D50648.1778263195.diff
No OneTemporary
Actions

Size

20 KB

Referenced Files

None

Subscribers

None

D50648.1778263195.diff
View Options

	diff --git a/sys/fs/msdosfs/msdosfs_lookup.c b/sys/fs/msdosfs/msdosfs_lookup.c
	--- a/sys/fs/msdosfs/msdosfs_lookup.c
	+++ b/sys/fs/msdosfs/msdosfs_lookup.c
	@@ -845,7 +845,6 @@
	*wait_scn = 0;

	pmp = target->de_pmp;
	- lockmgr_assert(&pmp->pm_checkpath_lock, KA_XLOCKED);
	KASSERT(pmp == source->de_pmp,
	("doscheckpath: source and target on different filesystems"));

	diff --git a/sys/fs/msdosfs/msdosfs_vfsops.c b/sys/fs/msdosfs/msdosfs_vfsops.c
	--- a/sys/fs/msdosfs/msdosfs_vfsops.c
	+++ b/sys/fs/msdosfs/msdosfs_vfsops.c
	@@ -575,7 +575,6 @@
	pmp->pm_bo = bo;

	lockinit(&pmp->pm_fatlock, 0, msdosfs_lock_msg, 0, 0);
	- lockinit(&pmp->pm_checkpath_lock, 0, "msdoscp", 0, 0);

	TASK_INIT(&pmp->pm_rw2ro_task, 0, msdosfs_remount_ro, pmp);

	@@ -871,7 +870,6 @@
	}
	if (pmp != NULL) {
	lockdestroy(&pmp->pm_fatlock);
	- lockdestroy(&pmp->pm_checkpath_lock);
	free(pmp->pm_inusemap, M_MSDOSFSFAT);
	free(pmp, M_MSDOSFSMNT);
	mp->mnt_data = NULL;
	@@ -971,7 +969,6 @@
	dev_rel(pmp->pm_dev);
	free(pmp->pm_inusemap, M_MSDOSFSFAT);
	lockdestroy(&pmp->pm_fatlock);
	- lockdestroy(&pmp->pm_checkpath_lock);
	free(pmp, M_MSDOSFSMNT);
	mp->mnt_data = NULL;
	return (error);
	diff --git a/sys/fs/msdosfs/msdosfs_vnops.c b/sys/fs/msdosfs/msdosfs_vnops.c
	--- a/sys/fs/msdosfs/msdosfs_vnops.c
	+++ b/sys/fs/msdosfs/msdosfs_vnops.c
	@@ -945,7 +945,7 @@
	struct denode fdip, fip, tdip, tip, *nip;
	u_char toname[12], oldname[11];
	u_long to_diroffset;
	- bool checkpath_locked, doingdirectory, newparent;
	+ bool doingdirectory, newparent;
	int error;
	u_long cn, pcl, blkoff;
	daddr_t bn, wait_scn, scn;
	@@ -986,8 +986,6 @@
	if (tvp != NULL && tvp != tdvp)
	VOP_UNLOCK(tvp);

	- checkpath_locked = false;
	-
	relock:
	doingdirectory = newparent = false;

	@@ -1108,12 +1106,8 @@
	if (doingdirectory && newparent) {
	if (error != 0) /* write access check above */
	goto unlock;
	- lockmgr(&pmp->pm_checkpath_lock, LK_EXCLUSIVE, NULL);
	- checkpath_locked = true;
	error = doscheckpath(fip, tdip, &wait_scn);
	if (wait_scn != 0) {
	- lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
	- checkpath_locked = false;
	VOP_UNLOCK(fdvp);
	VOP_UNLOCK(tdvp);
	VOP_UNLOCK(fvp);
	@@ -1276,8 +1270,6 @@
	cache_purge(fvp);

	unlock:
	- if (checkpath_locked)
	- lockmgr(&pmp->pm_checkpath_lock, LK_RELEASE, NULL);
	vput(fdvp);
	vput(fvp);
	if (tvp != NULL) {
	@@ -1289,7 +1281,6 @@
	vput(tdvp);
	return (error);
	releout:
	- MPASS(!checkpath_locked);
	vrele(tdvp);
	if (tvp != NULL)
	vrele(tvp);
	diff --git a/sys/fs/msdosfs/msdosfsmount.h b/sys/fs/msdosfs/msdosfsmount.h
	--- a/sys/fs/msdosfs/msdosfsmount.h
	+++ b/sys/fs/msdosfs/msdosfsmount.h
	@@ -118,7 +118,6 @@
	void pm_u2d; / Unicode->DOS iconv handle */
	void pm_d2u; / DOS->Local iconv handle */
	struct lock pm_fatlock; /* lockmgr protecting allocations */
	- struct lock pm_checkpath_lock; /* protects doscheckpath result */
	struct task pm_rw2ro_task; /* context for emergency remount ro */
	};

