNLOCK(ifq); return (ENOMEM); } tbr->tbr_rate = TBR_SCALE(profile->rate / 8) / machclk_freq; tbr->tbr_depth = TBR_SCALE(profile->depth); if (tbr->tbr_rate > 0) tbr->tbr_filluptime = tbr->tbr_depth / tbr->tbr_rate; else tbr->tbr_filluptime = LLONG_MAX; /* * The longest time between tbr_dequeue() calls will be about 1 * system tick, as the callout that drives it is scheduled once per * tick. The refill-time detection logic in tbr_dequeue() can only * properly detect the passage of up to LLONG_MAX machclk ticks. * Therefore, in order for this logic to function properly in the * extreme case, the maximum value of tbr_filluptime should be * LLONG_MAX less one system tick's worth of machclk ticks less * some additional slop factor (here one more system tick's worth * of machclk ticks). */ if (tbr->tbr_filluptime > (LLONG_MAX - 2 * machclk_per_tick)) tbr->tbr_filluptime = LLONG_MAX - 2 * machclk_per_tick; tbr->tbr_token = tbr->tbr_depth; tbr->tbr_last = read_machclk(); tbr->tbr_lastop = ALTDQ_REMOVE; otbr = ifq->altq_tbr; ifq->altq_tbr = tbr; /* set the new tbr */ if (otbr != NULL) free(otbr, M_DEVBUF); else { if (tbr_timer == 0) { CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0); tbr_timer = 1; } } IFQ_UNLOCK(ifq); return (0); } /* * tbr_timeout goes through the interface list, and kicks the drivers * if necessary. * * MPSAFE */ static void tbr_timeout(arg) void *arg; { VNET_ITERATOR_DECL(vnet_iter); struct ifnet *ifp; struct epoch_tracker et; int active; active = 0; NET_EPOCH_ENTER(et); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); for (ifp = CK_STAILQ_FIRST(&V_ifnet); ifp; ifp = CK_STAILQ_NEXT(ifp, if_link)) { /* read from if_snd unlocked */ if (!TBR_IS_ENABLED(&ifp->if_snd)) continue; active++; if (!IFQ_IS_EMPTY(&ifp->if_snd) && ifp->if_start != NULL) (*ifp->if_start)(ifp); } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); NET_EPOCH_EXIT(et); if (active > 0) CALLOUT_RESET(&tbr_callout, 1, tbr_timeout, (void *)0); else tbr_timer = 0; /* don't need tbr_timer anymore */ } /* * attach a discipline to the interface. if one already exists, it is * overridden. * Locking is done in the discipline specific attach functions. Basically * they call back to altq_attach which takes care of the attach and locking. */ int altq_pfattach(struct pf_altq *a) { int error = 0; switch (a->scheduler) { case ALTQT_NONE: break; #ifdef ALTQ_CBQ case ALTQT_CBQ: error = cbq_pfattach(a); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: error = priq_pfattach(a); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: error = hfsc_pfattach(a); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: error = fairq_pfattach(a); break; #endif #ifdef ALTQ_CODEL case ALTQT_CODEL: error = codel_pfattach(a); break; #endif default: error = ENXIO; } return (error); } /* * detach a discipline from the interface. * it is possible that the discipline was already overridden by another * discipline. */ int altq_pfdetach(struct pf_altq *a) { struct ifnet *ifp; int s, error = 0; if ((ifp = ifunit(a->ifname)) == NULL) return (EINVAL); /* if this discipline is no longer referenced, just return */ /* read unlocked from if_snd */ if (a->altq_disc == NULL || a->altq_disc != ifp->if_snd.altq_disc) return (0); s = splnet(); /* read unlocked from if_snd, _disable and _detach take care */ if (ALTQ_IS_ENABLED(&ifp->if_snd)) error = altq_disable(&ifp->if_snd); if (error == 0) error = altq_detach(&ifp->if_snd); splx(s); return (error); } /* * add a discipline or a queue * Locking is done in the discipline specific functions with regards to * malloc with WAITOK, also it is not yet clear which lock to use. */ int -altq_add(struct pf_altq *a) +altq_add(struct ifnet *ifp, struct pf_altq *a) { int error = 0; if (a->qname[0] != 0) return (altq_add_queue(a)); if (machclk_freq == 0) init_machclk(); if (machclk_freq == 0) panic("altq_add: no cpu clock"); switch (a->scheduler) { #ifdef ALTQ_CBQ case ALTQT_CBQ: - error = cbq_add_altq(a); + error = cbq_add_altq(ifp, a); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: - error = priq_add_altq(a); + error = priq_add_altq(ifp, a); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: - error = hfsc_add_altq(a); + error = hfsc_add_altq(ifp, a); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: - error = fairq_add_altq(a); + error = fairq_add_altq(ifp, a); break; #endif #ifdef ALTQ_CODEL case ALTQT_CODEL: - error = codel_add_altq(a); + error = codel_add_altq(ifp, a); break; #endif default: error = ENXIO; } return (error); } /* * remove a discipline or a queue * It is yet unclear what lock to use to protect this operation, the * discipline specific functions will determine and grab it */ int altq_remove(struct pf_altq *a) { int error = 0; if (a->qname[0] != 0) return (altq_remove_queue(a)); switch (a->scheduler) { #ifdef ALTQ_CBQ case ALTQT_CBQ: error = cbq_remove_altq(a); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: error = priq_remove_altq(a); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: error = hfsc_remove_altq(a); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: error = fairq_remove_altq(a); break; #endif #ifdef ALTQ_CODEL case ALTQT_CODEL: error = codel_remove_altq(a); break; #endif default: error = ENXIO; } return (error); } /* * add a queue to the discipline * It is yet unclear what lock to use to protect this operation, the * discipline specific functions will determine and grab it */ int altq_add_queue(struct pf_altq *a) { int error = 0; switch (a->scheduler) { #ifdef ALTQ_CBQ case ALTQT_CBQ: error = cbq_add_queue(a); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: error = priq_add_queue(a); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: error = hfsc_add_queue(a); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: error = fairq_add_queue(a); break; #endif default: error = ENXIO; } return (error); } /* * remove a queue from the discipline * It is yet unclear what lock to use to protect this operation, the * discipline specific functions will determine and grab it */ int altq_remove_queue(struct pf_altq *a) { int error = 0; switch (a->scheduler) { #ifdef ALTQ_CBQ case ALTQT_CBQ: error = cbq_remove_queue(a); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: error = priq_remove_queue(a); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: error = hfsc_remove_queue(a); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: error = fairq_remove_queue(a); break; #endif default: error = ENXIO; } return (error); } /* * get queue statistics * Locking is done in the discipline specific functions with regards to * copyout operations, also it is not yet clear which lock to use. */ int altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes, int version) { int error = 0; switch (a->scheduler) { #ifdef ALTQ_CBQ case ALTQT_CBQ: error = cbq_getqstats(a, ubuf, nbytes, version); break; #endif #ifdef ALTQ_PRIQ case ALTQT_PRIQ: error = priq_getqstats(a, ubuf, nbytes, version); break; #endif #ifdef ALTQ_HFSC case ALTQT_HFSC: error = hfsc_getqstats(a, ubuf, nbytes, version); break; #endif #ifdef ALTQ_FAIRQ case ALTQT_FAIRQ: error = fairq_getqstats(a, ubuf, nbytes, version); break; #endif #ifdef ALTQ_CODEL case ALTQT_CODEL: error = codel_getqstats(a, ubuf, nbytes, version); break; #endif default: error = ENXIO; } return (error); } /* * read and write diffserv field in IPv4 or IPv6 header */ u_int8_t read_dsfield(m, pktattr) struct mbuf *m; struct altq_pktattr *pktattr; { struct mbuf *m0; u_int8_t ds_field = 0; if (pktattr == NULL || (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) return ((u_int8_t)0); /* verify that pattr_hdr is within the mbuf data */ for (m0 = m; m0 != NULL; m0 = m0->m_next) if ((pktattr->pattr_hdr >= m0->m_data) && (pktattr->pattr_hdr < m0->m_data + m0->m_len)) break; if (m0 == NULL) { /* ick, pattr_hdr is stale */ pktattr->pattr_af = AF_UNSPEC; #ifdef ALTQ_DEBUG printf("read_dsfield: can't locate header!\n"); #endif return ((u_int8_t)0); } if (pktattr->pattr_af == AF_INET) { struct ip *ip = (struct ip *)pktattr->pattr_hdr; if (ip->ip_v != 4) return ((u_int8_t)0); /* version mismatch! */ ds_field = ip->ip_tos; } #ifdef INET6 else if (pktattr->pattr_af == AF_INET6) { struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; u_int32_t flowlabel; flowlabel = ntohl(ip6->ip6_flow); if ((flowlabel >> 28) != 6) return ((u_int8_t)0); /* version mismatch! */ ds_field = (flowlabel >> 20) & 0xff; } #endif return (ds_field); } void write_dsfield(struct mbuf *m, struct altq_pktattr *pktattr, u_int8_t dsfield) { struct mbuf *m0; if (pktattr == NULL || (pktattr->pattr_af != AF_INET && pktattr->pattr_af != AF_INET6)) return; /* verify that pattr_hdr is within the mbuf data */ for (m0 = m; m0 != NULL; m0 = m0->m_next) if ((pktattr->pattr_hdr >= m0->m_data) && (pktattr->pattr_hdr < m0->m_data + m0->m_len)) break; if (m0 == NULL) { /* ick, pattr_hdr is stale */ pktattr->pattr_af = AF_UNSPEC; #ifdef ALTQ_DEBUG printf("write_dsfield: can't locate header!\n"); #endif return; } if (pktattr->pattr_af == AF_INET) { struct ip *ip = (struct ip *)pktattr->pattr_hdr; u_int8_t old; int32_t sum; if (ip->ip_v != 4) return; /* version mismatch! */ old = ip->ip_tos; dsfield |= old & 3; /* leave CU bits */ if (old == dsfield) return; ip->ip_tos = dsfield; /* * update checksum (from RFC1624) * HC' = ~(~HC + ~m + m') */ sum = ~ntohs(ip->ip_sum) & 0xffff; sum += 0xff00 + (~old & 0xff) + dsfield; sum = (sum >> 16) + (sum & 0xffff); sum += (sum >> 16); /* add carry */ ip->ip_sum = htons(~sum & 0xffff); } #ifdef INET6 else if (pktattr->pattr_af == AF_INET6) { struct ip6_hdr *ip6 = (struct ip6_hdr *)pktattr->pattr_hdr; u_int32_t flowlabel; flowlabel = ntohl(ip6->ip6_flow); if ((flowlabel >> 28) != 6) return; /* version mismatch! */ flowlabel = (flowlabel & 0xf03fffff) | (dsfield << 20); ip6->ip6_flow = htonl(flowlabel); } #endif return; } /* * high resolution clock support taking advantage of a machine dependent * high resolution time counter (e.g., timestamp counter of intel pentium). * we assume * - 64-bit-long monotonically-increasing counter * - frequency range is 100M-4GHz (CPU speed) */ /* if pcc is not available or disabled, emulate 256MHz using microtime() */ #define MACHCLK_SHIFT 8 int machclk_usepcc; u_int32_t machclk_freq; u_int32_t machclk_per_tick; #if defined(__i386__) && defined(__NetBSD__) extern u_int64_t cpu_tsc_freq; #endif #if (__FreeBSD_version >= 700035) /* Update TSC freq with the value indicated by the caller. */ static void tsc_freq_changed(void *arg, const struct cf_level *level, int status) { /* If there was an error during the transition, don't do anything. */ if (status != 0) return; #if (__FreeBSD_version >= 701102) && (defined(__amd64__) || defined(__i386__)) /* If TSC is P-state invariant, don't do anything. */ if (tsc_is_invariant) return; #endif /* Total setting for this level gives the new frequency in MHz. */ init_machclk(); } EVENTHANDLER_DEFINE(cpufreq_post_change, tsc_freq_changed, NULL, EVENTHANDLER_PRI_LAST); #endif /* __FreeBSD_version >= 700035 */ static void init_machclk_setup(void) { #if (__FreeBSD_version >= 600000) callout_init(&tbr_callout, 0); #endif machclk_usepcc = 1; #if (!defined(__amd64__) && !defined(__i386__)) || defined(ALTQ_NOPCC) machclk_usepcc = 0; #endif #if defined(__FreeBSD__) && defined(SMP) machclk_usepcc = 0; #endif #if defined(__NetBSD__) && defined(MULTIPROCESSOR) machclk_usepcc = 0; #endif #if defined(__amd64__) || defined(__i386__) /* check if TSC is available */ if ((cpu_feature & CPUID_TSC) == 0 || atomic_load_acq_64(&tsc_freq) == 0) machclk_usepcc = 0; #endif } void init_machclk(void) { static int called; /* Call one-time initialization function. */ if (!called) { init_machclk_setup(); called = 1; } if (machclk_usepcc == 0) { /* emulate 256MHz using microtime() */ machclk_freq = 1000000 << MACHCLK_SHIFT; machclk_per_tick = machclk_freq / hz; #ifdef ALTQ_DEBUG printf("altq: emulate %uHz cpu clock\n", machclk_freq); #endif return; } /* * if the clock frequency (of Pentium TSC or Alpha PCC) is * accessible, just use it. */ #if defined(__amd64__) || defined(__i386__) machclk_freq = atomic_load_acq_64(&tsc_freq); #endif /* * if we don't know the clock frequency, measure it. */ if (machclk_freq == 0) { static int wait; struct timeval tv_start, tv_end; u_int64_t start, end, diff; int timo; microtime(&tv_start); start = read_machclk(); timo = hz; /* 1 sec */ (void)tsleep(&wait, PWAIT | PCATCH, "init_machclk", timo); microtime(&tv_end); end = read_machclk(); diff = (u_int64_t)(tv_end.tv_sec - tv_start.tv_sec) * 1000000 + tv_end.tv_usec - tv_start.tv_usec; if (diff != 0) machclk_freq = (u_int)((end - start) * 1000000 / diff); } machclk_per_tick = machclk_freq / hz; #ifdef ALTQ_DEBUG printf("altq: CPU clock: %uHz\n", machclk_freq); #endif } #if defined(__OpenBSD__) && defined(__i386__) static __inline u_int64_t rdtsc(void) { u_int64_t rv; __asm __volatile(".byte 0x0f, 0x31" : "=A" (rv)); return (rv); } #endif /* __OpenBSD__ && __i386__ */ u_int64_t read_machclk(void) { u_int64_t val; if (machclk_usepcc) { #if defined(__amd64__) || defined(__i386__) val = rdtsc(); #else panic("read_machclk"); #endif } else { struct timeval tv, boottime; microtime(&tv); getboottime(&boottime); val = (((u_int64_t)(tv.tv_sec - boottime.tv_sec) * 1000000 + tv.tv_usec) << MACHCLK_SHIFT); } return (val); } #ifdef ALTQ3_CLFIER_COMPAT #ifndef IPPROTO_ESP #define IPPROTO_ESP 50 /* encapsulating security payload */ #endif #ifndef IPPROTO_AH #define IPPROTO_AH 51 /* authentication header */ #endif /* * extract flow information from a given packet. * filt_mask shows flowinfo fields required. * we assume the ip header is in one mbuf, and addresses and ports are * in network byte order. */ int altq_extractflow(m, af, flow, filt_bmask) struct mbuf *m; int af; struct flowinfo *flow; u_int32_t filt_bmask; { switch (af) { case PF_INET: { struct flowinfo_in *fin; struct ip *ip; ip = mtod(m, struct ip *); if (ip->ip_v != 4) break; fin = (struct flowinfo_in *)flow; fin->fi_len = sizeof(struct flowinfo_in); fin->fi_family = AF_INET; fin->fi_proto = ip->ip_p; fin->fi_tos = ip->ip_tos; fin->fi_src.s_addr = ip->ip_src.s_addr; fin->fi_dst.s_addr = ip->ip_dst.s_addr; if (filt_bmask & FIMB4_PORTS) /* if port info is required, extract port numbers */ extract_ports4(m, ip, fin); else { fin->fi_sport = 0; fin->fi_dport = 0; fin->fi_gpi = 0; } return (1); } #ifdef INET6 case PF_INET6: { struct flowinfo_in6 *fin6; struct ip6_hdr *ip6; ip6 = mtod(m, struct ip6_hdr *); /* should we check the ip version? */ fin6 = (struct flowinfo_in6 *)flow; fin6->fi6_len = sizeof(struct flowinfo_in6); fin6->fi6_family = AF_INET6; fin6->fi6_proto = ip6->ip6_nxt; fin6->fi6_tclass = (ntohl(ip6->ip6_flow) >> 20) & 0xff; fin6->fi6_flowlabel = ip6->ip6_flow & htonl(0x000fffff); fin6->fi6_src = ip6->ip6_src; fin6->fi6_dst = ip6->ip6_dst; if ((filt_bmask & FIMB6_PORTS) || ((filt_bmask & FIMB6_PROTO) && ip6->ip6_nxt > IPPROTO_IPV6)) /* * if port info is required, or proto is required * but there are option headers, extract port * and protocol numbers. */ extract_ports6(m, ip6, fin6); else { fin6->fi6_sport = 0; fin6->fi6_dport = 0; fin6->fi6_gpi = 0; } return (1); } #endif /* INET6 */ default: break; } /* failed */ flow->fi_len = sizeof(struct flowinfo); flow->fi_family = AF_UNSPEC; return (0); } /* * helper routine to extract port numbers */ /* structure for ipsec and ipv6 option header template */ struct _opt6 { u_int8_t opt6_nxt; /* next header */ u_int8_t opt6_hlen; /* header extension length */ u_int16_t _pad; u_int32_t ah_spi; /* security parameter index for authentication header */ }; /* * extract port numbers from a ipv4 packet. */ static int extract_ports4(m, ip, fin) struct mbuf *m; struct ip *ip; struct flowinfo_in *fin; { struct mbuf *m0; u_short ip_off; u_int8_t proto; int off; fin->fi_sport = 0; fin->fi_dport = 0; fin->fi_gpi = 0; ip_off = ntohs(ip->ip_off); /* if it is a fragment, try cached fragment info */ if (ip_off & IP_OFFMASK) { ip4f_lookup(ip, fin); return (1); } /* locate the mbuf containing the protocol header */ for (m0 = m; m0 != NULL; m0 = m0->m_next) if (((caddr_t)ip >= m0->m_data) && ((caddr_t)ip < m0->m_data + m0->m_len)) break; if (m0 == NULL) { #ifdef ALTQ_DEBUG printf("extract_ports4: can't locate header! ip=%p\n", ip); #endif return (0); } off = ((caddr_t)ip - m0->m_data) + (ip->ip_hl << 2); proto = ip->ip_p; #ifdef ALTQ_IPSEC again: #endif while (off >= m0->m_len) { off -= m0->m_len; m0 = m0->m_next; if (m0 == NULL) return (0); /* bogus ip_hl! */ } if (m0->m_len < off + 4) return (0); switch (proto) { case IPPROTO_TCP: case IPPROTO_UDP: { struct udphdr *udp; udp = (struct udphdr *)(mtod(m0, caddr_t) + off); fin->fi_sport = udp->uh_sport; fin->fi_dport = udp->uh_dport; fin->fi_proto = proto; } break; #ifdef ALTQ_IPSEC case IPPROTO_ESP: if (fin->fi_gpi == 0){ u_int32_t *gpi; gpi = (u_int32_t *)(mtod(m0, caddr_t) + off); fin->fi_gpi = *gpi; } fin->fi_proto = proto; break; case IPPROTO_AH: { /* get next header and header length */ struct _opt6 *opt6; opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); proto = opt6->opt6_nxt; off += 8 + (opt6->opt6_hlen * 4); if (fin->fi_gpi == 0 && m0->m_len >= off + 8) fin->fi_gpi = opt6->ah_spi; } /* goto the next header */ goto again; #endif /* ALTQ_IPSEC */ default: fin->fi_proto = proto; return (0); } /* if this is a first fragment, cache it. */ if (ip_off & IP_MF) ip4f_cache(ip, fin); return (1); } #ifdef INET6 static int extract_ports6(m, ip6, fin6) struct mbuf *m; struct ip6_hdr *ip6; struct flowinfo_in6 *fin6; { struct mbuf *m0; int off; u_int8_t proto; fin6->fi6_gpi = 0; fin6->fi6_sport = 0; fin6->fi6_dport = 0; /* locate the mbuf containing the protocol header */ for (m0 = m; m0 != NULL; m0 = m0->m_next) if (((caddr_t)ip6 >= m0->m_data) && ((caddr_t)ip6 < m0->m_data + m0->m_len)) break; if (m0 == NULL) { #ifdef ALTQ_DEBUG printf("extract_ports6: can't locate header! ip6=%p\n", ip6); #endif return (0); } off = ((caddr_t)ip6 - m0->m_data) + sizeof(struct ip6_hdr); proto = ip6->ip6_nxt; do { while (off >= m0->m_len) { off -= m0->m_len; m0 = m0->m_next; if (m0 == NULL) return (0); } if (m0->m_len < off + 4) return (0); switch (proto) { case IPPROTO_TCP: case IPPROTO_UDP: { struct udphdr *udp; udp = (struct udphdr *)(mtod(m0, caddr_t) + off); fin6->fi6_sport = udp->uh_sport; fin6->fi6_dport = udp->uh_dport; fin6->fi6_proto = proto; } return (1); case IPPROTO_ESP: if (fin6->fi6_gpi == 0) { u_int32_t *gpi; gpi = (u_int32_t *)(mtod(m0, caddr_t) + off); fin6->fi6_gpi = *gpi; } fin6->fi6_proto = proto; return (1); case IPPROTO_AH: { /* get next header and header length */ struct _opt6 *opt6; opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); if (fin6->fi6_gpi == 0 && m0->m_len >= off + 8) fin6->fi6_gpi = opt6->ah_spi; proto = opt6->opt6_nxt; off += 8 + (opt6->opt6_hlen * 4); /* goto the next header */ break; } case IPPROTO_HOPOPTS: case IPPROTO_ROUTING: case IPPROTO_DSTOPTS: { /* get next header and header length */ struct _opt6 *opt6; opt6 = (struct _opt6 *)(mtod(m0, caddr_t) + off); proto = opt6->opt6_nxt; off += (opt6->opt6_hlen + 1) * 8; /* goto the next header */ break; } case IPPROTO_FRAGMENT: /* ipv6 fragmentations are not supported yet */ default: fin6->fi6_proto = proto; return (0); } } while (1); /*NOTREACHED*/ } #endif /* INET6 */ /* * altq common classifier */ int acc_add_filter(classifier, filter, class, phandle) struct acc_classifier *classifier; struct flow_filter *filter; void *class; u_long *phandle; { struct acc_filter *afp, *prev, *tmp; int i, s; #ifdef INET6 if (filter->ff_flow.fi_family != AF_INET && filter->ff_flow.fi_family != AF_INET6) return (EINVAL); #else if (filter->ff_flow.fi_family != AF_INET) return (EINVAL); #endif afp = malloc(sizeof(struct acc_filter), M_DEVBUF, M_WAITOK); if (afp == NULL) return (ENOMEM); bzero(afp, sizeof(struct acc_filter)); afp->f_filter = *filter; afp->f_class = class; i = ACC_WILDCARD_INDEX; if (filter->ff_flow.fi_family == AF_INET) { struct flow_filter *filter4 = &afp->f_filter; /* * if address is 0, it's a wildcard. if address mask * isn't set, use full mask. */ if (filter4->ff_flow.fi_dst.s_addr == 0) filter4->ff_mask.mask_dst.s_addr = 0; else if (filter4->ff_mask.mask_dst.s_addr == 0) filter4->ff_mask.mask_dst.s_addr = 0xffffffff; if (filter4->ff_flow.fi_src.s_addr == 0) filter4->ff_mask.mask_src.s_addr = 0; else if (filter4->ff_mask.mask_src.s_addr == 0) filter4->ff_mask.mask_src.s_addr = 0xffffffff; /* clear extra bits in addresses */ filter4->ff_flow.fi_dst.s_addr &= filter4->ff_mask.mask_dst.s_addr; filter4->ff_flow.fi_src.s_addr &= filter4->ff_mask.mask_src.s_addr; /* * if dst address is a wildcard, use hash-entry * ACC_WILDCARD_INDEX. */ if (filter4->ff_mask.mask_dst.s_addr != 0xffffffff) i = ACC_WILDCARD_INDEX; else i = ACC_GET_HASH_INDEX(filter4->ff_flow.fi_dst.s_addr); } #ifdef INET6 else if (filter->ff_flow.fi_family == AF_INET6) { struct flow_filter6 *filter6 = (struct flow_filter6 *)&afp->f_filter; #ifndef IN6MASK0 /* taken from kame ipv6 */ #define IN6MASK0 {{{ 0, 0, 0, 0 }}} #define IN6MASK128 {{{ 0xffffffff, 0xffffffff, 0xffffffff, 0xffffffff }}} const struct in6_addr in6mask0 = IN6MASK0; const struct in6_addr in6mask128 = IN6MASK128; #endif if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_dst)) filter6->ff_mask6.mask6_dst = in6mask0; else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_dst)) filter6->ff_mask6.mask6_dst = in6mask128; if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_flow6.fi6_src)) filter6->ff_mask6.mask6_src = in6mask0; else if (IN6_IS_ADDR_UNSPECIFIED(&filter6->ff_mask6.mask6_src)) filter6->ff_mask6.mask6_src = in6mask128; /* clear extra bits in addresses */ for (i = 0; i < 16; i++) filter6->ff_flow6.fi6_dst.s6_addr[i] &= filter6->ff_mask6.mask6_dst.s6_addr[i]; for (i = 0; i < 16; i++) filter6->ff_flow6.fi6_src.s6_addr[i] &= filter6->ff_mask6.mask6_src.s6_addr[i]; if (filter6->ff_flow6.fi6_flowlabel == 0) i = ACC_WILDCARD_INDEX; else i = ACC_GET_HASH_INDEX(filter6->ff_flow6.fi6_flowlabel); } #endif /* INET6 */ afp->f_handle = get_filt_handle(classifier, i); /* update filter bitmask */ afp->f_fbmask = filt2fibmask(filter); classifier->acc_fbmask |= afp->f_fbmask; /* * add this filter to the filter list. * filters are ordered from the highest rule number. */ s = splnet(); prev = NULL; LIST_FOREACH(tmp, &classifier->acc_filters[i], f_chain) { if (tmp->f_filter.ff_ruleno > afp->f_filter.ff_ruleno) prev = tmp; else break; } if (prev == NULL) LIST_INSERT_HEAD(&classifier->acc_filters[i], afp, f_chain); else LIST_INSERT_AFTER(prev, afp, f_chain); splx(s); *phandle = afp->f_handle; return (0); } int acc_delete_filter(classifier, handle) struct acc_classifier *classifier; u_long handle; { struct acc_filter *afp; int s; if ((afp = filth_to_filtp(classifier, handle)) == NULL) return (EINVAL); s = splnet(); LIST_REMOVE(afp, f_chain); splx(s); free(afp, M_DEVBUF); /* todo: update filt_bmask */ return (0); } /* * delete filters referencing to the specified class. * if the all flag is not 0, delete all the filters. */ int acc_discard_filters(classifier, class, all) struct acc_classifier *classifier; void *class; int all; { struct acc_filter *afp; int i, s; s = splnet(); for (i = 0; i < ACC_FILTER_TABLESIZE; i++) { do { LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) if (all || afp->f_class == class) { LIST_REMOVE(afp, f_chain); free(afp, M_DEVBUF); /* start again from the head */ break; } } while (afp != NULL); } splx(s); if (all) classifier->acc_fbmask = 0; return (0); } void * acc_classify(clfier, m, af) void *clfier; struct mbuf *m; int af; { struct acc_classifier *classifier; struct flowinfo flow; struct acc_filter *afp; int i; classifier = (struct acc_classifier *)clfier; altq_extractflow(m, af, &flow, classifier->acc_fbmask); if (flow.fi_family == AF_INET) { struct flowinfo_in *fp = (struct flowinfo_in *)&flow; if ((classifier->acc_fbmask & FIMB4_ALL) == FIMB4_TOS) { /* only tos is used */ LIST_FOREACH(afp, &classifier->acc_filters[ACC_WILDCARD_INDEX], f_chain) if (apply_tosfilter4(afp->f_fbmask, &afp->f_filter, fp)) /* filter matched */ return (afp->f_class); } else if ((classifier->acc_fbmask & (~(FIMB4_PROTO|FIMB4_SPORT|FIMB4_DPORT) & FIMB4_ALL)) == 0) { /* only proto and ports are used */ LIST_FOREACH(afp, &classifier->acc_filters[ACC_WILDCARD_INDEX], f_chain) if (apply_ppfilter4(afp->f_fbmask, &afp->f_filter, fp)) /* filter matched */ return (afp->f_class); } else { /* get the filter hash entry from its dest address */ i = ACC_GET_HASH_INDEX(fp->fi_dst.s_addr); do { /* * go through this loop twice. first for dst * hash, second for wildcards. */ LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) if (apply_filter4(afp->f_fbmask, &afp->f_filter, fp)) /* filter matched */ return (afp->f_class); /* * check again for filters with a dst addr * wildcard. * (daddr == 0 || dmask != 0xffffffff). */ if (i != ACC_WILDCARD_INDEX) i = ACC_WILDCARD_INDEX; else break; } while (1); } } #ifdef INET6 else if (flow.fi_family == AF_INET6) { struct flowinfo_in6 *fp6 = (struct flowinfo_in6 *)&flow; /* get the filter hash entry from its flow ID */ if (fp6->fi6_flowlabel != 0) i = ACC_GET_HASH_INDEX(fp6->fi6_flowlabel); else /* flowlable can be zero */ i = ACC_WILDCARD_INDEX; /* go through this loop twice. first for flow hash, second for wildcards. */ do { LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) if (apply_filter6(afp->f_fbmask, (struct flow_filter6 *)&afp->f_filter, fp6)) /* filter matched */ return (afp->f_class); /* * check again for filters with a wildcard. */ if (i != ACC_WILDCARD_INDEX) i = ACC_WILDCARD_INDEX; else break; } while (1); } #endif /* INET6 */ /* no filter matched */ return (NULL); } static int apply_filter4(fbmask, filt, pkt) u_int32_t fbmask; struct flow_filter *filt; struct flowinfo_in *pkt; { if (filt->ff_flow.fi_family != AF_INET) return (0); if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport) return (0); if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport) return (0); if ((fbmask & FIMB4_DADDR) && filt->ff_flow.fi_dst.s_addr != (pkt->fi_dst.s_addr & filt->ff_mask.mask_dst.s_addr)) return (0); if ((fbmask & FIMB4_SADDR) && filt->ff_flow.fi_src.s_addr != (pkt->fi_src.s_addr & filt->ff_mask.mask_src.s_addr)) return (0); if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto) return (0); if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos != (pkt->fi_tos & filt->ff_mask.mask_tos)) return (0); if ((fbmask & FIMB4_GPI) && filt->ff_flow.fi_gpi != (pkt->fi_gpi)) return (0); /* match */ return (1); } /* * filter matching function optimized for a common case that checks * only protocol and port numbers */ static int apply_ppfilter4(fbmask, filt, pkt) u_int32_t fbmask; struct flow_filter *filt; struct flowinfo_in *pkt; { if (filt->ff_flow.fi_family != AF_INET) return (0); if ((fbmask & FIMB4_SPORT) && filt->ff_flow.fi_sport != pkt->fi_sport) return (0); if ((fbmask & FIMB4_DPORT) && filt->ff_flow.fi_dport != pkt->fi_dport) return (0); if ((fbmask & FIMB4_PROTO) && filt->ff_flow.fi_proto != pkt->fi_proto) return (0); /* match */ return (1); } /* * filter matching function only for tos field. */ static int apply_tosfilter4(fbmask, filt, pkt) u_int32_t fbmask; struct flow_filter *filt; struct flowinfo_in *pkt; { if (filt->ff_flow.fi_family != AF_INET) return (0); if ((fbmask & FIMB4_TOS) && filt->ff_flow.fi_tos != (pkt->fi_tos & filt->ff_mask.mask_tos)) return (0); /* match */ return (1); } #ifdef INET6 static int apply_filter6(fbmask, filt, pkt) u_int32_t fbmask; struct flow_filter6 *filt; struct flowinfo_in6 *pkt; { int i; if (filt->ff_flow6.fi6_family != AF_INET6) return (0); if ((fbmask & FIMB6_FLABEL) && filt->ff_flow6.fi6_flowlabel != pkt->fi6_flowlabel) return (0); if ((fbmask & FIMB6_PROTO) && filt->ff_flow6.fi6_proto != pkt->fi6_proto) return (0); if ((fbmask & FIMB6_SPORT) && filt->ff_flow6.fi6_sport != pkt->fi6_sport) return (0); if ((fbmask & FIMB6_DPORT) && filt->ff_flow6.fi6_dport != pkt->fi6_dport) return (0); if (fbmask & FIMB6_SADDR) { for (i = 0; i < 4; i++) if (filt->ff_flow6.fi6_src.s6_addr32[i] != (pkt->fi6_src.s6_addr32[i] & filt->ff_mask6.mask6_src.s6_addr32[i])) return (0); } if (fbmask & FIMB6_DADDR) { for (i = 0; i < 4; i++) if (filt->ff_flow6.fi6_dst.s6_addr32[i] != (pkt->fi6_dst.s6_addr32[i] & filt->ff_mask6.mask6_dst.s6_addr32[i])) return (0); } if ((fbmask & FIMB6_TCLASS) && filt->ff_flow6.fi6_tclass != (pkt->fi6_tclass & filt->ff_mask6.mask6_tclass)) return (0); if ((fbmask & FIMB6_GPI) && filt->ff_flow6.fi6_gpi != pkt->fi6_gpi) return (0); /* match */ return (1); } #endif /* INET6 */ /* * filter handle: * bit 20-28: index to the filter hash table * bit 0-19: unique id in the hash bucket. */ static u_long get_filt_handle(classifier, i) struct acc_classifier *classifier; int i; { static u_long handle_number = 1; u_long handle; struct acc_filter *afp; while (1) { handle = handle_number++ & 0x000fffff; if (LIST_EMPTY(&classifier->acc_filters[i])) break; LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) if ((afp->f_handle & 0x000fffff) == handle) break; if (afp == NULL) break; /* this handle is already used, try again */ } return ((i << 20) | handle); } /* convert filter handle to filter pointer */ static struct acc_filter * filth_to_filtp(classifier, handle) struct acc_classifier *classifier; u_long handle; { struct acc_filter *afp; int i; i = ACC_GET_HINDEX(handle); LIST_FOREACH(afp, &classifier->acc_filters[i], f_chain) if (afp->f_handle == handle) return (afp); return (NULL); } /* create flowinfo bitmask */ static u_int32_t filt2fibmask(filt) struct flow_filter *filt; { u_int32_t mask = 0; #ifdef INET6 struct flow_filter6 *filt6; #endif switch (filt->ff_flow.fi_family) { case AF_INET: if (filt->ff_flow.fi_proto != 0) mask |= FIMB4_PROTO; if (filt->ff_flow.fi_tos != 0) mask |= FIMB4_TOS; if (filt->ff_flow.fi_dst.s_addr != 0) mask |= FIMB4_DADDR; if (filt->ff_flow.fi_src.s_addr != 0) mask |= FIMB4_SADDR; if (filt->ff_flow.fi_sport != 0) mask |= FIMB4_SPORT; if (filt->ff_flow.fi_dport != 0) mask |= FIMB4_DPORT; if (filt->ff_flow.fi_gpi != 0) mask |= FIMB4_GPI; break; #ifdef INET6 case AF_INET6: filt6 = (struct flow_filter6 *)filt; if (filt6->ff_flow6.fi6_proto != 0) mask |= FIMB6_PROTO; if (filt6->ff_flow6.fi6_tclass != 0) mask |= FIMB6_TCLASS; if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_dst)) mask |= FIMB6_DADDR; if (!IN6_IS_ADDR_UNSPECIFIED(&filt6->ff_flow6.fi6_src)) mask |= FIMB6_SADDR; if (filt6->ff_flow6.fi6_sport != 0) mask |= FIMB6_SPORT; if (filt6->ff_flow6.fi6_dport != 0) mask |= FIMB6_DPORT; if (filt6->ff_flow6.fi6_gpi != 0) mask |= FIMB6_GPI; if (filt6->ff_flow6.fi6_flowlabel != 0) mask |= FIMB6_FLABEL; break; #endif /* INET6 */ } return (mask); } /* * helper functions to handle IPv4 fragments. * currently only in-sequence fragments are handled. * - fragment info is cached in a LRU list. * - when a first fragment is found, cache its flow info. * - when a non-first fragment is found, lookup the cache. */ struct ip4_frag { TAILQ_ENTRY(ip4_frag) ip4f_chain; char ip4f_valid; u_short ip4f_id; struct flowinfo_in ip4f_info; }; static TAILQ_HEAD(ip4f_list, ip4_frag) ip4f_list; /* IPv4 fragment cache */ #define IP4F_TABSIZE 16 /* IPv4 fragment cache size */ static void ip4f_cache(ip, fin) struct ip *ip; struct flowinfo_in *fin; { struct ip4_frag *fp; if (TAILQ_EMPTY(&ip4f_list)) { /* first time call, allocate fragment cache entries. */ if (ip4f_init() < 0) /* allocation failed! */ return; } fp = ip4f_alloc(); fp->ip4f_id = ip->ip_id; fp->ip4f_info.fi_proto = ip->ip_p; fp->ip4f_info.fi_src.s_addr = ip->ip_src.s_addr; fp->ip4f_info.fi_dst.s_addr = ip->ip_dst.s_addr; /* save port numbers */ fp->ip4f_info.fi_sport = fin->fi_sport; fp->ip4f_info.fi_dport = fin->fi_dport; fp->ip4f_info.fi_gpi = fin->fi_gpi; } static int ip4f_lookup(ip, fin) struct ip *ip; struct flowinfo_in *fin; { struct ip4_frag *fp; for (fp = TAILQ_FIRST(&ip4f_list); fp != NULL && fp->ip4f_valid; fp = TAILQ_NEXT(fp, ip4f_chain)) if (ip->ip_id == fp->ip4f_id && ip->ip_src.s_addr == fp->ip4f_info.fi_src.s_addr && ip->ip_dst.s_addr == fp->ip4f_info.fi_dst.s_addr && ip->ip_p == fp->ip4f_info.fi_proto) { /* found the matching entry */ fin->fi_sport = fp->ip4f_info.fi_sport; fin->fi_dport = fp->ip4f_info.fi_dport; fin->fi_gpi = fp->ip4f_info.fi_gpi; if ((ntohs(ip->ip_off) & IP_MF) == 0) /* this is the last fragment, release the entry. */ ip4f_free(fp); return (1); } /* no matching entry found */ return (0); } static int ip4f_init(void) { struct ip4_frag *fp; int i; TAILQ_INIT(&ip4f_list); for (i=0; iip4f_valid = 0; TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain); } return (0); } static struct ip4_frag * ip4f_alloc(void) { struct ip4_frag *fp; /* reclaim an entry at the tail, put it at the head */ fp = TAILQ_LAST(&ip4f_list, ip4f_list); TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain); fp->ip4f_valid = 1; TAILQ_INSERT_HEAD(&ip4f_list, fp, ip4f_chain); return (fp); } static void ip4f_free(fp) struct ip4_frag *fp; { TAILQ_REMOVE(&ip4f_list, fp, ip4f_chain); fp->ip4f_valid = 0; TAILQ_INSERT_TAIL(&ip4f_list, fp, ip4f_chain); } #endif /* ALTQ3_CLFIER_COMPAT */ Index: head/sys/net/altq/altq_var.h =================================================================== --- head/sys/net/altq/altq_var.h (revision 343994) +++ head/sys/net/altq/altq_var.h (revision 343995) @@ -1,242 +1,242 @@ /*- * Copyright (C) 1998-2003 * Sony Computer Science Laboratories Inc. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $KAME: altq_var.h,v 1.16 2003/10/03 05:05:15 kjc Exp $ * $FreeBSD$ */ #ifndef _ALTQ_ALTQ_VAR_H_ #define _ALTQ_ALTQ_VAR_H_ #ifdef _KERNEL #include #include #include #ifdef ALTQ3_CLFIER_COMPAT /* * filter structure for altq common classifier */ struct acc_filter { LIST_ENTRY(acc_filter) f_chain; void *f_class; /* pointer to the class */ u_long f_handle; /* filter id */ u_int32_t f_fbmask; /* filter bitmask */ struct flow_filter f_filter; /* filter value */ }; /* * XXX ACC_FILTER_TABLESIZE can't be larger than 2048 unless we fix * the handle assignment. */ #define ACC_FILTER_TABLESIZE (256+1) #define ACC_FILTER_MASK (ACC_FILTER_TABLESIZE - 2) #define ACC_WILDCARD_INDEX (ACC_FILTER_TABLESIZE - 1) #ifdef __GNUC__ #define ACC_GET_HASH_INDEX(addr) \ ({int x = (addr) + ((addr) >> 16); (x + (x >> 8)) & ACC_FILTER_MASK;}) #else #define ACC_GET_HASH_INDEX(addr) \ (((addr) + ((addr) >> 8) + ((addr) >> 16) + ((addr) >> 24)) \ & ACC_FILTER_MASK) #endif #define ACC_GET_HINDEX(handle) ((handle) >> 20) #if (__FreeBSD_version > 500000) #define ACC_LOCK_INIT(ac) mtx_init(&(ac)->acc_mtx, "classifier", MTX_DEF) #define ACC_LOCK_DESTROY(ac) mtx_destroy(&(ac)->acc_mtx) #define ACC_LOCK(ac) mtx_lock(&(ac)->acc_mtx) #define ACC_UNLOCK(ac) mtx_unlock(&(ac)->acc_mtx) #else #define ACC_LOCK_INIT(ac) #define ACC_LOCK_DESTROY(ac) #define ACC_LOCK(ac) #define ACC_UNLOCK(ac) #endif struct acc_classifier { u_int32_t acc_fbmask; LIST_HEAD(filt, acc_filter) acc_filters[ACC_FILTER_TABLESIZE]; #if (__FreeBSD_version > 500000) struct mtx acc_mtx; #endif }; /* * flowinfo mask bits used by classifier */ /* for ipv4 */ #define FIMB4_PROTO 0x0001 #define FIMB4_TOS 0x0002 #define FIMB4_DADDR 0x0004 #define FIMB4_SADDR 0x0008 #define FIMB4_DPORT 0x0010 #define FIMB4_SPORT 0x0020 #define FIMB4_GPI 0x0040 #define FIMB4_ALL 0x007f /* for ipv6 */ #define FIMB6_PROTO 0x0100 #define FIMB6_TCLASS 0x0200 #define FIMB6_DADDR 0x0400 #define FIMB6_SADDR 0x0800 #define FIMB6_DPORT 0x1000 #define FIMB6_SPORT 0x2000 #define FIMB6_GPI 0x4000 #define FIMB6_FLABEL 0x8000 #define FIMB6_ALL 0xff00 #define FIMB_ALL (FIMB4_ALL|FIMB6_ALL) #define FIMB4_PORTS (FIMB4_DPORT|FIMB4_SPORT|FIMB4_GPI) #define FIMB6_PORTS (FIMB6_DPORT|FIMB6_SPORT|FIMB6_GPI) #endif /* ALTQ3_CLFIER_COMPAT */ /* * machine dependent clock * a 64bit high resolution time counter. */ extern int machclk_usepcc; extern u_int32_t machclk_freq; extern u_int32_t machclk_per_tick; extern void init_machclk(void); extern u_int64_t read_machclk(void); /* * debug support */ #ifdef ALTQ_DEBUG #ifdef __STDC__ #define ASSERT(e) ((e) ? (void)0 : altq_assert(__FILE__, __LINE__, #e)) #else /* PCC */ #define ASSERT(e) ((e) ? (void)0 : altq_assert(__FILE__, __LINE__, "e")) #endif #else #define ASSERT(e) ((void)0) #endif /* * misc stuff for compatibility */ /* ioctl cmd type */ typedef u_long ioctlcmd_t; /* * queue macros: * the interface of TAILQ_LAST macro changed after the introduction * of softupdate. redefine it here to make it work with pre-2.2.7. */ #undef TAILQ_LAST #define TAILQ_LAST(head, headname) \ (*(((struct headname *)((head)->tqh_last))->tqh_last)) #ifndef TAILQ_EMPTY #define TAILQ_EMPTY(head) ((head)->tqh_first == NULL) #endif #ifndef TAILQ_FOREACH #define TAILQ_FOREACH(var, head, field) \ for (var = TAILQ_FIRST(head); var; var = TAILQ_NEXT(var, field)) #endif /* macro for timeout/untimeout */ /* use callout */ #include #if (__FreeBSD_version > 500000) #define CALLOUT_INIT(c) callout_init((c), 0) #else #define CALLOUT_INIT(c) callout_init((c)) #endif #define CALLOUT_RESET(c,t,f,a) callout_reset((c),(t),(f),(a)) #define CALLOUT_STOP(c) callout_stop((c)) #if !defined(CALLOUT_INITIALIZER) && (__FreeBSD_version < 600000) #define CALLOUT_INITIALIZER { { { NULL } }, 0, NULL, NULL, 0 } #endif #define m_pktlen(m) ((m)->m_pkthdr.len) struct ifnet; struct mbuf; struct pf_altq; #ifdef ALTQ3_CLFIER_COMPAT struct flowinfo; #endif void *altq_lookup(char *, int); #ifdef ALTQ3_CLFIER_COMPAT int altq_extractflow(struct mbuf *, int, struct flowinfo *, u_int32_t); int acc_add_filter(struct acc_classifier *, struct flow_filter *, void *, u_long *); int acc_delete_filter(struct acc_classifier *, u_long); int acc_discard_filters(struct acc_classifier *, void *, int); void *acc_classify(void *, struct mbuf *, int); #endif u_int8_t read_dsfield(struct mbuf *, struct altq_pktattr *); void write_dsfield(struct mbuf *, struct altq_pktattr *, u_int8_t); void altq_assert(const char *, int, const char *); int tbr_set(struct ifaltq *, struct tb_profile *); int altq_pfattach(struct pf_altq *); int altq_pfdetach(struct pf_altq *); -int altq_add(struct pf_altq *); +int altq_add(struct ifnet *, struct pf_altq *); int altq_remove(struct pf_altq *); int altq_add_queue(struct pf_altq *); int altq_remove_queue(struct pf_altq *); int altq_getqstats(struct pf_altq *, void *, int *, int); int cbq_pfattach(struct pf_altq *); -int cbq_add_altq(struct pf_altq *); +int cbq_add_altq(struct ifnet *, struct pf_altq *); int cbq_remove_altq(struct pf_altq *); int cbq_add_queue(struct pf_altq *); int cbq_remove_queue(struct pf_altq *); int cbq_getqstats(struct pf_altq *, void *, int *, int); int codel_pfattach(struct pf_altq *); -int codel_add_altq(struct pf_altq *); +int codel_add_altq(struct ifnet *, struct pf_altq *); int codel_remove_altq(struct pf_altq *); int codel_getqstats(struct pf_altq *, void *, int *, int); int priq_pfattach(struct pf_altq *); -int priq_add_altq(struct pf_altq *); +int priq_add_altq(struct ifnet *, struct pf_altq *); int priq_remove_altq(struct pf_altq *); int priq_add_queue(struct pf_altq *); int priq_remove_queue(struct pf_altq *); int priq_getqstats(struct pf_altq *, void *, int *, int); int hfsc_pfattach(struct pf_altq *); -int hfsc_add_altq(struct pf_altq *); +int hfsc_add_altq(struct ifnet *, struct pf_altq *); int hfsc_remove_altq(struct pf_altq *); int hfsc_add_queue(struct pf_altq *); int hfsc_remove_queue(struct pf_altq *); int hfsc_getqstats(struct pf_altq *, void *, int *, int); int fairq_pfattach(struct pf_altq *); -int fairq_add_altq(struct pf_altq *); +int fairq_add_altq(struct ifnet *, struct pf_altq *); int fairq_remove_altq(struct pf_altq *); int fairq_add_queue(struct pf_altq *); int fairq_remove_queue(struct pf_altq *); int fairq_getqstats(struct pf_altq *, void *, int *, int); #endif /* _KERNEL */ #endif /* _ALTQ_ALTQ_VAR_H_ */ Index: head/sys/net/pfvar.h =================================================================== --- head/sys/net/pfvar.h (revision 343994) +++ head/sys/net/pfvar.h (revision 343995) @@ -1,1881 +1,1889 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * $OpenBSD: pfvar.h,v 1.282 2009/01/29 15:12:28 pyr Exp $ * $FreeBSD$ */ #ifndef _NET_PFVAR_H_ #define _NET_PFVAR_H_ #include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include struct pf_addr { union { struct in_addr v4; struct in6_addr v6; u_int8_t addr8[16]; u_int16_t addr16[8]; u_int32_t addr32[4]; } pfa; /* 128-bit address */ #define v4 pfa.v4 #define v6 pfa.v6 #define addr8 pfa.addr8 #define addr16 pfa.addr16 #define addr32 pfa.addr32 }; #define PFI_AFLAG_NETWORK 0x01 #define PFI_AFLAG_BROADCAST 0x02 #define PFI_AFLAG_PEER 0x04 #define PFI_AFLAG_MODEMASK 0x07 #define PFI_AFLAG_NOALIAS 0x08 struct pf_addr_wrap { union { struct { struct pf_addr addr; struct pf_addr mask; } a; char ifname[IFNAMSIZ]; char tblname[PF_TABLE_NAME_SIZE]; } v; union { struct pfi_dynaddr *dyn; struct pfr_ktable *tbl; int dyncnt; int tblcnt; } p; u_int8_t type; /* PF_ADDR_* */ u_int8_t iflags; /* PFI_AFLAG_* */ }; #ifdef _KERNEL +SYSCTL_DECL(_net_pf); +MALLOC_DECLARE(M_PFHASH); + struct pfi_dynaddr { TAILQ_ENTRY(pfi_dynaddr) entry; struct pf_addr pfid_addr4; struct pf_addr pfid_mask4; struct pf_addr pfid_addr6; struct pf_addr pfid_mask6; struct pfr_ktable *pfid_kt; struct pfi_kif *pfid_kif; int pfid_net; /* mask or 128 */ int pfid_acnt4; /* address count IPv4 */ int pfid_acnt6; /* address count IPv6 */ sa_family_t pfid_af; /* rule af */ u_int8_t pfid_iflags; /* PFI_AFLAG_* */ }; /* * Address manipulation macros */ #define HTONL(x) (x) = htonl((__uint32_t)(x)) #define HTONS(x) (x) = htons((__uint16_t)(x)) #define NTOHL(x) (x) = ntohl((__uint32_t)(x)) #define NTOHS(x) (x) = ntohs((__uint16_t)(x)) #define PF_NAME "pf" #define PF_HASHROW_ASSERT(h) mtx_assert(&(h)->lock, MA_OWNED) #define PF_HASHROW_LOCK(h) mtx_lock(&(h)->lock) #define PF_HASHROW_UNLOCK(h) mtx_unlock(&(h)->lock) #define PF_STATE_LOCK(s) \ do { \ struct pf_idhash *_ih = &V_pf_idhash[PF_IDHASH(s)]; \ PF_HASHROW_LOCK(_ih); \ } while (0) #define PF_STATE_UNLOCK(s) \ do { \ struct pf_idhash *_ih = &V_pf_idhash[PF_IDHASH((s))]; \ PF_HASHROW_UNLOCK(_ih); \ } while (0) #ifdef INVARIANTS #define PF_STATE_LOCK_ASSERT(s) \ do { \ struct pf_idhash *_ih = &V_pf_idhash[PF_IDHASH(s)]; \ PF_HASHROW_ASSERT(_ih); \ } while (0) #else /* !INVARIANTS */ #define PF_STATE_LOCK_ASSERT(s) do {} while (0) #endif /* INVARIANTS */ extern struct mtx pf_unlnkdrules_mtx; #define PF_UNLNKDRULES_LOCK() mtx_lock(&pf_unlnkdrules_mtx) #define PF_UNLNKDRULES_UNLOCK() mtx_unlock(&pf_unlnkdrules_mtx) extern struct rmlock pf_rules_lock; #define PF_RULES_RLOCK_TRACKER struct rm_priotracker _pf_rules_tracker #define PF_RULES_RLOCK() rm_rlock(&pf_rules_lock, &_pf_rules_tracker) #define PF_RULES_RUNLOCK() rm_runlock(&pf_rules_lock, &_pf_rules_tracker) #define PF_RULES_WLOCK() rm_wlock(&pf_rules_lock) #define PF_RULES_WUNLOCK() rm_wunlock(&pf_rules_lock) #define PF_RULES_ASSERT() rm_assert(&pf_rules_lock, RA_LOCKED) #define PF_RULES_RASSERT() rm_assert(&pf_rules_lock, RA_RLOCKED) #define PF_RULES_WASSERT() rm_assert(&pf_rules_lock, RA_WLOCKED) extern struct sx pf_end_lock; #define PF_MODVER 1 #define PFLOG_MODVER 1 #define PFSYNC_MODVER 1 #define PFLOG_MINVER 1 #define PFLOG_PREFVER PFLOG_MODVER #define PFLOG_MAXVER 1 #define PFSYNC_MINVER 1 #define PFSYNC_PREFVER PFSYNC_MODVER #define PFSYNC_MAXVER 1 #ifdef INET #ifndef INET6 #define PF_INET_ONLY #endif /* ! INET6 */ #endif /* INET */ #ifdef INET6 #ifndef INET #define PF_INET6_ONLY #endif /* ! INET */ #endif /* INET6 */ #ifdef INET #ifdef INET6 #define PF_INET_INET6 #endif /* INET6 */ #endif /* INET */ #else #define PF_INET_INET6 #endif /* _KERNEL */ /* Both IPv4 and IPv6 */ #ifdef PF_INET_INET6 #define PF_AEQ(a, b, c) \ ((c == AF_INET && (a)->addr32[0] == (b)->addr32[0]) || \ (c == AF_INET6 && (a)->addr32[3] == (b)->addr32[3] && \ (a)->addr32[2] == (b)->addr32[2] && \ (a)->addr32[1] == (b)->addr32[1] && \ (a)->addr32[0] == (b)->addr32[0])) \ #define PF_ANEQ(a, b, c) \ ((c == AF_INET && (a)->addr32[0] != (b)->addr32[0]) || \ (c == AF_INET6 && ((a)->addr32[0] != (b)->addr32[0] || \ (a)->addr32[1] != (b)->addr32[1] || \ (a)->addr32[2] != (b)->addr32[2] || \ (a)->addr32[3] != (b)->addr32[3]))) \ #define PF_AZERO(a, c) \ ((c == AF_INET && !