	diff --git a/sys/kern/vfs_lookup.c b/sys/kern/vfs_lookup.c
	--- a/sys/kern/vfs_lookup.c
	+++ b/sys/kern/vfs_lookup.c
	@@ -74,15 +74,21 @@
	#define NDVALIDATE(ndp)
	#endif

	+/*
	+ * Reset ndp to its original state.
	+ */
	+#define NDRESET(ndp) do { \
	+ NDREINIT_DBG(ndp); \
	+ ndp->ni_resflags = 0; \
	+ ndp->ni_cnd.cn_flags &= ~NAMEI_INTERNAL_FLAGS; \
	+} while (0)
	/*
	* Prepare namei() to restart. Reset components to its original state and set
	* ISRESTARTED flag which signals the underlying lookup code to change the root
	* from ABI root to actual root and prevents a further restarts.
	*/
	#define NDRESTART(ndp) do { \
	- NDREINIT_DBG(ndp); \
	- ndp->ni_resflags = 0; \
	- ndp->ni_cnd.cn_flags &= ~NAMEI_INTERNAL_FLAGS; \
	+ NDRESET(ndp); \
	ndp->ni_cnd.cn_flags \|= ISRESTARTED; \
	} while (0)

	@@ -162,8 +168,8 @@
	*/

	struct nameicap_tracker {
	- struct vnode *dp;
	TAILQ_ENTRY(nameicap_tracker) nm_link;
	+ struct mount *mp;
	};

	/* Zone for cap mode tracker elements used for dotdot capability checks. */
	@@ -192,49 +198,75 @@
	"enables \"..\" components in path lookup in capability mode "
	"on non-local mount");

	-static void
	+static int
	nameicap_tracker_add(struct nameidata ndp, struct vnode dp)
	{
	struct nameicap_tracker *nt;
	+ struct mount *mp;
	+ int error;

	if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0 \|\| dp->v_type != VDIR)
	- return;
	+ return (0);
	+ mp = NULL;
	+ error = VOP_GETWRITEMOUNT(dp, &mp);
	+ if (error != 0)
	+ return (error);
	nt = TAILQ_LAST(&ndp->ni_cap_tracker, nameicap_tracker_head);
	- if (nt != NULL && nt->dp == dp)
	- return;
	+ if (nt != NULL && nt->mp == mp) {
	+ vfs_rel(mp);
	+ return (0);
	+ }
	nt = malloc(sizeof(*nt), M_NAMEITRACKER, M_WAITOK);
	- vhold(dp);
	- nt->dp = dp;
	- TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
	+ nt->mp = mp;
	+ error = lockmgr(&mp->mnt_renamelock, LK_SHARED \| LK_NOWAIT, 0);
	+ if (error != 0) {
	+ MPASS(ndp->ni_nctrack_mnt == NULL);
	+ ndp->ni_nctrack_mnt = mp;
	+ free(nt, M_NAMEITRACKER);
	+ error = ERESTART;
	+ } else {
	+ TAILQ_INSERT_TAIL(&ndp->ni_cap_tracker, nt, nm_link);
	+ }
	+ return (error);
	}

	static void
	-nameicap_cleanup_from(struct nameidata ndp, struct nameicap_tracker first)
	+nameicap_cleanup(struct nameidata *ndp, int error)
	{
	struct nameicap_tracker nt, nt1;
	+ struct mount *mp;

	- nt = first;
	- TAILQ_FOREACH_FROM_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
	+ KASSERT((ndp->ni_nctrack_mnt == NULL &&
	+ TAILQ_EMPTY(&ndp->ni_cap_tracker)) \|\|
	+ (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0,
	+ ("tracker active and not strictrelative"));
	+
	+ TAILQ_FOREACH_SAFE(nt, &ndp->ni_cap_tracker, nm_link, nt1) {
	+ mp = nt->mp;
	+ lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0);
	+ vfs_rel(mp);
	TAILQ_REMOVE(&ndp->ni_cap_tracker, nt, nm_link);
	- vdrop(nt->dp);
	free(nt, M_NAMEITRACKER);
	}
	+
	+ mp = ndp->ni_nctrack_mnt;
	+ if (mp != NULL) {
	+ if (error == ERESTART) {
	+ lockmgr(&mp->mnt_renamelock, LK_EXCLUSIVE, 0);
	+ lockmgr(&mp->mnt_renamelock, LK_RELEASE, 0);
	+ }
	+ vfs_rel(mp);
	+ ndp->ni_nctrack_mnt = NULL;
	+ }
	}

	-static void
	-nameicap_cleanup(struct nameidata *ndp)
	-{
	- KASSERT(TAILQ_EMPTY(&ndp->ni_cap_tracker) \|\|
	- (ndp->ni_lcf & NI_LCF_CAP_DOTDOT) != 0, ("not strictrelative"));
	- nameicap_cleanup_from(ndp, NULL);
	-}
	-
	/*
	- * For dotdot lookups in capability mode, only allow the component
	- * lookup to succeed if the resulting directory was already traversed
	- * during the operation. This catches situations where already
	- * traversed directory is moved to different parent, and then we walk
	- * over it with dotdots.
	+ * For dotdot lookups in capability mode, disallow walking over the
	+ * directory no_rbeneath_dpp that was used as the starting point of
	+ * the lookup. Since we take the mnt_renamelocks of all mounts we
	+ * ever walked over during lookup, parallel renames are disabled.
	+ * This prevents the situation where we circumvent walk over
	+ * ni_rbeneath_dpp following dotdots.
	*
	* Also allow to force failure of dotdot lookups for non-local
	* filesystems, where external agents might assist local lookups to
	@@ -243,7 +275,6 @@
	static int
	nameicap_check_dotdot(struct nameidata ndp, struct vnode dp)
	{
	- struct nameicap_tracker *nt;
	struct mount *mp;

	if (dp == NULL \|\| dp->v_type != VDIR \|\| (ndp->ni_lcf &
	@@ -253,22 +284,16 @@
	NI_LCF_CAP_DOTDOT_KTR)) == NI_LCF_STRICTREL_KTR))
	NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
	if ((ndp->ni_lcf & NI_LCF_CAP_DOTDOT) == 0)
	- return (ENOTCAPABLE);
	+ goto violation;
	+ if (dp == ndp->ni_rbeneath_dpp)
	+ goto violation;
	mp = dp->v_mount;
	if (lookup_cap_dotdot_nonlocal == 0 && mp != NULL &&
	(mp->mnt_flag & MNT_LOCAL) == 0)
	- goto capfail;
	- TAILQ_FOREACH_REVERSE(nt, &ndp->ni_cap_tracker, nameicap_tracker_head,
	- nm_link) {
	- if (dp == nt->dp) {
	- nt = TAILQ_NEXT(nt, nm_link);
	- if (nt != NULL)
	- nameicap_cleanup_from(ndp, nt);
	- return (0);
	- }
	- }
	+ goto violation;
	+ return (0);

	-capfail:
	+violation:
	if (__predict_false((ndp->ni_lcf & NI_LCF_STRICTREL_KTR) != 0))
	NI_CAP_VIOLATION(ndp, ndp->ni_cnd.cn_pnbuf);
	return (ENOTCAPABLE);
	@@ -394,6 +419,8 @@
	NI_LCF_CAP_DOTDOT;
	}
	}
	+ if (error == 0 && (ndp->ni_lcf & NI_LCF_STRICTREL) != 0)
	+ ndp->ni_rbeneath_dpp = *dpp;