(a)->addr32[0]) || \ (c == AF_INET6 && !(a)->addr32[0] && !(a)->addr32[1] && \ !(a)->addr32[2] && !(a)->addr32[3] )) \ #define PF_MATCHA(n, a, m, b, f) \ pf_match_addr(n, a, m, b, f) #define PF_ACPY(a, b, f) \ pf_addrcpy(a, b, f) #define PF_AINC(a, f) \ pf_addr_inc(a, f) #define PF_POOLMASK(a, b, c, d, f) \ pf_poolmask(a, b, c, d, f) #else /* Just IPv6 */ #ifdef PF_INET6_ONLY #define PF_AEQ(a, b, c) \ ((a)->addr32[3] == (b)->addr32[3] && \ (a)->addr32[2] == (b)->addr32[2] && \ (a)->addr32[1] == (b)->addr32[1] && \ (a)->addr32[0] == (b)->addr32[0]) \ #define PF_ANEQ(a, b, c) \ ((a)->addr32[3] != (b)->addr32[3] || \ (a)->addr32[2] != (b)->addr32[2] || \ (a)->addr32[1] != (b)->addr32[1] || \ (a)->addr32[0] != (b)->addr32[0]) \ #define PF_AZERO(a, c) \ (!(a)->addr32[0] && \ !(a)->addr32[1] && \ !(a)->addr32[2] && \ !(a)->addr32[3] ) \ #define PF_MATCHA(n, a, m, b, f) \ pf_match_addr(n, a, m, b, f) #define PF_ACPY(a, b, f) \ pf_addrcpy(a, b, f) #define PF_AINC(a, f) \ pf_addr_inc(a, f) #define PF_POOLMASK(a, b, c, d, f) \ pf_poolmask(a, b, c, d, f) #else /* Just IPv4 */ #ifdef PF_INET_ONLY #define PF_AEQ(a, b, c) \ ((a)->addr32[0] == (b)->addr32[0]) #define PF_ANEQ(a, b, c) \ ((a)->addr32[0] != (b)->addr32[0]) #define PF_AZERO(a, c) \ (!(a)->addr32[0]) #define PF_MATCHA(n, a, m, b, f) \ pf_match_addr(n, a, m, b, f) #define PF_ACPY(a, b, f) \ (a)->v4.s_addr = (b)->v4.s_addr #define PF_AINC(a, f) \ do { \ (a)->addr32[0] = htonl(ntohl((a)->addr32[0]) + 1); \ } while (0) #define PF_POOLMASK(a, b, c, d, f) \ do { \ (a)->addr32[0] = ((b)->addr32[0] & (c)->addr32[0]) | \ (((c)->addr32[0] ^ 0xffffffff ) & (d)->addr32[0]); \ } while (0) #endif /* PF_INET_ONLY */ #endif /* PF_INET6_ONLY */ #endif /* PF_INET_INET6 */ /* * XXX callers not FIB-aware in our version of pf yet. * OpenBSD fixed it later it seems, 2010/05/07 13:33:16 claudio. */ #define PF_MISMATCHAW(aw, x, af, neg, ifp, rtid) \ ( \ (((aw)->type == PF_ADDR_NOROUTE && \ pf_routable((x), (af), NULL, (rtid))) || \ (((aw)->type == PF_ADDR_URPFFAILED && (ifp) != NULL && \ pf_routable((x), (af), (ifp), (rtid))) || \ ((aw)->type == PF_ADDR_TABLE && \ !pfr_match_addr((aw)->p.tbl, (x), (af))) || \ ((aw)->type == PF_ADDR_DYNIFTL && \ !pfi_match_addr((aw)->p.dyn, (x), (af))) || \ ((aw)->type == PF_ADDR_RANGE && \ !pf_match_addr_range(&(aw)->v.a.addr, \ &(aw)->v.a.mask, (x), (af))) || \ ((aw)->type == PF_ADDR_ADDRMASK && \ !PF_AZERO(&(aw)->v.a.mask, (af)) && \ !PF_MATCHA(0, &(aw)->v.a.addr, \ &(aw)->v.a.mask, (x), (af))))) != \ (neg) \ ) struct pf_rule_uid { uid_t uid[2]; u_int8_t op; }; struct pf_rule_gid { uid_t gid[2]; u_int8_t op; }; struct pf_rule_addr { struct pf_addr_wrap addr; u_int16_t port[2]; u_int8_t neg; u_int8_t port_op; }; struct pf_pooladdr { struct pf_addr_wrap addr; TAILQ_ENTRY(pf_pooladdr) entries; char ifname[IFNAMSIZ]; struct pfi_kif *kif; }; TAILQ_HEAD(pf_palist, pf_pooladdr); struct pf_poolhashkey { union { u_int8_t key8[16]; u_int16_t key16[8]; u_int32_t key32[4]; } pfk; /* 128-bit hash key */ #define key8 pfk.key8 #define key16 pfk.key16 #define key32 pfk.key32 }; struct pf_pool { struct pf_palist list; struct pf_pooladdr *cur; struct pf_poolhashkey key; struct pf_addr counter; int tblidx; u_int16_t proxy_port[2]; u_int8_t opts; }; /* A packed Operating System description for fingerprinting */ typedef u_int32_t pf_osfp_t; #define PF_OSFP_ANY ((pf_osfp_t)0) #define PF_OSFP_UNKNOWN ((pf_osfp_t)-1) #define PF_OSFP_NOMATCH ((pf_osfp_t)-2) struct pf_osfp_entry { SLIST_ENTRY(pf_osfp_entry) fp_entry; pf_osfp_t fp_os; int fp_enflags; #define PF_OSFP_EXPANDED 0x001 /* expanded entry */ #define PF_OSFP_GENERIC 0x002 /* generic signature */ #define PF_OSFP_NODETAIL 0x004 /* no p0f details */ #define PF_OSFP_LEN 32 char fp_class_nm[PF_OSFP_LEN]; char fp_version_nm[PF_OSFP_LEN]; char fp_subtype_nm[PF_OSFP_LEN]; }; #define PF_OSFP_ENTRY_EQ(a, b) \ ((a)->fp_os == (b)->fp_os && \ memcmp((a)->fp_class_nm, (b)->fp_class_nm, PF_OSFP_LEN) == 0 && \ memcmp((a)->fp_version_nm, (b)->fp_version_nm, PF_OSFP_LEN) == 0 && \ memcmp((a)->fp_subtype_nm, (b)->fp_subtype_nm, PF_OSFP_LEN) == 0) /* handle pf_osfp_t packing */ #define _FP_RESERVED_BIT 1 /* For the special negative #defines */ #define _FP_UNUSED_BITS 1 #define _FP_CLASS_BITS 10 /* OS Class (Windows, Linux) */ #define _FP_VERSION_BITS 10 /* OS version (95, 98, NT, 2.4.54, 3.2) */ #define _FP_SUBTYPE_BITS 10 /* patch level (NT SP4, SP3, ECN patch) */ #define PF_OSFP_UNPACK(osfp, class, version, subtype) do { \ (class) = ((osfp) >> (_FP_VERSION_BITS+_FP_SUBTYPE_BITS)) & \ ((1 << _FP_CLASS_BITS) - 1); \ (version) = ((osfp) >> _FP_SUBTYPE_BITS) & \ ((1 << _FP_VERSION_BITS) - 1);\ (subtype) = (osfp) & ((1 << _FP_SUBTYPE_BITS) - 1); \ } while(0) #define PF_OSFP_PACK(osfp, class, version, subtype) do { \ (osfp) = ((class) & ((1 << _FP_CLASS_BITS) - 1)) << (_FP_VERSION_BITS \ + _FP_SUBTYPE_BITS); \ (osfp) |= ((version) & ((1 << _FP_VERSION_BITS) - 1)) << \ _FP_SUBTYPE_BITS; \ (osfp) |= (subtype) & ((1 << _FP_SUBTYPE_BITS) - 1); \ } while(0) /* the fingerprint of an OSes TCP SYN packet */ typedef u_int64_t pf_tcpopts_t; struct pf_os_fingerprint { SLIST_HEAD(pf_osfp_enlist, pf_osfp_entry) fp_oses; /* list of matches */ pf_tcpopts_t fp_tcpopts; /* packed TCP options */ u_int16_t fp_wsize; /* TCP window size */ u_int16_t fp_psize; /* ip->ip_len */ u_int16_t fp_mss; /* TCP MSS */ u_int16_t fp_flags; #define PF_OSFP_WSIZE_MOD 0x0001 /* Window modulus */ #define PF_OSFP_WSIZE_DC 0x0002 /* Window don't care */ #define PF_OSFP_WSIZE_MSS 0x0004 /* Window multiple of MSS */ #define PF_OSFP_WSIZE_MTU 0x0008 /* Window multiple of MTU */ #define PF_OSFP_PSIZE_MOD 0x0010 /* packet size modulus */ #define PF_OSFP_PSIZE_DC 0x0020 /* packet size don't care */ #define PF_OSFP_WSCALE 0x0040 /* TCP window scaling */ #define PF_OSFP_WSCALE_MOD 0x0080 /* TCP window scale modulus */ #define PF_OSFP_WSCALE_DC 0x0100 /* TCP window scale dont-care */ #define PF_OSFP_MSS 0x0200 /* TCP MSS */ #define PF_OSFP_MSS_MOD 0x0400 /* TCP MSS modulus */ #define PF_OSFP_MSS_DC 0x0800 /* TCP MSS dont-care */ #define PF_OSFP_DF 0x1000 /* IPv4 don't fragment bit */ #define PF_OSFP_TS0 0x2000 /* Zero timestamp */ #define PF_OSFP_INET6 0x4000 /* IPv6 */ u_int8_t fp_optcnt; /* TCP option count */ u_int8_t fp_wscale; /* TCP window scaling */ u_int8_t fp_ttl; /* IPv4 TTL */ #define PF_OSFP_MAXTTL_OFFSET 40 /* TCP options packing */ #define PF_OSFP_TCPOPT_NOP 0x0 /* TCP NOP option */ #define PF_OSFP_TCPOPT_WSCALE 0x1 /* TCP window scaling option */ #define PF_OSFP_TCPOPT_MSS 0x2 /* TCP max segment size opt */ #define PF_OSFP_TCPOPT_SACK 0x3 /* TCP SACK OK option */ #define PF_OSFP_TCPOPT_TS 0x4 /* TCP timestamp option */ #define PF_OSFP_TCPOPT_BITS 3 /* bits used by each option */ #define PF_OSFP_MAX_OPTS \ (sizeof(((struct pf_os_fingerprint *)0)->fp_tcpopts) * 8) \ / PF_OSFP_TCPOPT_BITS SLIST_ENTRY(pf_os_fingerprint) fp_next; }; struct pf_osfp_ioctl { struct pf_osfp_entry fp_os; pf_tcpopts_t fp_tcpopts; /* packed TCP options */ u_int16_t fp_wsize; /* TCP window size */ u_int16_t fp_psize; /* ip->ip_len */ u_int16_t fp_mss; /* TCP MSS */ u_int16_t fp_flags; u_int8_t fp_optcnt; /* TCP option count */ u_int8_t fp_wscale; /* TCP window scaling */ u_int8_t fp_ttl; /* IPv4 TTL */ int fp_getnum; /* DIOCOSFPGET number */ }; union pf_rule_ptr { struct pf_rule *ptr; u_int32_t nr; }; #define PF_ANCHOR_NAME_SIZE 64 struct pf_rule { struct pf_rule_addr src; struct pf_rule_addr dst; #define PF_SKIP_IFP 0 #define PF_SKIP_DIR 1 #define PF_SKIP_AF 2 #define PF_SKIP_PROTO 3 #define PF_SKIP_SRC_ADDR 4 #define PF_SKIP_SRC_PORT 5 #define PF_SKIP_DST_ADDR 6 #define PF_SKIP_DST_PORT 7 #define PF_SKIP_COUNT 8 union pf_rule_ptr skip[PF_SKIP_COUNT]; #define PF_RULE_LABEL_SIZE 64 char label[PF_RULE_LABEL_SIZE]; char ifname[IFNAMSIZ]; char qname[PF_QNAME_SIZE]; char pqname[PF_QNAME_SIZE]; #define PF_TAG_NAME_SIZE 64 char tagname[PF_TAG_NAME_SIZE]; char match_tagname[PF_TAG_NAME_SIZE]; char overload_tblname[PF_TABLE_NAME_SIZE]; TAILQ_ENTRY(pf_rule) entries; struct pf_pool rpool; u_int64_t evaluations; u_int64_t packets[2]; u_int64_t bytes[2]; struct pfi_kif *kif; struct pf_anchor *anchor; struct pfr_ktable *overload_tbl; pf_osfp_t os_fingerprint; int rtableid; u_int32_t timeout[PFTM_MAX]; u_int32_t max_states; u_int32_t max_src_nodes; u_int32_t max_src_states; u_int32_t max_src_conn; struct { u_int32_t limit; u_int32_t seconds; } max_src_conn_rate; u_int32_t qid; u_int32_t pqid; u_int32_t rt_listid; u_int32_t nr; u_int32_t prob; uid_t cuid; pid_t cpid; counter_u64_t states_cur; counter_u64_t states_tot; counter_u64_t src_nodes; u_int16_t return_icmp; u_int16_t return_icmp6; u_int16_t max_mss; u_int16_t tag; u_int16_t match_tag; u_int16_t scrub_flags; struct pf_rule_uid uid; struct pf_rule_gid gid; u_int32_t rule_flag; u_int8_t action; u_int8_t direction; u_int8_t log; u_int8_t logif; u_int8_t quick; u_int8_t ifnot; u_int8_t match_tag_not; u_int8_t natpass; #define PF_STATE_NORMAL 0x1 #define PF_STATE_MODULATE 0x2 #define PF_STATE_SYNPROXY 0x3 u_int8_t keep_state; sa_family_t af; u_int8_t proto; u_int8_t type; u_int8_t code; u_int8_t flags; u_int8_t flagset; u_int8_t min_ttl; u_int8_t allow_opts; u_int8_t rt; u_int8_t return_ttl; u_int8_t tos; u_int8_t set_tos; u_int8_t anchor_relative; u_int8_t anchor_wildcard; #define PF_FLUSH 0x01 #define PF_FLUSH_GLOBAL 0x02 u_int8_t flush; #define PF_PRIO_ZERO 0xff /* match "prio 0" packets */ #define PF_PRIO_MAX 7 u_int8_t prio; u_int8_t set_prio[2]; struct { struct pf_addr addr; u_int16_t port; } divert; uint64_t u_states_cur; uint64_t u_states_tot; uint64_t u_src_nodes; }; /* rule flags */ #define PFRULE_DROP 0x0000 #define PFRULE_RETURNRST 0x0001 #define PFRULE_FRAGMENT 0x0002 #define PFRULE_RETURNICMP 0x0004 #define PFRULE_RETURN 0x0008 #define PFRULE_NOSYNC 0x0010 #define PFRULE_SRCTRACK 0x0020 /* track source states */ #define PFRULE_RULESRCTRACK 0x0040 /* per rule */ #define PFRULE_REFS 0x0080 /* rule has references */ /* scrub flags */ #define PFRULE_NODF 0x0100 #define PFRULE_RANDOMID 0x0800 #define PFRULE_REASSEMBLE_TCP 0x1000 #define PFRULE_SET_TOS 0x2000 /* rule flags again */ #define PFRULE_IFBOUND 0x00010000 /* if-bound */ #define PFRULE_STATESLOPPY 0x00020000 /* sloppy state tracking */ #define PFSTATE_HIWAT 100000 /* default state table size */ #define PFSTATE_ADAPT_START 60000 /* default adaptive timeout start */ #define PFSTATE_ADAPT_END 120000 /* default adaptive timeout end */ struct pf_threshold { u_int32_t limit; #define PF_THRESHOLD_MULT 1000 #define PF_THRESHOLD_MAX 0xffffffff / PF_THRESHOLD_MULT u_int32_t seconds; u_int32_t count; u_int32_t last; }; struct pf_src_node { LIST_ENTRY(pf_src_node) entry; struct pf_addr addr; struct pf_addr raddr; union pf_rule_ptr rule; struct pfi_kif *kif; u_int64_t bytes[2]; u_int64_t packets[2]; u_int32_t states; u_int32_t conn; struct pf_threshold conn_rate; u_int32_t creation; u_int32_t expire; sa_family_t af; u_int8_t ruletype; }; #define PFSNODE_HIWAT 10000 /* default source node table size */ struct pf_state_scrub { struct timeval pfss_last; /* time received last packet */ u_int32_t pfss_tsecr; /* last echoed timestamp */ u_int32_t pfss_tsval; /* largest timestamp */ u_int32_t pfss_tsval0; /* original timestamp */ u_int16_t pfss_flags; #define PFSS_TIMESTAMP 0x0001 /* modulate timestamp */ #define PFSS_PAWS 0x0010 /* stricter PAWS checks */ #define PFSS_PAWS_IDLED 0x0020 /* was idle too long. no PAWS */ #define PFSS_DATA_TS 0x0040 /* timestamp on data packets */ #define PFSS_DATA_NOTS 0x0080 /* no timestamp on data packets */ u_int8_t pfss_ttl; /* stashed TTL */ u_int8_t pad; u_int32_t pfss_ts_mod; /* timestamp modulation */ }; struct pf_state_host { struct pf_addr addr; u_int16_t port; u_int16_t pad; }; struct pf_state_peer { struct pf_state_scrub *scrub; /* state is scrubbed */ u_int32_t seqlo; /* Max sequence number sent */ u_int32_t seqhi; /* Max the other end ACKd + win */ u_int32_t seqdiff; /* Sequence number modulator */ u_int16_t max_win; /* largest window (pre scaling) */ u_int16_t mss; /* Maximum segment size option */ u_int8_t state; /* active state level */ u_int8_t wscale; /* window scaling factor */ u_int8_t tcp_est; /* Did we reach TCPS_ESTABLISHED */ u_int8_t pad[1]; }; /* Keep synced with struct pf_state_key. */ struct pf_state_key_cmp { struct pf_addr addr[2]; u_int16_t port[2]; sa_family_t af; u_int8_t proto; u_int8_t pad[2]; }; struct pf_state_key { struct pf_addr addr[2]; u_int16_t port[2]; sa_family_t af; u_int8_t proto; u_int8_t pad[2]; LIST_ENTRY(pf_state_key) entry; TAILQ_HEAD(, pf_state) states[2]; }; /* Keep synced with struct pf_state. */ struct pf_state_cmp { u_int64_t id; u_int32_t creatorid; u_int8_t direction; u_int8_t pad[3]; }; struct pf_state { u_int64_t id; u_int32_t creatorid; u_int8_t direction; u_int8_t pad[3]; u_int refs; TAILQ_ENTRY(pf_state) sync_list; TAILQ_ENTRY(pf_state) key_list[2]; LIST_ENTRY(pf_state) entry; struct pf_state_peer src; struct pf_state_peer dst; union pf_rule_ptr rule; union pf_rule_ptr anchor; union pf_rule_ptr nat_rule; struct pf_addr rt_addr; struct pf_state_key *key[2]; /* addresses stack and wire */ struct pfi_kif *kif; struct pfi_kif *rt_kif; struct pf_src_node *src_node; struct pf_src_node *nat_src_node; u_int64_t packets[2]; u_int64_t bytes[2]; u_int32_t creation; u_int32_t expire; u_int32_t pfsync_time; u_int16_t tag; u_int8_t log; u_int8_t state_flags; #define PFSTATE_ALLOWOPTS 0x01 #define PFSTATE_SLOPPY 0x02 /* was PFSTATE_PFLOW 0x04 */ #define PFSTATE_NOSYNC 0x08 #define PFSTATE_ACK 0x10 #define PFSTATE_SETPRIO 0x0200 #define PFSTATE_SETMASK (PFSTATE_SETPRIO) u_int8_t timeout; u_int8_t sync_state; /* PFSYNC_S_x */ /* XXX */ u_int8_t sync_updates; u_int8_t _tail[3]; }; /* * Unified state structures for pulling states out of the kernel * used by pfsync(4) and the pf(4) ioctl. */ struct pfsync_state_scrub { u_int16_t pfss_flags; u_int8_t pfss_ttl; /* stashed TTL */ #define PFSYNC_SCRUB_FLAG_VALID 0x01 u_int8_t scrub_flag; u_int32_t pfss_ts_mod; /* timestamp modulation */ } __packed; struct pfsync_state_peer { struct pfsync_state_scrub scrub; /* state is scrubbed */ u_int32_t seqlo; /* Max sequence number sent */ u_int32_t seqhi; /* Max the other end ACKd + win */ u_int32_t seqdiff; /* Sequence number modulator */ u_int16_t max_win; /* largest window (pre scaling) */ u_int16_t mss; /* Maximum segment size option */ u_int8_t state; /* active state level */ u_int8_t wscale; /* window scaling factor */ u_int8_t pad[6]; } __packed; struct pfsync_state_key { struct pf_addr addr[2]; u_int16_t port[2]; }; struct pfsync_state { u_int64_t id; char ifname[IFNAMSIZ]; struct pfsync_state_key key[2]; struct pfsync_state_peer src; struct pfsync_state_peer dst; struct pf_addr rt_addr; u_int32_t rule; u_int32_t anchor; u_int32_t nat_rule; u_int32_t creation; u_int32_t expire; u_int32_t packets[2][2]; u_int32_t bytes[2][2]; u_int32_t creatorid; sa_family_t af; u_int8_t proto; u_int8_t direction; u_int8_t __spare[2]; u_int8_t log; u_int8_t state_flags; u_int8_t timeout; u_int8_t sync_flags; u_int8_t updates; } __packed; #ifdef _KERNEL /* pfsync */ typedef int pfsync_state_import_t(struct pfsync_state *, u_int8_t); typedef void pfsync_insert_state_t(struct pf_state *); typedef void pfsync_update_state_t(struct pf_state *); typedef void pfsync_delete_state_t(struct pf_state *); typedef void pfsync_clear_states_t(u_int32_t, const char *); typedef int pfsync_defer_t(struct pf_state *, struct mbuf *); typedef void pfsync_detach_ifnet_t(struct ifnet *); VNET_DECLARE(pfsync_state_import_t *, pfsync_state_import_ptr); #define V_pfsync_state_import_ptr VNET(pfsync_state_import_ptr) VNET_DECLARE(pfsync_insert_state_t *, pfsync_insert_state_ptr); #define V_pfsync_insert_state_ptr VNET(pfsync_insert_state_ptr) VNET_DECLARE(pfsync_update_state_t *, pfsync_update_state_ptr); #define V_pfsync_update_state_ptr VNET(pfsync_update_state_ptr) VNET_DECLARE(pfsync_delete_state_t *, pfsync_delete_state_ptr); #define V_pfsync_delete_state_ptr VNET(pfsync_delete_state_ptr) VNET_DECLARE(pfsync_clear_states_t *, pfsync_clear_states_ptr); #define V_pfsync_clear_states_ptr VNET(pfsync_clear_states_ptr) VNET_DECLARE(pfsync_defer_t *, pfsync_defer_ptr); #define V_pfsync_defer_ptr VNET(pfsync_defer_ptr) extern pfsync_detach_ifnet_t *pfsync_detach_ifnet_ptr; void pfsync_state_export(struct pfsync_state *, struct pf_state *); /* pflog */ struct pf_ruleset; struct pf_pdesc; typedef int pflog_packet_t(struct pfi_kif *, struct mbuf *, sa_family_t, u_int8_t, u_int8_t, struct pf_rule *, struct pf_rule *, struct pf_ruleset *, struct pf_pdesc *, int); extern pflog_packet_t *pflog_packet_ptr; #endif /* _KERNEL */ #define PFSYNC_FLAG_SRCNODE 0x04 #define PFSYNC_FLAG_NATSRCNODE 0x08 /* for copies to/from network byte order */ /* ioctl interface also uses network byte order */ #define pf_state_peer_hton(s,d) do { \ (d)->seqlo = htonl((s)->seqlo); \ (d)->seqhi = htonl((s)->seqhi); \ (d)->seqdiff = htonl((s)->seqdiff); \ (d)->max_win = htons((s)->max_win); \ (d)->mss = htons((s)->mss); \ (d)->state = (s)->state; \ (d)->wscale = (s)->wscale; \ if ((s)->scrub) { \ (d)->scrub.pfss_flags = \ htons((s)->scrub->pfss_flags & PFSS_TIMESTAMP); \ (d)->scrub.pfss_ttl = (s)->scrub->pfss_ttl; \ (d)->scrub.pfss_ts_mod = htonl((s)->scrub->pfss_ts_mod);\ (d)->scrub.scrub_flag = PFSYNC_SCRUB_FLAG_VALID; \ } \ } while (0) #define pf_state_peer_ntoh(s,d) do { \ (d)->seqlo = ntohl((s)->seqlo); \ (d)->seqhi = ntohl((s)->seqhi); \ (d)->seqdiff = ntohl((s)->seqdiff); \ (d)->max_win = ntohs((s)->max_win); \ (d)->mss = ntohs((s)->mss); \ (d)->state = (s)->state; \ (d)->wscale = (s)->wscale; \ if ((s)->scrub.scrub_flag == PFSYNC_SCRUB_FLAG_VALID && \ (d)->scrub != NULL) { \ (d)->scrub->pfss_flags = \ ntohs((s)->scrub.pfss_flags) & PFSS_TIMESTAMP; \ (d)->scrub->pfss_ttl = (s)->scrub.pfss_ttl; \ (d)->scrub->pfss_ts_mod = ntohl((s)->scrub.pfss_ts_mod);\ } \ } while (0) #define pf_state_counter_hton(s,d) do { \ d[0] = htonl((s>>32)&0xffffffff); \ d[1] = htonl(s&0xffffffff); \ } while (0) #define pf_state_counter_from_pfsync(s) \ (((u_int64_t)(s[0])<<32) | (u_int64_t)(s[1])) #define pf_state_counter_ntoh(s,d) do { \ d = ntohl(s[0]); \ d = d<<32; \ d += ntohl(s[1]); \ } while (0) TAILQ_HEAD(pf_rulequeue, pf_rule); struct pf_anchor; struct pf_ruleset { struct { struct pf_rulequeue queues[2]; struct { struct pf_rulequeue *ptr; struct pf_rule **ptr_array; u_int32_t rcount; u_int32_t ticket; int open; } active, inactive; } rules[PF_RULESET_MAX]; struct pf_anchor *anchor; u_int32_t tticket; int tables; int topen; }; RB_HEAD(pf_anchor_global, pf_anchor); RB_HEAD(pf_anchor_node, pf_anchor); struct pf_anchor { RB_ENTRY(pf_anchor) entry_global; RB_ENTRY(pf_anchor) entry_node; struct pf_anchor *parent; struct pf_anchor_node children; char name[PF_ANCHOR_NAME_SIZE]; char path[MAXPATHLEN]; struct pf_ruleset ruleset; int refcnt; /* anchor rules */ int match; /* XXX: used for pfctl black magic */ }; RB_PROTOTYPE(pf_anchor_global, pf_anchor, entry_global, pf_anchor_compare); RB_PROTOTYPE(pf_anchor_node, pf_anchor, entry_node, pf_anchor_compare); #define PF_RESERVED_ANCHOR "_pf" #define PFR_TFLAG_PERSIST 0x00000001 #define PFR_TFLAG_CONST 0x00000002 #define PFR_TFLAG_ACTIVE 0x00000004 #define PFR_TFLAG_INACTIVE 0x00000008 #define PFR_TFLAG_REFERENCED 0x00000010 #define PFR_TFLAG_REFDANCHOR 0x00000020 #define PFR_TFLAG_COUNTERS 0x00000040 /* Adjust masks below when adding flags. */ #define PFR_TFLAG_USRMASK (PFR_TFLAG_PERSIST | \ PFR_TFLAG_CONST | \ PFR_TFLAG_COUNTERS) #define PFR_TFLAG_SETMASK (PFR_TFLAG_ACTIVE | \ PFR_TFLAG_INACTIVE | \ PFR_TFLAG_REFERENCED | \ PFR_TFLAG_REFDANCHOR) #define PFR_TFLAG_ALLMASK (PFR_TFLAG_PERSIST | \ PFR_TFLAG_CONST | \ PFR_TFLAG_ACTIVE | \ PFR_TFLAG_INACTIVE | \ PFR_TFLAG_REFERENCED | \ PFR_TFLAG_REFDANCHOR | \ PFR_TFLAG_COUNTERS) struct pf_anchor_stackframe; struct pfr_table { char pfrt_anchor[MAXPATHLEN]; char pfrt_name[PF_TABLE_NAME_SIZE]; u_int32_t pfrt_flags; u_int8_t pfrt_fback; }; enum { PFR_FB_NONE, PFR_FB_MATCH, PFR_FB_ADDED, PFR_FB_DELETED, PFR_FB_CHANGED, PFR_FB_CLEARED, PFR_FB_DUPLICATE, PFR_FB_NOTMATCH, PFR_FB_CONFLICT, PFR_FB_NOCOUNT, PFR_FB_MAX }; struct pfr_addr { union { struct in_addr _pfra_ip4addr; struct in6_addr _pfra_ip6addr; } pfra_u; u_int8_t pfra_af; u_int8_t pfra_net; u_int8_t pfra_not; u_int8_t pfra_fback; }; #define pfra_ip4addr pfra_u._pfra_ip4addr #define pfra_ip6addr pfra_u._pfra_ip6addr enum { PFR_DIR_IN, PFR_DIR_OUT, PFR_DIR_MAX }; enum { PFR_OP_BLOCK, PFR_OP_PASS, PFR_OP_ADDR_MAX, PFR_OP_TABLE_MAX }; #define PFR_OP_XPASS PFR_OP_ADDR_MAX struct pfr_astats { struct pfr_addr pfras_a; u_int64_t pfras_packets[PFR_DIR_MAX][PFR_OP_ADDR_MAX]; u_int64_t pfras_bytes[PFR_DIR_MAX][PFR_OP_ADDR_MAX]; long pfras_tzero; }; enum { PFR_REFCNT_RULE, PFR_REFCNT_ANCHOR, PFR_REFCNT_MAX }; struct pfr_tstats { struct pfr_table pfrts_t; u_int64_t pfrts_packets[PFR_DIR_MAX][PFR_OP_TABLE_MAX]; u_int64_t pfrts_bytes[PFR_DIR_MAX][PFR_OP_TABLE_MAX]; u_int64_t pfrts_match; u_int64_t pfrts_nomatch; long pfrts_tzero; int pfrts_cnt; int pfrts_refcnt[PFR_REFCNT_MAX]; }; #define pfrts_name pfrts_t.pfrt_name #define pfrts_flags pfrts_t.pfrt_flags #ifndef _SOCKADDR_UNION_DEFINED #define _SOCKADDR_UNION_DEFINED union sockaddr_union { struct sockaddr sa; struct sockaddr_in sin; struct sockaddr_in6 sin6; }; #endif /* _SOCKADDR_UNION_DEFINED */ struct pfr_kcounters { u_int64_t pfrkc_packets[PFR_DIR_MAX][PFR_OP_ADDR_MAX]; u_int64_t pfrkc_bytes[PFR_DIR_MAX][PFR_OP_ADDR_MAX]; }; SLIST_HEAD(pfr_kentryworkq, pfr_kentry); struct pfr_kentry { struct radix_node pfrke_node[2]; union sockaddr_union pfrke_sa; SLIST_ENTRY(pfr_kentry) pfrke_workq; struct pfr_kcounters *pfrke_counters; long pfrke_tzero; u_int8_t pfrke_af; u_int8_t pfrke_net; u_int8_t pfrke_not; u_int8_t pfrke_mark; }; SLIST_HEAD(pfr_ktableworkq, pfr_ktable); RB_HEAD(pfr_ktablehead, pfr_ktable); struct pfr_ktable { struct pfr_tstats pfrkt_ts; RB_ENTRY(pfr_ktable) pfrkt_tree; SLIST_ENTRY(pfr_ktable) pfrkt_workq; struct radix_node_head *pfrkt_ip4; struct radix_node_head *pfrkt_ip6; struct pfr_ktable *pfrkt_shadow; struct pfr_ktable *pfrkt_root; struct pf_ruleset *pfrkt_rs; long pfrkt_larg; int pfrkt_nflags; }; #define pfrkt_t pfrkt_ts.pfrts_t #define pfrkt_name pfrkt_t.pfrt_name #define pfrkt_anchor pfrkt_t.pfrt_anchor #define pfrkt_ruleset pfrkt_t.pfrt_ruleset #define pfrkt_flags pfrkt_t.pfrt_flags #define pfrkt_cnt pfrkt_ts.pfrts_cnt #define pfrkt_refcnt pfrkt_ts.pfrts_refcnt #define pfrkt_packets pfrkt_ts.pfrts_packets #define pfrkt_bytes pfrkt_ts.pfrts_bytes #define pfrkt_match pfrkt_ts.pfrts_match #define pfrkt_nomatch pfrkt_ts.pfrts_nomatch #define pfrkt_tzero pfrkt_ts.pfrts_tzero /* keep synced with pfi_kif, used in RB_FIND */ struct pfi_kif_cmp { char pfik_name[IFNAMSIZ]; }; struct pfi_kif { char pfik_name[IFNAMSIZ]; union { RB_ENTRY(pfi_kif) _pfik_tree; LIST_ENTRY(pfi_kif) _pfik_list; } _pfik_glue; #define pfik_tree _pfik_glue._pfik_tree #define pfik_list _pfik_glue._pfik_list u_int64_t pfik_packets[2][2][2]; u_int64_t pfik_bytes[2][2][2]; u_int32_t pfik_tzero; u_int pfik_flags; struct ifnet *pfik_ifp; struct ifg_group *pfik_group; u_int pfik_rulerefs; TAILQ_HEAD(, pfi_dynaddr) pfik_dynaddrs; }; #define PFI_IFLAG_REFS 0x0001 /* has state references */ #define PFI_IFLAG_SKIP 0x0100 /* skip filtering on interface */ struct pf_pdesc { struct { int done; uid_t uid; gid_t gid; } lookup; u_int64_t tot_len; /* Make Mickey money */ union { struct tcphdr *tcp; struct udphdr *udp; struct icmp *icmp; #ifdef INET6 struct icmp6_hdr *icmp6; #endif /* INET6 */ void *any; } hdr; struct pf_rule *nat_rule; /* nat/rdr rule applied to packet */ struct pf_addr *src; /* src address */ struct pf_addr *dst; /* dst address */ u_int16_t *sport; u_int16_t *dport; struct pf_mtag *pf_mtag; u_int32_t p_len; /* total length of payload */ u_int16_t *ip_sum; u_int16_t *proto_sum; u_int16_t flags; /* Let SCRUB trigger behavior in * state code. Easier than tags */ #define PFDESC_TCP_NORM 0x0001 /* TCP shall be statefully scrubbed */ #define PFDESC_IP_REAS 0x0002 /* IP frags would've been reassembled */ sa_family_t af; u_int8_t proto; u_int8_t tos; u_int8_t dir; /* direction */ u_int8_t sidx; /* key index for source */ u_int8_t didx; /* key index for destination */ }; /* flags for RDR options */ #define PF_DPORT_RANGE 0x01 /* Dest port uses range */ #define PF_RPORT_RANGE 0x02 /* RDR'ed port uses range */ /* UDP state enumeration */ #define PFUDPS_NO_TRAFFIC 0 #define PFUDPS_SINGLE 1 #define PFUDPS_MULTIPLE 2 #define PFUDPS_NSTATES 3 /* number of state levels */ #define PFUDPS_NAMES { \ "NO_TRAFFIC", \ "SINGLE", \ "MULTIPLE", \ NULL \ } /* Other protocol state enumeration */ #define PFOTHERS_NO_TRAFFIC 0 #define PFOTHERS_SINGLE 1 #define PFOTHERS_MULTIPLE 2 #define PFOTHERS_NSTATES 3 /* number of state levels */ #define PFOTHERS_NAMES { \ "NO_TRAFFIC", \ "SINGLE", \ "MULTIPLE", \ NULL \ } #define ACTION_SET(a, x) \ do { \ if ((a) != NULL) \ *(a) = (x); \ } while (0) #define REASON_SET(a, x) \ do { \ if ((a) != NULL) \ *(a) = (x); \ if (x < PFRES_MAX) \ counter_u64_add(V_pf_status.counters[x], 1); \ } while (0) struct pf_kstatus { counter_u64_t counters[PFRES_MAX]; /* reason for passing/dropping */ counter_u64_t lcounters[LCNT_MAX]; /* limit counters */ counter_u64_t fcounters[FCNT_MAX]; /* state operation counters */ counter_u64_t scounters[SCNT_MAX]; /* src_node operation counters */ uint32_t states; uint32_t src_nodes; uint32_t running; uint32_t since; uint32_t debug; uint32_t hostid; char ifname[IFNAMSIZ]; uint8_t pf_chksum[PF_MD5_DIGEST_LENGTH]; }; struct pf_divert { union { struct in_addr ipv4; struct in6_addr ipv6; } addr; u_int16_t port; }; #define PFFRAG_FRENT_HIWAT 5000 /* Number of fragment entries */ #define PFR_KENTRY_HIWAT 200000 /* Number of table entries */ /* * Limit the length of the fragment queue traversal. Remember * search entry points based on the fragment offset. */ #define PF_FRAG_ENTRY_POINTS 16 /* * The number of entries in the fragment queue must be limited * to avoid DoS by linear seaching. Instead of a global limit, * use a limit per entry point. For large packets these sum up. */ #define PF_FRAG_ENTRY_LIMIT 64 /* * ioctl parameter structures */ struct pfioc_pooladdr { u_int32_t action; u_int32_t ticket; u_int32_t nr; u_int32_t r_num; u_int8_t r_action; u_int8_t r_last; u_int8_t af; char anchor[MAXPATHLEN]; struct pf_pooladdr addr; }; struct pfioc_rule { u_int32_t action; u_int32_t ticket; u_int32_t pool_ticket; u_int32_t nr; char anchor[MAXPATHLEN]; char anchor_call[MAXPATHLEN]; struct pf_rule rule; }; struct pfioc_natlook { struct pf_addr saddr; struct pf_addr daddr; struct pf_addr rsaddr; struct pf_addr rdaddr; u_int16_t sport; u_int16_t dport; u_int16_t rsport; u_int16_t rdport; sa_family_t af; u_int8_t proto; u_int8_t direction; }; struct pfioc_state { struct pfsync_state state; }; struct pfioc_src_node_kill { sa_family_t psnk_af; struct pf_rule_addr psnk_src; struct pf_rule_addr psnk_dst; u_int psnk_killed; }; struct pfioc_state_kill { struct pf_state_cmp psk_pfcmp; sa_family_t psk_af; int psk_proto; struct pf_rule_addr psk_src; struct pf_rule_addr psk_dst; char psk_ifname[IFNAMSIZ]; char psk_label[PF_RULE_LABEL_SIZE]; u_int psk_killed; }; struct pfioc_states { int ps_len; union { caddr_t psu_buf; struct pfsync_state *psu_states; } ps_u; #define ps_buf ps_u.psu_buf #define ps_states ps_u.psu_states }; struct pfioc_src_nodes { int psn_len; union { caddr_t psu_buf; struct pf_src_node *psu_src_nodes; } psn_u; #define psn_buf psn_u.psu_buf #define psn_src_nodes psn_u.psu_src_nodes }; struct pfioc_if { char ifname[IFNAMSIZ]; }; struct pfioc_tm { int timeout; int seconds; }; struct pfioc_limit { int index; unsigned limit; }; struct pfioc_altq_v0 { u_int32_t action; u_int32_t ticket; u_int32_t nr; struct pf_altq_v0 altq; }; struct pfioc_altq_v1 { u_int32_t action; u_int32_t ticket; u_int32_t nr; /* * Placed here so code that only uses the above parameters can be * written entirely in terms of the v0 or v1 type. */ u_int32_t version; struct pf_altq_v1 altq; }; /* * Latest version of struct pfioc_altq_vX. This must move in lock-step with * the latest version of struct pf_altq_vX as it has that struct as a * member. */ #define PFIOC_ALTQ_VERSION PF_ALTQ_VERSION struct pfioc_qstats_v0 { u_int32_t ticket; u_int32_t nr; void *buf; int nbytes; u_int8_t scheduler; }; struct pfioc_qstats_v1 { u_int32_t ticket; u_int32_t nr; void *buf; int nbytes; u_int8_t scheduler; /* * Placed here so code that only uses the above parameters can be * written entirely in terms of the v0 or v1 type. */ u_int32_t version; /* Requested version of stats struct */ }; /* Latest version of struct pfioc_qstats_vX */ #define PFIOC_QSTATS_VERSION 1 struct pfioc_ruleset { u_int32_t nr; char path[MAXPATHLEN]; char name[PF_ANCHOR_NAME_SIZE]; }; #define PF_RULESET_ALTQ (PF_RULESET_MAX) #define PF_RULESET_TABLE (PF_RULESET_MAX+1) struct pfioc_trans { int size; /* number of elements */ int esize; /* size of each element in bytes */ struct pfioc_trans_e { int rs_num; char anchor[MAXPATHLEN]; u_int32_t ticket; } *array; }; #define PFR_FLAG_ATOMIC 0x00000001 /* unused */ #define PFR_FLAG_DUMMY 0x00000002 #define PFR_FLAG_FEEDBACK 0x00000004 #define PFR_FLAG_CLSTATS 0x00000008 #define PFR_FLAG_ADDRSTOO 0x00000010 #define PFR_FLAG_REPLACE 0x00000020 #define PFR_FLAG_ALLRSETS 0x00000040 #define PFR_FLAG_ALLMASK 0x0000007F #ifdef _KERNEL #define PFR_FLAG_USERIOCTL 0x10000000 #endif struct pfioc_table { struct pfr_table pfrio_table; void *pfrio_buffer; int pfrio_esize; int pfrio_size; int pfrio_size2; int pfrio_nadd; int pfrio_ndel; int pfrio_nchange; int pfrio_flags; u_int32_t pfrio_ticket; }; #define pfrio_exists pfrio_nadd #define pfrio_nzero pfrio_nadd #define pfrio_nmatch pfrio_nadd #define pfrio_naddr pfrio_size2 #define pfrio_setflag pfrio_size2 #define pfrio_clrflag pfrio_nadd struct pfioc_iface { char pfiio_name[IFNAMSIZ]; void *pfiio_buffer; int pfiio_esize; int pfiio_size; int pfiio_nzero; int pfiio_flags; }; /* * ioctl operations */ #define DIOCSTART _IO ('D', 1) #define DIOCSTOP _IO ('D', 2) #define DIOCADDRULE _IOWR('D', 4, struct pfioc_rule) #define DIOCGETRULES _IOWR('D', 6, struct pfioc_rule) #define DIOCGETRULE _IOWR('D', 7, struct pfioc_rule) /* XXX cut 8 - 17 */ #define DIOCCLRSTATES _IOWR('D', 18, struct pfioc_state_kill) #define DIOCGETSTATE _IOWR('D', 19, struct pfioc_state) #define DIOCSETSTATUSIF _IOWR('D', 20, struct pfioc_if) #define DIOCGETSTATUS _IOWR('D', 21, struct pf_status) #define DIOCCLRSTATUS _IO ('D', 22) #define DIOCNATLOOK _IOWR('D', 23, struct pfioc_natlook) #define DIOCSETDEBUG _IOWR('D', 24, u_int32_t) #define DIOCGETSTATES _IOWR('D', 25, struct pfioc_states) #define DIOCCHANGERULE _IOWR('D', 26, struct pfioc_rule) /* XXX cut 26 - 28 */ #define DIOCSETTIMEOUT _IOWR('D', 29, struct pfioc_tm) #define DIOCGETTIMEOUT _IOWR('D', 30, struct pfioc_tm) #define DIOCADDSTATE _IOWR('D', 37, struct pfioc_state) #define DIOCCLRRULECTRS _IO ('D', 38) #define DIOCGETLIMIT _IOWR('D', 39, struct pfioc_limit) #define DIOCSETLIMIT _IOWR('D', 40, struct pfioc_limit) #define DIOCKILLSTATES _IOWR('D', 41, struct pfioc_state_kill) #define DIOCSTARTALTQ _IO ('D', 42) #define DIOCSTOPALTQ _IO ('D', 43) #define DIOCADDALTQV0 _IOWR('D', 45, struct pfioc_altq_v0) #define DIOCADDALTQV1 _IOWR('D', 45, struct pfioc_altq_v1) #define DIOCGETALTQSV0 _IOWR('D', 47, struct pfioc_altq_v0) #define DIOCGETALTQSV1 _IOWR('D', 47, struct pfioc_altq_v1) #define DIOCGETALTQV0 _IOWR('D', 48, struct pfioc_altq_v0) #define DIOCGETALTQV1 _IOWR('D', 48, struct pfioc_altq_v1) #define DIOCCHANGEALTQV0 _IOWR('D', 49, struct pfioc_altq_v0) #define DIOCCHANGEALTQV1 _IOWR('D', 49, struct pfioc_altq_v1) #define DIOCGETQSTATSV0 _IOWR('D', 50, struct pfioc_qstats_v0) #define DIOCGETQSTATSV1 _IOWR('D', 50, struct pfioc_qstats_v1) #define DIOCBEGINADDRS _IOWR('D', 51, struct pfioc_pooladdr) #define DIOCADDADDR _IOWR('D', 52, struct pfioc_pooladdr) #define DIOCGETADDRS _IOWR('D', 53, struct pfioc_pooladdr) #define DIOCGETADDR _IOWR('D', 54, struct pfioc_pooladdr) #define DIOCCHANGEADDR _IOWR('D', 55, struct pfioc_pooladdr) /* XXX cut 55 - 57 */ #define DIOCGETRULESETS _IOWR('D', 58, struct pfioc_ruleset) #define DIOCGETRULESET _IOWR('D', 59, struct pfioc_ruleset) #define DIOCRCLRTABLES _IOWR('D', 60, struct pfioc_table) #define DIOCRADDTABLES _IOWR('D', 61, struct pfioc_table) #define DIOCRDELTABLES _IOWR('D', 62, struct pfioc_table) #define DIOCRGETTABLES _IOWR('D', 63, struct pfioc_table) #define DIOCRGETTSTATS _IOWR('D', 64, struct pfioc_table) #define DIOCRCLRTSTATS _IOWR('D', 65, struct pfioc_table) #define DIOCRCLRADDRS _IOWR('D', 66, struct pfioc_table) #define DIOCRADDADDRS _IOWR('D', 67, struct pfioc_table) #define DIOCRDELADDRS _IOWR('D', 68, struct pfioc_table) #define DIOCRSETADDRS _IOWR('D', 69, struct pfioc_table) #define DIOCRGETADDRS _IOWR('D', 70, struct pfioc_table) #define DIOCRGETASTATS _IOWR('D', 71, struct pfioc_table) #define DIOCRCLRASTATS _IOWR('D', 72, struct pfioc_table) #define DIOCRTSTADDRS _IOWR('D', 73, struct pfioc_table) #define DIOCRSETTFLAGS _IOWR('D', 74, struct pfioc_table) #define DIOCRINADEFINE _IOWR('D', 77, struct pfioc_table) #define DIOCOSFPFLUSH _IO('D', 78) #define DIOCOSFPADD _IOWR('D', 79, struct pf_osfp_ioctl) #define DIOCOSFPGET _IOWR('D', 80, struct pf_osfp_ioctl) #define DIOCXBEGIN _IOWR('D', 81, struct pfioc_trans) #define DIOCXCOMMIT _IOWR('D', 82, struct pfioc_trans) #define DIOCXROLLBACK _IOWR('D', 83, struct pfioc_trans) #define DIOCGETSRCNODES _IOWR('D', 84, struct pfioc_src_nodes) #define DIOCCLRSRCNODES _IO('D', 85) #define DIOCSETHOSTID _IOWR('D', 86, u_int32_t) #define DIOCIGETIFACES _IOWR('D', 87, struct pfioc_iface) #define DIOCSETIFFLAG _IOWR('D', 89, struct pfioc_iface) #define DIOCCLRIFFLAG _IOWR('D', 90, struct pfioc_iface) #define DIOCKILLSRCNODES _IOWR('D', 91, struct pfioc_src_node_kill) struct pf_ifspeed_v0 { char ifname[IFNAMSIZ]; u_int32_t baudrate; }; struct pf_ifspeed_v1 { char ifname[IFNAMSIZ]; u_int32_t baudrate32; /* layout identical to struct pf_ifspeed_v0 up to this point */ u_int64_t baudrate; }; /* Latest version of struct pf_ifspeed_vX */ #define PF_IFSPEED_VERSION 1 #define DIOCGIFSPEEDV0 _IOWR('D', 92, struct pf_ifspeed_v0) #define DIOCGIFSPEEDV1 _IOWR('D', 92, struct pf_ifspeed_v1) /* * Compatibility and convenience macros */ #ifndef _KERNEL #ifdef PFIOC_USE_LATEST /* * Maintaining in-tree consumers of the ioctl interface is easier when that * code can be written in terms old names that refer to the latest interface * version as that reduces the required changes in the consumers to those * that are functionally necessary to accommodate a new interface version. */ #define pfioc_altq __CONCAT(pfioc_altq_v, PFIOC_ALTQ_VERSION) #define pfioc_qstats __CONCAT(pfioc_qstats_v, PFIOC_QSTATS_VERSION) #define pf_ifspeed __CONCAT(pf_ifspeed_v, PF_IFSPEED_VERSION) #define DIOCADDALTQ __CONCAT(DIOCADDALTQV, PFIOC_ALTQ_VERSION) #define DIOCGETALTQS __CONCAT(DIOCGETALTQSV, PFIOC_ALTQ_VERSION) #define DIOCGETALTQ __CONCAT(DIOCGETALTQV, PFIOC_ALTQ_VERSION) #define DIOCCHANGEALTQ __CONCAT(DIOCCHANGEALTQV, PFIOC_ALTQ_VERSION) #define DIOCGETQSTATS __CONCAT(DIOCGETQSTATSV, PFIOC_QSTATS_VERSION) #define DIOCGIFSPEED __CONCAT(DIOCGIFSPEEDV, PF_IFSPEED_VERSION) #else /* * When building out-of-tree code that is written for the old interface, * such as may exist in ports for example, resolve the old struct tags and * ioctl command names to the v0 versions. */ #define pfioc_altq __CONCAT(pfioc_altq_v, 0) #define pfioc_qstats __CONCAT(pfioc_qstats_v, 0) #define pf_ifspeed __CONCAT(pf_ifspeed_v, 0) #define DIOCADDALTQ __CONCAT(DIOCADDALTQV, 0) #define DIOCGETALTQS __CONCAT(DIOCGETALTQSV, 0) #define DIOCGETALTQ __CONCAT(DIOCGETALTQV, 0) #define DIOCCHANGEALTQ __CONCAT(DIOCCHANGEALTQV, 0) #define DIOCGETQSTATS __CONCAT(DIOCGETQSTATSV, 0) #define DIOCGIFSPEED __CONCAT(DIOCGIFSPEEDV, 0) #endif /* PFIOC_USE_LATEST */ #endif /* _KERNEL */ #ifdef _KERNEL LIST_HEAD(pf_src_node_list, pf_src_node); struct pf_srchash { struct pf_src_node_list nodes; struct mtx lock; }; struct pf_keyhash { LIST_HEAD(, pf_state_key) keys; struct mtx lock; }; struct pf_idhash { LIST_HEAD(, pf_state) states; struct mtx lock; }; extern u_long pf_hashmask; extern u_long pf_srchashmask; #define PF_HASHSIZ (131072) #define PF_SRCHASHSIZ (PF_HASHSIZ/4) VNET_DECLARE(struct pf_keyhash *, pf_keyhash); VNET_DECLARE(struct pf_idhash *, pf_idhash); #define V_pf_keyhash VNET(pf_keyhash) #define V_pf_idhash VNET(pf_idhash) VNET_DECLARE(struct pf_srchash *, pf_srchash); #define V_pf_srchash VNET(pf_srchash) #define PF_IDHASH(s) (be64toh((s)->id) % (pf_hashmask + 1)) VNET_DECLARE(void *, pf_swi_cookie); #define V_pf_swi_cookie VNET(pf_swi_cookie) VNET_DECLARE(uint64_t, pf_stateid[MAXCPU]); #define V_pf_stateid VNET(pf_stateid) TAILQ_HEAD(pf_altqqueue, pf_altq); -VNET_DECLARE(struct pf_altqqueue, pf_altqs[2]); +VNET_DECLARE(struct pf_altqqueue, pf_altqs[4]); #define V_pf_altqs VNET(pf_altqs) VNET_DECLARE(struct pf_palist, pf_pabuf); #define V_pf_pabuf VNET(pf_pabuf) VNET_DECLARE(u_int32_t, ticket_altqs_active); #define V_ticket_altqs_active VNET(ticket_altqs_active) VNET_DECLARE(u_int32_t, ticket_altqs_inactive); #define V_ticket_altqs_inactive VNET(ticket_altqs_inactive) VNET_DECLARE(int, altqs_inactive_open); #define V_altqs_inactive_open VNET(altqs_inactive_open) VNET_DECLARE(u_int32_t, ticket_pabuf); #define V_ticket_pabuf