	/*
	* If we are auditing the kernel pathname, save the user pathname.
	@@ -631,6 +658,7 @@
	error = namei_getpath(ndp);
	if (__predict_false(error != 0)) {
	namei_cleanup_cnp(cnp);
	+ nameicap_cleanup(ndp, error);
	SDT_PROBE4(vfs, namei, lookup, return, error, NULL,
	false, ndp);
	return (error);
	@@ -661,12 +689,12 @@
	else if (__predict_false(pwd->pwd_adir != pwd->pwd_rdir &&
	(cnp->cn_flags & ISRESTARTED) == 0)) {
	namei_cleanup_cnp(cnp);
	+ nameicap_cleanup(ndp, ERESTART);
	NDRESTART(ndp);
	goto restart;
	}
	return (error);
	case CACHE_FPL_STATUS_PARTIAL:
	- TAILQ_INIT(&ndp->ni_cap_tracker);
	dp = ndp->ni_startdir;
	break;
	case CACHE_FPL_STATUS_DESTROYED:
	@@ -674,18 +702,21 @@
	error = namei_getpath(ndp);
	if (__predict_false(error != 0)) {
	namei_cleanup_cnp(cnp);
	+ nameicap_cleanup(ndp, error);
	return (error);
	}
	cnp->cn_nameptr = cnp->cn_pnbuf;
	/* FALLTHROUGH */
	case CACHE_FPL_STATUS_ABORTED:
	- TAILQ_INIT(&ndp->ni_cap_tracker);
	MPASS(ndp->ni_lcf == 0);
	if (*cnp->cn_pnbuf == '\0') {
	if ((cnp->cn_flags & EMPTYPATH) != 0) {
	- return (namei_emptypath(ndp));
	+ error = namei_emptypath(ndp);
	+ nameicap_cleanup(ndp, error);
	+ return (error);
	}
	namei_cleanup_cnp(cnp);
	+ nameicap_cleanup(ndp, ENOENT);
	SDT_PROBE4(vfs, namei, lookup, return, ENOENT, NULL,
	false, ndp);
	return (ENOENT);
	@@ -693,6 +724,7 @@
	error = namei_setup(ndp, &dp, &pwd);
	if (error != 0) {
	namei_cleanup_cnp(cnp);
	+ nameicap_cleanup(ndp, error);
	return (error);
	}
	break;
	@@ -705,16 +737,23 @@
	ndp->ni_startdir = dp;
	error = vfs_lookup(ndp);
	if (error != 0) {
	- if (__predict_false(pwd->pwd_adir != pwd->pwd_rdir &&
	- error == ENOENT &&
	- (cnp->cn_flags & ISRESTARTED) == 0)) {
	- nameicap_cleanup(ndp);
	- pwd_drop(pwd);
	- namei_cleanup_cnp(cnp);
	- NDRESTART(ndp);
	- goto restart;
	- } else
	+ uint64_t was_restarted;
	+ bool abi_restart;
	+
	+ was_restarted = ndp->ni_cnd.cn_flags &
	+ ISRESTARTED;
	+ abi_restart = pwd->pwd_adir != pwd->pwd_rdir &&
	+ error == ENOENT && was_restarted == 0;
	+ if (error != ERESTART && !abi_restart)
	goto out;
	+ nameicap_cleanup(ndp, error);
	+ pwd_drop(pwd);
	+ namei_cleanup_cnp(cnp);
	+ NDRESET(ndp);
	+ if (abi_restart)
	+ was_restarted = ISRESTARTED;
	+ ndp->ni_cnd.cn_flags \|= was_restarted;
	+ goto restart;
	}

	/*
	@@ -723,7 +762,7 @@
	if ((cnp->cn_flags & ISSYMLINK) == 0) {
	SDT_PROBE4(vfs, namei, lookup, return, error,
	ndp->ni_vp, false, ndp);
	- nameicap_cleanup(ndp);
	+ nameicap_cleanup(ndp, 0);
	pwd_drop(pwd);
	NDVALIDATE(ndp);
	return (0);
	@@ -756,10 +795,10 @@
	ndp->ni_vp = NULL;
	vrele(ndp->ni_dvp);
	out:
	- MPASS(error != 0);
	+ MPASS(error != ERESTART);
	SDT_PROBE4(vfs, namei, lookup, return, error, NULL, false, ndp);
	namei_cleanup_cnp(cnp);
	- nameicap_cleanup(ndp);
	+ nameicap_cleanup(ndp, error);
	pwd_drop(pwd);
	return (error);
	}
	@@ -1185,7 +1224,9 @@
	}
	}

	- nameicap_tracker_add(ndp, dp);
	+ error = nameicap_tracker_add(ndp, dp);
	+ if (error != 0)
	+ goto bad;