VNET(ticket_pabuf) VNET_DECLARE(struct pf_altqqueue *, pf_altqs_active); #define V_pf_altqs_active VNET(pf_altqs_active) +VNET_DECLARE(struct pf_altqqueue *, pf_altq_ifs_active); +#define V_pf_altq_ifs_active VNET(pf_altq_ifs_active) VNET_DECLARE(struct pf_altqqueue *, pf_altqs_inactive); #define V_pf_altqs_inactive VNET(pf_altqs_inactive) +VNET_DECLARE(struct pf_altqqueue *, pf_altq_ifs_inactive); +#define V_pf_altq_ifs_inactive VNET(pf_altq_ifs_inactive) VNET_DECLARE(struct pf_rulequeue, pf_unlinked_rules); #define V_pf_unlinked_rules VNET(pf_unlinked_rules) void pf_initialize(void); void pf_mtag_initialize(void); void pf_mtag_cleanup(void); void pf_cleanup(void); struct pf_mtag *pf_get_mtag(struct mbuf *); extern void pf_calc_skip_steps(struct pf_rulequeue *); #ifdef ALTQ extern void pf_altq_ifnet_event(struct ifnet *, int); #endif VNET_DECLARE(uma_zone_t, pf_state_z); #define V_pf_state_z VNET(pf_state_z) VNET_DECLARE(uma_zone_t, pf_state_key_z); #define V_pf_state_key_z VNET(pf_state_key_z) VNET_DECLARE(uma_zone_t, pf_state_scrub_z); #define V_pf_state_scrub_z VNET(pf_state_scrub_z) extern void pf_purge_thread(void *); extern void pf_unload_vnet_purge(void); extern void pf_intr(void *); extern void pf_purge_expired_src_nodes(void); extern int pf_unlink_state(struct pf_state *, u_int); #define PF_ENTER_LOCKED 0x00000001 #define PF_RETURN_LOCKED 0x00000002 extern int pf_state_insert(struct pfi_kif *, struct pf_state_key *, struct pf_state_key *, struct pf_state *); extern void pf_free_state(struct pf_state *); static __inline void pf_ref_state(struct pf_state *s) { refcount_acquire(&s->refs); } static __inline int pf_release_state(struct pf_state *s) { if (refcount_release(&s->refs)) { pf_free_state(s); return (1); } else return (0); } extern struct pf_state *pf_find_state_byid(uint64_t, uint32_t); extern struct pf_state *pf_find_state_all(struct pf_state_key_cmp *, u_int, int *); extern struct pf_src_node *pf_find_src_node(struct pf_addr *, struct pf_rule *, sa_family_t, int); extern void pf_unlink_src_node(struct pf_src_node *); extern u_int pf_free_src_nodes(struct pf_src_node_list *); extern void pf_print_state(struct pf_state *); extern void pf_print_flags(u_int8_t); extern u_int16_t pf_cksum_fixup(u_int16_t, u_int16_t, u_int16_t, u_int8_t); extern u_int16_t pf_proto_cksum_fixup(struct mbuf *, u_int16_t, u_int16_t, u_int16_t, u_int8_t); VNET_DECLARE(struct ifnet *, sync_ifp); #define V_sync_ifp VNET(sync_ifp); VNET_DECLARE(struct pf_rule, pf_default_rule); #define V_pf_default_rule VNET(pf_default_rule) extern void pf_addrcpy(struct pf_addr *, struct pf_addr *, u_int8_t); void pf_free_rule(struct pf_rule *); #ifdef INET int pf_test(int, int, struct ifnet *, struct mbuf **, struct inpcb *); int pf_normalize_ip(struct mbuf **, int, struct pfi_kif *, u_short *, struct pf_pdesc *); #endif /* INET */ #ifdef INET6 int pf_test6(int, int, struct ifnet *, struct mbuf **, struct inpcb *); int pf_normalize_ip6(struct mbuf **, int, struct pfi_kif *, u_short *, struct pf_pdesc *); void pf_poolmask(struct pf_addr *, struct pf_addr*, struct pf_addr *, struct pf_addr *, u_int8_t); void pf_addr_inc(struct pf_addr *, sa_family_t); int pf_refragment6(struct ifnet *, struct mbuf **, struct m_tag *); #endif /* INET6 */ u_int32_t pf_new_isn(struct pf_state *); void *pf_pull_hdr(struct mbuf *, int, void *, int, u_short *, u_short *, sa_family_t); void pf_change_a(void *, u_int16_t *, u_int32_t, u_int8_t); void pf_change_proto_a(struct mbuf *, void *, u_int16_t *, u_int32_t, u_int8_t); void pf_change_tcp_a(struct mbuf *, void *, u_int16_t *, u_int32_t); void pf_send_deferred_syn(struct pf_state *); int pf_match_addr(u_int8_t, struct pf_addr *, struct pf_addr *, struct pf_addr *, sa_family_t); int pf_match_addr_range(struct pf_addr *, struct pf_addr *, struct pf_addr *, sa_family_t); int pf_match_port(u_int8_t, u_int16_t, u_int16_t, u_int16_t); void pf_normalize_init(void); void pf_normalize_cleanup(void); int pf_normalize_tcp(int, struct pfi_kif *, struct mbuf *, int, int, void *, struct pf_pdesc *); void pf_normalize_tcp_cleanup(struct pf_state *); int pf_normalize_tcp_init(struct mbuf *, int, struct pf_pdesc *, struct tcphdr *, struct pf_state_peer *, struct pf_state_peer *); int pf_normalize_tcp_stateful(struct mbuf *, int, struct pf_pdesc *, u_short *, struct tcphdr *, struct pf_state *, struct pf_state_peer *, struct pf_state_peer *, int *); u_int32_t pf_state_expires(const struct pf_state *); void pf_purge_expired_fragments(void); void pf_purge_fragments(uint32_t); int pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *, int); int pf_socket_lookup(int, struct pf_pdesc *, struct mbuf *); struct pf_state_key *pf_alloc_state_key(int); void pfr_initialize(void); void pfr_cleanup(void); int pfr_match_addr(struct pfr_ktable *, struct pf_addr *, sa_family_t); void pfr_update_stats(struct pfr_ktable *, struct pf_addr *, sa_family_t, u_int64_t, int, int, int); int pfr_pool_get(struct pfr_ktable *, int *, struct pf_addr *, sa_family_t); void pfr_dynaddr_update(struct pfr_ktable *, struct pfi_dynaddr *); struct pfr_ktable * pfr_attach_table(struct pf_ruleset *, char *); void pfr_detach_table(struct pfr_ktable *); int pfr_clr_tables(struct pfr_table *, int *, int); int pfr_add_tables(struct pfr_table *, int, int *, int); int pfr_del_tables(struct pfr_table *, int, int *, int); int pfr_table_count(struct pfr_table *, int); int pfr_get_tables(struct pfr_table *, struct pfr_table *, int *, int); int pfr_get_tstats(struct pfr_table *, struct pfr_tstats *, int *, int); int pfr_clr_tstats(struct pfr_table *, int, int *, int); int pfr_set_tflags(struct pfr_table *, int, int, int, int *, int *, int); int pfr_clr_addrs(struct pfr_table *, int *, int); int pfr_insert_kentry(struct pfr_ktable *, struct pfr_addr *, long); int pfr_add_addrs(struct pfr_table *, struct pfr_addr *, int, int *, int); int pfr_del_addrs(struct pfr_table *, struct pfr_addr *, int, int *, int); int pfr_set_addrs(struct pfr_table *, struct pfr_addr *, int, int *, int *, int *, int *, int, u_int32_t); int pfr_get_addrs(struct pfr_table *, struct pfr_addr *, int *, int); int pfr_get_astats(struct pfr_table *, struct pfr_astats *, int *, int); int pfr_clr_astats(struct pfr_table *, struct pfr_addr *, int, int *, int); int pfr_tst_addrs(struct pfr_table *, struct pfr_addr *, int, int *, int); int pfr_ina_begin(struct pfr_table *, u_int32_t *, int *, int); int pfr_ina_rollback(struct pfr_table *, u_int32_t, int *, int); int pfr_ina_commit(struct pfr_table *, u_int32_t, int *, int *, int); int pfr_ina_define(struct pfr_table *, struct pfr_addr *, int, int *, int *, u_int32_t, int); MALLOC_DECLARE(PFI_MTYPE); VNET_DECLARE(struct pfi_kif *, pfi_all); #define V_pfi_all VNET(pfi_all) void pfi_initialize(void); void pfi_initialize_vnet(void); void pfi_cleanup(void); void pfi_cleanup_vnet(void); void pfi_kif_ref(struct pfi_kif *); void pfi_kif_unref(struct pfi_kif *); struct pfi_kif *pfi_kif_find(const char *); struct pfi_kif *pfi_kif_attach(struct pfi_kif *, const char *); int pfi_kif_match(struct pfi_kif *, struct pfi_kif *); void pfi_kif_purge(void); int pfi_match_addr(struct pfi_dynaddr *, struct pf_addr *, sa_family_t); int pfi_dynaddr_setup(struct pf_addr_wrap *, sa_family_t); void pfi_dynaddr_remove(struct pfi_dynaddr *); void pfi_dynaddr_copyout(struct pf_addr_wrap *); void pfi_update_status(const char *, struct pf_status *); void pfi_get_ifaces(const char *, struct pfi_kif *, int *); int pfi_set_flags(const char *, int); int pfi_clear_flags(const char *, int); int pf_match_tag(struct mbuf *, struct pf_rule *, int *, int); int pf_tag_packet(struct mbuf *, struct pf_pdesc *, int); int pf_addr_cmp(struct pf_addr *, struct pf_addr *, sa_family_t); void pf_qid2qname(u_int32_t, char *); VNET_DECLARE(struct pf_kstatus, pf_status); #define V_pf_status VNET(pf_status) struct pf_limit { uma_zone_t zone; u_int limit; }; VNET_DECLARE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); #define V_pf_limits VNET(pf_limits) #endif /* _KERNEL */ #ifdef _KERNEL VNET_DECLARE(struct pf_anchor_global, pf_anchors); #define V_pf_anchors VNET(pf_anchors) VNET_DECLARE(struct pf_anchor, pf_main_anchor); #define V_pf_main_anchor VNET(pf_main_anchor) #define pf_main_ruleset V_pf_main_anchor.ruleset #endif /* these ruleset functions can be linked into userland programs (pfctl) */ int pf_get_ruleset_number(u_int8_t); void pf_init_ruleset(struct pf_ruleset *); int pf_anchor_setup(struct pf_rule *, const struct pf_ruleset *, const char *); int pf_anchor_copyout(const struct pf_ruleset *, const struct pf_rule *, struct pfioc_rule *); void pf_anchor_remove(struct pf_rule *); void pf_remove_if_empty_ruleset(struct pf_ruleset *); struct pf_ruleset *pf_find_ruleset(const char *); struct pf_ruleset *pf_find_or_create_ruleset(const char *); void pf_rs_initialize(void); /* The fingerprint functions can be linked into userland programs (tcpdump) */ int pf_osfp_add(struct pf_osfp_ioctl *); #ifdef _KERNEL struct pf_osfp_enlist * pf_osfp_fingerprint(struct pf_pdesc *, struct mbuf *, int, const struct tcphdr *); #endif /* _KERNEL */ void pf_osfp_flush(void); int pf_osfp_get(struct pf_osfp_ioctl *); int pf_osfp_match(struct pf_osfp_enlist *, pf_osfp_t); #ifdef _KERNEL void pf_print_host(struct pf_addr *, u_int16_t, u_int8_t); void pf_step_into_anchor(struct pf_anchor_stackframe *, int *, struct pf_ruleset **, int, struct pf_rule **, struct pf_rule **, int *); int pf_step_out_of_anchor(struct pf_anchor_stackframe *, int *, struct pf_ruleset **, int, struct pf_rule **, struct pf_rule **, int *); int pf_map_addr(u_int8_t, struct pf_rule *, struct pf_addr *, struct pf_addr *, struct pf_addr *, struct pf_src_node **); struct pf_rule *pf_get_translation(struct pf_pdesc *, struct mbuf *, int, int, struct pfi_kif *, struct pf_src_node **, struct pf_state_key **, struct pf_state_key **, struct pf_addr *, struct pf_addr *, uint16_t, uint16_t, struct pf_anchor_stackframe *); struct pf_state_key *pf_state_key_setup(struct pf_pdesc *, struct pf_addr *, struct pf_addr *, u_int16_t, u_int16_t); struct pf_state_key *pf_state_key_clone(struct pf_state_key *); #endif /* _KERNEL */ #endif /* _NET_PFVAR_H_ */ Index: head/sys/netpfil/pf/pf.c =================================================================== --- head/sys/netpfil/pf/pf.c (revision 343994) +++ head/sys/netpfil/pf/pf.c (revision 343995) @@ -1,6682 +1,6688 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * Copyright (c) 2002 - 2008 Henning Brauer * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * Effort sponsored in part by the Defense Advanced Research Projects * Agency (DARPA) and Air Force Research Laboratory, Air Force * Materiel Command, USAF, under agreement number F30602-01-2-0537. * * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_bpf.h" #include "opt_pf.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* XXX: only for DIR_IN/DIR_OUT */ #ifdef INET6 #include #include #include #include #include #include #include #endif /* INET6 */ #include #include #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x /* * Global variables */ /* state tables */ -VNET_DEFINE(struct pf_altqqueue, pf_altqs[2]); +VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); VNET_DEFINE(struct pf_palist, pf_pabuf); VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); +VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); +VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); VNET_DEFINE(struct pf_kstatus, pf_status); VNET_DEFINE(u_int32_t, ticket_altqs_active); VNET_DEFINE(u_int32_t, ticket_altqs_inactive); VNET_DEFINE(int, altqs_inactive_open); VNET_DEFINE(u_int32_t, ticket_pabuf); VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) VNET_DEFINE(u_char, pf_tcp_secret[16]); #define V_pf_tcp_secret VNET(pf_tcp_secret) VNET_DEFINE(int, pf_tcp_secret_init); #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) VNET_DEFINE(int, pf_tcp_iss_off); #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) VNET_DECLARE(int, pf_vnet_active); #define V_pf_vnet_active VNET(pf_vnet_active) VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); #define V_pf_purge_idx VNET(pf_purge_idx) /* * Queue for pf_intr() sends. */ static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); struct pf_send_entry { STAILQ_ENTRY(pf_send_entry) pfse_next; struct mbuf *pfse_m; enum { PFSE_IP, PFSE_IP6, PFSE_ICMP, PFSE_ICMP6, } pfse_type; struct { int type; int code; int mtu; } icmpopts; }; STAILQ_HEAD(pf_send_head, pf_send_entry); VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); #define V_pf_sendqueue VNET(pf_sendqueue) static struct mtx pf_sendqueue_mtx; MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) /* * Queue for pf_overload_task() tasks. */ struct pf_overload_entry { SLIST_ENTRY(pf_overload_entry) next; struct pf_addr addr; sa_family_t af; uint8_t dir; struct pf_rule *rule; }; SLIST_HEAD(pf_overload_head, pf_overload_entry); VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); #define V_pf_overloadqueue VNET(pf_overloadqueue) VNET_DEFINE_STATIC(struct task, pf_overloadtask); #define V_pf_overloadtask VNET(pf_overloadtask) static struct mtx pf_overloadqueue_mtx; MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, "pf overload/flush queue", MTX_DEF); #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) VNET_DEFINE(struct pf_rulequeue, pf_unlinked_rules); struct mtx pf_unlnkdrules_mtx; MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", MTX_DEF); VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); #define V_pf_sources_z VNET(pf_sources_z) uma_zone_t pf_mtag_z; VNET_DEFINE(uma_zone_t, pf_state_z); VNET_DEFINE(uma_zone_t, pf_state_key_z); VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); #define PFID_CPUBITS 8 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) #define PFID_MAXID (~PFID_CPUMASK) CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); static void pf_src_tree_remove_state(struct pf_state *); static void pf_init_threshold(struct pf_threshold *, u_int32_t, u_int32_t); static void pf_add_threshold(struct pf_threshold *); static int pf_check_threshold(struct pf_threshold *); static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, u_int16_t *, u_int16_t *, struct pf_addr *, u_int16_t, u_int8_t, sa_family_t); static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, struct tcphdr *, struct pf_state_peer *); static void pf_change_icmp(struct pf_addr *, u_int16_t *, struct pf_addr *, struct pf_addr *, u_int16_t, u_int16_t *, u_int16_t *, u_int16_t *, u_int16_t *, u_int8_t, sa_family_t); static void pf_send_tcp(struct mbuf *, const struct pf_rule *, sa_family_t, const struct pf_addr *, const struct pf_addr *, u_int16_t, u_int16_t, u_int32_t, u_int32_t, u_int8_t, u_int16_t, u_int16_t, u_int8_t, int, u_int16_t, struct ifnet *); static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, sa_family_t, struct pf_rule *); static void pf_detach_state(struct pf_state *); static int pf_state_key_attach(struct pf_state_key *, struct pf_state_key *, struct pf_state *); static void pf_state_key_detach(struct pf_state *, int); static int pf_state_key_ctor(void *, int, void *, int); static u_int32_t pf_tcp_iss(struct pf_pdesc *); static int pf_test_rule(struct pf_rule **, struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, struct pf_pdesc *, struct pf_rule **, struct pf_ruleset **, struct inpcb *); static int pf_create_state(struct pf_rule *, struct pf_rule *, struct pf_rule *, struct pf_pdesc *, struct pf_src_node *, struct pf_state_key *, struct pf_state_key *, struct mbuf *, int, u_int16_t, u_int16_t, int *, struct pfi_kif *, struct pf_state **, int, u_int16_t, u_int16_t, int); static int pf_test_fragment(struct pf_rule **, int, struct pfi_kif *, struct mbuf *, void *, struct pf_pdesc *, struct pf_rule **, struct pf_ruleset **); static int pf_tcp_track_full(struct pf_state_peer *, struct pf_state_peer *, struct pf_state **, struct pfi_kif *, struct mbuf *, int, struct pf_pdesc *, u_short *, int *); static int pf_tcp_track_sloppy(struct pf_state_peer *, struct pf_state_peer *, struct pf_state **, struct pf_pdesc *, u_short *); static int pf_test_state_tcp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *, u_short *); static int pf_test_state_udp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *); static int pf_test_state_icmp(struct pf_state **, int, struct pfi_kif *, struct mbuf *, int, void *, struct pf_pdesc *, u_short *); static int pf_test_state_other(struct pf_state **, int, struct pfi_kif *, struct mbuf *, struct pf_pdesc *); static u_int8_t pf_get_wscale(struct mbuf *, int, u_int16_t, sa_family_t); static u_int16_t pf_get_mss(struct mbuf *, int, u_int16_t, sa_family_t); static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, int, u_int16_t); static int pf_check_proto_cksum(struct mbuf *, int, int, u_int8_t, sa_family_t); static void pf_print_state_parts(struct pf_state *, struct pf_state_key *, struct pf_state_key *); static int pf_addr_wrap_neq(struct pf_addr_wrap *, struct pf_addr_wrap *); static struct pf_state *pf_find_state(struct pfi_kif *, struct pf_state_key_cmp *, u_int); static int pf_src_connlimit(struct pf_state **); static void pf_overload_task(void *v, int pending); static int pf_insert_src_node(struct pf_src_node **, struct pf_rule *, struct pf_addr *, sa_family_t); static u_int pf_purge_expired_states(u_int, int); static void pf_purge_unlinked_rules(void); static int pf_mtag_uminit(void *, int, int); static void pf_mtag_free(struct m_tag *); #ifdef INET static void pf_route(struct mbuf **, struct pf_rule *, int, struct ifnet *, struct pf_state *, struct pf_pdesc *, struct inpcb *); #endif /* INET */ #ifdef INET6 static void pf_change_a6(struct pf_addr *, u_int16_t *, struct pf_addr *, u_int8_t); static void pf_route6(struct mbuf **, struct pf_rule *, int, struct ifnet *, struct pf_state *, struct pf_pdesc *, struct inpcb *); #endif /* INET6 */ int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); extern int pf_end_threads; extern struct proc *pf_purge_proc; VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ (pd)->pf_mtag->flags & PF_PACKET_LOOPED) #define STATE_LOOKUP(i, k, d, s, pd) \ do { \ (s) = pf_find_state((i), (k), (d)); \ if ((s) == NULL) \ return (PF_DROP); \ if (PACKET_LOOPED(pd)) \ return (PF_PASS); \ if ((d) == PF_OUT && \ (((s)->rule.ptr->rt == PF_ROUTETO && \ (s)->rule.ptr->direction == PF_OUT) || \ ((s)->rule.ptr->rt == PF_REPLYTO && \ (s)->rule.ptr->direction == PF_IN)) && \ (s)->rt_kif != NULL && \ (s)->rt_kif != (i)) \ return (PF_PASS); \ } while (0) #define BOUND_IFACE(r, k) \ ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all #define STATE_INC_COUNTERS(s) \ do { \ counter_u64_add(s->rule.ptr->states_cur, 1); \ counter_u64_add(s->rule.ptr->states_tot, 1); \ if (s->anchor.ptr != NULL) { \ counter_u64_add(s->anchor.ptr->states_cur, 1); \ counter_u64_add(s->anchor.ptr->states_tot, 1); \ } \ if (s->nat_rule.ptr != NULL) { \ counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ } \ } while (0) #define STATE_DEC_COUNTERS(s) \ do { \ if (s->nat_rule.ptr != NULL) \ counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ if (s->anchor.ptr != NULL) \ counter_u64_add(s->anchor.ptr->states_cur, -1); \ counter_u64_add(s->rule.ptr->states_cur, -1); \ } while (0) -static MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); +MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); VNET_DEFINE(struct pf_keyhash *, pf_keyhash); VNET_DEFINE(struct pf_idhash *, pf_idhash); VNET_DEFINE(struct pf_srchash *, pf_srchash); SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW, 0, "pf(4)"); u_long pf_hashmask; u_long pf_srchashmask; static u_long pf_hashsize; static u_long pf_srchashsize; u_long pf_ioctl_maxcount = 65535; SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, &pf_hashsize, 0, "Size of pf(4) states hashtable"); SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RDTUN, &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); VNET_DEFINE(void *, pf_swi_cookie); VNET_DEFINE(uint32_t, pf_hashseed); #define V_pf_hashseed VNET(pf_hashseed) int pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: if (a->addr32[0] > b->addr32[0]) return (1); if (a->addr32[0] < b->addr32[0]) return (-1); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (a->addr32[3] > b->addr32[3]) return (1); if (a->addr32[3] < b->addr32[3]) return (-1); if (a->addr32[2] > b->addr32[2]) return (1); if (a->addr32[2] < b->addr32[2]) return (-1); if (a->addr32[1] > b->addr32[1]) return (1); if (a->addr32[1] < b->addr32[1]) return (-1); if (a->addr32[0] > b->addr32[0]) return (1); if (a->addr32[0] < b->addr32[0]) return (-1); break; #endif /* INET6 */ default: panic("%s: unknown address family %u", __func__, af); } return (0); } static __inline uint32_t pf_hashkey(struct pf_state_key *sk) { uint32_t h; h = murmur3_32_hash32((uint32_t *)sk, sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), V_pf_hashseed); return (h & pf_hashmask); } static __inline uint32_t pf_hashsrc(struct pf_addr *addr, sa_family_t af) { uint32_t h; switch (af) { case AF_INET: h = murmur3_32_hash32((uint32_t *)&addr->v4, sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); break; case AF_INET6: h = murmur3_32_hash32((uint32_t *)&addr->v6, sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); break; default: panic("%s: unknown address family %u", __func__, af); } return (h & pf_srchashmask); } #ifdef ALTQ static int pf_state_hash(struct pf_state *s) { u_int32_t hv = (intptr_t)s / sizeof(*s); hv ^= crc32(&s->src, sizeof(s->src)); hv ^= crc32(&s->dst, sizeof(s->dst)); if (hv == 0) hv = 1; return (hv); } #endif #ifdef INET6 void pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: dst->addr32[0] = src->addr32[0]; break; #endif /* INET */ case AF_INET6: dst->addr32[0] = src->addr32[0]; dst->addr32[1] = src->addr32[1]; dst->addr32[2] = src->addr32[2]; dst->addr32[3] = src->addr32[3]; break; } } #endif /* INET6 */ static void pf_init_threshold(struct pf_threshold *threshold, u_int32_t limit, u_int32_t seconds) { threshold->limit = limit * PF_THRESHOLD_MULT; threshold->seconds = seconds; threshold->count = 0; threshold->last = time_uptime; } static void pf_add_threshold(struct pf_threshold *threshold) { u_int32_t t = time_uptime, diff = t - threshold->last; if (diff >= threshold->seconds) threshold->count = 0; else threshold->count -= threshold->count * diff / threshold->seconds; threshold->count += PF_THRESHOLD_MULT; threshold->last = t; } static int pf_check_threshold(struct pf_threshold *threshold) { return (threshold->count > threshold->limit); } static int pf_src_connlimit(struct pf_state **state) { struct pf_overload_entry *pfoe; int bad = 0; PF_STATE_LOCK_ASSERT(*state); (*state)->src_node->conn++; (*state)->src.tcp_est = 1; pf_add_threshold(&(*state)->src_node->conn_rate); if ((*state)->rule.ptr->max_src_conn && (*state)->rule.ptr->max_src_conn < (*state)->src_node->conn) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); bad++; } if ((*state)->rule.ptr->max_src_conn_rate.limit && pf_check_threshold(&(*state)->src_node->conn_rate)) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); bad++; } if (!bad) return (0); /* Kill this state. */ (*state)->timeout = PFTM_PURGE; (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; if ((*state)->rule.ptr->overload_tbl == NULL) return (1); /* Schedule overloading and flushing task. */ pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); if (pfoe == NULL) return (1); /* too bad :( */ bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); pfoe->af = (*state)->key[PF_SK_WIRE]->af; pfoe->rule = (*state)->rule.ptr; pfoe->dir = (*state)->direction; PF_OVERLOADQ_LOCK(); SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); PF_OVERLOADQ_UNLOCK(); taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); return (1); } static void pf_overload_task(void *v, int pending) { struct pf_overload_head queue; struct pfr_addr p; struct pf_overload_entry *pfoe, *pfoe1; uint32_t killed = 0; CURVNET_SET((struct vnet *)v); PF_OVERLOADQ_LOCK(); queue = V_pf_overloadqueue; SLIST_INIT(&V_pf_overloadqueue); PF_OVERLOADQ_UNLOCK(); bzero(&p, sizeof(p)); SLIST_FOREACH(pfoe, &queue, next) { counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("%s: blocking address ", __func__); pf_print_host(&pfoe->addr, 0, pfoe->af); printf("\n"); } p.pfra_af = pfoe->af; switch (pfoe->af) { #ifdef INET case AF_INET: p.pfra_net = 32; p.pfra_ip4addr = pfoe->addr.v4; break; #endif #ifdef INET6 case AF_INET6: p.pfra_net = 128; p.pfra_ip6addr = pfoe->addr.v6; break; #endif } PF_RULES_WLOCK(); pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); PF_RULES_WUNLOCK(); } /* * Remove those entries, that don't need flushing. */ SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) if (pfoe->rule->flush == 0) { SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); free(pfoe, M_PFTEMP); } else counter_u64_add( V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); /* If nothing to flush, return. */ if (SLIST_EMPTY(&queue)) { CURVNET_RESTORE(); return; } for (int i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state_key *sk; struct pf_state *s; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { sk = s->key[PF_SK_WIRE]; SLIST_FOREACH(pfoe, &queue, next) if (sk->af == pfoe->af && ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || pfoe->rule == s->rule.ptr) && ((pfoe->dir == PF_OUT && PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || (pfoe->dir == PF_IN && PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { s->timeout = PFTM_PURGE; s->src.state = s->dst.state = TCPS_CLOSED; killed++; } } PF_HASHROW_UNLOCK(ih); } SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) free(pfoe, M_PFTEMP); if (V_pf_status.debug >= PF_DEBUG_MISC) printf("%s: %u states killed", __func__, killed); CURVNET_RESTORE(); } /* * Can return locked on failure, so that we can consistently * allocate and insert a new one. */ struct pf_src_node * pf_find_src_node(struct pf_addr *src, struct pf_rule *rule, sa_family_t af, int returnlocked) { struct pf_srchash *sh; struct pf_src_node *n; counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); sh = &V_pf_srchash[pf_hashsrc(src, af)]; PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) if (n->rule.ptr == rule && n->af == af && ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) break; if (n != NULL) { n->states++; PF_HASHROW_UNLOCK(sh); } else if (returnlocked == 0) PF_HASHROW_UNLOCK(sh); return (n); } static int pf_insert_src_node(struct pf_src_node **sn, struct pf_rule *rule, struct pf_addr *src, sa_family_t af) { KASSERT((rule->rule_flag & PFRULE_RULESRCTRACK || rule->rpool.opts & PF_POOL_STICKYADDR), ("%s for non-tracking rule %p", __func__, rule)); if (*sn == NULL) *sn = pf_find_src_node(src, rule, af, 1); if (*sn == NULL) { struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; PF_HASHROW_ASSERT(sh); if (!rule->max_src_nodes || counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); else counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 1); if ((*sn) == NULL) { PF_HASHROW_UNLOCK(sh); return (-1); } pf_init_threshold(&(*sn)->conn_rate, rule->max_src_conn_rate.limit, rule->max_src_conn_rate.seconds); (*sn)->af = af; (*sn)->rule.ptr = rule; PF_ACPY(&(*sn)->addr, src, af); LIST_INSERT_HEAD(&sh->nodes, *sn, entry); (*sn)->creation = time_uptime; (*sn)->ruletype = rule->action; (*sn)->states = 1; if ((*sn)->rule.ptr != NULL) counter_u64_add((*sn)->rule.ptr->src_nodes, 1); PF_HASHROW_UNLOCK(sh); counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); } else { if (rule->max_src_states && (*sn)->states >= rule->max_src_states) { counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 1); return (-1); } } return (0); } void pf_unlink_src_node(struct pf_src_node *src) { PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); LIST_REMOVE(src, entry); if (src->rule.ptr) counter_u64_add(src->rule.ptr->src_nodes, -1); } u_int pf_free_src_nodes(struct pf_src_node_list *head) { struct pf_src_node *sn, *tmp; u_int count = 0; LIST_FOREACH_SAFE(sn, head, entry, tmp) { uma_zfree(V_pf_sources_z, sn); count++; } counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); return (count); } void pf_mtag_initialize() { pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, UMA_ALIGN_PTR, 0); } /* Per-vnet data storage structures initialization. */ void pf_initialize() { struct pf_keyhash *kh; struct pf_idhash *ih; struct pf_srchash *sh; u_int i; if (pf_hashsize == 0 || !powerof2(pf_hashsize)) pf_hashsize = PF_HASHSIZ; if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) pf_srchashsize = PF_SRCHASHSIZ; V_pf_hashseed = arc4random(); /* States and state keys storage. */ V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_state), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); V_pf_state_key_z = uma_zcreate("pf state keys", sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_NOWAIT | M_ZERO); V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { printf("pf: Unable to allocate memory for " "state_hashsize %lu.\n", pf_hashsize); free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); pf_hashsize = PF_HASHSIZ; V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); } pf_hashmask = pf_hashsize - 1; for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; i++, kh++, ih++) { mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); } /* Source nodes. */ V_pf_sources_z = uma_zcreate("pf source nodes", sizeof(struct pf_src_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); V_pf_srchash = mallocarray(pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); if (V_pf_srchash == NULL) { printf("pf: Unable to allocate memory for " "source_hashsize %lu.\n", pf_srchashsize); pf_srchashsize = PF_SRCHASHSIZ; V_pf_srchash = mallocarray(pf_srchashsize, sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); } pf_srchashmask = pf_srchashsize - 1; for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); /* ALTQ */ TAILQ_INIT(&V_pf_altqs[0]); TAILQ_INIT(&V_pf_altqs[1]); + TAILQ_INIT(&V_pf_altqs[2]); + TAILQ_INIT(&V_pf_altqs[3]); TAILQ_INIT(&V_pf_pabuf); V_pf_altqs_active = &V_pf_altqs[0]; - V_pf_altqs_inactive = &V_pf_altqs[1]; + V_pf_altq_ifs_active = &V_pf_altqs[1]; + V_pf_altqs_inactive = &V_pf_altqs[2]; + V_pf_altq_ifs_inactive = &V_pf_altqs[3]; /* Send & overload+flush queues. */ STAILQ_INIT(&V_pf_sendqueue); SLIST_INIT(&V_pf_overloadqueue); TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); /* Unlinked, but may be referenced rules. */ TAILQ_INIT(&V_pf_unlinked_rules); } void pf_mtag_cleanup() { uma_zdestroy(pf_mtag_z); } void pf_cleanup() { struct pf_keyhash *kh; struct pf_idhash *ih; struct pf_srchash *sh; struct pf_send_entry *pfse, *next; u_int i; for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; i++, kh++, ih++) { KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", __func__)); KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", __func__)); mtx_destroy(&kh->lock); mtx_destroy(&ih->lock); } free(V_pf_keyhash, M_PFHASH); free(V_pf_idhash, M_PFHASH); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { KASSERT(LIST_EMPTY(&sh->nodes), ("%s: source node hash not empty", __func__)); mtx_destroy(&sh->lock); } free(V_pf_srchash, M_PFHASH); STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { m_freem(pfse->pfse_m); free(pfse, M_PFTEMP); } uma_zdestroy(V_pf_sources_z); uma_zdestroy(V_pf_state_z); uma_zdestroy(V_pf_state_key_z); } static int pf_mtag_uminit(void *mem, int size, int how) { struct m_tag *t; t = (struct m_tag *)mem; t->m_tag_cookie = MTAG_ABI_COMPAT; t->m_tag_id = PACKET_TAG_PF; t->m_tag_len = sizeof(struct pf_mtag); t->m_tag_free = pf_mtag_free; return (0); } static void pf_mtag_free(struct m_tag *t) { uma_zfree(pf_mtag_z, t); } struct pf_mtag * pf_get_mtag(struct mbuf *m) { struct m_tag *mtag; if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) return ((struct pf_mtag *)(mtag + 1)); mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); if (mtag == NULL) return (NULL); bzero(mtag + 1, sizeof(struct pf_mtag)); m_tag_prepend(m, mtag); return ((struct pf_mtag *)(mtag + 1)); } static int pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, struct pf_state *s) { struct pf_keyhash *khs, *khw, *kh; struct pf_state_key *sk, *cur; struct pf_state *si, *olds = NULL; int idx; KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); /* * We need to lock hash slots of both keys. To avoid deadlock * we always lock the slot with lower address first. Unlock order * isn't important. * * We also need to lock ID hash slot before dropping key * locks. On success we return with ID hash slot locked. */ if (skw == sks) { khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; PF_HASHROW_LOCK(khs); } else { khs = &V_pf_keyhash[pf_hashkey(sks)]; khw = &V_pf_keyhash[pf_hashkey(skw)]; if (khs == khw) { PF_HASHROW_LOCK(khs); } else if (khs < khw) { PF_HASHROW_LOCK(khs); PF_HASHROW_LOCK(khw); } else { PF_HASHROW_LOCK(khw); PF_HASHROW_LOCK(khs); } } #define KEYS_UNLOCK() do { \ if (khs != khw) { \ PF_HASHROW_UNLOCK(khs); \ PF_HASHROW_UNLOCK(khw); \ } else \ PF_HASHROW_UNLOCK(khs); \ } while (0) /* * First run: start with wire key. */ sk = skw; kh = khw; idx = PF_SK_WIRE; keyattach: LIST_FOREACH(cur, &kh->keys, entry) if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) break; if (cur != NULL) { /* Key exists. Check for same kif, if none, add to key. */ TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; PF_HASHROW_LOCK(ih); if (si->kif == s->kif && si->direction == s->direction) { if (sk->proto == IPPROTO_TCP && si->src.state >= TCPS_FIN_WAIT_2 && si->dst.state >= TCPS_FIN_WAIT_2) { /* * New state matches an old >FIN_WAIT_2 * state. We can't drop key hash locks, * thus we can't unlink it properly. * * As a workaround we drop it into * TCPS_CLOSED state, schedule purge * ASAP and push it into the very end * of the slot TAILQ, so that it won't * conflict with our new state. */ si->src.state = si->dst.state = TCPS_CLOSED; si->timeout = PFTM_PURGE; olds = si; } else { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: %s key attach " "failed on %s: ", (idx == PF_SK_WIRE) ? "wire" : "stack", s->kif->pfik_name); pf_print_state_parts(s, (idx == PF_SK_WIRE) ? sk : NULL, (idx == PF_SK_STACK) ? sk : NULL); printf(", existing: "); pf_print_state_parts(si, (idx == PF_SK_WIRE) ? sk : NULL, (idx == PF_SK_STACK) ? sk : NULL); printf("\n"); } PF_HASHROW_UNLOCK(ih); KEYS_UNLOCK(); uma_zfree(V_pf_state_key_z, sk); if (idx == PF_SK_STACK) pf_detach_state(s); return (EEXIST); /* collision! */ } } PF_HASHROW_UNLOCK(ih); } uma_zfree(V_pf_state_key_z, sk); s->key[idx] = cur; } else { LIST_INSERT_HEAD(&kh->keys, sk, entry); s->key[idx] = sk; } stateattach: /* List is sorted, if-bound states before floating. */ if (s->kif == V_pfi_all) TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); else TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); if (olds) { TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, key_list[idx]); olds = NULL; } /* * Attach done. See how should we (or should not?) * attach a second key. */ if (sks == skw) { s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; idx = PF_SK_STACK; sks = NULL; goto stateattach; } else if (sks != NULL) { /* * Continue attaching with stack key. */ sk = sks; kh = khs; idx = PF_SK_STACK; sks = NULL; goto keyattach; } PF_STATE_LOCK(s); KEYS_UNLOCK(); KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, ("%s failure", __func__)); return (0); #undef KEYS_UNLOCK } static void pf_detach_state(struct pf_state *s) { struct pf_state_key *sks = s->key[PF_SK_STACK]; struct pf_keyhash *kh; if (sks != NULL) { kh = &V_pf_keyhash[pf_hashkey(sks)]; PF_HASHROW_LOCK(kh); if (s->key[PF_SK_STACK] != NULL) pf_state_key_detach(s, PF_SK_STACK); /* * If both point to same key, then we are done. */ if (sks == s->key[PF_SK_WIRE]) { pf_state_key_detach(s, PF_SK_WIRE); PF_HASHROW_UNLOCK(kh); return; } PF_HASHROW_UNLOCK(kh); } if (s->key[PF_SK_WIRE] != NULL) { kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; PF_HASHROW_LOCK(kh); if (s->key[PF_SK_WIRE] != NULL) pf_state_key_detach(s, PF_SK_WIRE); PF_HASHROW_UNLOCK(kh); } } static void pf_state_key_detach(struct pf_state *s, int idx) { struct pf_state_key *sk = s->key[idx]; #ifdef INVARIANTS struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; PF_HASHROW_ASSERT(kh); #endif TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); s->key[idx] = NULL; if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { LIST_REMOVE(sk, entry); uma_zfree(V_pf_state_key_z, sk); } } static int pf_state_key_ctor(void *mem, int size, void *arg, int flags) { struct pf_state_key *sk = mem; bzero(sk, sizeof(struct pf_state_key_cmp)); TAILQ_INIT(&sk->states[PF_SK_WIRE]); TAILQ_INIT(&sk->states[PF_SK_STACK]); return (0); } struct pf_state_key * pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) { struct pf_state_key *sk; sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); if (sk == NULL) return (NULL); PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); sk->port[pd->sidx] = sport; sk->port[pd->didx] = dport; sk->proto = pd->proto; sk->af = pd->af; return (sk); } struct pf_state_key * pf_state_key_clone(struct pf_state_key *orig) { struct pf_state_key *sk; sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); if (sk == NULL) return (NULL); bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); return (sk); } int pf_state_insert(struct pfi_kif *kif, struct pf_state_key *skw, struct pf_state_key *sks, struct pf_state *s) { struct pf_idhash *ih; struct pf_state *cur; int error; KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), ("%s: sks not pristine", __func__)); KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), ("%s: skw not pristine", __func__)); KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); s->kif = kif; if (s->id == 0 && s->creatorid == 0) { /* XXX: should be atomic, but probability of collision low */ if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) V_pf_stateid[curcpu] = 1; s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; s->id = htobe64(s->id); s->creatorid = V_pf_status.hostid; } /* Returns with ID locked on success. */ if ((error = pf_state_key_attach(skw, sks, s)) != 0) return (error); ih = &V_pf_idhash[PF_IDHASH(s)]; PF_HASHROW_ASSERT(ih); LIST_FOREACH(cur, &ih->states, entry) if (cur->id == s->id && cur->creatorid == s->creatorid) break; if (cur != NULL) { PF_HASHROW_UNLOCK(ih); if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: state ID collision: " "id: %016llx creatorid: %08x\n", (unsigned long long)be64toh(s->id), ntohl(s->creatorid)); } pf_detach_state(s); return (EEXIST); } LIST_INSERT_HEAD(&ih->states, s, entry); /* One for keys, one for ID hash. */ refcount_init(&s->refs, 2); counter_u64_add(V_pf_status.fcounters[FCNT_STATE_INSERT], 1); if (V_pfsync_insert_state_ptr != NULL) V_pfsync_insert_state_ptr(s); /* Returns locked. */ return (0); } /* * Find state by ID: returns with locked row on success. */ struct pf_state * pf_find_state_byid(uint64_t id, uint32_t creatorid) { struct pf_idhash *ih; struct pf_state *s; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) if (s->id == id && s->creatorid == creatorid) break; if (s == NULL) PF_HASHROW_UNLOCK(ih); return (s); } /* * Find state by key. * Returns with ID hash slot locked on success. */ static struct pf_state * pf_find_state(struct pfi_kif *kif, struct pf_state_key_cmp *key, u_int dir) { struct pf_keyhash *kh; struct pf_state_key *sk; struct pf_state *s; int idx; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; PF_HASHROW_LOCK(kh); LIST_FOREACH(sk, &kh->keys, entry) if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) break; if (sk == NULL) { PF_HASHROW_UNLOCK(kh); return (NULL); } idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); /* List is sorted, if-bound states before floating ones. */ TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) if (s->kif == V_pfi_all || s->kif == kif) { PF_STATE_LOCK(s); PF_HASHROW_UNLOCK(kh); if (s->timeout >= PFTM_MAX) { /* * State is either being processed by * pf_unlink_state() in an other thread, or * is scheduled for immediate expiry. */ PF_STATE_UNLOCK(s); return (NULL); } return (s); } PF_HASHROW_UNLOCK(kh); return (NULL); } struct pf_state * pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) { struct pf_keyhash *kh; struct pf_state_key *sk; struct pf_state *s, *ret = NULL; int idx, inout = 0; counter_u64_add(V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; PF_HASHROW_LOCK(kh); LIST_FOREACH(sk, &kh->keys, entry) if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) break; if (sk == NULL) { PF_HASHROW_UNLOCK(kh); return (NULL); } switch (dir) { case PF_IN: idx = PF_SK_WIRE; break; case PF_OUT: idx = PF_SK_STACK; break; case PF_INOUT: idx = PF_SK_WIRE; inout = 1; break; default: panic("%s: dir %u", __func__, dir); } second_run: TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { if (more == NULL) { PF_HASHROW_UNLOCK(kh); return (s); } if (ret) (*more)++; else ret = s; } if (inout == 1) { inout = 0; idx = PF_SK_STACK; goto second_run; } PF_HASHROW_UNLOCK(kh); return (ret); } /* END state table stuff */ static void pf_send(struct pf_send_entry *pfse) { PF_SENDQ_LOCK(); STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); PF_SENDQ_UNLOCK(); swi_sched(V_pf_swi_cookie, 0); } void pf_intr(void *v) { struct pf_send_head queue; struct pf_send_entry *pfse, *next; CURVNET_SET((struct vnet *)v); PF_SENDQ_LOCK(); queue = V_pf_sendqueue; STAILQ_INIT(&V_pf_sendqueue); PF_SENDQ_UNLOCK(); STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { switch (pfse->pfse_type) { #ifdef INET case PFSE_IP: ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL); break; case PFSE_ICMP: icmp_error(pfse->pfse_m, pfse->icmpopts.type, pfse->icmpopts.code, 0, pfse->icmpopts.mtu); break; #endif /* INET */ #ifdef INET6 case PFSE_IP6: ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, NULL, NULL); break; case PFSE_ICMP6: icmp6_error(pfse->pfse_m, pfse->icmpopts.type, pfse->icmpopts.code, pfse->icmpopts.mtu); break; #endif /* INET6 */ default: panic("%s: unknown type", __func__); } free(pfse, M_PFTEMP); } CURVNET_RESTORE(); } void pf_purge_thread(void *unused __unused) { VNET_ITERATOR_DECL(vnet_iter); sx_xlock(&pf_end_lock); while (pf_end_threads == 0) { sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", hz / 10); VNET_LIST_RLOCK(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); /* Wait until V_pf_default_rule is initialized. */ if (V_pf_vnet_active == 0) { CURVNET_RESTORE(); continue; } /* * Process 1/interval fraction of the state * table every run. */ V_pf_purge_idx = pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); /* * Purge other expired types every * PFTM_INTERVAL seconds. */ if (V_pf_purge_idx == 0) { /* * Order is important: * - states and src nodes reference rules * - states and rules reference kifs */ pf_purge_expired_fragments(); pf_purge_expired_src_nodes(); pf_purge_unlinked_rules(); pfi_kif_purge(); } CURVNET_RESTORE(); } VNET_LIST_RUNLOCK(); } pf_end_threads++; sx_xunlock(&pf_end_lock); kproc_exit(0); } void pf_unload_vnet_purge(void) { /* * To cleanse up all kifs and rules we need * two runs: first one clears reference flags, * then pf_purge_expired_states() doesn't * raise them, and then second run frees. */ pf_purge_unlinked_rules(); pfi_kif_purge(); /* * Now purge everything. */ pf_purge_expired_states(0, pf_hashmask); pf_purge_fragments(UINT_MAX); pf_purge_expired_src_nodes(); /* * Now all kifs & rules should be unreferenced, * thus should be successfully freed. */ pf_purge_unlinked_rules(); pfi_kif_purge(); } u_int32_t pf_state_expires(const struct pf_state *state) { u_int32_t timeout; u_int32_t start; u_int32_t end; u_int32_t states; /* handle all PFTM_* > PFTM_MAX here */ if (state->timeout == PFTM_PURGE) return (time_uptime); KASSERT(state->timeout != PFTM_UNLINKED, ("pf_state_expires: timeout == PFTM_UNLINKED")); KASSERT((state->timeout < PFTM_MAX), ("pf_state_expires: timeout > PFTM_MAX")); timeout = state->rule.ptr->timeout[state->timeout]; if (!timeout) timeout = V_pf_default_rule.timeout[state->timeout]; start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; if (start) { end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; states = counter_u64_fetch(state->rule.ptr->states_cur); } else { start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; states = V_pf_status.states; } if (end && states > start && start < end) { if (states < end) { timeout = (u_int64_t)timeout * (end - states) / (end - start); return (state->expire + timeout); } else return (time_uptime); } return (state->expire + timeout); } void pf_purge_expired_src_nodes() { struct pf_src_node_list freelist; struct pf_srchash *sh; struct pf_src_node *cur, *next; int i; LIST_INIT(&freelist); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) if (cur->states == 0 && cur->expire <= time_uptime) { pf_unlink_src_node(cur); LIST_INSERT_HEAD(&freelist, cur, entry); } else if (cur->rule.ptr != NULL) cur->rule.ptr->rule_flag |= PFRULE_REFS; PF_HASHROW_UNLOCK(sh); } pf_free_src_nodes(&freelist); V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); } static void pf_src_tree_remove_state(struct pf_state *s) { struct pf_src_node *sn; struct pf_srchash *sh; uint32_t timeout; timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? s->rule.ptr->timeout[PFTM_SRC_NODE] : V_pf_default_rule.timeout[PFTM_SRC_NODE]; if (s->src_node != NULL) { sn = s->src_node; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (s->src.tcp_est) --sn->conn; if (--sn->states == 0) sn->expire = time_uptime + timeout; PF_HASHROW_UNLOCK(sh); } if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { sn = s->nat_src_node; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (--sn->states == 0) sn->expire = time_uptime + timeout; PF_HASHROW_UNLOCK(sh); } s->src_node = s->nat_src_node = NULL; } /* * Unlink and potentilly free a state. Function may be * called with ID hash row locked, but always returns * unlocked, since it needs to go through key hash locking. */ int pf_unlink_state(struct pf_state *s, u_int flags) { struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; if ((flags & PF_ENTER_LOCKED) == 0) PF_HASHROW_LOCK(ih); else PF_HASHROW_ASSERT(ih); if (s->timeout == PFTM_UNLINKED) { /* * State is being processed * by pf_unlink_state() in * an other thread. */ PF_HASHROW_UNLOCK(ih); return (0); /* XXXGL: undefined actually */ } if (s->src.state == PF_TCPS_PROXY_DST) { /* XXX wire key the right one? */ pf_send_tcp(NULL, s->rule.ptr, s->key[PF_SK_WIRE]->af, &s->key[PF_SK_WIRE]->addr[1], &s->key[PF_SK_WIRE]->addr[0], s->key[PF_SK_WIRE]->port[1], s->key[PF_SK_WIRE]->port[0], s->src.seqhi, s->src.seqlo + 1, TH_RST|TH_ACK, 0, 0, 0, 1, s->tag, NULL); } LIST_REMOVE(s, entry); pf_src_tree_remove_state(s); if (V_pfsync_delete_state_ptr != NULL) V_pfsync_delete_state_ptr(s); STATE_DEC_COUNTERS(s); s->timeout = PFTM_UNLINKED; PF_HASHROW_UNLOCK(ih); pf_detach_state(s); /* pf_state_insert() initialises refs to 2, so we can never release the * last reference here, only in pf_release_state(). */ (void)refcount_release(&s->refs); return (pf_release_state(s)); } void pf_free_state(struct pf_state *cur) { KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, cur->timeout)); pf_normalize_tcp_cleanup(cur); uma_zfree(V_pf_state_z, cur); counter_u64_add(V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); } /* * Called only from pf_purge_thread(), thus serialized. */ static u_int pf_purge_expired_states(u_int i, int maxcheck) { struct pf_idhash *ih; struct pf_state *s; V_pf_status.states = uma_zone_get_cur(V_pf_state_z); /* * Go through hash and unlink states that expire now. */ while (maxcheck > 0) { ih = &V_pf_idhash[i]; /* only take the lock if we expect to do work */ if (!LIST_EMPTY(&ih->states)) { relock: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (pf_state_expires(s) <= time_uptime) { V_pf_status.states -= pf_unlink_state(s, PF_ENTER_LOCKED); goto relock; } s->rule.ptr->rule_flag |= PFRULE_REFS; if (s->nat_rule.ptr != NULL) s->nat_rule.ptr->rule_flag |= PFRULE_REFS; if (s->anchor.ptr != NULL) s->anchor.ptr->rule_flag |= PFRULE_REFS; s->kif->pfik_flags |= PFI_IFLAG_REFS; if (s->rt_kif) s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; } PF_HASHROW_UNLOCK(ih); } /* Return when we hit end of hash. */ if (++i > pf_hashmask) { V_pf_status.states = uma_zone_get_cur(V_pf_state_z); return (0); } maxcheck--; } V_pf_status.states = uma_zone_get_cur(V_pf_state_z); return (i); } static void pf_purge_unlinked_rules() { struct pf_rulequeue tmpq; struct pf_rule *r, *r1; /* * If we have overloading task pending, then we'd * better skip purging this time. There is a tiny * probability that overloading task references * an already unlinked rule. */ PF_OVERLOADQ_LOCK(); if (!SLIST_EMPTY(&V_pf_overloadqueue)) { PF_OVERLOADQ_UNLOCK(); return; } PF_OVERLOADQ_UNLOCK(); /* * Do naive mark-and-sweep garbage collecting of old rules. * Reference flag is raised by pf_purge_expired_states() * and pf_purge_expired_src_nodes(). * * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, * use a temporary queue. */ TAILQ_INIT(&tmpq); PF_UNLNKDRULES_LOCK(); TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { if (!(r->rule_flag & PFRULE_REFS)) { TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); TAILQ_INSERT_TAIL(&tmpq, r, entries); } else r->rule_flag &= ~PFRULE_REFS; } PF_UNLNKDRULES_UNLOCK(); if (!TAILQ_EMPTY(&tmpq)) { PF_RULES_WLOCK(); TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { TAILQ_REMOVE(&tmpq, r, entries); pf_free_rule(r); } PF_RULES_WUNLOCK(); } } void pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: { u_int32_t a = ntohl(addr->addr32[0]); printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, (a>>8)&255, a&255); if (p) { p = ntohs(p); printf(":%u", p); } break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { u_int16_t b; u_int8_t i, curstart, curend, maxstart, maxend; curstart = curend = maxstart = maxend = 255; for (i = 0; i < 8; i++) { if (!addr->addr16[i]) { if (curstart == 255) curstart = i; curend = i; } else { if ((curend - curstart) > (maxend - maxstart)) { maxstart = curstart; maxend = curend; } curstart = curend = 255; } } if ((curend - curstart) > (maxend - maxstart)) { maxstart = curstart; maxend = curend; } for (i = 0; i < 8; i++) { if (i >= maxstart && i <= maxend) { if (i == 0) printf(":"); if (i == maxend) printf(":"); } else { b = ntohs(addr->addr16[i]); printf("%x", b); if (i < 7) printf(":"); } } if (p) { p = ntohs(p); printf("[%u]", p); } break; } #endif /* INET6 */ } } void pf_print_state(struct pf_state *s) { pf_print_state_parts(s, NULL, NULL); } static void pf_print_state_parts(struct pf_state *s, struct pf_state_key *skwp, struct pf_state_key *sksp) { struct pf_state_key *skw, *sks; u_int8_t proto, dir; /* Do our best to fill these, but they're skipped if NULL */ skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); proto = skw ? skw->proto : (sks ? sks->proto : 0); dir = s ? s->direction : 0; switch (proto) { case IPPROTO_IPV4: printf("IPv4"); break; case IPPROTO_IPV6: printf("IPv6"); break; case IPPROTO_TCP: printf("TCP"); break; case IPPROTO_UDP: printf("UDP"); break; case IPPROTO_ICMP: printf("ICMP"); break; case IPPROTO_ICMPV6: printf("ICMPv6"); break; default: printf("%u", proto); break; } switch (dir) { case PF_IN: printf(" in"); break; case PF_OUT: printf(" out"); break; } if (skw) { printf(" wire: "); pf_print_host(&skw->addr[0], skw->port[0], skw->af); printf(" "); pf_print_host(&skw->addr[1], skw->port[1], skw->af); } if (sks) { printf(" stack: "); if (sks != skw) { pf_print_host(&sks->addr[0], sks->port[0], sks->af); printf(" "); pf_print_host(&sks->addr[1], sks->port[1], sks->af); } else printf("-"); } if (s) { if (proto == IPPROTO_TCP) { printf(" [lo=%u high=%u win=%u modulator=%u", s->src.seqlo, s->src.seqhi, s->src.max_win, s->src.seqdiff); if (s->src.wscale && s->dst.wscale) printf(" wscale=%u", s->src.wscale & PF_WSCALE_MASK); printf("]"); printf(" [lo=%u high=%u win=%u modulator=%u", s->dst.seqlo, s->dst.seqhi, s->dst.max_win, s->dst.seqdiff); if (s->src.wscale && s->dst.wscale) printf(" wscale=%u", s->dst.wscale & PF_WSCALE_MASK); printf("]"); } printf(" %u:%u", s->src.state, s->dst.state); } } void pf_print_flags(u_int8_t f) { if (f) printf(" "); if (f & TH_FIN) printf("F"); if (f & TH_SYN) printf("S"); if (f & TH_RST) printf("R"); if (f & TH_PUSH) printf("P"); if (f & TH_ACK) printf("A"); if (f & TH_URG) printf("U"); if (f & TH_ECE) printf("E"); if (f & TH_CWR) printf("W"); } #define PF_SET_SKIP_STEPS(i) \ do { \ while (head[i] != cur) { \ head[i]->skip[i].ptr = cur; \ head[i] = TAILQ_NEXT(head[i], entries); \ } \ } while (0) void pf_calc_skip_steps(struct pf_rulequeue *rules) { struct pf_rule *cur, *prev, *head[PF_SKIP_COUNT]; int i; cur = TAILQ_FIRST(rules); prev = cur; for (i = 0; i < PF_SKIP_COUNT; ++i) head[i] = cur; while (cur != NULL) { if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) PF_SET_SKIP_STEPS(PF_SKIP_IFP); if (cur->direction != prev->direction) PF_SET_SKIP_STEPS(PF_SKIP_DIR); if (cur->af != prev->af) PF_SET_SKIP_STEPS(PF_SKIP_AF); if (cur->proto != prev->proto) PF_SET_SKIP_STEPS(PF_SKIP_PROTO); if (cur->src.neg != prev->src.neg || pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); if (cur->src.port[0] != prev->src.port[0] || cur->src.port[1] != prev->src.port[1] || cur->src.port_op != prev->src.port_op) PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); if (cur->dst.neg != prev->dst.neg || pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); if (cur->dst.port[0] != prev->dst.port[0] || cur->dst.port[1] != prev->dst.port[1] || cur->dst.port_op != prev->dst.port_op) PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); prev = cur; cur = TAILQ_NEXT(cur, entries); } for (i = 0; i < PF_SKIP_COUNT; ++i) PF_SET_SKIP_STEPS(i); } static int pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) { if (aw1->type != aw2->type) return (1); switch (aw1->type) { case PF_ADDR_ADDRMASK: case PF_ADDR_RANGE: if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) return (1); if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) return (1); return (0); case PF_ADDR_DYNIFTL: return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); case PF_ADDR_NOROUTE: case PF_ADDR_URPFFAILED: return (0); case PF_ADDR_TABLE: return (aw1->p.tbl != aw2->p.tbl); default: printf("invalid address type: %d\n", aw1->type); return (1); } } /** * Checksum updates are a little complicated because the checksum in the TCP/UDP * header isn't always a full checksum. In some cases (i.e. output) it's a * pseudo-header checksum, which is a partial checksum over src/dst IP * addresses, protocol number and length. * * That means we have the following cases: * * Input or forwarding: we don't have TSO, the checksum fields are full * checksums, we need to update the checksum whenever we change anything. * * Output (i.e. the checksum is a pseudo-header checksum): * x The field being updated is src/dst address or affects the length of * the packet. We need to update the pseudo-header checksum (note that this * checksum is not ones' complement). * x Some other field is being modified (e.g. src/dst port numbers): We * don't have to update anything. **/ u_int16_t pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) { u_int32_t l; if (udp && !cksum) return (0x0000); l = cksum + old - new; l = (l >> 16) + (l & 65535); l = l & 65535; if (udp && !l) return (0xFFFF); return (l); } u_int16_t pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) { if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) return (cksum); return (pf_cksum_fixup(cksum, old, new, udp)); } static void pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, sa_family_t af) { struct pf_addr ao; u_int16_t po = *p; PF_ACPY(&ao, a, af); PF_ACPY(a, an, af); if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) *pc = ~*pc; *p = pn; switch (af) { #ifdef INET case AF_INET: *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, ao.addr16[0], an->addr16[0], 0), ao.addr16[1], an->addr16[1], 0); *p = pn; *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u); *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); break; #endif /* INET */ #ifdef INET6 case AF_INET6: *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*pc, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u), ao.addr16[2], an->addr16[2], u), ao.addr16[3], an->addr16[3], u), ao.addr16[4], an->addr16[4], u), ao.addr16[5], an->addr16[5], u), ao.addr16[6], an->addr16[6], u), ao.addr16[7], an->addr16[7], u); *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); break; #endif /* INET6 */ } if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) { *pc = ~*pc; if (! *pc) *pc = 0xffff; } } /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ void pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) { u_int32_t ao; memcpy(&ao, a, sizeof(ao)); memcpy(a, &an, sizeof(u_int32_t)); *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), ao % 65536, an % 65536, u); } void pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) { u_int32_t ao; memcpy(&ao, a, sizeof(ao)); memcpy(a, &an, sizeof(u_int32_t)); *c = pf_proto_cksum_fixup(m, pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), ao % 65536, an % 65536, udp); } #ifdef INET6 static void pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) { struct pf_addr ao; PF_ACPY(&ao, a, AF_INET6); PF_ACPY(a, an, AF_INET6); *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*c, ao.addr16[0], an->addr16[0], u), ao.addr16[1], an->addr16[1], u), ao.addr16[2], an->addr16[2], u), ao.addr16[3], an->addr16[3], u), ao.addr16[4], an->addr16[4], u), ao.addr16[5], an->addr16[5], u), ao.addr16[6], an->addr16[6], u), ao.addr16[7], an->addr16[7], u); } #endif /* INET6 */ static void pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) { struct pf_addr oia, ooa; PF_ACPY(&oia, ia, af); if (oa) PF_ACPY(&ooa, oa, af); /* Change inner protocol port, fix inner protocol checksum. */ if (ip != NULL) { u_int16_t oip = *ip; u_int32_t opc; if (pc != NULL) opc = *pc; *ip = np; if (pc != NULL) *pc = pf_cksum_fixup(*pc, oip, *ip, u); *ic = pf_cksum_fixup(*ic, oip, *ip, 0); if (pc != NULL) *ic = pf_cksum_fixup(*ic, opc, *pc, 0); } /* Change inner ip address, fix inner ip and icmp checksums. */ PF_ACPY(ia, na, af); switch (af) { #ifdef INET case AF_INET: { u_int32_t oh2c = *h2c; *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, oia.addr16[0], ia->addr16[0], 0), oia.addr16[1], ia->addr16[1], 0); *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, oia.addr16[0], ia->addr16[0], 0), oia.addr16[1], ia->addr16[1], 0); *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); break; } #endif /* INET */ #ifdef INET6 case AF_INET6: *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*ic, oia.addr16[0], ia->addr16[0], u), oia.addr16[1], ia->addr16[1], u), oia.addr16[2], ia->addr16[2], u), oia.addr16[3], ia->addr16[3], u), oia.addr16[4], ia->addr16[4], u), oia.addr16[5], ia->addr16[5], u), oia.addr16[6], ia->addr16[6], u), oia.addr16[7], ia->addr16[7], u); break; #endif /* INET6 */ } /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ if (oa) { PF_ACPY(oa, na, af); switch (af) { #ifdef INET case AF_INET: *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, ooa.addr16[0], oa->addr16[0], 0), ooa.addr16[1], oa->addr16[1], 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( pf_cksum_fixup(pf_cksum_fixup(*ic, ooa.addr16[0], oa->addr16[0], u), ooa.addr16[1], oa->addr16[1], u), ooa.addr16[2], oa->addr16[2], u), ooa.addr16[3], oa->addr16[3], u), ooa.addr16[4], oa->addr16[4], u), ooa.addr16[5], oa->addr16[5], u), ooa.addr16[6], oa->addr16[6], u), ooa.addr16[7], oa->addr16[7], u); break; #endif /* INET6 */ } } } /* * Need to modulate the sequence numbers in the TCP SACK option * (credits to Krzysztof Pfaff for report and patch) */ static int pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, struct tcphdr *th, struct pf_state_peer *dst) { int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; u_int8_t opts[TCP_MAXOLEN], *opt = opts; int copyback = 0, i, olen; struct sackblk sack; #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) if (hlen < TCPOLEN_SACKLEN || !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) return 0; while (hlen >= TCPOLEN_SACKLEN) { olen = opt[1]; switch (*opt) { case TCPOPT_EOL: /* FALLTHROUGH */ case TCPOPT_NOP: opt++; hlen--; break; case TCPOPT_SACK: if (olen > hlen) olen = hlen; if (olen >= TCPOLEN_SACKLEN) { for (i = 2; i + TCPOLEN_SACK <= olen; i += TCPOLEN_SACK) { memcpy(&sack, &opt[i], sizeof(sack)); pf_change_proto_a(m, &sack.start, &th->th_sum, htonl(ntohl(sack.start) - dst->seqdiff), 0); pf_change_proto_a(m, &sack.end, &th->th_sum, htonl(ntohl(sack.end) - dst->seqdiff), 0); memcpy(&opt[i], &sack, sizeof(sack)); } copyback = 1; } /* FALLTHROUGH */ default: if (olen < 2) olen = 2; hlen -= olen; opt += olen; } } if (copyback) m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); return (copyback); } static void pf_send_tcp(struct mbuf *replyto, const struct pf_rule *r, sa_family_t af, const struct pf_addr *saddr, const struct pf_addr *daddr, u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, u_int16_t rtag, struct ifnet *ifp) { struct pf_send_entry *pfse; struct mbuf *m; int len, tlen; #ifdef INET struct ip *h = NULL; #endif /* INET */ #ifdef INET6 struct ip6_hdr *h6 = NULL; #endif /* INET6 */ struct tcphdr *th; char *opt; struct pf_mtag *pf_mtag; len = 0; th = NULL; /* maximum segment size tcp option */ tlen = sizeof(struct tcphdr); if (mss) tlen += 4; switch (af) { #ifdef INET case AF_INET: len = sizeof(struct ip) + tlen; break; #endif /* INET */ #ifdef INET6 case AF_INET6: len = sizeof(struct ip6_hdr) + tlen; break; #endif /* INET6 */ default: panic("%s: unsupported af %d", __func__, af); } /* Allocate outgoing queue entry, mbuf and mbuf tag. */ pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); if (pfse == NULL) return; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { free(pfse, M_PFTEMP); return; } #ifdef MAC mac_netinet_firewall_send(m); #endif if ((pf_mtag = pf_get_mtag(m)) == NULL) { free(pfse, M_PFTEMP); m_freem(m); return; } if (tag) m->m_flags |= M_SKIP_FIREWALL; pf_mtag->tag = rtag; if (r != NULL && r->rtableid >= 0) M_SETFIB(m, r->rtableid); #ifdef ALTQ if (r != NULL && r->qid) { pf_mtag->qid = r->qid; /* add hints for ecn */ pf_mtag->hdr = mtod(m, struct ip *); } #endif /* ALTQ */ m->m_data += max_linkhdr; m->m_pkthdr.len = m->m_len = len; m->m_pkthdr.rcvif = NULL; bzero(m->m_data, len); switch (af) { #ifdef INET case AF_INET: h = mtod(m, struct ip *); /* IP header fields included in the TCP checksum */ h->ip_p = IPPROTO_TCP; h->ip_len = htons(tlen); h->ip_src.s_addr = saddr->v4.s_addr; h->ip_dst.s_addr = daddr->v4.s_addr; th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); break; #endif /* INET */ #ifdef INET6 case AF_INET6: h6 = mtod(m, struct ip6_hdr *); /* IP header fields included in the TCP checksum */ h6->ip6_nxt = IPPROTO_TCP; h6->ip6_plen = htons(tlen); memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); break; #endif /* INET6 */ } /* TCP header */ th->th_sport = sport; th->th_dport = dport; th->th_seq = htonl(seq); th->th_ack = htonl(ack); th->th_off = tlen >> 2; th->th_flags = flags; th->th_win = htons(win); if (mss) { opt = (char *)(th + 1); opt[0] = TCPOPT_MAXSEG; opt[1] = 4; HTONS(mss); bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); } switch (af) { #ifdef INET case AF_INET: /* TCP checksum */ th->th_sum = in_cksum(m, len); /* Finish the IP header */ h->ip_v = 4; h->ip_hl = sizeof(*h) >> 2; h->ip_tos = IPTOS_LOWDELAY; h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); h->ip_len = htons(len); h->ip_ttl = ttl ? ttl : V_ip_defttl; h->ip_sum = 0; pfse->pfse_type = PFSE_IP; break; #endif /* INET */ #ifdef INET6 case AF_INET6: /* TCP checksum */ th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen); h6->ip6_vfc |= IPV6_VERSION; h6->ip6_hlim = IPV6_DEFHLIM; pfse->pfse_type = PFSE_IP6; break; #endif /* INET6 */ } pfse->pfse_m = m; pf_send(pfse); } static void pf_return(struct pf_rule *r, struct pf_rule *nr, struct pf_pdesc *pd, struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, struct pfi_kif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, u_short *reason) { struct pf_addr * const saddr = pd->src; struct pf_addr * const daddr = pd->dst; sa_family_t af = pd->af; /* undo NAT changes, if they have taken place */ if (nr != NULL) { PF_ACPY(saddr, &sk->addr[pd->sidx], af); PF_ACPY(daddr, &sk->addr[pd->didx], af); if (pd->sport) *pd->sport = sk->port[pd->sidx]; if (pd->dport) *pd->dport = sk->port[pd->didx]; if (pd->proto_sum) *pd->proto_sum = bproto_sum; if (pd->ip_sum) *pd->ip_sum = bip_sum; m_copyback(m, off, hdrlen, pd->hdr.any); } if (pd->proto == IPPROTO_TCP && ((r->rule_flag & PFRULE_RETURNRST) || (r->rule_flag & PFRULE_RETURN)) && !(th->th_flags & TH_RST)) { u_int32_t ack = ntohl(th->th_seq) + pd->p_len; int len = 0; #ifdef INET struct ip *h4; #endif #ifdef INET6 struct ip6_hdr *h6; #endif switch (af) { #ifdef INET case AF_INET: h4 = mtod(m, struct ip *); len = ntohs(h4->ip_len) - off; break; #endif #ifdef INET6 case AF_INET6: h6 = mtod(m, struct ip6_hdr *); len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); break; #endif } if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) REASON_SET(reason, PFRES_PROTCKSUM); else { if (th->th_flags & TH_SYN) ack++; if (th->th_flags & TH_FIN) ack++; pf_send_tcp(m, r, af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, r->return_ttl, 1, 0, kif->pfik_ifp); } } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && r->return_icmp) pf_send_icmp(m, r->return_icmp >> 8, r->return_icmp & 255, af, r); else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && r->return_icmp6) pf_send_icmp(m, r->return_icmp6 >> 8, r->return_icmp6 & 255, af, r); } static int pf_ieee8021q_setpcp(struct mbuf *m, u_int8_t prio) { struct m_tag *mtag; KASSERT(prio <= PF_PRIO_MAX, ("%s with invalid pcp", __func__)); mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_OUT, NULL); if (mtag == NULL) { mtag = m_tag_alloc(MTAG_8021Q, MTAG_8021Q_PCP_OUT, sizeof(uint8_t), M_NOWAIT); if (mtag == NULL) return (ENOMEM); m_tag_prepend(m, mtag); } *(uint8_t *)(mtag + 1) = prio; return (0); } static int pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) { struct m_tag *mtag; u_int8_t mpcp; mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); if (mtag == NULL) return (0); if (prio == PF_PRIO_ZERO) prio = 0; mpcp = *(uint8_t *)(mtag + 1); return (mpcp == prio); } static void pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, struct pf_rule *r) { struct pf_send_entry *pfse; struct mbuf *m0; struct pf_mtag *pf_mtag; /* Allocate outgoing queue entry, mbuf and mbuf tag. */ pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); if (pfse == NULL) return; if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { free(pfse, M_PFTEMP); return; } if ((pf_mtag = pf_get_mtag(m0)) == NULL) { free(pfse, M_PFTEMP); return; } /* XXX: revisit */ m0->m_flags |= M_SKIP_FIREWALL; if (r->rtableid >= 0) M_SETFIB(m0, r->rtableid); #ifdef ALTQ if (r->qid) { pf_mtag->qid = r->qid; /* add hints for ecn */ pf_mtag->hdr = mtod(m0, struct ip *); } #endif /* ALTQ */ switch (af) { #ifdef INET case AF_INET: pfse->pfse_type = PFSE_ICMP; break; #endif /* INET */ #ifdef INET6 case AF_INET6: pfse->pfse_type = PFSE_ICMP6; break; #endif /* INET6 */ } pfse->pfse_m = m0; pfse->icmpopts.type = type; pfse->icmpopts.code = code; pf_send(pfse); } /* * Return 1 if the addresses a and b match (with mask m), otherwise return 0. * If n is 0, they match if they are equal. If n is != 0, they match if they * are different. */ int pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, struct pf_addr *b, sa_family_t af) { int match = 0; switch (af) { #ifdef INET case AF_INET: if ((a->addr32[0] & m->addr32[0]) == (b->addr32[0] & m->addr32[0])) match++; break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (((a->addr32[0] & m->addr32[0]) == (b->addr32[0] & m->addr32[0])) && ((a->addr32[1] & m->addr32[1]) == (b->addr32[1] & m->addr32[1])) && ((a->addr32[2] & m->addr32[2]) == (b->addr32[2] & m->addr32[2])) && ((a->addr32[3] & m->addr32[3]) == (b->addr32[3] & m->addr32[3]))) match++; break; #endif /* INET6 */ } if (match) { if (n) return (0); else return (1); } else { if (n) return (1); else return (0); } } /* * Return 1 if b <= a <= e, otherwise return 0. */ int pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, struct pf_addr *a, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) return (0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: { int i; /* check a >= b */ for (i = 0; i < 4; ++i) if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) break; else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) return (0); /* check a <= e */ for (i = 0; i < 4; ++i) if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) break; else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) return (0); break; } #endif /* INET6 */ } return (1); } static int pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) { switch (op) { case PF_OP_IRG: return ((p > a1) && (p < a2)); case PF_OP_XRG: return ((p < a1) || (p > a2)); case PF_OP_RRG: return ((p >= a1) && (p <= a2)); case PF_OP_EQ: return (p == a1); case PF_OP_NE: return (p != a1); case PF_OP_LT: return (p < a1); case PF_OP_LE: return (p <= a1); case PF_OP_GT: return (p > a1); case PF_OP_GE: return (p >= a1); } return (0); /* never reached */ } int pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) { NTOHS(a1); NTOHS(a2); NTOHS(p); return (pf_match(op, a1, a2, p)); } static int pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) { if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) return (0); return (pf_match(op, a1, a2, u)); } static int pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) { if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) return (0); return (pf_match(op, a1, a2, g)); } int pf_match_tag(struct mbuf *m, struct pf_rule *r, int *tag, int mtag) { if (*tag == -1) *tag = mtag; return ((!r->match_tag_not && r->match_tag == *tag) || (r->match_tag_not && r->match_tag != *tag)); } int pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) { KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) return (ENOMEM); pd->pf_mtag->tag = tag; return (0); } #define PF_ANCHOR_STACKSIZE 32 struct pf_anchor_stackframe { struct pf_ruleset *rs; struct pf_rule *r; /* XXX: + match bit */ struct pf_anchor *child; }; /* * XXX: We rely on malloc(9) returning pointer aligned addresses. */ #define PF_ANCHORSTACK_MATCH 0x00000001 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) #define PF_ANCHOR_RULE(f) (struct pf_rule *) \ ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ } while (0) void pf_step_into_anchor(struct pf_anchor_stackframe *stack, int *depth, struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, int *match) { struct pf_anchor_stackframe *f; PF_RULES_RASSERT(); if (match) *match = 0; if (*depth >= PF_ANCHOR_STACKSIZE) { printf("%s: anchor stack overflow on %s\n", __func__, (*r)->anchor->name); *r = TAILQ_NEXT(*r, entries); return; } else if (*depth == 0 && a != NULL) *a = *r; f = stack + (*depth)++; f->rs = *rs; f->r = *r; if ((*r)->anchor_wildcard) { struct pf_anchor_node *parent = &(*r)->anchor->children; if ((f->child = RB_MIN(pf_anchor_node, parent)) == NULL) { *r = NULL; return; } *rs = &f->child->ruleset; } else { f->child = NULL; *rs = &(*r)->anchor->ruleset; } *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); } int pf_step_out_of_anchor(struct pf_anchor_stackframe *stack, int *depth, struct pf_ruleset **rs, int n, struct pf_rule **r, struct pf_rule **a, int *match) { struct pf_anchor_stackframe *f; struct pf_rule *fr; int quick = 0; PF_RULES_RASSERT(); do { if (*depth <= 0) break; f = stack + *depth - 1; fr = PF_ANCHOR_RULE(f); if (f->child != NULL) { struct pf_anchor_node *parent; /* * This block traverses through * a wildcard anchor. */ parent = &fr->anchor->children; if (match != NULL && *match) { /* * If any of "*" matched, then * "foo/ *" matched, mark frame * appropriately. */ PF_ANCHOR_SET_MATCH(f); *match = 0; } f->child = RB_NEXT(pf_anchor_node, parent, f->child); if (f->child != NULL) { *rs = &f->child->ruleset; *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); if (*r == NULL) continue; else break; } } (*depth)--; if (*depth == 0 && a != NULL) *a = NULL; *rs = f->rs; if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) quick = fr->quick; *r = TAILQ_NEXT(fr, entries); } while (*r == NULL); return (quick); } #ifdef INET6 void pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); break; #endif /* INET */ case AF_INET6: naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); break; } } void pf_addr_inc(struct pf_addr *addr, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); break; #endif /* INET */ case AF_INET6: if (addr->addr32[3] == 0xffffffff) { addr->addr32[3] = 0; if (addr->addr32[2] == 0xffffffff) { addr->addr32[2] = 0; if (addr->addr32[1] == 0xffffffff) { addr->addr32[1] = 0; addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); } else addr->addr32[1] = htonl(ntohl(addr->addr32[1]) + 1); } else addr->addr32[2] = htonl(ntohl(addr->addr32[2]) + 1); } else addr->addr32[3] = htonl(ntohl(addr->addr32[3]) + 1); break; } } #endif /* INET6 */ int pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) { struct pf_addr *saddr, *daddr; u_int16_t sport, dport; struct inpcbinfo *pi; struct inpcb *inp; pd->lookup.uid = UID_MAX; pd->lookup.gid = GID_MAX; switch (pd->proto) { case IPPROTO_TCP: if (pd->hdr.tcp == NULL) return (-1); sport = pd->hdr.tcp->th_sport; dport = pd->hdr.tcp->th_dport; pi = &V_tcbinfo; break; case IPPROTO_UDP: if (pd->hdr.udp == NULL) return (-1); sport = pd->hdr.udp->uh_sport; dport = pd->hdr.udp->uh_dport; pi = &V_udbinfo; break; default: return (-1); } if (direction == PF_IN) { saddr = pd->src; daddr = pd->dst; } else { u_int16_t p; p = sport; sport = dport; dport = p; saddr = pd->dst; daddr = pd->src; } switch (pd->af) { #ifdef INET case AF_INET: inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, dport, INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) { inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) return (-1); } break; #endif /* INET */ #ifdef INET6 case AF_INET6: inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, dport, INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) { inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, dport, INPLOOKUP_WILDCARD | INPLOOKUP_RLOCKPCB, NULL, m); if (inp == NULL) return (-1); } break; #endif /* INET6 */ default: return (-1); } INP_RLOCK_ASSERT(inp); pd->lookup.uid = inp->inp_cred->cr_uid; pd->lookup.gid = inp->inp_cred->cr_groups[0]; INP_RUNLOCK(inp); return (1); } static u_int8_t pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) { int hlen; u_int8_t hdr[60]; u_int8_t *opt, optlen; u_int8_t wscale = 0; hlen = th_off << 2; /* hlen <= sizeof(hdr) */ if (hlen <= sizeof(struct tcphdr)) return (0); if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) return (0); opt = hdr + sizeof(struct tcphdr); hlen -= sizeof(struct tcphdr); while (hlen >= 3) { switch (*opt) { case TCPOPT_EOL: case TCPOPT_NOP: ++opt; --hlen; break; case TCPOPT_WINDOW: wscale = opt[2]; if (wscale > TCP_MAX_WINSHIFT) wscale = TCP_MAX_WINSHIFT; wscale |= PF_WSCALE_FLAG; /* FALLTHROUGH */ default: optlen = opt[1]; if (optlen < 2) optlen = 2; hlen -= optlen; opt += optlen; break; } } return (wscale); } static u_int16_t pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) { int hlen; u_int8_t hdr[60]; u_int8_t *opt, optlen; u_int16_t mss = V_tcp_mssdflt; hlen = th_off << 2; /* hlen <= sizeof(hdr) */ if (hlen <= sizeof(struct tcphdr)) return (0); if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) return (0); opt = hdr + sizeof(struct tcphdr); hlen -= sizeof(struct tcphdr); while (hlen >= TCPOLEN_MAXSEG) { switch (*opt) { case TCPOPT_EOL: case TCPOPT_NOP: ++opt; --hlen; break; case TCPOPT_MAXSEG: bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); NTOHS(mss); /* FALLTHROUGH */ default: optlen = opt[1]; if (optlen < 2) optlen = 2; hlen -= optlen; opt += optlen; break; } } return (mss); } static u_int16_t pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) { #ifdef INET struct nhop4_basic nh4; #endif /* INET */ #ifdef INET6 struct nhop6_basic nh6; struct in6_addr dst6; uint32_t scopeid; #endif /* INET6 */ int hlen = 0; uint16_t mss = 0; switch (af) { #ifdef INET case AF_INET: hlen = sizeof(struct ip); if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) == 0) mss = nh4.nh_mtu - hlen - sizeof(struct tcphdr); break; #endif /* INET */ #ifdef INET6 case AF_INET6: hlen = sizeof(struct ip6_hdr); in6_splitscope(&addr->v6, &dst6, &scopeid); if (fib6_lookup_nh_basic(rtableid, &dst6, scopeid, 0,0,&nh6)==0) mss = nh6.nh_mtu - hlen - sizeof(struct tcphdr); break; #endif /* INET6 */ } mss = max(V_tcp_mssdflt, mss); mss = min(mss, offer); mss = max(mss, 64); /* sanity - at least max opt space */ return (mss); } static u_int32_t pf_tcp_iss(struct pf_pdesc *pd) { MD5_CTX ctx; u_int32_t digest[4]; if (V_pf_tcp_secret_init == 0) { read_random(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); MD5Init(&V_pf_tcp_secret_ctx); MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); V_pf_tcp_secret_init = 1; } ctx = V_pf_tcp_secret_ctx; MD5Update(&ctx, (char *)&pd->hdr.tcp->th_sport, sizeof(u_short)); MD5Update(&ctx, (char *)&pd->hdr.tcp->th_dport, sizeof(u_short)); if (pd->af == AF_INET6) { MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); } else { MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); } MD5Final((u_char *)digest, &ctx); V_pf_tcp_iss_off += 4096; #define ISN_RANDOM_INCREMENT (4096 - 1) return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + V_pf_tcp_iss_off); #undef ISN_RANDOM_INCREMENT } static int pf_test_rule(struct pf_rule **rm, struct pf_state **sm, int direction, struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm, struct inpcb *inp) { struct pf_rule *nr = NULL; struct pf_addr * const saddr = pd->src; struct pf_addr * const daddr = pd->dst; sa_family_t af = pd->af; struct pf_rule *r, *a = NULL; struct pf_ruleset *ruleset = NULL; struct pf_src_node *nsn = NULL; struct tcphdr *th = pd->hdr.tcp; struct pf_state_key *sk = NULL, *nk = NULL; u_short reason; int rewrite = 0, hdrlen = 0; int tag = -1, rtableid = -1; int asd = 0; int match = 0; int state_icmp = 0; u_int16_t sport = 0, dport = 0; u_int16_t bproto_sum = 0, bip_sum = 0; u_int8_t icmptype = 0, icmpcode = 0; struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; PF_RULES_RASSERT(); if (inp != NULL) { INP_LOCK_ASSERT(inp); pd->lookup.uid = inp->inp_cred->cr_uid; pd->lookup.gid = inp->inp_cred->cr_groups[0]; pd->lookup.done = 1; } switch (pd->proto) { case IPPROTO_TCP: sport = th->th_sport; dport = th->th_dport; hdrlen = sizeof(*th); break; case IPPROTO_UDP: sport = pd->hdr.udp->uh_sport; dport = pd->hdr.udp->uh_dport; hdrlen = sizeof(*pd->hdr.udp); break; #ifdef INET case IPPROTO_ICMP: if (pd->af != AF_INET) break; sport = dport = pd->hdr.icmp->icmp_id; hdrlen = sizeof(*pd->hdr.icmp); icmptype = pd->hdr.icmp->icmp_type; icmpcode = pd->hdr.icmp->icmp_code; if (icmptype == ICMP_UNREACH || icmptype == ICMP_SOURCEQUENCH || icmptype == ICMP_REDIRECT || icmptype == ICMP_TIMXCEED || icmptype == ICMP_PARAMPROB) state_icmp++; break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: if (af != AF_INET6) break; sport = dport = pd->hdr.icmp6->icmp6_id; hdrlen = sizeof(*pd->hdr.icmp6); icmptype = pd->hdr.icmp6->icmp6_type; icmpcode = pd->hdr.icmp6->icmp6_code; if (icmptype == ICMP6_DST_UNREACH || icmptype == ICMP6_PACKET_TOO_BIG || icmptype == ICMP6_TIME_EXCEEDED || icmptype == ICMP6_PARAM_PROB) state_icmp++; break; #endif /* INET6 */ default: sport = dport = hdrlen = 0; break; } r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); /* check packet for BINAT/NAT/RDR */ if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { KASSERT(sk != NULL, ("%s: null sk", __func__)); KASSERT(nk != NULL, ("%s: null nk", __func__)); if (pd->ip_sum) bip_sum = *pd->ip_sum; switch (pd->proto) { case IPPROTO_TCP: bproto_sum = th->th_sum; pd->proto_sum = &th->th_sum; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || nk->port[pd->sidx] != sport) { pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 0, af); pd->sport = &th->th_sport; sport = th->th_sport; } if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || nk->port[pd->didx] != dport) { pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], nk->port[pd->didx], 0, af); dport = th->th_dport; pd->dport = &th->th_dport; } rewrite++; break; case IPPROTO_UDP: bproto_sum = pd->hdr.udp->uh_sum; pd->proto_sum = &pd->hdr.udp->uh_sum; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || nk->port[pd->sidx] != sport) { pf_change_ap(m, saddr, &pd->hdr.udp->uh_sport, pd->ip_sum, &pd->hdr.udp->uh_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 1, af); sport = pd->hdr.udp->uh_sport; pd->sport = &pd->hdr.udp->uh_sport; } if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || nk->port[pd->didx] != dport) { pf_change_ap(m, daddr, &pd->hdr.udp->uh_dport, pd->ip_sum, &pd->hdr.udp->uh_sum, &nk->addr[pd->didx], nk->port[pd->didx], 1, af); dport = pd->hdr.udp->uh_dport; pd->dport = &pd->hdr.udp->uh_dport; } rewrite++; break; #ifdef INET case IPPROTO_ICMP: nk->port[0] = nk->port[1]; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); if (nk->port[1] != pd->hdr.icmp->icmp_id) { pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( pd->hdr.icmp->icmp_cksum, sport, nk->port[1], 0); pd->hdr.icmp->icmp_id = nk->port[1]; pd->sport = &pd->hdr.icmp->icmp_id; } m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: nk->port[0] = nk->port[1]; if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->sidx], 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->didx], 0); rewrite++; break; #endif /* INET */ default: switch (af) { #ifdef INET case AF_INET: if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) PF_ACPY(saddr, &nk->addr[pd->sidx], af); if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) PF_ACPY(daddr, &nk->addr[pd->didx], af); break; #endif /* INET */ } break; } if (nr->natpass) r = NULL; pd->nat_rule = nr; } while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != direction) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != af) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != pd->proto) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, saddr, af, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; /* tcp/udp only. port_op always 0 in other cases */ else if (r->src.port_op && !pf_match_port(r->src.port_op, r->src.port[0], r->src.port[1], sport)) r = r->skip[PF_SKIP_SRC_PORT].ptr; else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; /* tcp/udp only. port_op always 0 in other cases */ else if (r->dst.port_op && !pf_match_port(r->dst.port_op, r->dst.port[0], r->dst.port[1], dport)) r = r->skip[PF_SKIP_DST_PORT].ptr; /* icmp only. type always 0 in other cases */ else if (r->type && r->type != icmptype + 1) r = TAILQ_NEXT(r, entries); /* icmp only. type always 0 in other cases */ else if (r->code && r->code != icmpcode + 1) r = TAILQ_NEXT(r, entries); else if (r->tos && !(r->tos == pd->tos)) r = TAILQ_NEXT(r, entries); else if (r->rule_flag & PFRULE_FRAGMENT) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_TCP && (r->flagset & th->th_flags) != r->flags) r = TAILQ_NEXT(r, entries); /* tcp/udp only. uid.op always 0 in other cases */ else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = pf_socket_lookup(direction, pd, m), 1)) && !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], pd->lookup.uid)) r = TAILQ_NEXT(r, entries); /* tcp/udp only. gid.op always 0 in other cases */ else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = pf_socket_lookup(direction, pd, m), 1)) && !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], pd->lookup.gid)) r = TAILQ_NEXT(r, entries); else if (r->prio && !pf_match_ieee8021q_pcp(r->prio, m)) r = TAILQ_NEXT(r, entries); else if (r->prob && r->prob <= arc4random()) r = TAILQ_NEXT(r, entries); else if (r->match_tag && !pf_match_tag(m, r, &tag, pd->pf_mtag ? pd->pf_mtag->tag : 0)) r = TAILQ_NEXT(r, entries); else if (r->os_fingerprint != PF_OSFP_ANY && (pd->proto != IPPROTO_TCP || !pf_osfp_match( pf_osfp_fingerprint(pd, m, off, th), r->os_fingerprint))) r = TAILQ_NEXT(r, entries); else { if (r->tag) tag = r->tag; if (r->rtableid >= 0) rtableid = r->rtableid; if (r->anchor == NULL) { match = 1; *rm = r; *am = a; *rsm = ruleset; if ((*rm)->quick) break; r = TAILQ_NEXT(r, entries); } else pf_step_into_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match); } if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match)) break; } r = *rm; a = *am; ruleset = *rsm; REASON_SET(&reason, PFRES_MATCH); if (r->log || (nr != NULL && nr->log)) { if (rewrite) m_copyback(m, off, hdrlen, pd->hdr.any); PFLOG_PACKET(kif, m, af, direction, reason, r->log ? r : nr, a, ruleset, pd, 1); } if ((r->action == PF_DROP) && ((r->rule_flag & PFRULE_RETURNRST) || (r->rule_flag & PFRULE_RETURNICMP) || (r->rule_flag & PFRULE_RETURN))) { pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, bip_sum, hdrlen, &reason); } if (r->action == PF_DROP) goto cleanup; if (tag > 0 && pf_tag_packet(m, pd, tag)) { REASON_SET(&reason, PFRES_MEMORY); goto cleanup; } if (rtableid >= 0) M_SETFIB(m, rtableid); if (!state_icmp && (r->keep_state || nr != NULL || (pd->flags & PFDESC_TCP_NORM))) { int action; action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, hdrlen); if (action != PF_PASS) { if (action == PF_DROP && (r->rule_flag & PFRULE_RETURN)) pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, bip_sum, hdrlen, &reason); return (action); } } else { if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); } /* copy back packet headers if we performed NAT operations */ if (rewrite) m_copyback(m, off, hdrlen, pd->hdr.any); if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && direction == PF_OUT && V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) /* * We want the state created, but we dont * want to send this in case a partner * firewall has to know about it to allow * replies through it. */ return (PF_DEFER); return (PF_PASS); cleanup: if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); return (PF_DROP); } static int pf_create_state(struct pf_rule *r, struct pf_rule *nr, struct pf_rule *a, struct pf_pdesc *pd, struct pf_src_node *nsn, struct pf_state_key *nk, struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, u_int16_t dport, int *rewrite, struct pfi_kif *kif, struct pf_state **sm, int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) { struct pf_state *s = NULL; struct pf_src_node *sn = NULL; struct tcphdr *th = pd->hdr.tcp; u_int16_t mss = V_tcp_mssdflt; u_short reason; /* check maximums */ if (r->max_states && (counter_u64_fetch(r->states_cur) >= r->max_states)) { counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); REASON_SET(&reason, PFRES_MAXSTATES); goto csfailed; } /* src node for filter rule */ if ((r->rule_flag & PFRULE_SRCTRACK || r->rpool.opts & PF_POOL_STICKYADDR) && pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { REASON_SET(&reason, PFRES_SRCLIMIT); goto csfailed; } /* src node for translation rule */ if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { REASON_SET(&reason, PFRES_SRCLIMIT); goto csfailed; } s = uma_zalloc(V_pf_state_z, M_NOWAIT | M_ZERO); if (s == NULL) { REASON_SET(&reason, PFRES_MEMORY); goto csfailed; } s->rule.ptr = r; s->nat_rule.ptr = nr; s->anchor.ptr = a; STATE_INC_COUNTERS(s); if (r->allow_opts) s->state_flags |= PFSTATE_ALLOWOPTS; if (r->rule_flag & PFRULE_STATESLOPPY) s->state_flags |= PFSTATE_SLOPPY; s->log = r->log & PF_LOG_ALL; s->sync_state = PFSYNC_S_NONE; if (nr != NULL) s->log |= nr->log & PF_LOG_ALL; switch (pd->proto) { case IPPROTO_TCP: s->src.seqlo = ntohl(th->th_seq); s->src.seqhi = s->src.seqlo + pd->p_len + 1; if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && r->keep_state == PF_STATE_MODULATE) { /* Generate sequence number modulator */ if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 0) s->src.seqdiff = 1; pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(s->src.seqlo + s->src.seqdiff), 0); *rewrite = 1; } else s->src.seqdiff = 0; if (th->th_flags & TH_SYN) { s->src.seqhi++; s->src.wscale = pf_get_wscale(m, off, th->th_off, pd->af); } s->src.max_win = MAX(ntohs(th->th_win), 1); if (s->src.wscale & PF_WSCALE_MASK) { /* Remove scale factor from initial window */ int win = s->src.max_win; win += 1 << (s->src.wscale & PF_WSCALE_MASK); s->src.max_win = (win - 1) >> (s->src.wscale & PF_WSCALE_MASK); } if (th->th_flags & TH_FIN) s->src.seqhi++; s->dst.seqhi = 1; s->dst.max_win = 1; s->src.state = TCPS_SYN_SENT; s->dst.state = TCPS_CLOSED; s->timeout = PFTM_TCP_FIRST_PACKET; break; case IPPROTO_UDP: s->src.state = PFUDPS_SINGLE; s->dst.state = PFUDPS_NO_TRAFFIC; s->timeout = PFTM_UDP_FIRST_PACKET; break; case IPPROTO_ICMP: #ifdef INET6 case IPPROTO_ICMPV6: #endif s->timeout = PFTM_ICMP_FIRST_PACKET; break; default: s->src.state = PFOTHERS_SINGLE; s->dst.state = PFOTHERS_NO_TRAFFIC; s->timeout = PFTM_OTHER_FIRST_PACKET; } if (r->rt) { if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { REASON_SET(&reason, PFRES_MAPFAILED); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); goto csfailed; } s->rt_kif = r->rpool.cur->kif; } s->creation = time_uptime; s->expire = time_uptime; if (sn != NULL) s->src_node = sn; if (nsn != NULL) { /* XXX We only modify one side for now. */ PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); s->nat_src_node = nsn; } if (pd->proto == IPPROTO_TCP) { if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, off, pd, th, &s->src, &s->dst)) { REASON_SET(&reason, PFRES_MEMORY); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, &s->src, &s->dst, rewrite)) { /* This really shouldn't happen!!! */ DPFPRINTF(PF_DEBUG_URGENT, ("pf_normalize_tcp_stateful failed on first pkt")); pf_normalize_tcp_cleanup(s); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } } s->direction = pd->dir; /* * sk/nk could already been setup by pf_get_translation(). */ if (nr == NULL) { KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", __func__, nr, sk, nk)); sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); if (sk == NULL) goto csfailed; nk = sk; } else KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", __func__, nr, sk, nk)); /* Swap sk/nk for PF_OUT. */ if (pf_state_insert(BOUND_IFACE(r, kif), (pd->dir == PF_IN) ? sk : nk, (pd->dir == PF_IN) ? nk : sk, s)) { if (pd->proto == IPPROTO_TCP) pf_normalize_tcp_cleanup(s); REASON_SET(&reason, PFRES_STATEINS); pf_src_tree_remove_state(s); STATE_DEC_COUNTERS(s); uma_zfree(V_pf_state_z, s); return (PF_DROP); } else *sm = s; if (tag > 0) s->tag = tag; if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { s->src.state = PF_TCPS_PROXY_SRC; /* undo NAT changes, if they have taken place */ if (nr != NULL) { struct pf_state_key *skt = s->key[PF_SK_WIRE]; if (pd->dir == PF_OUT) skt = s->key[PF_SK_STACK]; PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); if (pd->sport) *pd->sport = skt->port[pd->sidx]; if (pd->dport) *pd->dport = skt->port[pd->didx]; if (pd->proto_sum) *pd->proto_sum = bproto_sum; if (pd->ip_sum) *pd->ip_sum = bip_sum; m_copyback(m, off, hdrlen, pd->hdr.any); } s->src.seqhi = htonl(arc4random()); /* Find mss option */ int rtid = M_GETFIB(m); mss = pf_get_mss(m, off, th->th_off, pd->af); mss = pf_calc_mss(pd->src, pd->af, rtid, mss); mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); s->src.mss = mss; pf_send_tcp(NULL, r, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0, NULL); REASON_SET(&reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } return (PF_PASS); csfailed: if (sk != NULL) uma_zfree(V_pf_state_key_z, sk); if (nk != NULL) uma_zfree(V_pf_state_key_z, nk); if (sn != NULL) { struct pf_srchash *sh; sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; PF_HASHROW_LOCK(sh); if (--sn->states == 0 && sn->expire == 0) { pf_unlink_src_node(sn); uma_zfree(V_pf_sources_z, sn); counter_u64_add( V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); } PF_HASHROW_UNLOCK(sh); } if (nsn != sn && nsn != NULL) { struct pf_srchash *sh; sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; PF_HASHROW_LOCK(sh); if (--nsn->states == 0 && nsn->expire == 0) { pf_unlink_src_node(nsn); uma_zfree(V_pf_sources_z, nsn); counter_u64_add( V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); } PF_HASHROW_UNLOCK(sh); } return (PF_DROP); } static int pf_test_fragment(struct pf_rule **rm, int direction, struct pfi_kif *kif, struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_rule **am, struct pf_ruleset **rsm) { struct pf_rule *r, *a = NULL; struct pf_ruleset *ruleset = NULL; sa_family_t af = pd->af; u_short reason; int tag = -1; int asd = 0; int match = 0; struct pf_anchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; PF_RULES_RASSERT(); r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); while (r != NULL) { r->evaluations++; if (pfi_kif_match(r->kif, kif) == r->ifnot) r = r->skip[PF_SKIP_IFP].ptr; else if (r->direction && r->direction != direction) r = r->skip[PF_SKIP_DIR].ptr; else if (r->af && r->af != af) r = r->skip[PF_SKIP_AF].ptr; else if (r->proto && r->proto != pd->proto) r = r->skip[PF_SKIP_PROTO].ptr; else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, r->src.neg, kif, M_GETFIB(m))) r = r->skip[PF_SKIP_SRC_ADDR].ptr; else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, r->dst.neg, NULL, M_GETFIB(m))) r = r->skip[PF_SKIP_DST_ADDR].ptr; else if (r->tos && !(r->tos == pd->tos)) r = TAILQ_NEXT(r, entries); else if (r->os_fingerprint != PF_OSFP_ANY) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_UDP && (r->src.port_op || r->dst.port_op)) r = TAILQ_NEXT(r, entries); else if (pd->proto == IPPROTO_TCP && (r->src.port_op || r->dst.port_op || r->flagset)) r = TAILQ_NEXT(r, entries); else if ((pd->proto == IPPROTO_ICMP || pd->proto == IPPROTO_ICMPV6) && (r->type || r->code)) r = TAILQ_NEXT(r, entries); else if (r->prio && !pf_match_ieee8021q_pcp(r->prio, m)) r = TAILQ_NEXT(r, entries); else if (r->prob && r->prob <= (arc4random() % (UINT_MAX - 1) + 1)) r = TAILQ_NEXT(r, entries); else if (r->match_tag && !pf_match_tag(m, r, &tag, pd->pf_mtag ? pd->pf_mtag->tag : 0)) r = TAILQ_NEXT(r, entries); else { if (r->anchor == NULL) { match = 1; *rm = r; *am = a; *rsm = ruleset; if ((*rm)->quick) break; r = TAILQ_NEXT(r, entries); } else pf_step_into_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match); } if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, &ruleset, PF_RULESET_FILTER, &r, &a, &match)) break; } r = *rm; a = *am; ruleset = *rsm; REASON_SET(&reason, PFRES_MATCH); if (r->log) PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 1); if (r->action != PF_PASS) return (PF_DROP); if (tag > 0 && pf_tag_packet(m, pd, tag)) { REASON_SET(&reason, PFRES_MEMORY); return (PF_DROP); } return (PF_PASS); } static int pf_tcp_track_full(struct pf_state_peer *src, struct pf_state_peer *dst, struct pf_state **state, struct pfi_kif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, int *copyback) { struct tcphdr *th = pd->hdr.tcp; u_int16_t win = ntohs(th->th_win); u_int32_t ack, end, seq, orig_seq; u_int8_t sws, dws; int ackskew; if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { sws = src->wscale & PF_WSCALE_MASK; dws = dst->wscale & PF_WSCALE_MASK; } else sws = dws = 0; /* * Sequence tracking algorithm from Guido van Rooij's paper: * http://www.madison-gurkha.com/publications/tcp_filtering/ * tcp_filtering.ps */ orig_seq = seq = ntohl(th->th_seq); if (src->seqlo == 0) { /* First packet from this end. Set its state */ if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && src->scrub == NULL) { if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { REASON_SET(reason, PFRES_MEMORY); return (PF_DROP); } } /* Deferred generation of sequence number modulator */ if (dst->seqdiff && !src->seqdiff) { /* use random iss for the TCP server */ while ((src->seqdiff = arc4random() - seq) == 0) ; ack = ntohl(th->th_ack) - dst->seqdiff; pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + src->seqdiff), 0); pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); *copyback = 1; } else { ack = ntohl(th->th_ack); } end = seq + pd->p_len; if (th->th_flags & TH_SYN) { end++; if (dst->wscale & PF_WSCALE_FLAG) { src->wscale = pf_get_wscale(m, off, th->th_off, pd->af); if (src->wscale & PF_WSCALE_FLAG) { /* Remove scale factor from initial * window */ sws = src->wscale & PF_WSCALE_MASK; win = ((u_int32_t)win + (1 << sws) - 1) >> sws; dws = dst->wscale & PF_WSCALE_MASK; } else { /* fixup other window */ dst->max_win <<= dst->wscale & PF_WSCALE_MASK; /* in case of a retrans SYN|ACK */ dst->wscale = 0; } } } if (th->th_flags & TH_FIN) end++; src->seqlo = seq; if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; /* * May need to slide the window (seqhi may have been set by * the crappy stack check or if we picked up the connection * after establishment) */ if (src->seqhi == 1 || SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) src->seqhi = end + MAX(1, dst->max_win << dws); if (win > src->max_win) src->max_win = win; } else { ack = ntohl(th->th_ack) - dst->seqdiff; if (src->seqdiff) { /* Modulate sequence numbers */ pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + src->seqdiff), 0); pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); *copyback = 1; } end = seq + pd->p_len; if (th->th_flags & TH_SYN) end++; if (th->th_flags & TH_FIN) end++; } if ((th->th_flags & TH_ACK) == 0) { /* Let it pass through the ack skew check */ ack = dst->seqlo; } else if ((ack == 0 && (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || /* broken tcp stacks do not set ack */ (dst->state < TCPS_SYN_SENT)) { /* * Many stacks (ours included) will set the ACK number in an * FIN|ACK if the SYN times out -- no sequence to ACK. */ ack = dst->seqlo; } if (seq == end) { /* Ease sequencing restrictions on no data packets */ seq = src->seqlo; end = seq; } ackskew = dst->seqlo - ack; /* * Need to demodulate the sequence numbers in any TCP SACK options * (Selective ACK). We could optionally validate the SACK values * against the current ACK window, either forwards or backwards, but * I'm not confident that SACK has been implemented properly * everywhere. It wouldn't surprise me if several stacks accidentally * SACK too far backwards of previously ACKed data. There really aren't * any security implications of bad SACKing unless the target stack * doesn't validate the option length correctly. Someone trying to * spoof into a TCP connection won't bother blindly sending SACK * options anyway. */ if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { if (pf_modulate_sack(m, off, pd, th, dst)) *copyback = 1; } #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ if (SEQ_GEQ(src->seqhi, end) && /* Last octet inside other's window space */ SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && /* Retrans: not more than one window back */ (ackskew >= -MAXACKWINDOW) && /* Acking not more than one reassembled fragment backwards */ (ackskew <= (MAXACKWINDOW << sws)) && /* Acking not more than one window forward */ ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || (pd->flags & PFDESC_IP_REAS) == 0)) { /* Require an exact/+1 sequence match on resets when possible */ if (dst->scrub || src->scrub) { if (pf_normalize_tcp_stateful(m, off, pd, reason, th, *state, src, dst, copyback)) return (PF_DROP); } /* update max window */ if (src->max_win < win) src->max_win = win; /* synchronize sequencing */ if (SEQ_GT(end, src->seqlo)) src->seqlo = end; /* slide the window of what the other end can send */ if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) dst->seqhi = ack + MAX((win << sws), 1); /* update states */ if (th->th_flags & TH_SYN) if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_ACK) { if (dst->state == TCPS_SYN_SENT) { dst->state = TCPS_ESTABLISHED; if (src->state == TCPS_ESTABLISHED && (*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (dst->state == TCPS_CLOSING) dst->state = TCPS_FIN_WAIT_2; } if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* update expire time */ (*state)->expire = time_uptime; if (src->state >= TCPS_FIN_WAIT_2 && dst->state >= TCPS_FIN_WAIT_2) (*state)->timeout = PFTM_TCP_CLOSED; else if (src->state >= TCPS_CLOSING && dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_FIN_WAIT; else if (src->state < TCPS_ESTABLISHED || dst->state < TCPS_ESTABLISHED) (*state)->timeout = PFTM_TCP_OPENING; else if (src->state >= TCPS_CLOSING || dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_CLOSING; else (*state)->timeout = PFTM_TCP_ESTABLISHED; /* Fall through to PASS packet */ } else if ((dst->state < TCPS_SYN_SENT || dst->state >= TCPS_FIN_WAIT_2 || src->state >= TCPS_FIN_WAIT_2) && SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && /* Within a window forward of the originating packet */ SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { /* Within a window backward of the originating packet */ /* * This currently handles three situations: * 1) Stupid stacks will shotgun SYNs before their peer * replies. * 2) When PF catches an already established stream (the * firewall rebooted, the state table was flushed, routes * changed...) * 3) Packets get funky immediately after the connection * closes (this should catch Solaris spurious ACK|FINs * that web servers like to spew after a close) * * This must be a little more careful than the above code * since packet floods will also be caught here. We don't * update the TTL here to mitigate the damage of a packet * flood and so the same code can handle awkward establishment * and a loosened connection close. * In the establishment case, a correct peer response will * validate the connection, go through the normal state code * and keep updating the state TTL. */ if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: loose state match: "); pf_print_state(*state); pf_print_flags(th->th_flags); printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], (unsigned long long)(*state)->packets[1], pd->dir == PF_IN ? "in" : "out", pd->dir == (*state)->direction ? "fwd" : "rev"); } if (dst->scrub || src->scrub) { if (pf_normalize_tcp_stateful(m, off, pd, reason, th, *state, src, dst, copyback)) return (PF_DROP); } /* update max window */ if (src->max_win < win) src->max_win = win; /* synchronize sequencing */ if (SEQ_GT(end, src->seqlo)) src->seqlo = end; /* slide the window of what the other end can send */ if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) dst->seqhi = ack + MAX((win << sws), 1); /* * Cannot set dst->seqhi here since this could be a shotgunned * SYN and not an already established connection. */ if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* Fall through to PASS packet */ } else { if ((*state)->dst.state == TCPS_SYN_SENT && (*state)->src.state == TCPS_SYN_SENT) { /* Send RST for state mismatches during handshake */ if (!(th->th_flags & TH_RST)) pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), 0, TH_RST, 0, 0, (*state)->rule.ptr->return_ttl, 1, 0, kif->pfik_ifp); src->seqlo = 0; src->seqhi = 1; src->max_win = 1; } else if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: BAD state: "); pf_print_state(*state); pf_print_flags(th->th_flags); printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], (unsigned long long)(*state)->packets[1], pd->dir == PF_IN ? "in" : "out", pd->dir == (*state)->direction ? "fwd" : "rev"); printf("pf: State failure on: %c %c %c %c | %c %c\n", SEQ_GEQ(src->seqhi, end) ? ' ' : '1', SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? ' ': '2', (ackskew >= -MAXACKWINDOW) ? ' ' : '3', (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); } REASON_SET(reason, PFRES_BADSTATE); return (PF_DROP); } return (PF_PASS); } static int pf_tcp_track_sloppy(struct pf_state_peer *src, struct pf_state_peer *dst, struct pf_state **state, struct pf_pdesc *pd, u_short *reason) { struct tcphdr *th = pd->hdr.tcp; if (th->th_flags & TH_SYN) if (src->state < TCPS_SYN_SENT) src->state = TCPS_SYN_SENT; if (th->th_flags & TH_FIN) if (src->state < TCPS_CLOSING) src->state = TCPS_CLOSING; if (th->th_flags & TH_ACK) { if (dst->state == TCPS_SYN_SENT) { dst->state = TCPS_ESTABLISHED; if (src->state == TCPS_ESTABLISHED && (*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (dst->state == TCPS_CLOSING) { dst->state = TCPS_FIN_WAIT_2; } else if (src->state == TCPS_SYN_SENT && dst->state < TCPS_SYN_SENT) { /* * Handle a special sloppy case where we only see one * half of the connection. If there is a ACK after * the initial SYN without ever seeing a packet from * the destination, set the connection to established. */ dst->state = src->state = TCPS_ESTABLISHED; if ((*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } } else if (src->state == TCPS_CLOSING && dst->state == TCPS_ESTABLISHED && dst->seqlo == 0) { /* * Handle the closing of half connections where we * don't see the full bidirectional FIN/ACK+ACK * handshake. */ dst->state = TCPS_CLOSING; } } if (th->th_flags & TH_RST) src->state = dst->state = TCPS_TIME_WAIT; /* update expire time */ (*state)->expire = time_uptime; if (src->state >= TCPS_FIN_WAIT_2 && dst->state >= TCPS_FIN_WAIT_2) (*state)->timeout = PFTM_TCP_CLOSED; else if (src->state >= TCPS_CLOSING && dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_FIN_WAIT; else if (src->state < TCPS_ESTABLISHED || dst->state < TCPS_ESTABLISHED) (*state)->timeout = PFTM_TCP_OPENING; else if (src->state >= TCPS_CLOSING || dst->state >= TCPS_CLOSING) (*state)->timeout = PFTM_TCP_CLOSING; else (*state)->timeout = PFTM_TCP_ESTABLISHED; return (PF_PASS); } static int pf_test_state_tcp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) { struct pf_state_key_cmp key; struct tcphdr *th = pd->hdr.tcp; int copyback = 0; struct pf_state_peer *src, *dst; struct pf_state_key *sk; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = IPPROTO_TCP; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = th->th_sport; key.port[1] = th->th_dport; } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = th->th_sport; key.port[0] = th->th_dport; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } sk = (*state)->key[pd->didx]; if ((*state)->src.state == PF_TCPS_PROXY_SRC) { if (direction != (*state)->direction) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } if (th->th_flags & TH_SYN) { if (ntohl(th->th_seq) != (*state)->src.seqlo) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, (*state)->src.seqhi, ntohl(th->th_seq) + 1, TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0, NULL); REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } else if ((*state)->src_node != NULL && pf_src_connlimit(state)) { REASON_SET(reason, PFRES_SRCLIMIT); return (PF_DROP); } else (*state)->src.state = PF_TCPS_PROXY_DST; } if ((*state)->src.state == PF_TCPS_PROXY_DST) { if (direction == (*state)->direction) { if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } (*state)->src.max_win = MAX(ntohs(th->th_win), 1); if ((*state)->dst.seqhi == 1) (*state)->dst.seqhi = htonl(arc4random()); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, &sk->addr[pd->sidx], &sk->addr[pd->didx], sk->port[pd->sidx], sk->port[pd->didx], (*state)->dst.seqhi, 0, TH_SYN, 0, (*state)->src.mss, 0, 0, (*state)->tag, NULL); REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } else if (((th->th_flags & (TH_SYN|TH_ACK)) != (TH_SYN|TH_ACK)) || (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { REASON_SET(reason, PFRES_SYNPROXY); return (PF_DROP); } else { (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); (*state)->dst.seqlo = ntohl(th->th_seq); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, pd->dst, pd->src, th->th_dport, th->th_sport, ntohl(th->th_ack), ntohl(th->th_seq) + 1, TH_ACK, (*state)->src.max_win, 0, 0, 0, (*state)->tag, NULL); pf_send_tcp(NULL, (*state)->rule.ptr, pd->af, &sk->addr[pd->sidx], &sk->addr[pd->didx], sk->port[pd->sidx], sk->port[pd->didx], (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0, NULL); (*state)->src.seqdiff = (*state)->dst.seqhi - (*state)->src.seqlo; (*state)->dst.seqdiff = (*state)->src.seqhi - (*state)->dst.seqlo; (*state)->src.seqhi = (*state)->src.seqlo + (*state)->dst.max_win; (*state)->dst.seqhi = (*state)->dst.seqlo + (*state)->src.max_win; (*state)->src.wscale = (*state)->dst.wscale = 0; (*state)->src.state = (*state)->dst.state = TCPS_ESTABLISHED; REASON_SET(reason, PFRES_SYNPROXY); return (PF_SYNPROXY_DROP); } } if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && dst->state >= TCPS_FIN_WAIT_2 && src->state >= TCPS_FIN_WAIT_2) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: state reuse "); pf_print_state(*state); pf_print_flags(th->th_flags); printf("\n"); } /* XXX make sure it's the same direction ?? */ (*state)->src.state = (*state)->dst.state = TCPS_CLOSED; pf_unlink_state(*state, PF_ENTER_LOCKED); *state = NULL; return (PF_DROP); } if ((*state)->state_flags & PFSTATE_SLOPPY) { if (pf_tcp_track_sloppy(src, dst, state, pd, reason) == PF_DROP) return (PF_DROP); } else { if (pf_tcp_track_full(src, dst, state, kif, m, off, pd, reason, ©back) == PF_DROP) return (PF_DROP); } /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || nk->port[pd->sidx] != th->th_sport) pf_change_ap(m, pd->src, &th->th_sport, pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 0, pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || nk->port[pd->didx] != th->th_dport) pf_change_ap(m, pd->dst, &th->th_dport, pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], nk->port[pd->didx], 0, pd->af); copyback = 1; } /* Copyback sequence modulation or stateful scrub changes if needed */ if (copyback) m_copyback(m, off, sizeof(*th), (caddr_t)th); return (PF_PASS); } static int pf_test_state_udp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd) { struct pf_state_peer *src, *dst; struct pf_state_key_cmp key; struct udphdr *uh = pd->hdr.udp; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = IPPROTO_UDP; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = uh->uh_sport; key.port[1] = uh->uh_dport; } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = uh->uh_sport; key.port[0] = uh->uh_dport; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } /* update states */ if (src->state < PFUDPS_SINGLE) src->state = PFUDPS_SINGLE; if (dst->state == PFUDPS_SINGLE) dst->state = PFUDPS_MULTIPLE; /* update expire time */ (*state)->expire = time_uptime; if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) (*state)->timeout = PFTM_UDP_MULTIPLE; else (*state)->timeout = PFTM_UDP_SINGLE; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || nk->port[pd->sidx] != uh->uh_sport) pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, &uh->uh_sum, &nk->addr[pd->sidx], nk->port[pd->sidx], 1, pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || nk->port[pd->didx] != uh->uh_dport) pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, &uh->uh_sum, &nk->addr[pd->didx], nk->port[pd->didx], 1, pd->af); m_copyback(m, off, sizeof(*uh), (caddr_t)uh); } return (PF_PASS); } static int pf_test_state_icmp(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) { struct pf_addr *saddr = pd->src, *daddr = pd->dst; u_int16_t icmpid = 0, *icmpsum; u_int8_t icmptype; int state_icmp = 0; struct pf_state_key_cmp key; bzero(&key, sizeof(key)); switch (pd->proto) { #ifdef INET case IPPROTO_ICMP: icmptype = pd->hdr.icmp->icmp_type; icmpid = pd->hdr.icmp->icmp_id; icmpsum = &pd->hdr.icmp->icmp_cksum; if (icmptype == ICMP_UNREACH || icmptype == ICMP_SOURCEQUENCH || icmptype == ICMP_REDIRECT || icmptype == ICMP_TIMXCEED || icmptype == ICMP_PARAMPROB) state_icmp++; break; #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: icmptype = pd->hdr.icmp6->icmp6_type; icmpid = pd->hdr.icmp6->icmp6_id; icmpsum = &pd->hdr.icmp6->icmp6_cksum; if (icmptype == ICMP6_DST_UNREACH || icmptype == ICMP6_PACKET_TOO_BIG || icmptype == ICMP6_TIME_EXCEEDED || icmptype == ICMP6_PARAM_PROB) state_icmp++; break; #endif /* INET6 */ } if (!state_icmp) { /* * ICMP query/reply message not related to a TCP/UDP packet. * Search for an ICMP state. */ key.af = pd->af; key.proto = pd->proto; key.port[0] = key.port[1] = icmpid; if (direction == PF_IN) { /* wire side, straight */ PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); } else { /* stack side, reverse */ PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); } STATE_LOOKUP(kif, &key, direction, *state, pd); (*state)->expire = time_uptime; (*state)->timeout = PFTM_ICMP_ERROR_REPLY; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; switch (pd->af) { #ifdef INET case AF_INET: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&saddr->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) pf_change_a(&daddr->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); if (nk->port[0] != pd->hdr.icmp->icmp_id) { pd->hdr.icmp->icmp_cksum = pf_cksum_fixup( pd->hdr.icmp->icmp_cksum, icmpid, nk->port[pd->sidx], 0); pd->hdr.icmp->icmp_id = nk->port[pd->sidx]; } m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET6)) pf_change_a6(saddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->sidx], 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET6)) pf_change_a6(daddr, &pd->hdr.icmp6->icmp6_cksum, &nk->addr[pd->didx], 0); m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); break; #endif /* INET6 */ } } return (PF_PASS); } else { /* * ICMP error message in response to a TCP/UDP packet. * Extract the inner TCP/UDP header and search for that state. */ struct pf_pdesc pd2; bzero(&pd2, sizeof pd2); #ifdef INET struct ip h2; #endif /* INET */ #ifdef INET6 struct ip6_hdr h2_6; int terminal = 0; #endif /* INET6 */ int ipoff2 = 0; int off2 = 0; pd2.af = pd->af; /* Payload packet is from the opposite direction. */ pd2.sidx = (direction == PF_IN) ? 1 : 0; pd2.didx = (direction == PF_IN) ? 0 : 1; switch (pd->af) { #ifdef INET case AF_INET: /* offset of h2 in mbuf chain */ ipoff2 = off + ICMP_MINLEN; if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(ip)\n")); return (PF_DROP); } /* * ICMP error messages don't refer to non-first * fragments */ if (h2.ip_off & htons(IP_OFFMASK)) { REASON_SET(reason, PFRES_FRAG); return (PF_DROP); } /* offset of protocol header that follows h2 */ off2 = ipoff2 + (h2.ip_hl << 2); pd2.proto = h2.ip_p; pd2.src = (struct pf_addr *)&h2.ip_src; pd2.dst = (struct pf_addr *)&h2.ip_dst; pd2.ip_sum = &h2.ip_sum; break; #endif /* INET */ #ifdef INET6 case AF_INET6: ipoff2 = off + sizeof(struct icmp6_hdr); if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(ip6)\n")); return (PF_DROP); } pd2.proto = h2_6.ip6_nxt; pd2.src = (struct pf_addr *)&h2_6.ip6_src; pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; pd2.ip_sum = NULL; off2 = ipoff2 + sizeof(h2_6); do { switch (pd2.proto) { case IPPROTO_FRAGMENT: /* * ICMPv6 error messages for * non-first fragments */ REASON_SET(reason, PFRES_FRAG); return (PF_DROP); case IPPROTO_AH: case IPPROTO_HOPOPTS: case IPPROTO_ROUTING: case IPPROTO_DSTOPTS: { /* get next header and header length */ struct ip6_ext opt6; if (!pf_pull_hdr(m, off2, &opt6, sizeof(opt6), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMPv6 short opt\n")); return (PF_DROP); } if (pd2.proto == IPPROTO_AH) off2 += (opt6.ip6e_len + 2) * 4; else off2 += (opt6.ip6e_len + 1) * 8; pd2.proto = opt6.ip6e_nxt; /* goto the next header */ break; } default: terminal++; break; } } while (!terminal); break; #endif /* INET6 */ } switch (pd2.proto) { case IPPROTO_TCP: { struct tcphdr th; u_int32_t seq; struct pf_state_peer *src, *dst; u_int8_t dws; int copyback = 0; /* * Only the first 8 bytes of the TCP header can be * expected. Don't access any TCP header fields after * th_seq, an ackskew test is not possible. */ if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(tcp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_TCP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[pd2.sidx] = th.th_sport; key.port[pd2.didx] = th.th_dport; STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->dst; dst = &(*state)->src; } else { src = &(*state)->src; dst = &(*state)->dst; } if (src->wscale && dst->wscale) dws = dst->wscale & PF_WSCALE_MASK; else dws = 0; /* Demodulate sequence number */ seq = ntohl(th.th_seq) - src->seqdiff; if (src->seqdiff) { pf_change_a(&th.th_seq, icmpsum, htonl(seq), 0); copyback = 1; } if (!((*state)->state_flags & PFSTATE_SLOPPY) && (!SEQ_GEQ(src->seqhi, seq) || !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: BAD ICMP %d:%d ", icmptype, pd->hdr.icmp->icmp_code); pf_print_host(pd->src, 0, pd->af); printf(" -> "); pf_print_host(pd->dst, 0, pd->af); printf(" state: "); pf_print_state(*state); printf(" seq=%u\n", seq); } REASON_SET(reason, PFRES_BADSTATE); return (PF_DROP); } else { if (V_pf_status.debug >= PF_DEBUG_MISC) { printf("pf: OK ICMP %d:%d ", icmptype, pd->hdr.icmp->icmp_code); pf_print_host(pd->src, 0, pd->af); printf(" -> "); pf_print_host(pd->dst, 0, pd->af); printf(" state: "); pf_print_state(*state); printf(" seq=%u\n", seq); } } /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != th.th_sport) pf_change_icmp(pd2.src, &th.th_sport, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != th.th_dport) pf_change_icmp(pd2.dst, &th.th_dport, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); copyback = 1; } if (copyback) { switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t )&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } m_copyback(m, off2, 8, (caddr_t)&th); } return (PF_PASS); break; } case IPPROTO_UDP: { struct udphdr uh; if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(udp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_UDP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[pd2.sidx] = uh.uh_sport; key.port[pd2.didx] = uh.uh_dport; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != uh.uh_sport) pf_change_icmp(pd2.src, &uh.uh_sport, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], &uh.uh_sum, pd2.ip_sum, icmpsum, pd->ip_sum, 1, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != uh.uh_dport) pf_change_icmp(pd2.dst, &uh.uh_dport, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], &uh.uh_sum, pd2.ip_sum, icmpsum, pd->ip_sum, 1, pd2.af); switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t )pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); } return (PF_PASS); break; } #ifdef INET case IPPROTO_ICMP: { struct icmp iih; if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short i" "(icmp)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_ICMP; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = iih.icmp_id; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != iih.icmp_id) pf_change_icmp(pd2.src, &iih.icmp_id, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != iih.icmp_id) pf_change_icmp(pd2.dst, &iih.icmp_id, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET); m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); } return (PF_PASS); break; } #endif /* INET */ #ifdef INET6 case IPPROTO_ICMPV6: { struct icmp6_hdr iih; if (!pf_pull_hdr(m, off2, &iih, sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: ICMP error message too short " "(icmp6)\n")); return (PF_DROP); } key.af = pd2.af; key.proto = IPPROTO_ICMPV6; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = iih.icmp6_id; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af) || nk->port[pd2.sidx] != iih.icmp6_id) pf_change_icmp(pd2.src, &iih.icmp6_id, daddr, &nk->addr[pd2.sidx], nk->port[pd2.sidx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET6); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af) || nk->port[pd2.didx] != iih.icmp6_id) pf_change_icmp(pd2.dst, &iih.icmp6_id, saddr, &nk->addr[pd2.didx], nk->port[pd2.didx], NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, AF_INET6); m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t)pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); m_copyback(m, off2, sizeof(struct icmp6_hdr), (caddr_t)&iih); } return (PF_PASS); break; } #endif /* INET6 */ default: { key.af = pd2.af; key.proto = pd2.proto; PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); key.port[0] = key.port[1] = 0; STATE_LOOKUP(kif, &key, direction, *state, pd); /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; if (PF_ANEQ(pd2.src, &nk->addr[pd2.sidx], pd2.af)) pf_change_icmp(pd2.src, NULL, daddr, &nk->addr[pd2.sidx], 0, NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); if (PF_ANEQ(pd2.dst, &nk->addr[pd2.didx], pd2.af)) pf_change_icmp(pd2.dst, NULL, saddr, &nk->addr[pd2.didx], 0, NULL, pd2.ip_sum, icmpsum, pd->ip_sum, 0, pd2.af); switch (pd2.af) { #ifdef INET case AF_INET: m_copyback(m, off, ICMP_MINLEN, (caddr_t)pd->hdr.icmp); m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); break; #endif /* INET */ #ifdef INET6 case AF_INET6: m_copyback(m, off, sizeof(struct icmp6_hdr), (caddr_t )pd->hdr.icmp6); m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t )&h2_6); break; #endif /* INET6 */ } } return (PF_PASS); break; } } } } static int pf_test_state_other(struct pf_state **state, int direction, struct pfi_kif *kif, struct mbuf *m, struct pf_pdesc *pd) { struct pf_state_peer *src, *dst; struct pf_state_key_cmp key; bzero(&key, sizeof(key)); key.af = pd->af; key.proto = pd->proto; if (direction == PF_IN) { PF_ACPY(&key.addr[0], pd->src, key.af); PF_ACPY(&key.addr[1], pd->dst, key.af); key.port[0] = key.port[1] = 0; } else { PF_ACPY(&key.addr[1], pd->src, key.af); PF_ACPY(&key.addr[0], pd->dst, key.af); key.port[1] = key.port[0] = 0; } STATE_LOOKUP(kif, &key, direction, *state, pd); if (direction == (*state)->direction) { src = &(*state)->src; dst = &(*state)->dst; } else { src = &(*state)->dst; dst = &(*state)->src; } /* update states */ if (src->state < PFOTHERS_SINGLE) src->state = PFOTHERS_SINGLE; if (dst->state == PFOTHERS_SINGLE) dst->state = PFOTHERS_MULTIPLE; /* update expire time */ (*state)->expire = time_uptime; if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) (*state)->timeout = PFTM_OTHER_MULTIPLE; else (*state)->timeout = PFTM_OTHER_SINGLE; /* translate source/destination address, if necessary */ if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { struct pf_state_key *nk = (*state)->key[pd->didx]; KASSERT(nk, ("%s: nk is null", __func__)); KASSERT(pd, ("%s: pd is null", __func__)); KASSERT(pd->src, ("%s: pd->src is null", __func__)); KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); switch (pd->af) { #ifdef INET case AF_INET: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) pf_change_a(&pd->src->v4.s_addr, pd->ip_sum, nk->addr[pd->sidx].v4.s_addr, 0); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) pf_change_a(&pd->dst->v4.s_addr, pd->ip_sum, nk->addr[pd->didx].v4.s_addr, 0); break; #endif /* INET */ #ifdef INET6 case AF_INET6: if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); #endif /* INET6 */ } } return (PF_PASS); } /* * ipoff and off are measured from the start of the mbuf chain. * h must be at "ipoff" on the mbuf chain. */ void * pf_pull_hdr(struct mbuf *m, int off, void *p, int len, u_short *actionp, u_short *reasonp, sa_family_t af) { switch (af) { #ifdef INET case AF_INET: { struct ip *h = mtod(m, struct ip *); u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; if (fragoff) { if (fragoff >= len) ACTION_SET(actionp, PF_PASS); else { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_FRAG); } return (NULL); } if (m->m_pkthdr.len < off + len || ntohs(h->ip_len) < off + len) { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_SHORT); return (NULL); } break; } #endif /* INET */ #ifdef INET6 case AF_INET6: { struct ip6_hdr *h = mtod(m, struct ip6_hdr *); if (m->m_pkthdr.len < off + len || (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < (unsigned)(off + len)) { ACTION_SET(actionp, PF_DROP); REASON_SET(reasonp, PFRES_SHORT); return (NULL); } break; } #endif /* INET6 */ } m_copydata(m, off, len, p); return (p); } #ifdef RADIX_MPATH static int pf_routable_oldmpath(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, int rtableid) { struct radix_node_head *rnh; struct sockaddr_in *dst; int ret = 1; int check_mpath; #ifdef INET6 struct sockaddr_in6 *dst6; struct route_in6 ro; #else struct route ro; #endif struct radix_node *rn; struct rtentry *rt; struct ifnet *ifp; check_mpath = 0; /* XXX: stick to table 0 for now */ rnh = rt_tables_get_rnh(0, af); if (rnh != NULL && rn_mpath_capable(rnh)) check_mpath = 1; bzero(&ro, sizeof(ro)); switch (af) { case AF_INET: dst = satosin(&ro.ro_dst); dst->sin_family = AF_INET; dst->sin_len = sizeof(*dst); dst->sin_addr = addr->v4; break; #ifdef INET6 case AF_INET6: /* * Skip check for addresses with embedded interface scope, * as they would always match anyway. */ if (IN6_IS_SCOPE_EMBED(&addr->v6)) goto out; dst6 = (struct sockaddr_in6 *)&ro.ro_dst; dst6->sin6_family = AF_INET6; dst6->sin6_len = sizeof(*dst6); dst6->sin6_addr = addr->v6; break; #endif /* INET6 */ default: return (0); } /* Skip checks for ipsec interfaces */ if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) goto out; switch (af) { #ifdef INET6 case AF_INET6: in6_rtalloc_ign(&ro, 0, rtableid); break; #endif #ifdef INET case AF_INET: in_rtalloc_ign((struct route *)&ro, 0, rtableid); break; #endif } if (ro.ro_rt != NULL) { /* No interface given, this is a no-route check */ if (kif == NULL) goto out; if (kif->pfik_ifp == NULL) { ret = 0; goto out; } /* Perform uRPF check if passed input interface */ ret = 0; rn = (struct radix_node *)ro.ro_rt; do { rt = (struct rtentry *)rn; ifp = rt->rt_ifp; if (kif->pfik_ifp == ifp) ret = 1; rn = rn_mpath_next(rn); } while (check_mpath == 1 && rn != NULL && ret == 0); } else ret = 0; out: if (ro.ro_rt != NULL) RTFREE(ro.ro_rt); return (ret); } #endif int pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kif *kif, int rtableid) { #ifdef INET struct nhop4_basic nh4; #endif #ifdef INET6 struct nhop6_basic nh6; #endif struct ifnet *ifp; #ifdef RADIX_MPATH struct radix_node_head *rnh; /* XXX: stick to table 0 for now */ rnh = rt_tables_get_rnh(0, af); if (rnh != NULL && rn_mpath_capable(rnh)) return (pf_routable_oldmpath(addr, af, kif, rtableid)); #endif /* * Skip check for addresses with embedded interface scope, * as they would always match anyway. */ if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) return (1); if (af != AF_INET && af != AF_INET6) return (0); /* Skip checks for ipsec interfaces */ if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) return (1); ifp = NULL; switch (af) { #ifdef INET6 case AF_INET6: if (fib6_lookup_nh_basic(rtableid, &addr->v6, 0, 0, 0, &nh6)!=0) return (0); ifp = nh6.nh_ifp; break; #endif #ifdef INET case AF_INET: if (fib4_lookup_nh_basic(rtableid, addr->v4, 0, 0, &nh4) != 0) return (0); ifp = nh4.nh_ifp; break; #endif } /* No interface given, this is a no-route check */ if (kif == NULL) return (1); if (kif->pfik_ifp == NULL) return (0); /* Perform uRPF check if passed input interface */ if (kif->pfik_ifp == ifp) return (1); return (0); } #ifdef INET static void pf_route(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) { struct mbuf *m0, *m1; struct sockaddr_in dst; struct ip *ip; struct ifnet *ifp = NULL; struct pf_addr naddr; struct pf_src_node *sn = NULL; int error = 0; uint16_t ip_len, ip_off; KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", __func__)); if ((pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || pd->pf_mtag->routed++ > 3) { m0 = *m; *m = NULL; goto bad_locked; } if (r->rt == PF_DUPTO) { if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { if (s) PF_STATE_UNLOCK(s); return; } } else { if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { if (s) PF_STATE_UNLOCK(s); return; } m0 = *m; } ip = mtod(m0, struct ip *); bzero(&dst, sizeof(dst)); dst.sin_family = AF_INET; dst.sin_len = sizeof(dst); dst.sin_addr = ip->ip_dst; bzero(&naddr, sizeof(naddr)); if (TAILQ_EMPTY(&r->rpool.list)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); goto bad_locked; } if (s == NULL) { pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, &naddr, NULL, &sn); if (!PF_AZERO(&naddr, AF_INET)) dst.sin_addr.s_addr = naddr.v4.s_addr; ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; } else { if (!PF_AZERO(&s->rt_addr, AF_INET)) dst.sin_addr.s_addr = s->rt_addr.v4.s_addr; ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; PF_STATE_UNLOCK(s); } if (ifp == NULL) goto bad; if (oifp != ifp) { if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) goto bad; else if (m0 == NULL) goto done; if (m0->m_len < sizeof(struct ip)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); goto bad; } ip = mtod(m0, struct ip *); } if (ifp->if_flags & IFF_LOOPBACK) m0->m_flags |= M_SKIP_FIREWALL; ip_len = ntohs(ip->ip_len); ip_off = ntohs(ip->ip_off); /* Copied from FreeBSD 10.0-CURRENT ip_output. */ m0->m_pkthdr.csum_flags |= CSUM_IP; if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { in_delayed_cksum(m0); m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; } #ifdef SCTP if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; } #endif /* * If small enough for interface, or the interface will take * care of the fragmentation for us, we can just send directly. */ if (ip_len <= ifp->if_mtu || (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { ip->ip_sum = 0; if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); m0->m_pkthdr.csum_flags &= ~CSUM_IP; } m_clrprotoflags(m0); /* Avoid confusing lower layers. */ error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); goto done; } /* Balk when DF bit is set or the interface didn't support TSO. */ if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { error = EMSGSIZE; KMOD_IPSTAT_INC(ips_cantfrag); if (r->rt != PF_DUPTO) { icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, ifp->if_mtu); goto done; } else goto bad; } error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); if (error) goto bad; for (; m0; m0 = m1) { m1 = m0->m_nextpkt; m0->m_nextpkt = NULL; if (error == 0) { m_clrprotoflags(m0); error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); } else m_freem(m0); } if (error == 0) KMOD_IPSTAT_INC(ips_fragmented); done: if (r->rt != PF_DUPTO) *m = NULL; return; bad_locked: if (s) PF_STATE_UNLOCK(s); bad: m_freem(m0); goto done; } #endif /* INET */ #ifdef INET6 static void pf_route6(struct mbuf **m, struct pf_rule *r, int dir, struct ifnet *oifp, struct pf_state *s, struct pf_pdesc *pd, struct inpcb *inp) { struct mbuf *m0; struct sockaddr_in6 dst; struct ip6_hdr *ip6; struct ifnet *ifp = NULL; struct pf_addr naddr; struct pf_src_node *sn = NULL; KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", __func__)); if ((pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || pd->pf_mtag->routed++ > 3) { m0 = *m; *m = NULL; goto bad_locked; } if (r->rt == PF_DUPTO) { if ((m0 = m_dup(*m, M_NOWAIT)) == NULL) { if (s) PF_STATE_UNLOCK(s); return; } } else { if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { if (s) PF_STATE_UNLOCK(s); return; } m0 = *m; } ip6 = mtod(m0, struct ip6_hdr *); bzero(&dst, sizeof(dst)); dst.sin6_family = AF_INET6; dst.sin6_len = sizeof(dst); dst.sin6_addr = ip6->ip6_dst; bzero(&naddr, sizeof(naddr)); if (TAILQ_EMPTY(&r->rpool.list)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); goto bad_locked; } if (s == NULL) { pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, &naddr, NULL, &sn); if (!PF_AZERO(&naddr, AF_INET6)) PF_ACPY((struct pf_addr *)&dst.sin6_addr, &naddr, AF_INET6); ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; } else { if (!PF_AZERO(&s->rt_addr, AF_INET6)) PF_ACPY((struct pf_addr *)&dst.sin6_addr, &s->rt_addr, AF_INET6); ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; } if (s) PF_STATE_UNLOCK(s); if (ifp == NULL) goto bad; if (oifp != ifp) { if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS) goto bad; else if (m0 == NULL) goto done; if (m0->m_len < sizeof(struct ip6_hdr)) { DPFPRINTF(PF_DEBUG_URGENT, ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", __func__)); goto bad; } ip6 = mtod(m0, struct ip6_hdr *); } if (ifp->if_flags & IFF_LOOPBACK) m0->m_flags |= M_SKIP_FIREWALL; if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & ~ifp->if_hwassist) { uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; } /* * If the packet is too large for the outgoing interface, * send back an icmp6 error. */ if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) nd6_output_ifp(ifp, ifp, m0, &dst, NULL); else { in6_ifstat_inc(ifp, ifs6_in_toobig); if (r->rt != PF_DUPTO) icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); else goto bad; } done: if (r->rt != PF_DUPTO) *m = NULL; return; bad_locked: if (s) PF_STATE_UNLOCK(s); bad: m_freem(m0); goto done; } #endif /* INET6 */ /* * FreeBSD supports cksum offloads for the following drivers. * em(4), fxp(4), lge(4), ndis(4), nge(4), re(4), ti(4), txp(4), xl(4) * * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : * network driver performed cksum including pseudo header, need to verify * csum_data * CSUM_DATA_VALID : * network driver performed cksum, needs to additional pseudo header * cksum computation with partial csum_data(i.e. lack of H/W support for * pseudo header, for instance hme(4), sk(4) and possibly gem(4)) * * After validating the cksum of packet, set both flag CSUM_DATA_VALID and * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper * TCP/UDP layer. * Also, set csum_data to 0xffff to force cksum validation. */ static int pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) { u_int16_t sum = 0; int hw_assist = 0; struct ip *ip; if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) return (1); if (m->m_pkthdr.len < off + len) return (1); switch (p) { case IPPROTO_TCP: if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { sum = m->m_pkthdr.csum_data; } else { ip = mtod(m, struct ip *); sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + IPPROTO_TCP)); } sum ^= 0xffff; ++hw_assist; } break; case IPPROTO_UDP: if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { sum = m->m_pkthdr.csum_data; } else { ip = mtod(m, struct ip *); sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl((u_short)len + m->m_pkthdr.csum_data + IPPROTO_UDP)); } sum ^= 0xffff; ++hw_assist; } break; case IPPROTO_ICMP: #ifdef INET6 case IPPROTO_ICMPV6: #endif /* INET6 */ break; default: return (1); } if (!hw_assist) { switch (af) { case AF_INET: if (p == IPPROTO_ICMP) { if (m->m_len < off) return (1); m->m_data += off; m->m_len -= off; sum = in_cksum(m, len); m->m_data -= off; m->m_len += off; } else { if (m->m_len < sizeof(struct ip)) return (1); sum = in4_cksum(m, p, off, len); } break; #ifdef INET6 case AF_INET6: if (m->m_len < sizeof(struct ip6_hdr)) return (1); sum = in6_cksum(m, p, off, len); break; #endif /* INET6 */ default: return (1); } } if (sum) { switch (p) { case IPPROTO_TCP: { KMOD_TCPSTAT_INC(tcps_rcvbadsum); break; } case IPPROTO_UDP: { KMOD_UDPSTAT_INC(udps_badsum); break; } #ifdef INET case IPPROTO_ICMP: { KMOD_ICMPSTAT_INC(icps_checksum); break; } #endif #ifdef INET6 case IPPROTO_ICMPV6: { KMOD_ICMP6STAT_INC(icp6s_checksum); break; } #endif /* INET6 */ } return (1); } else { if (p == IPPROTO_TCP || p == IPPROTO_UDP) { m->m_pkthdr.csum_flags |= (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m->m_pkthdr.csum_data = 0xffff; } } return (0); } #ifdef INET int pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) { struct pfi_kif *kif; u_short action, reason = 0, log = 0; struct mbuf *m = *m0; struct ip *h = NULL; struct m_tag *ipfwtag; struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; struct pf_state *s = NULL; struct pf_ruleset *ruleset = NULL; struct pf_pdesc pd; int off, dirndx, pqid = 0; PF_RULES_RLOCK_TRACKER; M_ASSERTPKTHDR(m); if (!V_pf_status.running) return (PF_PASS); memset(&pd, 0, sizeof(pd)); kif = (struct pfi_kif *)ifp->if_pf_kif; if (kif == NULL) { DPFPRINTF(PF_DEBUG_URGENT, ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); return (PF_DROP); } if (kif->pfik_flags & PFI_IFLAG_SKIP) return (PF_PASS); if (m->m_flags & M_SKIP_FIREWALL) return (PF_PASS); pd.pf_mtag = pf_find_mtag(m); PF_RULES_RLOCK(); if (ip_divert_ptr != NULL && ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); if (rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; goto done; } pd.pf_mtag->flags |= PF_PACKET_LOOPED; m_tag_delete(m, ipfwtag); } if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { m->m_flags |= M_FASTFWD_OURS; pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; } } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { /* We do IP header normalization and packet reassembly here */ action = PF_DROP; goto done; } m = *m0; /* pf_normalize messes with m0 */ h = mtod(m, struct ip *); off = h->ip_hl << 2; if (off < (int)sizeof(struct ip)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); log = 1; goto done; } pd.src = (struct pf_addr *)&h->ip_src; pd.dst = (struct pf_addr *)&h->ip_dst; pd.sport = pd.dport = NULL; pd.ip_sum = &h->ip_sum; pd.proto_sum = NULL; pd.proto = h->ip_p; pd.dir = dir; pd.sidx = (dir == PF_IN) ? 0 : 1; pd.didx = (dir == PF_IN) ? 1 : 0; pd.af = AF_INET; pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; pd.tot_len = ntohs(h->ip_len); /* handle fragments that didn't get reassembled by normalization */ if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { action = pf_test_fragment(&r, dir, kif, m, h, &pd, &a, &ruleset); goto done; } switch (h->ip_p) { case IPPROTO_TCP: { struct tcphdr th; pd.hdr.tcp = &th; if (!pf_pull_hdr(m, off, &th, sizeof(th), &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } pd.p_len = pd.tot_len - off - (th.th_off << 2); if ((th.th_flags & TH_ACK) && pd.p_len == 0) pqid = 1; action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); if (action == PF_DROP) goto done; action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_UDP: { struct udphdr uh; pd.hdr.udp = &uh; if (!pf_pull_hdr(m, off, &uh, sizeof(uh), &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } if (uh.uh_dport == 0 || ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); goto done; } action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_ICMP: { struct icmp ih; pd.hdr.icmp = &ih; if (!pf_pull_hdr(m, off, &ih, ICMP_MINLEN, &action, &reason, AF_INET)) { log = action != PF_PASS; goto done; } action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } #ifdef INET6 case IPPROTO_ICMPV6: { action = PF_DROP; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping IPv4 packet with ICMPv6 payload\n")); goto done; } #endif default: action = pf_test_state_other(&s, dir, kif, m, &pd); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } done: PF_RULES_RUNLOCK(); if (action == PF_PASS && h->ip_hl > 5 && !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = r->log; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping packet with ip options\n")); } if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } if (r->rtableid >= 0) M_SETFIB(m, r->rtableid); if (r->scrub_flags & PFSTATE_SETPRIO) { if (pd.tos & IPTOS_LOWDELAY) pqid = 1; if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate 802.1q mtag\n")); } } #ifdef ALTQ if (action == PF_PASS && r->qid) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } else { if (s != NULL) pd.pf_mtag->qid_hash = pf_state_hash(s); if (pqid || (pd.tos & IPTOS_LOWDELAY)) pd.pf_mtag->qid = r->pqid; else pd.pf_mtag->qid = r->qid; /* Add hints for ecn. */ pd.pf_mtag->hdr = h; } } #endif /* ALTQ */ /* * connections redirected to loopback should not match sockets * bound specifically to loopback due to security implications, * see tcp_input() and in_pcblookup_listen(). */ if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && (s->nat_rule.ptr->action == PF_RDR || s->nat_rule.ptr->action == PF_BINAT) && (ntohl(pd.dst->v4.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) m->m_flags |= M_SKIP_FIREWALL; if (action == PF_PASS && r->divert.port && ip_divert_ptr != NULL && !PACKET_LOOPED(&pd)) { ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); if (ipfwtag != NULL) { ((struct ipfw_rule_ref *)(ipfwtag+1))->info = ntohs(r->divert.port); ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; if (s) PF_STATE_UNLOCK(s); m_tag_prepend(m, ipfwtag); if (m->m_flags & M_FASTFWD_OURS) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate tag\n")); } else { pd.pf_mtag->flags |= PF_FASTFWD_OURS_PRESENT; m->m_flags &= ~M_FASTFWD_OURS; } } ip_divert_ptr(*m0, dir == PF_IN ? DIR_IN : DIR_OUT); *m0 = NULL; return (action); } else { /* XXX: ipfw has the same behaviour! */ action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate divert tag\n")); } } if (log) { struct pf_rule *lr; if (s != NULL && s->nat_rule.ptr != NULL && s->nat_rule.ptr->log & PF_LOG_ALL) lr = s->nat_rule.ptr; else lr = r; PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, (s == NULL)); } kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS] += pd.tot_len; kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS]++; if (action == PF_PASS || r->action == PF_DROP) { dirndx = (dir == PF_OUT); r->packets[dirndx]++; r->bytes[dirndx] += pd.tot_len; if (a != NULL) { a->packets[dirndx]++; a->bytes[dirndx] += pd.tot_len; } if (s != NULL) { if (s->nat_rule.ptr != NULL) { s->nat_rule.ptr->packets[dirndx]++; s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; } if (s->src_node != NULL) { s->src_node->packets[dirndx]++; s->src_node->bytes[dirndx] += pd.tot_len; } if (s->nat_src_node != NULL) { s->nat_src_node->packets[dirndx]++; s->nat_src_node->bytes[dirndx] += pd.tot_len; } dirndx = (dir == s->direction) ? 0 : 1; s->packets[dirndx]++; s->bytes[dirndx] += pd.tot_len; } tr = r; nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; if (nr != NULL && r == &V_pf_default_rule) tr = nr; if (tr->src.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->src.addr.p.tbl, (s == NULL) ? pd.src : &s->key[(s->direction == PF_IN)]-> addr[(s->direction == PF_OUT)], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->src.neg); if (tr->dst.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->dst.addr.p.tbl, (s == NULL) ? pd.dst : &s->key[(s->direction == PF_IN)]-> addr[(s->direction == PF_IN)], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->dst.neg); } switch (action) { case PF_SYNPROXY_DROP: m_freem(*m0); case PF_DEFER: *m0 = NULL; action = PF_PASS; break; case PF_DROP: m_freem(*m0); *m0 = NULL; break; default: /* pf_route() returns unlocked. */ if (r->rt) { pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); return (action); } break; } if (s) PF_STATE_UNLOCK(s); return (action); } #endif /* INET */ #ifdef INET6 int pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) { struct pfi_kif *kif; u_short action, reason = 0, log = 0; struct mbuf *m = *m0, *n = NULL; struct m_tag *mtag; struct ip6_hdr *h = NULL; struct pf_rule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; struct pf_state *s = NULL; struct pf_ruleset *ruleset = NULL; struct pf_pdesc pd; int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; PF_RULES_RLOCK_TRACKER; M_ASSERTPKTHDR(m); if (!V_pf_status.running) return (PF_PASS); memset(&pd, 0, sizeof(pd)); pd.pf_mtag = pf_find_mtag(m); if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) return (PF_PASS); kif = (struct pfi_kif *)ifp->if_pf_kif; if (kif == NULL) { DPFPRINTF(PF_DEBUG_URGENT, ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); return (PF_DROP); } if (kif->pfik_flags & PFI_IFLAG_SKIP) return (PF_PASS); if (m->m_flags & M_SKIP_FIREWALL) return (PF_PASS); PF_RULES_RLOCK(); /* We do IP header normalization and packet reassembly here */ if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { action = PF_DROP; goto done; } m = *m0; /* pf_normalize messes with m0 */ h = mtod(m, struct ip6_hdr *); #if 1 /* * we do not support jumbogram yet. if we keep going, zero ip6_plen * will do something bad, so drop the packet for now. */ if (htons(h->ip6_plen) == 0) { action = PF_DROP; REASON_SET(&reason, PFRES_NORM); /*XXX*/ goto done; } #endif pd.src = (struct pf_addr *)&h->ip6_src; pd.dst = (struct pf_addr *)&h->ip6_dst; pd.sport = pd.dport = NULL; pd.ip_sum = NULL; pd.proto_sum = NULL; pd.dir = dir; pd.sidx = (dir == PF_IN) ? 0 : 1; pd.didx = (dir == PF_IN) ? 1 : 0; pd.af = AF_INET6; pd.tos = 0; pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); pd.proto = h->ip6_nxt; do { switch (pd.proto) { case IPPROTO_FRAGMENT: action = pf_test_fragment(&r, dir, kif, m, h, &pd, &a, &ruleset); if (action == PF_DROP) REASON_SET(&reason, PFRES_FRAG); goto done; case IPPROTO_ROUTING: { struct ip6_rthdr rthdr; if (rh_cnt++) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 more than one rthdr\n")); action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = 1; goto done; } if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, &reason, pd.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 short rthdr\n")); action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); log = 1; goto done; } if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 rthdr0\n")); action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = 1; goto done; } /* FALLTHROUGH */ } case IPPROTO_AH: case IPPROTO_HOPOPTS: case IPPROTO_DSTOPTS: { /* get next header and header length */ struct ip6_ext opt6; if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), NULL, &reason, pd.af)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: IPv6 short opt\n")); action = PF_DROP; log = 1; goto done; } if (pd.proto == IPPROTO_AH) off += (opt6.ip6e_len + 2) * 4; else off += (opt6.ip6e_len + 1) * 8; pd.proto = opt6.ip6e_nxt; /* goto the next header */ break; } default: terminal++; break; } } while (!terminal); /* if there's no routing header, use unmodified mbuf for checksumming */ if (!n) n = m; switch (pd.proto) { case IPPROTO_TCP: { struct tcphdr th; pd.hdr.tcp = &th; if (!pf_pull_hdr(m, off, &th, sizeof(th), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } pd.p_len = pd.tot_len - off - (th.th_off << 2); action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); if (action == PF_DROP) goto done; action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_UDP: { struct udphdr uh; pd.hdr.udp = &uh; if (!pf_pull_hdr(m, off, &uh, sizeof(uh), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } if (uh.uh_dport == 0 || ntohs(uh.uh_ulen) > m->m_pkthdr.len - off || ntohs(uh.uh_ulen) < sizeof(struct udphdr)) { action = PF_DROP; REASON_SET(&reason, PFRES_SHORT); goto done; } action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } case IPPROTO_ICMP: { action = PF_DROP; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping IPv6 packet with ICMPv4 payload\n")); goto done; } case IPPROTO_ICMPV6: { struct icmp6_hdr ih; pd.hdr.icmp6 = &ih; if (!pf_pull_hdr(m, off, &ih, sizeof(ih), &action, &reason, AF_INET6)) { log = action != PF_PASS; goto done; } action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, &reason); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } default: action = pf_test_state_other(&s, dir, kif, m, &pd); if (action == PF_PASS) { if (V_pfsync_update_state_ptr != NULL) V_pfsync_update_state_ptr(s); r = s->rule.ptr; a = s->anchor.ptr; log = s->log; } else if (s == NULL) action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, &a, &ruleset, inp); break; } done: PF_RULES_RUNLOCK(); if (n != m) { m_freem(n); n = NULL; } /* handle dangerous IPv6 extension headers. */ if (action == PF_PASS && rh_cnt && !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { action = PF_DROP; REASON_SET(&reason, PFRES_IPOPTIONS); log = r->log; DPFPRINTF(PF_DEBUG_MISC, ("pf: dropping packet with dangerous v6 headers\n")); } if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } if (r->rtableid >= 0) M_SETFIB(m, r->rtableid); if (r->scrub_flags & PFSTATE_SETPRIO) { if (pd.tos & IPTOS_LOWDELAY) pqid = 1; if (pf_ieee8021q_setpcp(m, r->set_prio[pqid])) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); log = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: failed to allocate 802.1q mtag\n")); } } #ifdef ALTQ if (action == PF_PASS && r->qid) { if (pd.pf_mtag == NULL && ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { action = PF_DROP; REASON_SET(&reason, PFRES_MEMORY); } else { if (s != NULL) pd.pf_mtag->qid_hash = pf_state_hash(s); if (pd.tos & IPTOS_LOWDELAY) pd.pf_mtag->qid = r->pqid; else pd.pf_mtag->qid = r->qid; /* Add hints for ecn. */ pd.pf_mtag->hdr = h; } } #endif /* ALTQ */ if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && (s->nat_rule.ptr->action == PF_RDR || s->nat_rule.ptr->action == PF_BINAT) && IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) m->m_flags |= M_SKIP_FIREWALL; /* XXX: Anybody working on it?! */ if (r->divert.port) printf("pf: divert(9) is not supported for IPv6\n"); if (log) { struct pf_rule *lr; if (s != NULL && s->nat_rule.ptr != NULL && s->nat_rule.ptr->log & PF_LOG_ALL) lr = s->nat_rule.ptr; else lr = r; PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, &pd, (s == NULL)); } kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS] += pd.tot_len; kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS]++; if (action == PF_PASS || r->action == PF_DROP) { dirndx = (dir == PF_OUT); r->packets[dirndx]++; r->bytes[dirndx] += pd.tot_len; if (a != NULL) { a->packets[dirndx]++; a->bytes[dirndx] += pd.tot_len; } if (s != NULL) { if (s->nat_rule.ptr != NULL) { s->nat_rule.ptr->packets[dirndx]++; s->nat_rule.ptr->bytes[dirndx] += pd.tot_len; } if (s->src_node != NULL) { s->src_node->packets[dirndx]++; s->src_node->bytes[dirndx] += pd.tot_len; } if (s->nat_src_node != NULL) { s->nat_src_node->packets[dirndx]++; s->nat_src_node->bytes[dirndx] += pd.tot_len; } dirndx = (dir == s->direction) ? 0 : 1; s->packets[dirndx]++; s->bytes[dirndx] += pd.tot_len; } tr = r; nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; if (nr != NULL && r == &V_pf_default_rule) tr = nr; if (tr->src.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->src.addr.p.tbl, (s == NULL) ? pd.src : &s->key[(s->direction == PF_IN)]->addr[0], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->src.neg); if (tr->dst.addr.type == PF_ADDR_TABLE) pfr_update_stats(tr->dst.addr.p.tbl, (s == NULL) ? pd.dst : &s->key[(s->direction == PF_IN)]->addr[1], pd.af, pd.tot_len, dir == PF_OUT, r->action == PF_PASS, tr->dst.neg); } switch (action) { case PF_SYNPROXY_DROP: m_freem(*m0); case PF_DEFER: *m0 = NULL; action = PF_PASS; break; case PF_DROP: m_freem(*m0); *m0 = NULL; break; default: /* pf_route6() returns unlocked. */ if (r->rt) { pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); return (action); } break; } if (s) PF_STATE_UNLOCK(s); /* If reassembled packet passed, create new fragments. */ if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) && (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) action = pf_refragment6(ifp, m0, mtag); return (action); } #endif /* INET6 */ Index: head/sys/netpfil/pf/pf_ioctl.c =================================================================== --- head/sys/netpfil/pf/pf_ioctl.c (revision 343994) +++ head/sys/netpfil/pf/pf_ioctl.c (revision 343995) @@ -1,4329 +1,4495 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (c) 2001 Daniel Hartmeier * Copyright (c) 2002,2003 Henning Brauer * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * - Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following * disclaimer in the documentation and/or other materials provided * with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. * * Effort sponsored in part by the Defense Advanced Research Projects * Agency (DARPA) and Air Force Research Laboratory, Air Force * Materiel Command, USAF, under agreement number F30602-01-2-0537. * * $OpenBSD: pf_ioctl.c,v 1.213 2009/02/15 21:46:12 mbalmer Exp $ */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_bpf.h" #include "opt_pf.h" #include +#include +#include #include #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif /* INET6 */ #ifdef ALTQ #include #endif static struct pf_pool *pf_get_pool(char *, u_int32_t, u_int8_t, u_int32_t, u_int8_t, u_int8_t, u_int8_t); static void pf_mv_pool(struct pf_palist *, struct pf_palist *); static void pf_empty_pool(struct pf_palist *); static int pfioctl(struct cdev *, u_long, caddr_t, int, struct thread *); #ifdef ALTQ static int pf_begin_altq(u_int32_t *); static int pf_rollback_altq(u_int32_t); static int pf_commit_altq(u_int32_t); static int pf_enable_altq(struct pf_altq *); static int pf_disable_altq(struct pf_altq *); static u_int32_t pf_qname2qid(char *); static void pf_qid_unref(u_int32_t); #endif /* ALTQ */ static int pf_begin_rules(u_int32_t *, int, const char *); static int pf_rollback_rules(u_int32_t, int, char *); static int pf_setup_pfsync_matching(struct pf_ruleset *); static void pf_hash_rule(MD5_CTX *, struct pf_rule *); static void pf_hash_rule_addr(MD5_CTX *, struct pf_rule_addr *); static int pf_commit_rules(u_int32_t, int, char *); static int pf_addr_setup(struct pf_ruleset *, struct pf_addr_wrap *, sa_family_t); static void pf_addr_copyout(struct pf_addr_wrap *); #ifdef ALTQ static int pf_export_kaltq(struct pf_altq *, struct pfioc_altq_v1 *, size_t); static int pf_import_kaltq(struct pfioc_altq_v1 *, struct pf_altq *, size_t); #endif /* ALTQ */ VNET_DEFINE(struct pf_rule, pf_default_rule); #ifdef ALTQ VNET_DEFINE_STATIC(int, pf_altq_running); #define V_pf_altq_running VNET(pf_altq_running) #endif #define TAGID_MAX 50000 struct pf_tagname { - TAILQ_ENTRY(pf_tagname) entries; + TAILQ_ENTRY(pf_tagname) namehash_entries; + TAILQ_ENTRY(pf_tagname) taghash_entries; char name[PF_TAG_NAME_SIZE]; uint16_t tag; int ref; }; -TAILQ_HEAD(pf_tags, pf_tagname); -#define V_pf_tags VNET(pf_tags) -VNET_DEFINE(struct pf_tags, pf_tags); -#define V_pf_qids VNET(pf_qids) -VNET_DEFINE(struct pf_tags, pf_qids); -static MALLOC_DEFINE(M_PFTAG, "pf_tag", "pf(4) tag names"); +struct pf_tagset { + TAILQ_HEAD(, pf_tagname) *namehash; + TAILQ_HEAD(, pf_tagname) *taghash; + unsigned int mask; + uint32_t seed; + BITSET_DEFINE(, TAGID_MAX) avail; +}; + +VNET_DEFINE(struct pf_tagset, pf_tags); +#define V_pf_tags VNET(pf_tags) +static unsigned int pf_rule_tag_hashsize; +#define PF_RULE_TAG_HASH_SIZE_DEFAULT 128 +SYSCTL_UINT(_net_pf, OID_AUTO, rule_tag_hashsize, CTLFLAG_RDTUN, + &pf_rule_tag_hashsize, PF_RULE_TAG_HASH_SIZE_DEFAULT, + "Size of pf(4) rule tag hashtable"); + +#ifdef ALTQ +VNET_DEFINE(struct pf_tagset, pf_qids); +#define V_pf_qids VNET(pf_qids) +static unsigned int pf_queue_tag_hashsize; +#define PF_QUEUE_TAG_HASH_SIZE_DEFAULT 128 +SYSCTL_UINT(_net_pf, OID_AUTO, queue_tag_hashsize, CTLFLAG_RDTUN, + &pf_queue_tag_hashsize, PF_QUEUE_TAG_HASH_SIZE_DEFAULT, + "Size of pf(4) queue tag hashtable"); +#endif +VNET_DEFINE(uma_zone_t, pf_tag_z); +#define V_pf_tag_z VNET(pf_tag_z) static MALLOC_DEFINE(M_PFALTQ, "pf_altq", "pf(4) altq configuration db"); static MALLOC_DEFINE(M_PFRULE, "pf_rule", "pf(4) rules"); #if (PF_QNAME_SIZE != PF_TAG_NAME_SIZE) #error PF_QNAME_SIZE must be equal to PF_TAG_NAME_SIZE #endif -static u_int16_t tagname2tag(struct pf_tags *, char *); +static void pf_init_tagset(struct pf_tagset *, unsigned int *, + unsigned int); +static void pf_cleanup_tagset(struct pf_tagset *); +static uint16_t tagname2hashindex(const struct pf_tagset *, const char *); +static uint16_t tag2hashindex(const struct pf_tagset *, uint16_t); +static u_int16_t tagname2tag(struct pf_tagset *, char *); static u_int16_t pf_tagname2tag(char *); -static void tag_unref(struct pf_tags *, u_int16_t); +static void tag_unref(struct pf_tagset *, u_int16_t); #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x struct cdev *pf_dev; /* * XXX - These are new and need to be checked when moveing to a new version */ static void pf_clear_states(void); static int pf_clear_tables(void); static void pf_clear_srcnodes(struct pf_src_node *); static void pf_kill_srcnodes(struct pfioc_src_node_kill *); static void pf_tbladdr_copyout(struct pf_addr_wrap *); /* * Wrapper functions for pfil(9) hooks */ #ifdef INET static pfil_return_t pf_check_in(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp); static pfil_return_t pf_check_out(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp); #endif #ifdef INET6 static pfil_return_t pf_check6_in(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp); static pfil_return_t pf_check6_out(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp); #endif static int hook_pf(void); static int dehook_pf(void); static int shutdown_pf(void); static int pf_load(void); static void pf_unload(void); static struct cdevsw pf_cdevsw = { .d_ioctl = pfioctl, .d_name = PF_NAME, .d_version = D_VERSION, }; volatile VNET_DEFINE_STATIC(int, pf_pfil_hooked); #define V_pf_pfil_hooked VNET(pf_pfil_hooked) /* * We need a flag that is neither hooked nor running to know when * the VNET is "valid". We primarily need this to control (global) * external event, e.g., eventhandlers. */ VNET_DEFINE(int, pf_vnet_active); #define V_pf_vnet_active VNET(pf_vnet_active) int pf_end_threads; struct proc *pf_purge_proc; struct rmlock pf_rules_lock; struct sx pf_ioctl_lock; struct sx pf_end_lock; /* pfsync */ VNET_DEFINE(pfsync_state_import_t *, pfsync_state_import_ptr); VNET_DEFINE(pfsync_insert_state_t *, pfsync_insert_state_ptr); VNET_DEFINE(pfsync_update_state_t *, pfsync_update_state_ptr); VNET_DEFINE(pfsync_delete_state_t *, pfsync_delete_state_ptr); VNET_DEFINE(pfsync_clear_states_t *, pfsync_clear_states_ptr); VNET_DEFINE(pfsync_defer_t *, pfsync_defer_ptr); pfsync_detach_ifnet_t *pfsync_detach_ifnet_ptr; /* pflog */ pflog_packet_t *pflog_packet_ptr = NULL; extern u_long pf_ioctl_maxcount; static void pfattach_vnet(void) { u_int32_t *my_timeout = V_pf_default_rule.timeout; pf_initialize(); pfr_initialize(); pfi_initialize_vnet(); pf_normalize_init(); V_pf_limits[PF_LIMIT_STATES].limit = PFSTATE_HIWAT; V_pf_limits[PF_LIMIT_SRC_NODES].limit = PFSNODE_HIWAT; RB_INIT(&V_pf_anchors); pf_init_ruleset(&pf_main_ruleset); /* default rule should never be garbage collected */ V_pf_default_rule.entries.tqe_prev = &V_pf_default_rule.entries.tqe_next; #ifdef PF_DEFAULT_TO_DROP V_pf_default_rule.action = PF_DROP; #else V_pf_default_rule.action = PF_PASS; #endif V_pf_default_rule.nr = -1; V_pf_default_rule.rtableid = -1; V_pf_default_rule.states_cur = counter_u64_alloc(M_WAITOK); V_pf_default_rule.states_tot = counter_u64_alloc(M_WAITOK); V_pf_default_rule.src_nodes = counter_u64_alloc(M_WAITOK); /* initialize default timeouts */ my_timeout[PFTM_TCP_FIRST_PACKET] = PFTM_TCP_FIRST_PACKET_VAL; my_timeout[PFTM_TCP_OPENING] = PFTM_TCP_OPENING_VAL; my_timeout[PFTM_TCP_ESTABLISHED] = PFTM_TCP_ESTABLISHED_VAL; my_timeout[PFTM_TCP_CLOSING] = PFTM_TCP_CLOSING_VAL; my_timeout[PFTM_TCP_FIN_WAIT] = PFTM_TCP_FIN_WAIT_VAL; my_timeout[PFTM_TCP_CLOSED] = PFTM_TCP_CLOSED_VAL; my_timeout[PFTM_UDP_FIRST_PACKET] = PFTM_UDP_FIRST_PACKET_VAL; my_timeout[PFTM_UDP_SINGLE] = PFTM_UDP_SINGLE_VAL; my_timeout[PFTM_UDP_MULTIPLE] = PFTM_UDP_MULTIPLE_VAL; my_timeout[PFTM_ICMP_FIRST_PACKET] = PFTM_ICMP_FIRST_PACKET_VAL; my_timeout[PFTM_ICMP_ERROR_REPLY] = PFTM_ICMP_ERROR_REPLY_VAL; my_timeout[PFTM_OTHER_FIRST_PACKET] = PFTM_OTHER_FIRST_PACKET_VAL; my_timeout[PFTM_OTHER_SINGLE] = PFTM_OTHER_SINGLE_VAL; my_timeout[PFTM_OTHER_MULTIPLE] = PFTM_OTHER_MULTIPLE_VAL; my_timeout[PFTM_FRAG] = PFTM_FRAG_VAL; my_timeout[PFTM_INTERVAL] = PFTM_INTERVAL_VAL; my_timeout[PFTM_SRC_NODE] = PFTM_SRC_NODE_VAL; my_timeout[PFTM_TS_DIFF] = PFTM_TS_DIFF_VAL; my_timeout[PFTM_ADAPTIVE_START] = PFSTATE_ADAPT_START; my_timeout[PFTM_ADAPTIVE_END] = PFSTATE_ADAPT_END; bzero(&V_pf_status, sizeof(V_pf_status)); V_pf_status.debug = PF_DEBUG_URGENT; V_pf_pfil_hooked = 0; /* XXX do our best to avoid a conflict */ V_pf_status.hostid = arc4random(); for (int i = 0; i < PFRES_MAX; i++) V_pf_status.counters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < LCNT_MAX; i++) V_pf_status.lcounters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < FCNT_MAX; i++) V_pf_status.fcounters[i] = counter_u64_alloc(M_WAITOK); for (int i = 0; i < SCNT_MAX; i++) V_pf_status.scounters[i] = counter_u64_alloc(M_WAITOK); if (swi_add(NULL, "pf send", pf_intr, curvnet, SWI_NET, INTR_MPSAFE, &V_pf_swi_cookie) != 0) /* XXXGL: leaked all above. */ return; } static struct pf_pool * pf_get_pool(char *anchor, u_int32_t ticket, u_int8_t rule_action, u_int32_t rule_number, u_int8_t r_last, u_int8_t active, u_int8_t check_ticket) { struct pf_ruleset *ruleset; struct pf_rule *rule; int rs_num; ruleset = pf_find_ruleset(anchor); if (ruleset == NULL) return (NULL); rs_num = pf_get_ruleset_number(rule_action); if (rs_num >= PF_RULESET_MAX) return (NULL); if (active) { if (check_ticket && ticket != ruleset->rules[rs_num].active.ticket) return (NULL); if (r_last) rule = TAILQ_LAST(ruleset->rules[rs_num].active.ptr, pf_rulequeue); else rule = TAILQ_FIRST(ruleset->rules[rs_num].active.ptr); } else { if (check_ticket && ticket != ruleset->rules[rs_num].inactive.ticket) return (NULL); if (r_last) rule = TAILQ_LAST(ruleset->rules[rs_num].inactive.ptr, pf_rulequeue); else rule = TAILQ_FIRST(ruleset->rules[rs_num].inactive.ptr); } if (!r_last) { while ((rule != NULL) && (rule->nr != rule_number)) rule = TAILQ_NEXT(rule, entries); } if (rule == NULL) return (NULL); return (&rule->rpool); } static void pf_mv_pool(struct pf_palist *poola, struct pf_palist *poolb) { struct pf_pooladdr *mv_pool_pa; while ((mv_pool_pa = TAILQ_FIRST(poola)) != NULL) { TAILQ_REMOVE(poola, mv_pool_pa, entries); TAILQ_INSERT_TAIL(poolb, mv_pool_pa, entries); } } static void pf_empty_pool(struct pf_palist *poola) { struct pf_pooladdr *pa; while ((pa = TAILQ_FIRST(poola)) != NULL) { switch (pa->addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(pa->addr.p.dyn); break; case PF_ADDR_TABLE: /* XXX: this could be unfinished pooladdr on pabuf */ if (pa->addr.p.tbl != NULL) pfr_detach_table(pa->addr.p.tbl); break; } if (pa->kif) pfi_kif_unref(pa->kif); TAILQ_REMOVE(poola, pa, entries); free(pa, M_PFRULE); } } static void pf_unlink_rule(struct pf_rulequeue *rulequeue, struct pf_rule *rule) { PF_RULES_WASSERT(); TAILQ_REMOVE(rulequeue, rule, entries); PF_UNLNKDRULES_LOCK(); rule->rule_flag |= PFRULE_REFS; TAILQ_INSERT_TAIL(&V_pf_unlinked_rules, rule, entries); PF_UNLNKDRULES_UNLOCK(); } void pf_free_rule(struct pf_rule *rule) { PF_RULES_WASSERT(); if (rule->tag) tag_unref(&V_pf_tags, rule->tag); if (rule->match_tag) tag_unref(&V_pf_tags, rule->match_tag); #ifdef ALTQ if (rule->pqid != rule->qid) pf_qid_unref(rule->pqid); pf_qid_unref(rule->qid); #endif switch (rule->src.addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(rule->src.addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(rule->src.addr.p.tbl); break; } switch (rule->dst.addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(rule->dst.addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(rule->dst.addr.p.tbl); break; } if (rule->overload_tbl) pfr_detach_table(rule->overload_tbl); if (rule->kif) pfi_kif_unref(rule->kif); pf_anchor_remove(rule); pf_empty_pool(&rule->rpool.list); counter_u64_free(rule->states_cur); counter_u64_free(rule->states_tot); counter_u64_free(rule->src_nodes); free(rule, M_PFRULE); } +static