	/*
	* Make sure degenerate names don't get here, their handling was
	@@ -1210,9 +1251,7 @@
	* the jail or chroot, don't let them out.
	* 5. If doing a capability lookup and lookup_cap_dotdot is
	* enabled, return ENOTCAPABLE if the lookup would escape
	- * from the initial file descriptor directory. Checks are
	- * done by ensuring that namei() already traversed the
	- * result of dotdot lookup.
	+ * from the initial file descriptor directory.
	*/
	if (cnp->cn_flags & ISDOTDOT) {
	if (__predict_false((ndp->ni_lcf & (NI_LCF_STRICTREL_KTR \|
	@@ -1238,7 +1277,7 @@
	NI_CAP_VIOLATION(ndp, cnp->cn_pnbuf);
	if ((ndp->ni_lcf & NI_LCF_STRICTREL) != 0) {
	error = ENOTCAPABLE;
	- goto capdotdot;
	+ goto bad;
	}
	}
	if (isroot \|\| ((dp->v_vflag & VV_ROOT) != 0 &&
	@@ -1261,11 +1300,6 @@
	vn_lock(dp,
	enforce_lkflags(dp->v_mount, cnp->cn_lkflags \|
	LK_RETRY));
	- error = nameicap_check_dotdot(ndp, dp);
	- if (error != 0) {
	-capdotdot:
	- goto bad;
	- }
	}
	}

	@@ -1314,7 +1348,9 @@
	vn_lock(dp,
	enforce_lkflags(dp->v_mount, cnp->cn_lkflags \|
	LK_RETRY));
	- nameicap_tracker_add(ndp, dp);
	+ error = nameicap_tracker_add(ndp, dp);
	+ if (error != 0)
	+ goto bad;
	goto unionlookup;
	}

	@@ -1415,7 +1451,7 @@
	goto dirloop;
	}
	if (cnp->cn_flags & ISDOTDOT) {
	- error = nameicap_check_dotdot(ndp, ndp->ni_vp);
	+ error = nameicap_check_dotdot(ndp, ndp->ni_dvp);
	if (error != 0)
	goto bad2;
	}
	@@ -1485,8 +1521,11 @@
	}
	success_right_lock:
	if (ndp->ni_vp != NULL) {
	- if ((cnp->cn_flags & ISDOTDOT) == 0)
	- nameicap_tracker_add(ndp, ndp->ni_vp);
	+ if ((cnp->cn_flags & ISDOTDOT) == 0) {
	+ error = nameicap_tracker_add(ndp, ndp->ni_vp);
	+ if (error != 0)
	+ goto bad2;
	+ }
	if ((cnp->cn_flags & (FAILIFEXISTS \| ISSYMLINK)) == FAILIFEXISTS)
	return (vfs_lookup_failifexists(ndp));
	}
	diff --git a/sys/kern/vfs_mount.c b/sys/kern/vfs_mount.c
	--- a/sys/kern/vfs_mount.c
	+++ b/sys/kern/vfs_mount.c
	@@ -156,6 +156,7 @@
	mtx_init(&mp->mnt_mtx, "struct mount mtx", NULL, MTX_DEF);
	mtx_init(&mp->mnt_listmtx, "struct mount vlist mtx", NULL, MTX_DEF);
	lockinit(&mp->mnt_explock, PVFS, "explock", 0, 0);
	+ lockinit(&mp->mnt_renamelock, PVFS, "rename", 0, 0);
	mp->mnt_pcpu = uma_zalloc_pcpu(pcpu_zone_16, M_WAITOK \| M_ZERO);
	mp->mnt_ref = 0;
	mp->mnt_vfs_ops = 1;
	@@ -170,6 +171,7 @@

	mp = (struct mount *)mem;
	uma_zfree_pcpu(pcpu_zone_16, mp->mnt_pcpu);
	+ lockdestroy(&mp->mnt_renamelock);
	lockdestroy(&mp->mnt_explock);
	mtx_destroy(&mp->mnt_listmtx);
	mtx_destroy(&mp->mnt_mtx);
	diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
	--- a/sys/kern/vfs_subr.c
	+++ b/sys/kern/vfs_subr.c
	@@ -5851,6 +5851,8 @@
	struct vop_rename_args *a = ap;