void +pf_init_tagset(struct pf_tagset *ts, unsigned int *tunable_size, + unsigned int default_size) +{ + unsigned int i; + unsigned int hashsize; + + if (*tunable_size == 0 || !powerof2(*tunable_size)) + *tunable_size = default_size; + + hashsize = *tunable_size; + ts->namehash = mallocarray(hashsize, sizeof(*ts->namehash), M_PFHASH, + M_WAITOK); + ts->taghash = mallocarray(hashsize, sizeof(*ts->taghash), M_PFHASH, + M_WAITOK); + ts->mask = hashsize - 1; + ts->seed = arc4random(); + for (i = 0; i < hashsize; i++) { + TAILQ_INIT(&ts->namehash[i]); + TAILQ_INIT(&ts->taghash[i]); + } + BIT_FILL(TAGID_MAX, &ts->avail); +} + +static void +pf_cleanup_tagset(struct pf_tagset *ts) +{ + unsigned int i; + unsigned int hashsize; + struct pf_tagname *t, *tmp; + + /* + * Only need to clean up one of the hashes as each tag is hashed + * into each table. + */ + hashsize = ts->mask + 1; + for (i = 0; i < hashsize; i++) + TAILQ_FOREACH_SAFE(t, &ts->namehash[i], namehash_entries, tmp) + uma_zfree(V_pf_tag_z, t); + + free(ts->namehash, M_PFHASH); + free(ts->taghash, M_PFHASH); +} + +static uint16_t +tagname2hashindex(const struct pf_tagset *ts, const char *tagname) +{ + + return (murmur3_32_hash(tagname, strlen(tagname), ts->seed) & ts->mask); +} + +static uint16_t +tag2hashindex(const struct pf_tagset *ts, uint16_t tag) +{ + + return (tag & ts->mask); +} + static u_int16_t -tagname2tag(struct pf_tags *head, char *tagname) +tagname2tag(struct pf_tagset *ts, char *tagname) { - struct pf_tagname *tag, *p = NULL; - u_int16_t new_tagid = 1; + struct pf_tagname *tag; + u_int32_t index; + u_int16_t new_tagid; PF_RULES_WASSERT(); - TAILQ_FOREACH(tag, head, entries) + index = tagname2hashindex(ts, tagname); + TAILQ_FOREACH(tag, &ts->namehash[index], namehash_entries) if (strcmp(tagname, tag->name) == 0) { tag->ref++; return (tag->tag); } /* + * new entry + * * to avoid fragmentation, we do a linear search from the beginning - * and take the first free slot we find. if there is none or the list - * is empty, append a new entry at the end. + * and take the first free slot we find. */ - - /* new entry */ - if (!TAILQ_EMPTY(head)) - for (p = TAILQ_FIRST(head); p != NULL && - p->tag == new_tagid; p = TAILQ_NEXT(p, entries)) - new_tagid = p->tag + 1; - - if (new_tagid > TAGID_MAX) + new_tagid = BIT_FFS(TAGID_MAX, &ts->avail); + /* + * Tags are 1-based, with valid tags in the range [1..TAGID_MAX]. + * BIT_FFS() returns a 1-based bit number, with 0 indicating no bits + * set. It may also return a bit number greater than TAGID_MAX due + * to rounding of the number of bits in the vector up to a multiple + * of the vector word size at declaration/allocation time. + */ + if ((new_tagid == 0) || (new_tagid > TAGID_MAX)) return (0); + /* Mark the tag as in use. Bits are 0-based for BIT_CLR() */ + BIT_CLR(TAGID_MAX, new_tagid - 1, &ts->avail); + /* allocate and fill new struct pf_tagname */ - tag = malloc(sizeof(*tag), M_PFTAG, M_NOWAIT|M_ZERO); + tag = uma_zalloc(V_pf_tag_z, M_NOWAIT); if (tag == NULL) return (0); strlcpy(tag->name, tagname, sizeof(tag->name)); tag->tag = new_tagid; - tag->ref++; + tag->ref = 1; - if (p != NULL) /* insert new entry before p */ - TAILQ_INSERT_BEFORE(p, tag, entries); - else /* either list empty or no free slot in between */ - TAILQ_INSERT_TAIL(head, tag, entries); + /* Insert into namehash */ + TAILQ_INSERT_TAIL(&ts->namehash[index], tag, namehash_entries); + /* Insert into taghash */ + index = tag2hashindex(ts, new_tagid); + TAILQ_INSERT_TAIL(&ts->taghash[index], tag, taghash_entries); + return (tag->tag); } static void -tag_unref(struct pf_tags *head, u_int16_t tag) +tag_unref(struct pf_tagset *ts, u_int16_t tag) { - struct pf_tagname *p, *next; - + struct pf_tagname *t; + uint16_t index; + PF_RULES_WASSERT(); - for (p = TAILQ_FIRST(head); p != NULL; p = next) { - next = TAILQ_NEXT(p, entries); - if (tag == p->tag) { - if (--p->ref == 0) { - TAILQ_REMOVE(head, p, entries); - free(p, M_PFTAG); + index = tag2hashindex(ts, tag); + TAILQ_FOREACH(t, &ts->taghash[index], taghash_entries) + if (tag == t->tag) { + if (--t->ref == 0) { + TAILQ_REMOVE(&ts->taghash[index], t, + taghash_entries); + index = tagname2hashindex(ts, t->name); + TAILQ_REMOVE(&ts->namehash[index], t, + namehash_entries); + /* Bits are 0-based for BIT_SET() */ + BIT_SET(TAGID_MAX, tag - 1, &ts->avail); + uma_zfree(V_pf_tag_z, t); } break; } - } } static u_int16_t pf_tagname2tag(char *tagname) { return (tagname2tag(&V_pf_tags, tagname)); } #ifdef ALTQ static u_int32_t pf_qname2qid(char *qname) { return ((u_int32_t)tagname2tag(&V_pf_qids, qname)); } static void pf_qid_unref(u_int32_t qid) { tag_unref(&V_pf_qids, (u_int16_t)qid); } static int pf_begin_altq(u_int32_t *ticket) { - struct pf_altq *altq; + struct pf_altq *altq, *tmp; int error = 0; PF_RULES_WASSERT(); - /* Purge the old altq list */ - while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { - TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); - if (altq->qname[0] == 0 && - (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { + /* Purge the old altq lists */ + TAILQ_FOREACH_SAFE(altq, V_pf_altq_ifs_inactive, entries, tmp) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ error = altq_remove(altq); - } else - pf_qid_unref(altq->qid); + } free(altq, M_PFALTQ); } + TAILQ_INIT(V_pf_altq_ifs_inactive); + TAILQ_FOREACH_SAFE(altq, V_pf_altqs_inactive, entries, tmp) { + pf_qid_unref(altq->qid); + free(altq, M_PFALTQ); + } + TAILQ_INIT(V_pf_altqs_inactive); if (error) return (error); *ticket = ++V_ticket_altqs_inactive; V_altqs_inactive_open = 1; return (0); } static int pf_rollback_altq(u_int32_t ticket) { - struct pf_altq *altq; + struct pf_altq *altq, *tmp; int error = 0; PF_RULES_WASSERT(); if (!V_altqs_inactive_open || ticket != V_ticket_altqs_inactive) return (0); - /* Purge the old altq list */ - while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { - TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); - if (altq->qname[0] == 0 && - (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { + /* Purge the old altq lists */ + TAILQ_FOREACH_SAFE(altq, V_pf_altq_ifs_inactive, entries, tmp) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ error = altq_remove(altq); - } else - pf_qid_unref(altq->qid); + } free(altq, M_PFALTQ); } + TAILQ_INIT(V_pf_altq_ifs_inactive); + TAILQ_FOREACH_SAFE(altq, V_pf_altqs_inactive, entries, tmp) { + pf_qid_unref(altq->qid); + free(altq, M_PFALTQ); + } + TAILQ_INIT(V_pf_altqs_inactive); V_altqs_inactive_open = 0; return (error); } static int pf_commit_altq(u_int32_t ticket) { - struct pf_altqqueue *old_altqs; - struct pf_altq *altq; + struct pf_altqqueue *old_altqs, *old_altq_ifs; + struct pf_altq *altq, *tmp; int err, error = 0; PF_RULES_WASSERT(); if (!V_altqs_inactive_open || ticket != V_ticket_altqs_inactive) return (EBUSY); /* swap altqs, keep the old. */ old_altqs = V_pf_altqs_active; + old_altq_ifs = V_pf_altq_ifs_active; V_pf_altqs_active = V_pf_altqs_inactive; + V_pf_altq_ifs_active = V_pf_altq_ifs_inactive; V_pf_altqs_inactive = old_altqs; + V_pf_altq_ifs_inactive = old_altq_ifs; V_ticket_altqs_active = V_ticket_altqs_inactive; /* Attach new disciplines */ - TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { - if (altq->qname[0] == 0 && - (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { + TAILQ_FOREACH(altq, V_pf_altq_ifs_active, entries) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* attach the discipline */ error = altq_pfattach(altq); if (error == 0 && V_pf_altq_running) error = pf_enable_altq(altq); if (error != 0) return (error); } } - /* Purge the old altq list */ - while ((altq = TAILQ_FIRST(V_pf_altqs_inactive)) != NULL) { - TAILQ_REMOVE(V_pf_altqs_inactive, altq, entries); - if (altq->qname[0] == 0 && - (altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { + /* Purge the old altq lists */ + TAILQ_FOREACH_SAFE(altq, V_pf_altq_ifs_inactive, entries, tmp) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { /* detach and destroy the discipline */ if (V_pf_altq_running) error = pf_disable_altq(altq); err = altq_pfdetach(altq); if (err != 0 && error == 0) error = err; err = altq_remove(altq); if (err != 0 && error == 0) error = err; - } else - pf_qid_unref(altq->qid); + } free(altq, M_PFALTQ); } + TAILQ_INIT(V_pf_altq_ifs_inactive); + TAILQ_FOREACH_SAFE(altq, V_pf_altqs_inactive, entries, tmp) { + pf_qid_unref(altq->qid); + free(altq, M_PFALTQ); + } + TAILQ_INIT(V_pf_altqs_inactive); V_altqs_inactive_open = 0; return (error); } static int pf_enable_altq(struct pf_altq *altq) { struct ifnet *ifp; struct tb_profile tb; int error = 0; if ((ifp = ifunit(altq->ifname)) == NULL) return (EINVAL); if (ifp->if_snd.altq_type != ALTQT_NONE) error = altq_enable(&ifp->if_snd); /* set tokenbucket regulator */ if (error == 0 && ifp != NULL && ALTQ_IS_ENABLED(&ifp->if_snd)) { tb.rate = altq->ifbandwidth; tb.depth = altq->tbrsize; error = tbr_set(&ifp->if_snd, &tb); } return (error); } static int pf_disable_altq(struct pf_altq *altq) { struct ifnet *ifp; struct tb_profile tb; int error; if ((ifp = ifunit(altq->ifname)) == NULL) return (EINVAL); /* * when the discipline is no longer referenced, it was overridden * by a new one. if so, just return. */ if (altq->altq_disc != ifp->if_snd.altq_disc) return (0); error = altq_disable(&ifp->if_snd); if (error == 0) { /* clear tokenbucket regulator */ tb.rate = 0; error = tbr_set(&ifp->if_snd, &tb); } return (error); } +static int +pf_altq_ifnet_event_add(struct ifnet *ifp, int remove, u_int32_t ticket, + struct pf_altq *altq) +{ + struct ifnet *ifp1; + int error = 0; + + /* Deactivate the interface in question */ + altq->local_flags &= ~PFALTQ_FLAG_IF_REMOVED; + if ((ifp1 = ifunit(altq->ifname)) == NULL || + (remove && ifp1 == ifp)) { + altq->local_flags |= PFALTQ_FLAG_IF_REMOVED; + } else { + error = altq_add(ifp1, altq); + + if (ticket != V_ticket_altqs_inactive) + error = EBUSY; + + if (error) + free(altq, M_PFALTQ); + } + + return (error); +} + void pf_altq_ifnet_event(struct ifnet *ifp, int remove) { - struct ifnet *ifp1; struct pf_altq *a1, *a2, *a3; u_int32_t ticket; int error = 0; /* * No need to re-evaluate the configuration for events on interfaces * that do not support ALTQ, as it's not possible for such * interfaces to be part of the configuration. */ if (!ALTQ_IS_READY(&ifp->if_snd)) return; /* Interrupt userland queue modifications */ if (V_altqs_inactive_open) pf_rollback_altq(V_ticket_altqs_inactive); /* Start new altq ruleset */ if (pf_begin_altq(&ticket)) return; /* Copy the current active set */ - TAILQ_FOREACH(a1, V_pf_altqs_active, entries) { + TAILQ_FOREACH(a1, V_pf_altq_ifs_active, entries) { a2 = malloc(sizeof(*a2), M_PFALTQ, M_NOWAIT); if (a2 == NULL) { error = ENOMEM; break; } bcopy(a1, a2, sizeof(struct pf_altq)); - if (a2->qname[0] != 0) { - if ((a2->qid = pf_qname2qid(a2->qname)) == 0) { - error = EBUSY; - free(a2, M_PFALTQ); - break; - } - a2->altq_disc = NULL; - TAILQ_FOREACH(a3, V_pf_altqs_inactive, entries) { - if (strncmp(a3->ifname, a2->ifname, - IFNAMSIZ) == 0 && a3->qname[0] == 0) { - a2->altq_disc = a3->altq_disc; - break; - } - } + error = pf_altq_ifnet_event_add(ifp, remove, ticket, a2); + if (error) + break; + + TAILQ_INSERT_TAIL(V_pf_altq_ifs_inactive, a2, entries); + } + if (error) + goto out; + TAILQ_FOREACH(a1, V_pf_altqs_active, entries) { + a2 = malloc(sizeof(*a2), M_PFALTQ, M_NOWAIT); + if (a2 == NULL) { + error = ENOMEM; + break; } - /* Deactivate the interface in question */ - a2->local_flags &= ~PFALTQ_FLAG_IF_REMOVED; - if ((ifp1 = ifunit(a2->ifname)) == NULL || - (remove && ifp1 == ifp)) { - a2->local_flags |= PFALTQ_FLAG_IF_REMOVED; - } else { - error = altq_add(a2); + bcopy(a1, a2, sizeof(struct pf_altq)); - if (ticket != V_ticket_altqs_inactive) - error = EBUSY; - - if (error) { - free(a2, M_PFALTQ); + if ((a2->qid = pf_qname2qid(a2->qname)) == 0) { + error = EBUSY; + free(a2, M_PFALTQ); + break; + } + a2->altq_disc = NULL; + TAILQ_FOREACH(a3, V_pf_altq_ifs_inactive, entries) { + if (strncmp(a3->ifname, a2->ifname, + IFNAMSIZ) == 0) { + a2->altq_disc = a3->altq_disc; break; } } + error = pf_altq_ifnet_event_add(ifp, remove, ticket, a2); + if (error) + break; TAILQ_INSERT_TAIL(V_pf_altqs_inactive, a2, entries); } +out: if (error != 0) pf_rollback_altq(ticket); else pf_commit_altq(ticket); } #endif /* ALTQ */ static int pf_begin_rules(u_int32_t *ticket, int rs_num, const char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_or_create_ruleset(anchor); if (rs == NULL) return (EINVAL); while ((rule = TAILQ_FIRST(rs->rules[rs_num].inactive.ptr)) != NULL) { pf_unlink_rule(rs->rules[rs_num].inactive.ptr, rule); rs->rules[rs_num].inactive.rcount--; } *ticket = ++rs->rules[rs_num].inactive.ticket; rs->rules[rs_num].inactive.open = 1; return (0); } static int pf_rollback_rules(u_int32_t ticket, int rs_num, char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_ruleset(anchor); if (rs == NULL || !rs->rules[rs_num].inactive.open || rs->rules[rs_num].inactive.ticket != ticket) return (0); while ((rule = TAILQ_FIRST(rs->rules[rs_num].inactive.ptr)) != NULL) { pf_unlink_rule(rs->rules[rs_num].inactive.ptr, rule); rs->rules[rs_num].inactive.rcount--; } rs->rules[rs_num].inactive.open = 0; return (0); } #define PF_MD5_UPD(st, elm) \ MD5Update(ctx, (u_int8_t *) &(st)->elm, sizeof((st)->elm)) #define PF_MD5_UPD_STR(st, elm) \ MD5Update(ctx, (u_int8_t *) (st)->elm, strlen((st)->elm)) #define PF_MD5_UPD_HTONL(st, elm, stor) do { \ (stor) = htonl((st)->elm); \ MD5Update(ctx, (u_int8_t *) &(stor), sizeof(u_int32_t));\ } while (0) #define PF_MD5_UPD_HTONS(st, elm, stor) do { \ (stor) = htons((st)->elm); \ MD5Update(ctx, (u_int8_t *) &(stor), sizeof(u_int16_t));\ } while (0) static void pf_hash_rule_addr(MD5_CTX *ctx, struct pf_rule_addr *pfr) { PF_MD5_UPD(pfr, addr.type); switch (pfr->addr.type) { case PF_ADDR_DYNIFTL: PF_MD5_UPD(pfr, addr.v.ifname); PF_MD5_UPD(pfr, addr.iflags); break; case PF_ADDR_TABLE: PF_MD5_UPD(pfr, addr.v.tblname); break; case PF_ADDR_ADDRMASK: /* XXX ignore af? */ PF_MD5_UPD(pfr, addr.v.a.addr.addr32); PF_MD5_UPD(pfr, addr.v.a.mask.addr32); break; } PF_MD5_UPD(pfr, port[0]); PF_MD5_UPD(pfr, port[1]); PF_MD5_UPD(pfr, neg); PF_MD5_UPD(pfr, port_op); } static void pf_hash_rule(MD5_CTX *ctx, struct pf_rule *rule) { u_int16_t x; u_int32_t y; pf_hash_rule_addr(ctx, &rule->src); pf_hash_rule_addr(ctx, &rule->dst); PF_MD5_UPD_STR(rule, label); PF_MD5_UPD_STR(rule, ifname); PF_MD5_UPD_STR(rule, match_tagname); PF_MD5_UPD_HTONS(rule, match_tag, x); /* dup? */ PF_MD5_UPD_HTONL(rule, os_fingerprint, y); PF_MD5_UPD_HTONL(rule, prob, y); PF_MD5_UPD_HTONL(rule, uid.uid[0], y); PF_MD5_UPD_HTONL(rule, uid.uid[1], y); PF_MD5_UPD(rule, uid.op); PF_MD5_UPD_HTONL(rule, gid.gid[0], y); PF_MD5_UPD_HTONL(rule, gid.gid[1], y); PF_MD5_UPD(rule, gid.op); PF_MD5_UPD_HTONL(rule, rule_flag, y); PF_MD5_UPD(rule, action); PF_MD5_UPD(rule, direction); PF_MD5_UPD(rule, af); PF_MD5_UPD(rule, quick); PF_MD5_UPD(rule, ifnot); PF_MD5_UPD(rule, match_tag_not); PF_MD5_UPD(rule, natpass); PF_MD5_UPD(rule, keep_state); PF_MD5_UPD(rule, proto); PF_MD5_UPD(rule, type); PF_MD5_UPD(rule, code); PF_MD5_UPD(rule, flags); PF_MD5_UPD(rule, flagset); PF_MD5_UPD(rule, allow_opts); PF_MD5_UPD(rule, rt); PF_MD5_UPD(rule, tos); } static int pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) { struct pf_ruleset *rs; struct pf_rule *rule, **old_array; struct pf_rulequeue *old_rules; int error; u_int32_t old_rcount; PF_RULES_WASSERT(); if (rs_num < 0 || rs_num >= PF_RULESET_MAX) return (EINVAL); rs = pf_find_ruleset(anchor); if (rs == NULL || !rs->rules[rs_num].inactive.open || ticket != rs->rules[rs_num].inactive.ticket) return (EBUSY); /* Calculate checksum for the main ruleset */ if (rs == &pf_main_ruleset) { error = pf_setup_pfsync_matching(rs); if (error != 0) return (error); } /* Swap rules, keep the old. */ old_rules = rs->rules[rs_num].active.ptr; old_rcount = rs->rules[rs_num].active.rcount; old_array = rs->rules[rs_num].active.ptr_array; rs->rules[rs_num].active.ptr = rs->rules[rs_num].inactive.ptr; rs->rules[rs_num].active.ptr_array = rs->rules[rs_num].inactive.ptr_array; rs->rules[rs_num].active.rcount = rs->rules[rs_num].inactive.rcount; rs->rules[rs_num].inactive.ptr = old_rules; rs->rules[rs_num].inactive.ptr_array = old_array; rs->rules[rs_num].inactive.rcount = old_rcount; rs->rules[rs_num].active.ticket = rs->rules[rs_num].inactive.ticket; pf_calc_skip_steps(rs->rules[rs_num].active.ptr); /* Purge the old rule list. */ while ((rule = TAILQ_FIRST(old_rules)) != NULL) pf_unlink_rule(old_rules, rule); if (rs->rules[rs_num].inactive.ptr_array) free(rs->rules[rs_num].inactive.ptr_array, M_TEMP); rs->rules[rs_num].inactive.ptr_array = NULL; rs->rules[rs_num].inactive.rcount = 0; rs->rules[rs_num].inactive.open = 0; pf_remove_if_empty_ruleset(rs); return (0); } static int pf_setup_pfsync_matching(struct pf_ruleset *rs) { MD5_CTX ctx; struct pf_rule *rule; int rs_cnt; u_int8_t digest[PF_MD5_DIGEST_LENGTH]; MD5Init(&ctx); for (rs_cnt = 0; rs_cnt < PF_RULESET_MAX; rs_cnt++) { /* XXX PF_RULESET_SCRUB as well? */ if (rs_cnt == PF_RULESET_SCRUB) continue; if (rs->rules[rs_cnt].inactive.ptr_array) free(rs->rules[rs_cnt].inactive.ptr_array, M_TEMP); rs->rules[rs_cnt].inactive.ptr_array = NULL; if (rs->rules[rs_cnt].inactive.rcount) { rs->rules[rs_cnt].inactive.ptr_array = malloc(sizeof(caddr_t) * rs->rules[rs_cnt].inactive.rcount, M_TEMP, M_NOWAIT); if (!rs->rules[rs_cnt].inactive.ptr_array) return (ENOMEM); } TAILQ_FOREACH(rule, rs->rules[rs_cnt].inactive.ptr, entries) { pf_hash_rule(&ctx, rule); (rs->rules[rs_cnt].inactive.ptr_array)[rule->nr] = rule; } } MD5Final(digest, &ctx); memcpy(V_pf_status.pf_chksum, digest, sizeof(V_pf_status.pf_chksum)); return (0); } static int pf_addr_setup(struct pf_ruleset *ruleset, struct pf_addr_wrap *addr, sa_family_t af) { int error = 0; switch (addr->type) { case PF_ADDR_TABLE: addr->p.tbl = pfr_attach_table(ruleset, addr->v.tblname); if (addr->p.tbl == NULL) error = ENOMEM; break; case PF_ADDR_DYNIFTL: error = pfi_dynaddr_setup(addr, af); break; } return (error); } static void pf_addr_copyout(struct pf_addr_wrap *addr) { switch (addr->type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_copyout(addr); break; case PF_ADDR_TABLE: pf_tbladdr_copyout(addr); break; } } #ifdef ALTQ /* * Handle export of struct pf_kaltq to user binaries that may be using any * version of struct pf_altq. */ static int pf_export_kaltq(struct pf_altq *q, struct pfioc_altq_v1 *pa, size_t ioc_size) { u_int32_t version; if (ioc_size == sizeof(struct pfioc_altq_v0)) version = 0; else version = pa->version; if (version > PFIOC_ALTQ_VERSION) return (EINVAL); #define ASSIGN(x) exported_q->x = q->x #define COPY(x) \ bcopy(&q->x, &exported_q->x, min(sizeof(q->x), sizeof(exported_q->x))) #define SATU16(x) (u_int32_t)uqmin((x), USHRT_MAX) #define SATU32(x) (u_int32_t)uqmin((x), UINT_MAX) switch (version) { case 0: { struct pf_altq_v0 *exported_q = &((struct pfioc_altq_v0 *)pa)->altq; COPY(ifname); ASSIGN(scheduler); ASSIGN(tbrsize); exported_q->tbrsize = SATU16(q->tbrsize); exported_q->ifbandwidth = SATU32(q->ifbandwidth); COPY(qname); COPY(parent); ASSIGN(parent_qid); exported_q->bandwidth = SATU32(q->bandwidth); ASSIGN(priority); ASSIGN(local_flags); ASSIGN(qlimit); ASSIGN(flags); if (q->scheduler == ALTQT_HFSC) { #define ASSIGN_OPT(x) exported_q->pq_u.hfsc_opts.x = q->pq_u.hfsc_opts.x #define ASSIGN_OPT_SATU32(x) exported_q->pq_u.hfsc_opts.x = \ SATU32(q->pq_u.hfsc_opts.x) ASSIGN_OPT_SATU32(rtsc_m1); ASSIGN_OPT(rtsc_d); ASSIGN_OPT_SATU32(rtsc_m2); ASSIGN_OPT_SATU32(lssc_m1); ASSIGN_OPT(lssc_d); ASSIGN_OPT_SATU32(lssc_m2); ASSIGN_OPT_SATU32(ulsc_m1); ASSIGN_OPT(ulsc_d); ASSIGN_OPT_SATU32(ulsc_m2); ASSIGN_OPT(flags); #undef ASSIGN_OPT #undef ASSIGN_OPT_SATU32 } else COPY(pq_u); ASSIGN(qid); break; } case 1: { struct pf_altq_v1 *exported_q = &((struct pfioc_altq_v1 *)pa)->altq; COPY(ifname); ASSIGN(scheduler); ASSIGN(tbrsize); ASSIGN(ifbandwidth); COPY(qname); COPY(parent); ASSIGN(parent_qid); ASSIGN(bandwidth); ASSIGN(priority); ASSIGN(local_flags); ASSIGN(qlimit); ASSIGN(flags); COPY(pq_u); ASSIGN(qid); break; } default: panic("%s: unhandled struct pfioc_altq version", __func__); break; } #undef ASSIGN #undef COPY #undef SATU16 #undef SATU32 return (0); } /* * Handle import to struct pf_kaltq of struct pf_altq from user binaries * that may be using any version of it. */ static int pf_import_kaltq(struct pfioc_altq_v1 *pa, struct pf_altq *q, size_t ioc_size) { u_int32_t version; if (ioc_size == sizeof(struct pfioc_altq_v0)) version = 0; else version = pa->version; if (version > PFIOC_ALTQ_VERSION) return (EINVAL); #define ASSIGN(x) q->x = imported_q->x #define COPY(x) \ bcopy(&imported_q->x, &q->x, min(sizeof(imported_q->x), sizeof(q->x))) switch (version) { case 0: { struct pf_altq_v0 *imported_q = &((struct pfioc_altq_v0 *)pa)->altq; COPY(ifname); ASSIGN(scheduler); ASSIGN(tbrsize); /* 16-bit -> 32-bit */ ASSIGN(ifbandwidth); /* 32-bit -> 64-bit */ COPY(qname); COPY(parent); ASSIGN(parent_qid); ASSIGN(bandwidth); /* 32-bit -> 64-bit */ ASSIGN(priority); ASSIGN(local_flags); ASSIGN(qlimit); ASSIGN(flags); if (imported_q->scheduler == ALTQT_HFSC) { #define ASSIGN_OPT(x) q->pq_u.hfsc_opts.x = imported_q->pq_u.hfsc_opts.x /* * The m1 and m2 parameters are being copied from * 32-bit to 64-bit. */ ASSIGN_OPT(rtsc_m1); ASSIGN_OPT(rtsc_d); ASSIGN_OPT(rtsc_m2); ASSIGN_OPT(lssc_m1); ASSIGN_OPT(lssc_d); ASSIGN_OPT(lssc_m2); ASSIGN_OPT(ulsc_m1); ASSIGN_OPT(ulsc_d); ASSIGN_OPT(ulsc_m2); ASSIGN_OPT(flags); #undef ASSIGN_OPT } else COPY(pq_u); ASSIGN(qid); break; } case 1: { struct pf_altq_v1 *imported_q = &((struct pfioc_altq_v1 *)pa)->altq; COPY(ifname); ASSIGN(scheduler); ASSIGN(tbrsize); ASSIGN(ifbandwidth); COPY(qname); COPY(parent); ASSIGN(parent_qid); ASSIGN(bandwidth); ASSIGN(priority); ASSIGN(local_flags); ASSIGN(qlimit); ASSIGN(flags); COPY(pq_u); ASSIGN(qid); break; } default: panic("%s: unhandled struct pfioc_altq version", __func__); break; } #undef ASSIGN #undef COPY return (0); } #endif /* ALTQ */ +static struct pf_altq * +pf_altq_get_nth_active(u_int32_t n) +{ + struct pf_altq *altq; + u_int32_t nr; + + nr = 0; + TAILQ_FOREACH(altq, V_pf_altq_ifs_active, entries) { + if (nr == n) + return (altq); + nr++; + } + + TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { + if (nr == n) + return (altq); + nr++; + } + + return (NULL); +} + static int pfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td) { int error = 0; PF_RULES_RLOCK_TRACKER; /* XXX keep in sync with switch() below */ if (securelevel_gt(td->td_ucred, 2)) switch (cmd) { case DIOCGETRULES: case DIOCGETRULE: case DIOCGETADDRS: case DIOCGETADDR: case DIOCGETSTATE: case DIOCSETSTATUSIF: case DIOCGETSTATUS: case DIOCCLRSTATUS: case DIOCNATLOOK: case DIOCSETDEBUG: case DIOCGETSTATES: case DIOCGETTIMEOUT: case DIOCCLRRULECTRS: case DIOCGETLIMIT: case DIOCGETALTQSV0: case DIOCGETALTQSV1: case DIOCGETALTQV0: case DIOCGETALTQV1: case DIOCGETQSTATSV0: case DIOCGETQSTATSV1: case DIOCGETRULESETS: case DIOCGETRULESET: case DIOCRGETTABLES: case DIOCRGETTSTATS: case DIOCRCLRTSTATS: case DIOCRCLRADDRS: case DIOCRADDADDRS: case DIOCRDELADDRS: case DIOCRSETADDRS: case DIOCRGETADDRS: case DIOCRGETASTATS: case DIOCRCLRASTATS: case DIOCRTSTADDRS: case DIOCOSFPGET: case DIOCGETSRCNODES: case DIOCCLRSRCNODES: case DIOCIGETIFACES: case DIOCGIFSPEEDV0: case DIOCGIFSPEEDV1: case DIOCSETIFFLAG: case DIOCCLRIFFLAG: break; case DIOCRCLRTABLES: case DIOCRADDTABLES: case DIOCRDELTABLES: case DIOCRSETTFLAGS: if (((struct pfioc_table *)addr)->pfrio_flags & PFR_FLAG_DUMMY) break; /* dummy operation ok */ return (EPERM); default: return (EPERM); } if (!(flags & FWRITE)) switch (cmd) { case DIOCGETRULES: case DIOCGETADDRS: case DIOCGETADDR: case DIOCGETSTATE: case DIOCGETSTATUS: case DIOCGETSTATES: case DIOCGETTIMEOUT: case DIOCGETLIMIT: case DIOCGETALTQSV0: case DIOCGETALTQSV1: case DIOCGETALTQV0: case DIOCGETALTQV1: case DIOCGETQSTATSV0: case DIOCGETQSTATSV1: case DIOCGETRULESETS: case DIOCGETRULESET: case DIOCNATLOOK: case DIOCRGETTABLES: case DIOCRGETTSTATS: case DIOCRGETADDRS: case DIOCRGETASTATS: case DIOCRTSTADDRS: case DIOCOSFPGET: case DIOCGETSRCNODES: case DIOCIGETIFACES: case DIOCGIFSPEEDV1: case DIOCGIFSPEEDV0: break; case DIOCRCLRTABLES: case DIOCRADDTABLES: case DIOCRDELTABLES: case DIOCRCLRTSTATS: case DIOCRCLRADDRS: case DIOCRADDADDRS: case DIOCRDELADDRS: case DIOCRSETADDRS: case DIOCRSETTFLAGS: if (((struct pfioc_table *)addr)->pfrio_flags & PFR_FLAG_DUMMY) { flags |= FWRITE; /* need write lock for dummy */ break; /* dummy operation ok */ } return (EACCES); case DIOCGETRULE: if (((struct pfioc_rule *)addr)->action == PF_GET_CLR_CNTR) return (EACCES); break; default: return (EACCES); } CURVNET_SET(TD_TO_VNET(td)); switch (cmd) { case DIOCSTART: sx_xlock(&pf_ioctl_lock); if (V_pf_status.running) error = EEXIST; else { int cpu; error = hook_pf(); if (error) { DPFPRINTF(PF_DEBUG_MISC, ("pf: pfil registration failed\n")); break; } V_pf_status.running = 1; V_pf_status.since = time_second; CPU_FOREACH(cpu) V_pf_stateid[cpu] = time_second; DPFPRINTF(PF_DEBUG_MISC, ("pf: started\n")); } break; case DIOCSTOP: sx_xlock(&pf_ioctl_lock); if (!V_pf_status.running) error = ENOENT; else { V_pf_status.running = 0; error = dehook_pf(); if (error) { V_pf_status.running = 1; DPFPRINTF(PF_DEBUG_MISC, ("pf: pfil unregistration failed\n")); } V_pf_status.since = time_second; DPFPRINTF(PF_DEBUG_MISC, ("pf: stopped\n")); } break; case DIOCADDRULE: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *rule, *tail; struct pf_pooladdr *pa; struct pfi_kif *kif = NULL; int rs_num; if (pr->rule.return_icmp >> 8 > ICMP_MAXTYPE) { error = EINVAL; break; } #ifndef INET if (pr->rule.af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pr->rule.af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ rule = malloc(sizeof(*rule), M_PFRULE, M_WAITOK); bcopy(&pr->rule, rule, sizeof(struct pf_rule)); if (rule->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); rule->states_cur = counter_u64_alloc(M_WAITOK); rule->states_tot = counter_u64_alloc(M_WAITOK); rule->src_nodes = counter_u64_alloc(M_WAITOK); rule->cuid = td->td_ucred->cr_ruid; rule->cpid = td->td_proc ? td->td_proc->p_pid : 0; TAILQ_INIT(&rule->rpool.list); #define ERROUT(x) { error = (x); goto DIOCADDRULE_error; } PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) ERROUT(EINVAL); rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) ERROUT(EINVAL); if (pr->ticket != ruleset->rules[rs_num].inactive.ticket) { DPFPRINTF(PF_DEBUG_MISC, ("ticket: %d != [%d]%d\n", pr->ticket, rs_num, ruleset->rules[rs_num].inactive.ticket)); ERROUT(EBUSY); } if (pr->pool_ticket != V_ticket_pabuf) { DPFPRINTF(PF_DEBUG_MISC, ("pool_ticket: %d != %d\n", pr->pool_ticket, V_ticket_pabuf)); ERROUT(EBUSY); } tail = TAILQ_LAST(ruleset->rules[rs_num].inactive.ptr, pf_rulequeue); if (tail) rule->nr = tail->nr + 1; else rule->nr = 0; if (rule->ifname[0]) { rule->kif = pfi_kif_attach(kif, rule->ifname); pfi_kif_ref(rule->kif); } else rule->kif = NULL; if (rule->rtableid > 0 && rule->rtableid >= rt_numfibs) error = EBUSY; #ifdef ALTQ /* set queue IDs */ if (rule->qname[0] != 0) { if ((rule->qid = pf_qname2qid(rule->qname)) == 0) error = EBUSY; else if (rule->pqname[0] != 0) { if ((rule->pqid = pf_qname2qid(rule->pqname)) == 0) error = EBUSY; } else rule->pqid = rule->qid; } #endif if (rule->tagname[0]) if ((rule->tag = pf_tagname2tag(rule->tagname)) == 0) error = EBUSY; if (rule->match_tagname[0]) if ((rule->match_tag = pf_tagname2tag(rule->match_tagname)) == 0) error = EBUSY; if (rule->rt && !rule->direction) error = EINVAL; if (!rule->log) rule->logif = 0; if (rule->logif >= PFLOGIFS_MAX) error = EINVAL; if (pf_addr_setup(ruleset, &rule->src.addr, rule->af)) error = ENOMEM; if (pf_addr_setup(ruleset, &rule->dst.addr, rule->af)) error = ENOMEM; if (pf_anchor_setup(rule, ruleset, pr->anchor_call)) error = EINVAL; if (rule->scrub_flags & PFSTATE_SETPRIO && (rule->set_prio[0] > PF_PRIO_MAX || rule->set_prio[1] > PF_PRIO_MAX)) error = EINVAL; TAILQ_FOREACH(pa, &V_pf_pabuf, entries) if (pa->addr.type == PF_ADDR_TABLE) { pa->addr.p.tbl = pfr_attach_table(ruleset, pa->addr.v.tblname); if (pa->addr.p.tbl == NULL) error = ENOMEM; } rule->overload_tbl = NULL; if (rule->overload_tblname[0]) { if ((rule->overload_tbl = pfr_attach_table(ruleset, rule->overload_tblname)) == NULL) error = EINVAL; else rule->overload_tbl->pfrkt_flags |= PFR_TFLAG_ACTIVE; } pf_mv_pool(&V_pf_pabuf, &rule->rpool.list); if (((((rule->action == PF_NAT) || (rule->action == PF_RDR) || (rule->action == PF_BINAT)) && rule->anchor == NULL) || (rule->rt > PF_NOPFROUTE)) && (TAILQ_FIRST(&rule->rpool.list) == NULL)) error = EINVAL; if (error) { pf_free_rule(rule); PF_RULES_WUNLOCK(); break; } rule->rpool.cur = TAILQ_FIRST(&rule->rpool.list); rule->evaluations = rule->packets[0] = rule->packets[1] = rule->bytes[0] = rule->bytes[1] = 0; TAILQ_INSERT_TAIL(ruleset->rules[rs_num].inactive.ptr, rule, entries); ruleset->rules[rs_num].inactive.rcount++; PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCADDRULE_error: PF_RULES_WUNLOCK(); counter_u64_free(rule->states_cur); counter_u64_free(rule->states_tot); counter_u64_free(rule->src_nodes); free(rule, M_PFRULE); if (kif) free(kif, PFI_MTYPE); break; } case DIOCGETRULES: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *tail; int rs_num; PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); error = EINVAL; break; } tail = TAILQ_LAST(ruleset->rules[rs_num].active.ptr, pf_rulequeue); if (tail) pr->nr = tail->nr + 1; else pr->nr = 0; pr->ticket = ruleset->rules[rs_num].active.ticket; PF_RULES_WUNLOCK(); break; } case DIOCGETRULE: { struct pfioc_rule *pr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *rule; int rs_num, i; PF_RULES_WLOCK(); pr->anchor[sizeof(pr->anchor) - 1] = 0; ruleset = pf_find_ruleset(pr->anchor); if (ruleset == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } rs_num = pf_get_ruleset_number(pr->rule.action); if (rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); error = EINVAL; break; } if (pr->ticket != ruleset->rules[rs_num].active.ticket) { PF_RULES_WUNLOCK(); error = EBUSY; break; } rule = TAILQ_FIRST(ruleset->rules[rs_num].active.ptr); while ((rule != NULL) && (rule->nr != pr->nr)) rule = TAILQ_NEXT(rule, entries); if (rule == NULL) { PF_RULES_WUNLOCK(); error = EBUSY; break; } bcopy(rule, &pr->rule, sizeof(struct pf_rule)); pr->rule.u_states_cur = counter_u64_fetch(rule->states_cur); pr->rule.u_states_tot = counter_u64_fetch(rule->states_tot); pr->rule.u_src_nodes = counter_u64_fetch(rule->src_nodes); if (pf_anchor_copyout(ruleset, rule, pr)) { PF_RULES_WUNLOCK(); error = EBUSY; break; } pf_addr_copyout(&pr->rule.src.addr); pf_addr_copyout(&pr->rule.dst.addr); for (i = 0; i < PF_SKIP_COUNT; ++i) if (rule->skip[i].ptr == NULL) pr->rule.skip[i].nr = -1; else pr->rule.skip[i].nr = rule->skip[i].ptr->nr; if (pr->action == PF_GET_CLR_CNTR) { rule->evaluations = 0; rule->packets[0] = rule->packets[1] = 0; rule->bytes[0] = rule->bytes[1] = 0; counter_u64_zero(rule->states_tot); } PF_RULES_WUNLOCK(); break; } case DIOCCHANGERULE: { struct pfioc_rule *pcr = (struct pfioc_rule *)addr; struct pf_ruleset *ruleset; struct pf_rule *oldrule = NULL, *newrule = NULL; struct pfi_kif *kif = NULL; struct pf_pooladdr *pa; u_int32_t nr = 0; int rs_num; if (pcr->action < PF_CHANGE_ADD_HEAD || pcr->action > PF_CHANGE_GET_TICKET) { error = EINVAL; break; } if (pcr->rule.return_icmp >> 8 > ICMP_MAXTYPE) { error = EINVAL; break; } if (pcr->action != PF_CHANGE_REMOVE) { #ifndef INET if (pcr->rule.af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pcr->rule.af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ newrule = malloc(sizeof(*newrule), M_PFRULE, M_WAITOK); bcopy(&pcr->rule, newrule, sizeof(struct pf_rule)); if (newrule->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); newrule->states_cur = counter_u64_alloc(M_WAITOK); newrule->states_tot = counter_u64_alloc(M_WAITOK); newrule->src_nodes = counter_u64_alloc(M_WAITOK); newrule->cuid = td->td_ucred->cr_ruid; newrule->cpid = td->td_proc ? td->td_proc->p_pid : 0; TAILQ_INIT(&newrule->rpool.list); } #define ERROUT(x) { error = (x); goto DIOCCHANGERULE_error; } PF_RULES_WLOCK(); if (!(pcr->action == PF_CHANGE_REMOVE || pcr->action == PF_CHANGE_GET_TICKET) && pcr->pool_ticket != V_ticket_pabuf) ERROUT(EBUSY); ruleset = pf_find_ruleset(pcr->anchor); if (ruleset == NULL) ERROUT(EINVAL); rs_num = pf_get_ruleset_number(pcr->rule.action); if (rs_num >= PF_RULESET_MAX) ERROUT(EINVAL); if (pcr->action == PF_CHANGE_GET_TICKET) { pcr->ticket = ++ruleset->rules[rs_num].active.ticket; ERROUT(0); } else if (pcr->ticket != ruleset->rules[rs_num].active.ticket) ERROUT(EINVAL); if (pcr->action != PF_CHANGE_REMOVE) { if (newrule->ifname[0]) { newrule->kif = pfi_kif_attach(kif, newrule->ifname); pfi_kif_ref(newrule->kif); } else newrule->kif = NULL; if (newrule->rtableid > 0 && newrule->rtableid >= rt_numfibs) error = EBUSY; #ifdef ALTQ /* set queue IDs */ if (newrule->qname[0] != 0) { if ((newrule->qid = pf_qname2qid(newrule->qname)) == 0) error = EBUSY; else if (newrule->pqname[0] != 0) { if ((newrule->pqid = pf_qname2qid(newrule->pqname)) == 0) error = EBUSY; } else newrule->pqid = newrule->qid; } #endif /* ALTQ */ if (newrule->tagname[0]) if ((newrule->tag = pf_tagname2tag(newrule->tagname)) == 0) error = EBUSY; if (newrule->match_tagname[0]) if ((newrule->match_tag = pf_tagname2tag( newrule->match_tagname)) == 0) error = EBUSY; if (newrule->rt && !newrule->direction) error = EINVAL; if (!newrule->log) newrule->logif = 0; if (newrule->logif >= PFLOGIFS_MAX) error = EINVAL; if (pf_addr_setup(ruleset, &newrule->src.addr, newrule->af)) error = ENOMEM; if (pf_addr_setup(ruleset, &newrule->dst.addr, newrule->af)) error = ENOMEM; if (pf_anchor_setup(newrule, ruleset, pcr->anchor_call)) error = EINVAL; TAILQ_FOREACH(pa, &V_pf_pabuf, entries) if (pa->addr.type == PF_ADDR_TABLE) { pa->addr.p.tbl = pfr_attach_table(ruleset, pa->addr.v.tblname); if (pa->addr.p.tbl == NULL) error = ENOMEM; } newrule->overload_tbl = NULL; if (newrule->overload_tblname[0]) { if ((newrule->overload_tbl = pfr_attach_table( ruleset, newrule->overload_tblname)) == NULL) error = EINVAL; else newrule->overload_tbl->pfrkt_flags |= PFR_TFLAG_ACTIVE; } pf_mv_pool(&V_pf_pabuf, &newrule->rpool.list); if (((((newrule->action == PF_NAT) || (newrule->action == PF_RDR) || (newrule->action == PF_BINAT) || (newrule->rt > PF_NOPFROUTE)) && !newrule->anchor)) && (TAILQ_FIRST(&newrule->rpool.list) == NULL)) error = EINVAL; if (error) { pf_free_rule(newrule); PF_RULES_WUNLOCK(); break; } newrule->rpool.cur = TAILQ_FIRST(&newrule->rpool.list); newrule->evaluations = 0; newrule->packets[0] = newrule->packets[1] = 0; newrule->bytes[0] = newrule->bytes[1] = 0; } pf_empty_pool(&V_pf_pabuf); if (pcr->action == PF_CHANGE_ADD_HEAD) oldrule = TAILQ_FIRST( ruleset->rules[rs_num].active.ptr); else if (pcr->action == PF_CHANGE_ADD_TAIL) oldrule = TAILQ_LAST( ruleset->rules[rs_num].active.ptr, pf_rulequeue); else { oldrule = TAILQ_FIRST( ruleset->rules[rs_num].active.ptr); while ((oldrule != NULL) && (oldrule->nr != pcr->nr)) oldrule = TAILQ_NEXT(oldrule, entries); if (oldrule == NULL) { if (newrule != NULL) pf_free_rule(newrule); PF_RULES_WUNLOCK(); error = EINVAL; break; } } if (pcr->action == PF_CHANGE_REMOVE) { pf_unlink_rule(ruleset->rules[rs_num].active.ptr, oldrule); ruleset->rules[rs_num].active.rcount--; } else { if (oldrule == NULL) TAILQ_INSERT_TAIL( ruleset->rules[rs_num].active.ptr, newrule, entries); else if (pcr->action == PF_CHANGE_ADD_HEAD || pcr->action == PF_CHANGE_ADD_BEFORE) TAILQ_INSERT_BEFORE(oldrule, newrule, entries); else TAILQ_INSERT_AFTER( ruleset->rules[rs_num].active.ptr, oldrule, newrule, entries); ruleset->rules[rs_num].active.rcount++; } nr = 0; TAILQ_FOREACH(oldrule, ruleset->rules[rs_num].active.ptr, entries) oldrule->nr = nr++; ruleset->rules[rs_num].active.ticket++; pf_calc_skip_steps(ruleset->rules[rs_num].active.ptr); pf_remove_if_empty_ruleset(ruleset); PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCCHANGERULE_error: PF_RULES_WUNLOCK(); if (newrule != NULL) { counter_u64_free(newrule->states_cur); counter_u64_free(newrule->states_tot); counter_u64_free(newrule->src_nodes); free(newrule, M_PFRULE); } if (kif != NULL) free(kif, PFI_MTYPE); break; } case DIOCCLRSTATES: { struct pf_state *s; struct pfioc_state_kill *psk = (struct pfioc_state_kill *)addr; u_int i, killed = 0; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock_DIOCCLRSTATES: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) if (!psk->psk_ifname[0] || !strcmp(psk->psk_ifname, s->kif->pfik_name)) { /* * Don't send out individual * delete messages. */ s->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(s, PF_ENTER_LOCKED); killed++; goto relock_DIOCCLRSTATES; } PF_HASHROW_UNLOCK(ih); } psk->psk_killed = killed; if (V_pfsync_clear_states_ptr != NULL) V_pfsync_clear_states_ptr(V_pf_status.hostid, psk->psk_ifname); break; } case DIOCKILLSTATES: { struct pf_state *s; struct pf_state_key *sk; struct pf_addr *srcaddr, *dstaddr; u_int16_t srcport, dstport; struct pfioc_state_kill *psk = (struct pfioc_state_kill *)addr; u_int i, killed = 0; if (psk->psk_pfcmp.id) { if (psk->psk_pfcmp.creatorid == 0) psk->psk_pfcmp.creatorid = V_pf_status.hostid; if ((s = pf_find_state_byid(psk->psk_pfcmp.id, psk->psk_pfcmp.creatorid))) { pf_unlink_state(s, PF_ENTER_LOCKED); psk->psk_killed = 1; } break; } for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock_DIOCKILLSTATES: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { sk = s->key[PF_SK_WIRE]; if (s->direction == PF_OUT) { srcaddr = &sk->addr[1]; dstaddr = &sk->addr[0]; srcport = sk->port[1]; dstport = sk->port[0]; } else { srcaddr = &sk->addr[0]; dstaddr = &sk->addr[1]; srcport = sk->port[0]; dstport = sk->port[1]; } if ((!psk->psk_af || sk->af == psk->psk_af) && (!psk->psk_proto || psk->psk_proto == sk->proto) && PF_MATCHA(psk->psk_src.neg, &psk->psk_src.addr.v.a.addr, &psk->psk_src.addr.v.a.mask, srcaddr, sk->af) && PF_MATCHA(psk->psk_dst.neg, &psk->psk_dst.addr.v.a.addr, &psk->psk_dst.addr.v.a.mask, dstaddr, sk->af) && (psk->psk_src.port_op == 0 || pf_match_port(psk->psk_src.port_op, psk->psk_src.port[0], psk->psk_src.port[1], srcport)) && (psk->psk_dst.port_op == 0 || pf_match_port(psk->psk_dst.port_op, psk->psk_dst.port[0], psk->psk_dst.port[1], dstport)) && (!psk->psk_label[0] || (s->rule.ptr->label[0] && !strcmp(psk->psk_label, s->rule.ptr->label))) && (!psk->psk_ifname[0] || !strcmp(psk->psk_ifname, s->kif->pfik_name))) { pf_unlink_state(s, PF_ENTER_LOCKED); killed++; goto relock_DIOCKILLSTATES; } } PF_HASHROW_UNLOCK(ih); } psk->psk_killed = killed; break; } case DIOCADDSTATE: { struct pfioc_state *ps = (struct pfioc_state *)addr; struct pfsync_state *sp = &ps->state; if (sp->timeout >= PFTM_MAX) { error = EINVAL; break; } if (V_pfsync_state_import_ptr != NULL) { PF_RULES_RLOCK(); error = V_pfsync_state_import_ptr(sp, PFSYNC_SI_IOCTL); PF_RULES_RUNLOCK(); } else error = EOPNOTSUPP; break; } case DIOCGETSTATE: { struct pfioc_state *ps = (struct pfioc_state *)addr; struct pf_state *s; s = pf_find_state_byid(ps->state.id, ps->state.creatorid); if (s == NULL) { error = ENOENT; break; } pfsync_state_export(&ps->state, s); PF_STATE_UNLOCK(s); break; } case DIOCGETSTATES: { struct pfioc_states *ps = (struct pfioc_states *)addr; struct pf_state *s; struct pfsync_state *pstore, *p; int i, nr; if (ps->ps_len == 0) { nr = uma_zone_get_cur(V_pf_state_z); ps->ps_len = sizeof(struct pfsync_state) * nr; break; } p = pstore = malloc(ps->ps_len, M_TEMP, M_WAITOK); nr = 0; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (s->timeout == PFTM_UNLINKED) continue; if ((nr+1) * sizeof(*p) > ps->ps_len) { PF_HASHROW_UNLOCK(ih); goto DIOCGETSTATES_full; } pfsync_state_export(p, s); p++; nr++; } PF_HASHROW_UNLOCK(ih); } DIOCGETSTATES_full: error = copyout(pstore, ps->ps_states, sizeof(struct pfsync_state) * nr); if (error) { free(pstore, M_TEMP); break; } ps->ps_len = sizeof(struct pfsync_state) * nr; free(pstore, M_TEMP); break; } case DIOCGETSTATUS: { struct pf_status *s = (struct pf_status *)addr; PF_RULES_RLOCK(); s->running = V_pf_status.running; s->since = V_pf_status.since; s->debug = V_pf_status.debug; s->hostid = V_pf_status.hostid; s->states = V_pf_status.states; s->src_nodes = V_pf_status.src_nodes; for (int i = 0; i < PFRES_MAX; i++) s->counters[i] = counter_u64_fetch(V_pf_status.counters[i]); for (int i = 0; i < LCNT_MAX; i++) s->lcounters[i] = counter_u64_fetch(V_pf_status.lcounters[i]); for (int i = 0; i < FCNT_MAX; i++) s->fcounters[i] = counter_u64_fetch(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) s->scounters[i] = counter_u64_fetch(V_pf_status.scounters[i]); bcopy(V_pf_status.ifname, s->ifname, IFNAMSIZ); bcopy(V_pf_status.pf_chksum, s->pf_chksum, PF_MD5_DIGEST_LENGTH); pfi_update_status(s->ifname, s); PF_RULES_RUNLOCK(); break; } case DIOCSETSTATUSIF: { struct pfioc_if *pi = (struct pfioc_if *)addr; if (pi->ifname[0] == 0) { bzero(V_pf_status.ifname, IFNAMSIZ); break; } PF_RULES_WLOCK(); strlcpy(V_pf_status.ifname, pi->ifname, IFNAMSIZ); PF_RULES_WUNLOCK(); break; } case DIOCCLRSTATUS: { PF_RULES_WLOCK(); for (int i = 0; i < PFRES_MAX; i++) counter_u64_zero(V_pf_status.counters[i]); for (int i = 0; i < FCNT_MAX; i++) counter_u64_zero(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) counter_u64_zero(V_pf_status.scounters[i]); for (int i = 0; i < LCNT_MAX; i++) counter_u64_zero(V_pf_status.lcounters[i]); V_pf_status.since = time_second; if (*V_pf_status.ifname) pfi_update_status(V_pf_status.ifname, NULL); PF_RULES_WUNLOCK(); break; } case DIOCNATLOOK: { struct pfioc_natlook *pnl = (struct pfioc_natlook *)addr; struct pf_state_key *sk; struct pf_state *state; struct pf_state_key_cmp key; int m = 0, direction = pnl->direction; int sidx, didx; /* NATLOOK src and dst are reversed, so reverse sidx/didx */ sidx = (direction == PF_IN) ? 