	#ifdef DEBUG_VFS_LOCKS
	+ struct mount *tmp;
	+
	if (a->a_tvp)
	ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
	ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
	@@ -5868,6 +5870,11 @@
	if (a->a_tvp)
	ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
	ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
	+
	+ tmp = NULL;
	+ VOP_GETWRITEMOUNT(a->a_tdvp, &tmp);
	+ lockmgr_assert(&tmp->mnt_renamelock, KA_XLOCKED);
	+ vfs_rel(tmp);
	#endif
	/*
	* It may be tempting to add vn_seqc_write_begin/end calls here and
	diff --git a/sys/kern/vfs_syscalls.c b/sys/kern/vfs_syscalls.c
	--- a/sys/kern/vfs_syscalls.c
	+++ b/sys/kern/vfs_syscalls.c
	@@ -3766,7 +3766,7 @@
	kern_renameat(struct thread td, int oldfd, const char old, int newfd,
	const char *new, enum uio_seg pathseg)
	{
	- struct mount *mp = NULL;
	+ struct mount mp, tmp;
	struct vnode tvp, fvp, *tdvp;
	struct nameidata fromnd, tond;
	uint64_t tondflags;
	@@ -3774,6 +3774,7 @@
	short irflag;

	again:
	+ tmp = mp = NULL;
	bwillwrite();
	#ifdef MAC
	if (mac_vnode_check_rename_from_enabled()) {
	@@ -3809,6 +3810,7 @@
	tvp = tond.ni_vp;
	error = vn_start_write(fvp, &mp, V_NOWAIT);
	if (error != 0) {
	+again1:
	NDFREE_PNBUF(&fromnd);
	NDFREE_PNBUF(&tond);
	if (tvp != NULL)
	@@ -3819,11 +3821,25 @@
	vput(tdvp);
	vrele(fromnd.ni_dvp);
	vrele(fvp);
	+ if (tmp != NULL) {
	+ lockmgr(&tmp->mnt_renamelock, LK_EXCLUSIVE, NULL);
	+ lockmgr(&tmp->mnt_renamelock, LK_RELEASE, NULL);
	+ vfs_rel(tmp);
	+ tmp = NULL;
	+ }
	error = vn_start_write(NULL, &mp, V_XSLEEP \| V_PCATCH);
	if (error != 0)
	return (error);
	goto again;
	}
	+ error = VOP_GETWRITEMOUNT(tdvp, &tmp);
	+ if (error != 0 \|\| tmp == NULL)
	+ goto again1;
	+ error = lockmgr(&tmp->mnt_renamelock, LK_EXCLUSIVE \| LK_NOWAIT, NULL);
	+ if (error != 0) {
	+ vn_finished_write(mp);
	+ goto again1;
	+ }
	irflag = vn_irflag_read(fvp);
	if (((irflag & VIRF_NAMEDATTR) != 0 && tdvp != fromnd.ni_dvp) \|\|
	(irflag & VIRF_NAMEDDIR) != 0) {
	@@ -3884,6 +3900,8 @@
	vrele(fromnd.ni_dvp);
	vrele(fvp);
	}
	+ lockmgr(&tmp->mnt_renamelock, LK_RELEASE, 0);
	+ vfs_rel(tmp);
	vn_finished_write(mp);
	out1:
	if (error == ERESTART)
	diff --git a/sys/sys/mount.h b/sys/sys/mount.h
	--- a/sys/sys/mount.h
	+++ b/sys/sys/mount.h
	@@ -267,6 +267,7 @@
	int mnt_lazyvnodelistsize; /* (l) # of lazy vnodes */
	int mnt_upper_pending; /* (i) # of pending ops on mnt_uppers */
	struct lock mnt_explock; /* vfs_export walkers lock */
	+ struct lock mnt_renamelock; /* renames and O_RESOLVE_BENEATH */
	TAILQ_HEAD(, mount_upper_node) mnt_uppers; /* (i) upper mounts over us */
	TAILQ_HEAD(, mount_upper_node) mnt_notify; /* (i) upper mounts for notification */
	STAILQ_ENTRY(mount) mnt_taskqueue_link; /* (d) our place in deferred unmount list */
	diff --git a/sys/sys/namei.h b/sys/sys/namei.h
	--- a/sys/sys/namei.h
	+++ b/sys/sys/namei.h
	@@ -108,7 +108,12 @@
	* through the VOP interface.
	*/
	struct componentname ni_cnd;
	+
	+ /* Serving RBENEATH. */
	struct nameicap_tracker_head ni_cap_tracker;
	+ struct vnode *ni_rbeneath_dpp;
	+ struct mount *ni_nctrack_mnt;
	+
	/*
	* Private helper data for UFS, must be at the end. See
	* NDINIT_PREFILL().
	@@ -235,6 +240,10 @@
	panic("namei data not inited"); \
	if (((arg)->ni_debugflags & NAMEI_DBG_HADSTARTDIR) != 0) \
	panic("NDREINIT on namei data with NAMEI_DBG_HADSTARTDIR"); \
	+ if ((arg)->ni_nctrack_mnt != NULL) \
	+ panic("NDREINIT on namei data with leaked ni_nctrack_mnt"); \
	+ if (!TAILQ_EMPTY(&(arg)->ni_cap_tracker)) \
	+ panic("NDREINIT on namei data with leaked ni_cap_tracker"); \
	(arg)->ni_debugflags = NAMEI_DBG_INITED; \
	}
	#else
	@@ -259,6 +268,9 @@
	_ndp->ni_resflags = 0; \
	filecaps_init(&_ndp->ni_filecaps); \
	_ndp->ni_rightsneeded = _rightsp; \
	+ _ndp->ni_rbeneath_dpp = NULL; \
	+ _ndp->ni_nctrack_mnt = NULL; \
	+ TAILQ_INIT(&_ndp->ni_cap_tracker); \
	} while (0)