1 : 0; didx = (direction == PF_IN) ? 0 : 1; if (!pnl->proto || PF_AZERO(&pnl->saddr, pnl->af) || PF_AZERO(&pnl->daddr, pnl->af) || ((pnl->proto == IPPROTO_TCP || pnl->proto == IPPROTO_UDP) && (!pnl->dport || !pnl->sport))) error = EINVAL; else { bzero(&key, sizeof(key)); key.af = pnl->af; key.proto = pnl->proto; PF_ACPY(&key.addr[sidx], &pnl->saddr, pnl->af); key.port[sidx] = pnl->sport; PF_ACPY(&key.addr[didx], &pnl->daddr, pnl->af); key.port[didx] = pnl->dport; state = pf_find_state_all(&key, direction, &m); if (m > 1) error = E2BIG; /* more than one state */ else if (state != NULL) { /* XXXGL: not locked read */ sk = state->key[sidx]; PF_ACPY(&pnl->rsaddr, &sk->addr[sidx], sk->af); pnl->rsport = sk->port[sidx]; PF_ACPY(&pnl->rdaddr, &sk->addr[didx], sk->af); pnl->rdport = sk->port[didx]; } else error = ENOENT; } break; } case DIOCSETTIMEOUT: { struct pfioc_tm *pt = (struct pfioc_tm *)addr; int old; if (pt->timeout < 0 || pt->timeout >= PFTM_MAX || pt->seconds < 0) { error = EINVAL; break; } PF_RULES_WLOCK(); old = V_pf_default_rule.timeout[pt->timeout]; if (pt->timeout == PFTM_INTERVAL && pt->seconds == 0) pt->seconds = 1; V_pf_default_rule.timeout[pt->timeout] = pt->seconds; if (pt->timeout == PFTM_INTERVAL && pt->seconds < old) wakeup(pf_purge_thread); pt->seconds = old; PF_RULES_WUNLOCK(); break; } case DIOCGETTIMEOUT: { struct pfioc_tm *pt = (struct pfioc_tm *)addr; if (pt->timeout < 0 || pt->timeout >= PFTM_MAX) { error = EINVAL; break; } PF_RULES_RLOCK(); pt->seconds = V_pf_default_rule.timeout[pt->timeout]; PF_RULES_RUNLOCK(); break; } case DIOCGETLIMIT: { struct pfioc_limit *pl = (struct pfioc_limit *)addr; if (pl->index < 0 || pl->index >= PF_LIMIT_MAX) { error = EINVAL; break; } PF_RULES_RLOCK(); pl->limit = V_pf_limits[pl->index].limit; PF_RULES_RUNLOCK(); break; } case DIOCSETLIMIT: { struct pfioc_limit *pl = (struct pfioc_limit *)addr; int old_limit; PF_RULES_WLOCK(); if (pl->index < 0 || pl->index >= PF_LIMIT_MAX || V_pf_limits[pl->index].zone == NULL) { PF_RULES_WUNLOCK(); error = EINVAL; break; } uma_zone_set_max(V_pf_limits[pl->index].zone, pl->limit); old_limit = V_pf_limits[pl->index].limit; V_pf_limits[pl->index].limit = pl->limit; pl->limit = old_limit; PF_RULES_WUNLOCK(); break; } case DIOCSETDEBUG: { u_int32_t *level = (u_int32_t *)addr; PF_RULES_WLOCK(); V_pf_status.debug = *level; PF_RULES_WUNLOCK(); break; } case DIOCCLRRULECTRS: { /* obsoleted by DIOCGETRULE with action=PF_GET_CLR_CNTR */ struct pf_ruleset *ruleset = &pf_main_ruleset; struct pf_rule *rule; PF_RULES_WLOCK(); TAILQ_FOREACH(rule, ruleset->rules[PF_RULESET_FILTER].active.ptr, entries) { rule->evaluations = 0; rule->packets[0] = rule->packets[1] = 0; rule->bytes[0] = rule->bytes[1] = 0; } PF_RULES_WUNLOCK(); break; } case DIOCGIFSPEEDV0: case DIOCGIFSPEEDV1: { struct pf_ifspeed_v1 *psp = (struct pf_ifspeed_v1 *)addr; struct pf_ifspeed_v1 ps; struct ifnet *ifp; if (psp->ifname[0] != 0) { /* Can we completely trust user-land? */ strlcpy(ps.ifname, psp->ifname, IFNAMSIZ); ifp = ifunit(ps.ifname); if (ifp != NULL) { psp->baudrate32 = (u_int32_t)uqmin(ifp->if_baudrate, UINT_MAX); if (cmd == DIOCGIFSPEEDV1) psp->baudrate = ifp->if_baudrate; } else error = EINVAL; } else error = EINVAL; break; } #ifdef ALTQ case DIOCSTARTALTQ: { struct pf_altq *altq; PF_RULES_WLOCK(); /* enable all altq interfaces on active list */ - TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { - if (altq->qname[0] == 0 && (altq->local_flags & - PFALTQ_FLAG_IF_REMOVED) == 0) { + TAILQ_FOREACH(altq, V_pf_altq_ifs_active, entries) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { error = pf_enable_altq(altq); if (error != 0) break; } } if (error == 0) V_pf_altq_running = 1; PF_RULES_WUNLOCK(); DPFPRINTF(PF_DEBUG_MISC, ("altq: started\n")); break; } case DIOCSTOPALTQ: { struct pf_altq *altq; PF_RULES_WLOCK(); /* disable all altq interfaces on active list */ - TAILQ_FOREACH(altq, V_pf_altqs_active, entries) { - if (altq->qname[0] == 0 && (altq->local_flags & - PFALTQ_FLAG_IF_REMOVED) == 0) { + TAILQ_FOREACH(altq, V_pf_altq_ifs_active, entries) { + if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) == 0) { error = pf_disable_altq(altq); if (error != 0) break; } } if (error == 0) V_pf_altq_running = 0; PF_RULES_WUNLOCK(); DPFPRINTF(PF_DEBUG_MISC, ("altq: stopped\n")); break; } case DIOCADDALTQV0: case DIOCADDALTQV1: { struct pfioc_altq_v1 *pa = (struct pfioc_altq_v1 *)addr; struct pf_altq *altq, *a; struct ifnet *ifp; altq = malloc(sizeof(*altq), M_PFALTQ, M_WAITOK | M_ZERO); error = pf_import_kaltq(pa, altq, IOCPARM_LEN(cmd)); if (error) break; altq->local_flags = 0; PF_RULES_WLOCK(); if (pa->ticket != V_ticket_altqs_inactive) { PF_RULES_WUNLOCK(); free(altq, M_PFALTQ); error = EBUSY; break; } /* * if this is for a queue, find the discipline and * copy the necessary fields */ if (altq->qname[0] != 0) { if ((altq->qid = pf_qname2qid(altq->qname)) == 0) { PF_RULES_WUNLOCK(); error = EBUSY; free(altq, M_PFALTQ); break; } altq->altq_disc = NULL; - TAILQ_FOREACH(a, V_pf_altqs_inactive, entries) { + TAILQ_FOREACH(a, V_pf_altq_ifs_inactive, entries) { if (strncmp(a->ifname, altq->ifname, - IFNAMSIZ) == 0 && a->qname[0] == 0) { + IFNAMSIZ) == 0) { altq->altq_disc = a->altq_disc; break; } } } if ((ifp = ifunit(altq->ifname)) == NULL) altq->local_flags |= PFALTQ_FLAG_IF_REMOVED; else - error = altq_add(altq); + error = altq_add(ifp, altq); if (error) { PF_RULES_WUNLOCK(); free(altq, M_PFALTQ); break; } - TAILQ_INSERT_TAIL(V_pf_altqs_inactive, altq, entries); + if (altq->qname[0] != 0) + TAILQ_INSERT_TAIL(V_pf_altqs_inactive, altq, entries); + else + TAILQ_INSERT_TAIL(V_pf_altq_ifs_inactive, altq, entries); /* version error check done on import above */ pf_export_kaltq(altq, pa, IOCPARM_LEN(cmd)); PF_RULES_WUNLOCK(); break; } case DIOCGETALTQSV0: case DIOCGETALTQSV1: { struct pfioc_altq_v1 *pa = (struct pfioc_altq_v1 *)addr; struct pf_altq *altq; PF_RULES_RLOCK(); pa->nr = 0; + TAILQ_FOREACH(altq, V_pf_altq_ifs_active, entries) + pa->nr++; TAILQ_FOREACH(altq, V_pf_altqs_active, entries) pa->nr++; pa->ticket = V_ticket_altqs_active; PF_RULES_RUNLOCK(); break; } case DIOCGETALTQV0: case DIOCGETALTQV1: { struct pfioc_altq_v1 *pa = (struct pfioc_altq_v1 *)addr; struct pf_altq *altq; - u_int32_t nr; PF_RULES_RLOCK(); if (pa->ticket != V_ticket_altqs_active) { PF_RULES_RUNLOCK(); error = EBUSY; break; } - nr = 0; - altq = TAILQ_FIRST(V_pf_altqs_active); - while ((altq != NULL) && (nr < pa->nr)) { - altq = TAILQ_NEXT(altq, entries); - nr++; - } + altq = pf_altq_get_nth_active(pa->nr); if (altq == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } pf_export_kaltq(altq, pa, IOCPARM_LEN(cmd)); PF_RULES_RUNLOCK(); break; } case DIOCCHANGEALTQV0: case DIOCCHANGEALTQV1: /* CHANGEALTQ not supported yet! */ error = ENODEV; break; case DIOCGETQSTATSV0: case DIOCGETQSTATSV1: { struct pfioc_qstats_v1 *pq = (struct pfioc_qstats_v1 *)addr; struct pf_altq *altq; - u_int32_t nr; int nbytes; u_int32_t version; PF_RULES_RLOCK(); if (pq->ticket != V_ticket_altqs_active) { PF_RULES_RUNLOCK(); error = EBUSY; break; } nbytes = pq->nbytes; - nr = 0; - altq = TAILQ_FIRST(V_pf_altqs_active); - while ((altq != NULL) && (nr < pq->nr)) { - altq = TAILQ_NEXT(altq, entries); - nr++; - } + altq = pf_altq_get_nth_active(pq->nr); if (altq == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } if ((altq->local_flags & PFALTQ_FLAG_IF_REMOVED) != 0) { PF_RULES_RUNLOCK(); error = ENXIO; break; } PF_RULES_RUNLOCK(); if (cmd == DIOCGETQSTATSV0) version = 0; /* DIOCGETQSTATSV0 means stats struct v0 */ else version = pq->version; error = altq_getqstats(altq, pq->buf, &nbytes, version); if (error == 0) { pq->scheduler = altq->scheduler; pq->nbytes = nbytes; } break; } #endif /* ALTQ */ case DIOCBEGINADDRS: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; PF_RULES_WLOCK(); pf_empty_pool(&V_pf_pabuf); pp->ticket = ++V_ticket_pabuf; PF_RULES_WUNLOCK(); break; } case DIOCADDADDR: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pooladdr *pa; struct pfi_kif *kif = NULL; #ifndef INET if (pp->af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pp->af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ if (pp->addr.addr.type != PF_ADDR_ADDRMASK && pp->addr.addr.type != PF_ADDR_DYNIFTL && pp->addr.addr.type != PF_ADDR_TABLE) { error = EINVAL; break; } pa = malloc(sizeof(*pa), M_PFRULE, M_WAITOK); bcopy(&pp->addr, pa, sizeof(struct pf_pooladdr)); if (pa->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); PF_RULES_WLOCK(); if (pp->ticket != V_ticket_pabuf) { PF_RULES_WUNLOCK(); if (pa->ifname[0]) free(kif, PFI_MTYPE); free(pa, M_PFRULE); error = EBUSY; break; } if (pa->ifname[0]) { pa->kif = pfi_kif_attach(kif, pa->ifname); pfi_kif_ref(pa->kif); } else pa->kif = NULL; if (pa->addr.type == PF_ADDR_DYNIFTL && ((error = pfi_dynaddr_setup(&pa->addr, pp->af)) != 0)) { if (pa->ifname[0]) pfi_kif_unref(pa->kif); PF_RULES_WUNLOCK(); free(pa, M_PFRULE); break; } TAILQ_INSERT_TAIL(&V_pf_pabuf, pa, entries); PF_RULES_WUNLOCK(); break; } case DIOCGETADDRS: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *pa; PF_RULES_RLOCK(); pp->nr = 0; pool = pf_get_pool(pp->anchor, pp->ticket, pp->r_action, pp->r_num, 0, 1, 0); if (pool == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } TAILQ_FOREACH(pa, &pool->list, entries) pp->nr++; PF_RULES_RUNLOCK(); break; } case DIOCGETADDR: { struct pfioc_pooladdr *pp = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *pa; u_int32_t nr = 0; PF_RULES_RLOCK(); pool = pf_get_pool(pp->anchor, pp->ticket, pp->r_action, pp->r_num, 0, 1, 1); if (pool == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } pa = TAILQ_FIRST(&pool->list); while ((pa != NULL) && (nr < pp->nr)) { pa = TAILQ_NEXT(pa, entries); nr++; } if (pa == NULL) { PF_RULES_RUNLOCK(); error = EBUSY; break; } bcopy(pa, &pp->addr, sizeof(struct pf_pooladdr)); pf_addr_copyout(&pp->addr.addr); PF_RULES_RUNLOCK(); break; } case DIOCCHANGEADDR: { struct pfioc_pooladdr *pca = (struct pfioc_pooladdr *)addr; struct pf_pool *pool; struct pf_pooladdr *oldpa = NULL, *newpa = NULL; struct pf_ruleset *ruleset; struct pfi_kif *kif = NULL; if (pca->action < PF_CHANGE_ADD_HEAD || pca->action > PF_CHANGE_REMOVE) { error = EINVAL; break; } if (pca->addr.addr.type != PF_ADDR_ADDRMASK && pca->addr.addr.type != PF_ADDR_DYNIFTL && pca->addr.addr.type != PF_ADDR_TABLE) { error = EINVAL; break; } if (pca->action != PF_CHANGE_REMOVE) { #ifndef INET if (pca->af == AF_INET) { error = EAFNOSUPPORT; break; } #endif /* INET */ #ifndef INET6 if (pca->af == AF_INET6) { error = EAFNOSUPPORT; break; } #endif /* INET6 */ newpa = malloc(sizeof(*newpa), M_PFRULE, M_WAITOK); bcopy(&pca->addr, newpa, sizeof(struct pf_pooladdr)); if (newpa->ifname[0]) kif = malloc(sizeof(*kif), PFI_MTYPE, M_WAITOK); newpa->kif = NULL; } #define ERROUT(x) { error = (x); goto DIOCCHANGEADDR_error; } PF_RULES_WLOCK(); ruleset = pf_find_ruleset(pca->anchor); if (ruleset == NULL) ERROUT(EBUSY); pool = pf_get_pool(pca->anchor, pca->ticket, pca->r_action, pca->r_num, pca->r_last, 1, 1); if (pool == NULL) ERROUT(EBUSY); if (pca->action != PF_CHANGE_REMOVE) { if (newpa->ifname[0]) { newpa->kif = pfi_kif_attach(kif, newpa->ifname); pfi_kif_ref(newpa->kif); kif = NULL; } switch (newpa->addr.type) { case PF_ADDR_DYNIFTL: error = pfi_dynaddr_setup(&newpa->addr, pca->af); break; case PF_ADDR_TABLE: newpa->addr.p.tbl = pfr_attach_table(ruleset, newpa->addr.v.tblname); if (newpa->addr.p.tbl == NULL) error = ENOMEM; break; } if (error) goto DIOCCHANGEADDR_error; } switch (pca->action) { case PF_CHANGE_ADD_HEAD: oldpa = TAILQ_FIRST(&pool->list); break; case PF_CHANGE_ADD_TAIL: oldpa = TAILQ_LAST(&pool->list, pf_palist); break; default: oldpa = TAILQ_FIRST(&pool->list); for (int i = 0; oldpa && i < pca->nr; i++) oldpa = TAILQ_NEXT(oldpa, entries); if (oldpa == NULL) ERROUT(EINVAL); } if (pca->action == PF_CHANGE_REMOVE) { TAILQ_REMOVE(&pool->list, oldpa, entries); switch (oldpa->addr.type) { case PF_ADDR_DYNIFTL: pfi_dynaddr_remove(oldpa->addr.p.dyn); break; case PF_ADDR_TABLE: pfr_detach_table(oldpa->addr.p.tbl); break; } if (oldpa->kif) pfi_kif_unref(oldpa->kif); free(oldpa, M_PFRULE); } else { if (oldpa == NULL) TAILQ_INSERT_TAIL(&pool->list, newpa, entries); else if (pca->action == PF_CHANGE_ADD_HEAD || pca->action == PF_CHANGE_ADD_BEFORE) TAILQ_INSERT_BEFORE(oldpa, newpa, entries); else TAILQ_INSERT_AFTER(&pool->list, oldpa, newpa, entries); } pool->cur = TAILQ_FIRST(&pool->list); PF_ACPY(&pool->counter, &pool->cur->addr.v.a.addr, pca->af); PF_RULES_WUNLOCK(); break; #undef ERROUT DIOCCHANGEADDR_error: if (newpa != NULL) { if (newpa->kif) pfi_kif_unref(newpa->kif); free(newpa, M_PFRULE); } PF_RULES_WUNLOCK(); if (kif != NULL) free(kif, PFI_MTYPE); break; } case DIOCGETRULESETS: { struct pfioc_ruleset *pr = (struct pfioc_ruleset *)addr; struct pf_ruleset *ruleset; struct pf_anchor *anchor; PF_RULES_RLOCK(); pr->path[sizeof(pr->path) - 1] = 0; if ((ruleset = pf_find_ruleset(pr->path)) == NULL) { PF_RULES_RUNLOCK(); error = ENOENT; break; } pr->nr = 0; if (ruleset->anchor == NULL) { /* XXX kludge for pf_main_ruleset */ RB_FOREACH(anchor, pf_anchor_global, &V_pf_anchors) if (anchor->parent == NULL) pr->nr++; } else { RB_FOREACH(anchor, pf_anchor_node, &ruleset->anchor->children) pr->nr++; } PF_RULES_RUNLOCK(); break; } case DIOCGETRULESET: { struct pfioc_ruleset *pr = (struct pfioc_ruleset *)addr; struct pf_ruleset *ruleset; struct pf_anchor *anchor; u_int32_t nr = 0; PF_RULES_RLOCK(); pr->path[sizeof(pr->path) - 1] = 0; if ((ruleset = pf_find_ruleset(pr->path)) == NULL) { PF_RULES_RUNLOCK(); error = ENOENT; break; } pr->name[0] = 0; if (ruleset->anchor == NULL) { /* XXX kludge for pf_main_ruleset */ RB_FOREACH(anchor, pf_anchor_global, &V_pf_anchors) if (anchor->parent == NULL && nr++ == pr->nr) { strlcpy(pr->name, anchor->name, sizeof(pr->name)); break; } } else { RB_FOREACH(anchor, pf_anchor_node, &ruleset->anchor->children) if (nr++ == pr->nr) { strlcpy(pr->name, anchor->name, sizeof(pr->name)); break; } } if (!pr->name[0]) error = EBUSY; PF_RULES_RUNLOCK(); break; } case DIOCRCLRTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; if (io->pfrio_esize != 0) { error = ENODEV; break; } PF_RULES_WLOCK(); error = pfr_clr_tables(&io->pfrio_table, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); break; } case DIOCRADDTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_table))) { error = ENOMEM; break; } totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_WAITOK); error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_add_tables(pfrts, io->pfrio_size, &io->pfrio_nadd, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRDELTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_table))) { error = ENOMEM; break; } totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_WAITOK); error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_del_tables(pfrts, io->pfrio_size, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRGETTABLES: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_RLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_RUNLOCK(); break; } error = pfr_get_tables(&io->pfrio_table, pfrts, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfrts, io->pfrio_buffer, totlen); free(pfrts, M_TEMP); break; } case DIOCRGETTSTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_tstats *pfrtstats; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_tstats)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_tstats); pfrtstats = mallocarray(io->pfrio_size, sizeof(struct pfr_tstats), M_TEMP, M_NOWAIT); if (pfrtstats == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = pfr_get_tstats(&io->pfrio_table, pfrtstats, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0) error = copyout(pfrtstats, io->pfrio_buffer, totlen); free(pfrtstats, M_TEMP); break; } case DIOCRCLRTSTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); PF_RULES_WUNLOCK(); break; } error = pfr_clr_tstats(pfrts, io->pfrio_size, &io->pfrio_nzero, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRSETTFLAGS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_table *pfrts; size_t totlen, n; if (io->pfrio_esize != sizeof(struct pfr_table)) { error = ENODEV; break; } PF_RULES_WLOCK(); n = pfr_table_count(&io->pfrio_table, io->pfrio_flags); io->pfrio_size = min(io->pfrio_size, n); totlen = io->pfrio_size * sizeof(struct pfr_table); pfrts = mallocarray(io->pfrio_size, sizeof(struct pfr_table), M_TEMP, M_NOWAIT); if (pfrts == NULL) { error = ENOMEM; PF_RULES_WUNLOCK(); break; } error = copyin(io->pfrio_buffer, pfrts, totlen); if (error) { free(pfrts, M_TEMP); PF_RULES_WUNLOCK(); break; } error = pfr_set_tflags(pfrts, io->pfrio_size, io->pfrio_setflag, io->pfrio_clrflag, &io->pfrio_nchange, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfrts, M_TEMP); break; } case DIOCRCLRADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; if (io->pfrio_esize != 0) { error = ENODEV; break; } PF_RULES_WLOCK(); error = pfr_clr_addrs(&io->pfrio_table, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); break; } case DIOCRADDADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_add_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nadd, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRDELADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_del_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_ndel, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRSETADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen, count; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size2 < 0) { error = EINVAL; break; } count = max(io->pfrio_size, io->pfrio_size2); if (count > pf_ioctl_maxcount || WOULD_OVERFLOW(count, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = count * sizeof(struct pfr_addr); pfras = mallocarray(count, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_set_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_size2, &io->pfrio_nadd, &io->pfrio_ndel, &io->pfrio_nchange, io->pfrio_flags | PFR_FLAG_USERIOCTL, 0); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRGETADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } PF_RULES_RLOCK(); error = pfr_get_addrs(&io->pfrio_table, pfras, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRGETASTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_astats *pfrastats; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_astats)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_astats))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_astats); pfrastats = mallocarray(io->pfrio_size, sizeof(struct pfr_astats), M_TEMP, M_NOWAIT); if (! pfrastats) { error = ENOMEM; break; } PF_RULES_RLOCK(); error = pfr_get_astats(&io->pfrio_table, pfrastats, &io->pfrio_size, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfrastats, io->pfrio_buffer, totlen); free(pfrastats, M_TEMP); break; } case DIOCRCLRASTATS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_clr_astats(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nzero, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); if (error == 0 && io->pfrio_flags & PFR_FLAG_FEEDBACK) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRTSTADDRS: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_RLOCK(); error = pfr_tst_addrs(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nmatch, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_RUNLOCK(); if (error == 0) error = copyout(pfras, io->pfrio_buffer, totlen); free(pfras, M_TEMP); break; } case DIOCRINADEFINE: { struct pfioc_table *io = (struct pfioc_table *)addr; struct pfr_addr *pfras; size_t totlen; if (io->pfrio_esize != sizeof(struct pfr_addr)) { error = ENODEV; break; } if (io->pfrio_size < 0 || io->pfrio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfrio_size, sizeof(struct pfr_addr))) { error = EINVAL; break; } totlen = io->pfrio_size * sizeof(struct pfr_addr); pfras = mallocarray(io->pfrio_size, sizeof(struct pfr_addr), M_TEMP, M_NOWAIT); if (! pfras) { error = ENOMEM; break; } error = copyin(io->pfrio_buffer, pfras, totlen); if (error) { free(pfras, M_TEMP); break; } PF_RULES_WLOCK(); error = pfr_ina_define(&io->pfrio_table, pfras, io->pfrio_size, &io->pfrio_nadd, &io->pfrio_naddr, io->pfrio_ticket, io->pfrio_flags | PFR_FLAG_USERIOCTL); PF_RULES_WUNLOCK(); free(pfras, M_TEMP); break; } case DIOCOSFPADD: { struct pf_osfp_ioctl *io = (struct pf_osfp_ioctl *)addr; PF_RULES_WLOCK(); error = pf_osfp_add(io); PF_RULES_WUNLOCK(); break; } case DIOCOSFPGET: { struct pf_osfp_ioctl *io = (struct pf_osfp_ioctl *)addr; PF_RULES_RLOCK(); error = pf_osfp_get(io); PF_RULES_RUNLOCK(); break; } case DIOCXBEGIN: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioes, *ioe; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (! ioes) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if ((error = pf_begin_altq(&ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_begin(&table, &ioe->ticket, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; } default: if ((error = pf_begin_rules(&ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; } break; } } PF_RULES_WUNLOCK(); error = copyout(ioes, io->array, totlen); free(ioes, M_TEMP); break; } case DIOCXROLLBACK: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioe, *ioes; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (! ioes) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if ((error = pf_rollback_altq(ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_rollback(&table, ioe->ticket, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } default: if ((error = pf_rollback_rules(ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } } PF_RULES_WUNLOCK(); free(ioes, M_TEMP); break; } case DIOCXCOMMIT: { struct pfioc_trans *io = (struct pfioc_trans *)addr; struct pfioc_trans_e *ioe, *ioes; struct pf_ruleset *rs; size_t totlen; int i; if (io->esize != sizeof(*ioe)) { error = ENODEV; break; } if (io->size < 0 || io->size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->size, sizeof(struct pfioc_trans_e))) { error = EINVAL; break; } totlen = sizeof(struct pfioc_trans_e) * io->size; ioes = mallocarray(io->size, sizeof(struct pfioc_trans_e), M_TEMP, M_NOWAIT); if (ioes == NULL) { error = ENOMEM; break; } error = copyin(io->array, ioes, totlen); if (error) { free(ioes, M_TEMP); break; } PF_RULES_WLOCK(); /* First makes sure everything will succeed. */ for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if (ioe->anchor[0]) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } if (!V_altqs_inactive_open || ioe->ticket != V_ticket_altqs_inactive) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; #endif /* ALTQ */ case PF_RULESET_TABLE: rs = pf_find_ruleset(ioe->anchor); if (rs == NULL || !rs->topen || ioe->ticket != rs->tticket) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; default: if (ioe->rs_num < 0 || ioe->rs_num >= PF_RULESET_MAX) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EINVAL; goto fail; } rs = pf_find_ruleset(ioe->anchor); if (rs == NULL || !rs->rules[ioe->rs_num].inactive.open || rs->rules[ioe->rs_num].inactive.ticket != ioe->ticket) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); error = EBUSY; goto fail; } break; } } /* Now do the commit - no errors should happen here. */ for (i = 0, ioe = ioes; i < io->size; i++, ioe++) { switch (ioe->rs_num) { #ifdef ALTQ case PF_RULESET_ALTQ: if ((error = pf_commit_altq(ioe->ticket))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; #endif /* ALTQ */ case PF_RULESET_TABLE: { struct pfr_table table; bzero(&table, sizeof(table)); strlcpy(table.pfrt_anchor, ioe->anchor, sizeof(table.pfrt_anchor)); if ((error = pfr_ina_commit(&table, ioe->ticket, NULL, NULL, 0))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } default: if ((error = pf_commit_rules(ioe->ticket, ioe->rs_num, ioe->anchor))) { PF_RULES_WUNLOCK(); free(ioes, M_TEMP); goto fail; /* really bad */ } break; } } PF_RULES_WUNLOCK(); free(ioes, M_TEMP); break; } case DIOCGETSRCNODES: { struct pfioc_src_nodes *psn = (struct pfioc_src_nodes *)addr; struct pf_srchash *sh; struct pf_src_node *n, *p, *pstore; uint32_t i, nr = 0; for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) nr++; PF_HASHROW_UNLOCK(sh); } psn->psn_len = min(psn->psn_len, sizeof(struct pf_src_node) * nr); if (psn->psn_len == 0) { psn->psn_len = sizeof(struct pf_src_node) * nr; break; } p = pstore = malloc(psn->psn_len, M_TEMP, M_WAITOK); for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { int secs = time_uptime, diff; if ((nr + 1) * sizeof(*p) > (unsigned)psn->psn_len) break; bcopy(n, p, sizeof(struct pf_src_node)); if (n->rule.ptr != NULL) p->rule.nr = n->rule.ptr->nr; p->creation = secs - p->creation; if (p->expire > secs) p->expire -= secs; else p->expire = 0; /* Adjust the connection rate estimate. */ diff = secs - n->conn_rate.last; if (diff >= n->conn_rate.seconds) p->conn_rate.count = 0; else p->conn_rate.count -= n->conn_rate.count * diff / n->conn_rate.seconds; p++; nr++; } PF_HASHROW_UNLOCK(sh); } error = copyout(pstore, psn->psn_src_nodes, sizeof(struct pf_src_node) * nr); if (error) { free(pstore, M_TEMP); break; } psn->psn_len = sizeof(struct pf_src_node) * nr; free(pstore, M_TEMP); break; } case DIOCCLRSRCNODES: { pf_clear_srcnodes(NULL); pf_purge_expired_src_nodes(); break; } case DIOCKILLSRCNODES: pf_kill_srcnodes((struct pfioc_src_node_kill *)addr); break; case DIOCSETHOSTID: { u_int32_t *hostid = (u_int32_t *)addr; PF_RULES_WLOCK(); if (*hostid == 0) V_pf_status.hostid = arc4random(); else V_pf_status.hostid = *hostid; PF_RULES_WUNLOCK(); break; } case DIOCOSFPFLUSH: PF_RULES_WLOCK(); pf_osfp_flush(); PF_RULES_WUNLOCK(); break; case DIOCIGETIFACES: { struct pfioc_iface *io = (struct pfioc_iface *)addr; struct pfi_kif *ifstore; size_t bufsiz; if (io->pfiio_esize != sizeof(struct pfi_kif)) { error = ENODEV; break; } if (io->pfiio_size < 0 || io->pfiio_size > pf_ioctl_maxcount || WOULD_OVERFLOW(io->pfiio_size, sizeof(struct pfi_kif))) { error = EINVAL; break; } bufsiz = io->pfiio_size * sizeof(struct pfi_kif); ifstore = mallocarray(io->pfiio_size, sizeof(struct pfi_kif), M_TEMP, M_NOWAIT); if (ifstore == NULL) { error = ENOMEM; break; } PF_RULES_RLOCK(); pfi_get_ifaces(io->pfiio_name, ifstore, &io->pfiio_size); PF_RULES_RUNLOCK(); error = copyout(ifstore, io->pfiio_buffer, bufsiz); free(ifstore, M_TEMP); break; } case DIOCSETIFFLAG: { struct pfioc_iface *io = (struct pfioc_iface *)addr; PF_RULES_WLOCK(); error = pfi_set_flags(io->pfiio_name, io->pfiio_flags); PF_RULES_WUNLOCK(); break; } case DIOCCLRIFFLAG: { struct pfioc_iface *io = (struct pfioc_iface *)addr; PF_RULES_WLOCK(); error = pfi_clear_flags(io->pfiio_name, io->pfiio_flags); PF_RULES_WUNLOCK(); break; } default: error = ENODEV; break; } fail: if (sx_xlocked(&pf_ioctl_lock)) sx_xunlock(&pf_ioctl_lock); CURVNET_RESTORE(); return (error); } void pfsync_state_export(struct pfsync_state *sp, struct pf_state *st) { bzero(sp, sizeof(struct pfsync_state)); /* copy from state key */ sp->key[PF_SK_WIRE].addr[0] = st->key[PF_SK_WIRE]->addr[0]; sp->key[PF_SK_WIRE].addr[1] = st->key[PF_SK_WIRE]->addr[1]; sp->key[PF_SK_WIRE].port[0] = st->key[PF_SK_WIRE]->port[0]; sp->key[PF_SK_WIRE].port[1] = st->key[PF_SK_WIRE]->port[1]; sp->key[PF_SK_STACK].addr[0] = st->key[PF_SK_STACK]->addr[0]; sp->key[PF_SK_STACK].addr[1] = st->key[PF_SK_STACK]->addr[1]; sp->key[PF_SK_STACK].port[0] = st->key[PF_SK_STACK]->port[0]; sp->key[PF_SK_STACK].port[1] = st->key[PF_SK_STACK]->port[1]; sp->proto = st->key[PF_SK_WIRE]->proto; sp->af = st->key[PF_SK_WIRE]->af; /* copy from state */ strlcpy(sp->ifname, st->kif->pfik_name, sizeof(sp->ifname)); bcopy(&st->rt_addr, &sp->rt_addr, sizeof(sp->rt_addr)); sp->creation = htonl(time_uptime - st->creation); sp->expire = pf_state_expires(st); if (sp->expire <= time_uptime) sp->expire = htonl(0); else sp->expire = htonl(sp->expire - time_uptime); sp->direction = st->direction; sp->log = st->log; sp->timeout = st->timeout; sp->state_flags = st->state_flags; if (st->src_node) sp->sync_flags |= PFSYNC_FLAG_SRCNODE; if (st->nat_src_node) sp->sync_flags |= PFSYNC_FLAG_NATSRCNODE; sp->id = st->id; sp->creatorid = st->creatorid; pf_state_peer_hton(&st->src, &sp->src); pf_state_peer_hton(&st->dst, &sp->dst); if (st->rule.ptr == NULL) sp->rule = htonl(-1); else sp->rule = htonl(st->rule.ptr->nr); if (st->anchor.ptr == NULL) sp->anchor = htonl(-1); else sp->anchor = htonl(st->anchor.ptr->nr); if (st->nat_rule.ptr == NULL) sp->nat_rule = htonl(-1); else sp->nat_rule = htonl(st->nat_rule.ptr->nr); pf_state_counter_hton(st->packets[0], sp->packets[0]); pf_state_counter_hton(st->packets[1], sp->packets[1]); pf_state_counter_hton(st->bytes[0], sp->bytes[0]); pf_state_counter_hton(st->bytes[1], sp->bytes[1]); } static void pf_tbladdr_copyout(struct pf_addr_wrap *aw) { struct pfr_ktable *kt; KASSERT(aw->type == PF_ADDR_TABLE, ("%s: type %u", __func__, aw->type)); kt = aw->p.tbl; if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) kt = kt->pfrkt_root; aw->p.tbl = NULL; aw->p.tblcnt = (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) ? kt->pfrkt_cnt : -1; } /* * XXX - Check for version missmatch!!! */ static void pf_clear_states(void) { struct pf_state *s; u_int i; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; relock: PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { s->timeout = PFTM_PURGE; /* Don't send out individual delete messages. */ s->state_flags |= PFSTATE_NOSYNC; pf_unlink_state(s, PF_ENTER_LOCKED); goto relock; } PF_HASHROW_UNLOCK(ih); } } static int pf_clear_tables(void) { struct pfioc_table io; int error; bzero(&io, sizeof(io)); error = pfr_clr_tables(&io.pfrio_table, &io.pfrio_ndel, io.pfrio_flags); return (error); } static void pf_clear_srcnodes(struct pf_src_node *n) { struct pf_state *s; int i; for (i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (n == NULL || n == s->src_node) s->src_node = NULL; if (n == NULL || n == s->nat_src_node) s->nat_src_node = NULL; } PF_HASHROW_UNLOCK(ih); } if (n == NULL) { struct pf_srchash *sh; for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { PF_HASHROW_LOCK(sh); LIST_FOREACH(n, &sh->nodes, entry) { n->expire = 1; n->states = 0; } PF_HASHROW_UNLOCK(sh); } } else { /* XXX: hash slot should already be locked here. */ n->expire = 1; n->states = 0; } } static void pf_kill_srcnodes(struct pfioc_src_node_kill *psnk) { struct pf_src_node_list kill; LIST_INIT(&kill); for (int i = 0; i <= pf_srchashmask; i++) { struct pf_srchash *sh = &V_pf_srchash[i]; struct pf_src_node *sn, *tmp; PF_HASHROW_LOCK(sh); LIST_FOREACH_SAFE(sn, &sh->nodes, entry, tmp) if (PF_MATCHA(psnk->psnk_src.neg, &psnk->psnk_src.addr.v.a.addr, &psnk->psnk_src.addr.v.a.mask, &sn->addr, sn->af) && PF_MATCHA(psnk->psnk_dst.neg, &psnk->psnk_dst.addr.v.a.addr, &psnk->psnk_dst.addr.v.a.mask, &sn->raddr, sn->af)) { pf_unlink_src_node(sn); LIST_INSERT_HEAD(&kill, sn, entry); sn->expire = 1; } PF_HASHROW_UNLOCK(sh); } for (int i = 0; i <= pf_hashmask; i++) { struct pf_idhash *ih = &V_pf_idhash[i]; struct pf_state *s; PF_HASHROW_LOCK(ih); LIST_FOREACH(s, &ih->states, entry) { if (s->src_node && s->src_node->expire == 1) s->src_node = NULL; if (s->nat_src_node && s->nat_src_node->expire == 1) s->nat_src_node = NULL; } PF_HASHROW_UNLOCK(ih); } psnk->psnk_killed = pf_free_src_nodes(&kill); } /* * XXX - Check for version missmatch!!! */ /* * Duplicate pfctl -Fa operation to get rid of as much as we can. */ static int shutdown_pf(void) { int error = 0; u_int32_t t[5]; char nn = '\0'; do { if ((error = pf_begin_rules(&t[0], PF_RULESET_SCRUB, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: SCRUB\n")); break; } if ((error = pf_begin_rules(&t[1], PF_RULESET_FILTER, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: FILTER\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[2], PF_RULESET_NAT, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: NAT\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[3], PF_RULESET_BINAT, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: BINAT\n")); break; /* XXX: rollback? */ } if ((error = pf_begin_rules(&t[4], PF_RULESET_RDR, &nn)) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: RDR\n")); break; /* XXX: rollback? */ } /* XXX: these should always succeed here */ pf_commit_rules(t[0], PF_RULESET_SCRUB, &nn); pf_commit_rules(t[1], PF_RULESET_FILTER, &nn); pf_commit_rules(t[2], PF_RULESET_NAT, &nn); pf_commit_rules(t[3], PF_RULESET_BINAT, &nn); pf_commit_rules(t[4], PF_RULESET_RDR, &nn); if ((error = pf_clear_tables()) != 0) break; #ifdef ALTQ if ((error = pf_begin_altq(&t[0])) != 0) { DPFPRINTF(PF_DEBUG_MISC, ("shutdown_pf: ALTQ\n")); break; } pf_commit_altq(t[0]); #endif pf_clear_states(); pf_clear_srcnodes(NULL); /* status does not use malloced mem so no need to cleanup */ /* fingerprints and interfaces have their own cleanup code */ } while(0); return (error); } static pfil_return_t pf_check_return(int chk, struct mbuf **m) { switch (chk) { case PF_PASS: if (*m == NULL) return (PFIL_CONSUMED); else return (PFIL_PASS); break; default: if (*m != NULL) { m_freem(*m); *m = NULL; } return (PFIL_DROPPED); } } #ifdef INET static pfil_return_t pf_check_in(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp) { int chk; chk = pf_test(PF_IN, flags, ifp, m, inp); return (pf_check_return(chk, m)); } static pfil_return_t pf_check_out(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp) { int chk; chk = pf_test(PF_OUT, flags, ifp, m, inp); return (pf_check_return(chk, m)); } #endif #ifdef INET6 static pfil_return_t pf_check6_in(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp) { int chk; /* * In case of loopback traffic IPv6 uses the real interface in * order to support scoped addresses. In order to support stateful * filtering we have change this to lo0 as it is the case in IPv4. */ CURVNET_SET(ifp->if_vnet); chk = pf_test6(PF_IN, flags, (*m)->m_flags & M_LOOP ? V_loif : ifp, m, inp); CURVNET_RESTORE(); return (pf_check_return(chk, m)); } static pfil_return_t pf_check6_out(struct mbuf **m, struct ifnet *ifp, int flags, void *ruleset __unused, struct inpcb *inp) { int chk; CURVNET_SET(ifp->if_vnet); chk = pf_test6(PF_OUT, flags, ifp, m, inp); CURVNET_RESTORE(); return (pf_check_return(chk, m)); } #endif /* INET6 */ #ifdef INET VNET_DEFINE_STATIC(pfil_hook_t, pf_ip4_in_hook); VNET_DEFINE_STATIC(pfil_hook_t, pf_ip4_out_hook); #define V_pf_ip4_in_hook VNET(pf_ip4_in_hook) #define V_pf_ip4_out_hook VNET(pf_ip4_out_hook) #endif #ifdef INET6 VNET_DEFINE_STATIC(pfil_hook_t, pf_ip6_in_hook); VNET_DEFINE_STATIC(pfil_hook_t, pf_ip6_out_hook); #define V_pf_ip6_in_hook VNET(pf_ip6_in_hook) #define V_pf_ip6_out_hook VNET(pf_ip6_out_hook) #endif static int hook_pf(void) { struct pfil_hook_args pha; struct pfil_link_args pla; if (V_pf_pfil_hooked) return (0); pha.pa_version = PFIL_VERSION; pha.pa_modname = "pf"; pha.pa_ruleset = NULL; pla.pa_version = PFIL_VERSION; #ifdef INET pha.pa_type = PFIL_TYPE_IP4; pha.pa_func = pf_check_in; pha.pa_flags = PFIL_IN; pha.pa_rulname = "default-in"; V_pf_ip4_in_hook = pfil_add_hook(&pha); pla.pa_flags = PFIL_IN | PFIL_HEADPTR | PFIL_HOOKPTR; pla.pa_head = V_inet_pfil_head; pla.pa_hook = V_pf_ip4_in_hook; (void)pfil_link(&pla); pha.pa_func = pf_check_out; pha.pa_flags = PFIL_OUT; pha.pa_rulname = "default-out"; V_pf_ip4_out_hook = pfil_add_hook(&pha); pla.pa_flags = PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR; pla.pa_head = V_inet_pfil_head; pla.pa_hook = V_pf_ip4_out_hook; (void)pfil_link(&pla); #endif #ifdef INET6 pha.pa_type = PFIL_TYPE_IP6; pha.pa_func = pf_check6_in; pha.pa_flags = PFIL_IN; pha.pa_rulname = "default-in6"; V_pf_ip6_in_hook = pfil_add_hook(&pha); pla.pa_flags = PFIL_IN | PFIL_HEADPTR | PFIL_HOOKPTR; pla.pa_head = V_inet6_pfil_head; pla.pa_hook = V_pf_ip6_in_hook; (void)pfil_link(&pla); pha.pa_func = pf_check6_out; pha.pa_rulname = "default-out6"; pha.pa_flags = PFIL_OUT; V_pf_ip6_out_hook = pfil_add_hook(&pha); pla.pa_flags = PFIL_OUT | PFIL_HEADPTR | PFIL_HOOKPTR; pla.pa_head = V_inet6_pfil_head; pla.pa_hook = V_pf_ip6_out_hook; (void)pfil_link(&pla); #endif V_pf_pfil_hooked = 1; return (0); } static int dehook_pf(void) { if (V_pf_pfil_hooked == 0) return (0); #ifdef INET pfil_remove_hook(V_pf_ip4_in_hook); pfil_remove_hook(V_pf_ip4_out_hook); #endif #ifdef INET6 pfil_remove_hook(V_pf_ip6_in_hook); pfil_remove_hook(V_pf_ip6_out_hook); #endif V_pf_pfil_hooked = 0; return (0); } static void pf_load_vnet(void) { - TAILQ_INIT(&V_pf_tags); - TAILQ_INIT(&V_pf_qids); + V_pf_tag_z = uma_zcreate("pf tags", sizeof(struct pf_tagname), + NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); + pf_init_tagset(&V_pf_tags, &pf_rule_tag_hashsize, + PF_RULE_TAG_HASH_SIZE_DEFAULT); +#ifdef ALTQ + pf_init_tagset(&V_pf_qids, &pf_queue_tag_hashsize, + PF_QUEUE_TAG_HASH_SIZE_DEFAULT); +#endif + pfattach_vnet(); V_pf_vnet_active = 1; } static int pf_load(void) { int error; rm_init(&pf_rules_lock, "pf rulesets"); sx_init(&pf_ioctl_lock, "pf ioctl"); sx_init(&pf_end_lock, "pf end thread"); pf_mtag_initialize(); pf_dev = make_dev(&pf_cdevsw, 0, 0, 0, 0600, PF_NAME); if (pf_dev == NULL) return (ENOMEM); pf_end_threads = 0; error = kproc_create(pf_purge_thread, NULL, &pf_purge_proc, 0, 0, "pf purge"); if (error != 0) return (error); pfi_initialize(); return (0); } static void pf_unload_vnet(void) { int error; V_pf_vnet_active = 0; V_pf_status.running = 0; error = dehook_pf(); if (error) { /* * Should not happen! * XXX Due to error code ESRCH, kldunload will show * a message like 'No such process'. */ printf("%s : pfil unregisteration fail\n", __FUNCTION__); return; } PF_RULES_WLOCK(); shutdown_pf(); PF_RULES_WUNLOCK(); swi_remove(V_pf_swi_cookie); pf_unload_vnet_purge(); pf_normalize_cleanup(); PF_RULES_WLOCK(); pfi_cleanup_vnet(); PF_RULES_WUNLOCK(); pfr_cleanup(); pf_osfp_flush(); pf_cleanup(); if (IS_DEFAULT_VNET(curvnet)) pf_mtag_cleanup(); + + pf_cleanup_tagset(&V_pf_tags); +#ifdef ALTQ + pf_cleanup_tagset(&V_pf_qids); +#endif + uma_zdestroy(V_pf_tag_z); /* Free counters last as we updated them during shutdown. */ counter_u64_free(V_pf_default_rule.states_cur); counter_u64_free(V_pf_default_rule.states_tot); counter_u64_free(V_pf_default_rule.src_nodes); for (int i = 0; i < PFRES_MAX; i++) counter_u64_free(V_pf_status.counters[i]); for (int i = 0; i < LCNT_MAX; i++) counter_u64_free(V_pf_status.lcounters[i]); for (int i = 0; i < FCNT_MAX; i++) counter_u64_free(V_pf_status.fcounters[i]); for (int i = 0; i < SCNT_MAX; i++) counter_u64_free(V_pf_status.scounters[i]); } static void pf_unload(void) { sx_xlock(&pf_end_lock); pf_end_threads = 1; while (pf_end_threads < 2) { wakeup_one(pf_purge_thread); sx_sleep(pf_purge_proc, &pf_end_lock, 0, "pftmo", 0); } sx_xunlock(&pf_end_lock); if (pf_dev != NULL) destroy_dev(pf_dev); pfi_cleanup(); rm_destroy(&pf_rules_lock); sx_destroy(&pf_ioctl_lock); sx_destroy(&pf_end_lock); } static void vnet_pf_init(void *unused __unused) { pf_load_vnet(); } VNET_SYSINIT(vnet_pf_init, SI_SUB_PROTO_FIREWALL, SI_ORDER_THIRD, vnet_pf_init, NULL); static void vnet_pf_uninit(const void *unused __unused) { pf_unload_vnet(); } SYSUNINIT(pf_unload, SI_SUB_PROTO_FIREWALL, SI_ORDER_SECOND, pf_unload, NULL); VNET_SYSUNINIT(vnet_pf_uninit, SI_SUB_PROTO_FIREWALL, SI_ORDER_THIRD, vnet_pf_uninit, NULL); static int pf_modevent(module_t mod, int type, void *data) { int error = 0; switch(type) { case MOD_LOAD: error = pf_load(); break; case MOD_UNLOAD: /* Handled in SYSUNINIT(pf_unload) to ensure it's done after * the vnet_pf_uninit()s */ break; default: error = EINVAL; break; } return (error); } static moduledata_t pf_mod = { "pf", pf_modevent, 0 }; DECLARE_MODULE(pf, pf_mod, SI_SUB_PROTO_FIREWALL, SI_ORDER_SECOND); MODULE_VERSION(pf, PF_MODVER);