	#define NDREINIT(ndp) do { \
	diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c
	--- a/sys/ufs/ffs/ffs_vfsops.c
	+++ b/sys/ufs/ffs/ffs_vfsops.c
	@@ -1012,7 +1012,6 @@
	else
	ump->um_check_blkno = NULL;
	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
	- sx_init(&ump->um_checkpath_lock, "uchpth");
	fs->fs_ronly = ronly;
	fs->fs_active = NULL;
	mp->mnt_data = ump;
	@@ -1182,7 +1181,6 @@
	}
	if (ump != NULL) {
	mtx_destroy(UFS_MTX(ump));
	- sx_destroy(&ump->um_checkpath_lock);
	if (mp->mnt_gjprovider != NULL) {
	free(mp->mnt_gjprovider, M_UFSMNT);
	mp->mnt_gjprovider = NULL;
	@@ -1306,7 +1304,6 @@
	vrele(ump->um_odevvp);
	dev_rel(ump->um_dev);
	mtx_destroy(UFS_MTX(ump));
	- sx_destroy(&ump->um_checkpath_lock);
	if (mp->mnt_gjprovider != NULL) {
	free(mp->mnt_gjprovider, M_UFSMNT);
	mp->mnt_gjprovider = NULL;
	diff --git a/sys/ufs/ufs/ufs_lookup.c b/sys/ufs/ufs/ufs_lookup.c
	--- a/sys/ufs/ufs/ufs_lookup.c
	+++ b/sys/ufs/ufs/ufs_lookup.c
	@@ -1412,7 +1412,6 @@
	vp = tvp = ITOV(target);
	mp = vp->v_mount;
	*wait_ino = 0;
	- sx_assert(&VFSTOUFS(mp)->um_checkpath_lock, SA_XLOCKED);

	if (target->i_number == source_ino)
	return (EEXIST);
	diff --git a/sys/ufs/ufs/ufs_vnops.c b/sys/ufs/ufs/ufs_vnops.c
	--- a/sys/ufs/ufs/ufs_vnops.c
	+++ b/sys/ufs/ufs/ufs_vnops.c
	@@ -1273,9 +1273,9 @@
	struct mount *mp;
	ino_t ino;
	seqc_t fdvp_s, fvp_s, tdvp_s, tvp_s;
	- bool checkpath_locked, want_seqc_end;
	+ bool want_seqc_end;

	- checkpath_locked = want_seqc_end = false;
	+ want_seqc_end = false;

	endoff = 0;
	mp = tdvp->v_mount;
	@@ -1427,10 +1427,6 @@
	}
	vfs_ref(mp);
	MPASS(!want_seqc_end);
	- if (checkpath_locked) {
	- sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
	- checkpath_locked = false;
	- }
	VOP_UNLOCK(fdvp);
	VOP_UNLOCK(fvp);
	vref(tdvp);
	@@ -1484,8 +1480,6 @@
	if (error)
	goto unlockout;

	- sx_xlock(&VFSTOUFS(mp)->um_checkpath_lock);
	- checkpath_locked = true;
	error = ufs_checkpath(ino, fdp->i_number, tdp, tcnp->cn_cred,
	&ino);
	/*
	@@ -1493,8 +1487,6 @@
	* everything else and VGET before restarting.
	*/
	if (ino) {
	- sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
	- checkpath_locked = false;
	VOP_UNLOCK(fdvp);
	VOP_UNLOCK(fvp);
	VOP_UNLOCK(tdvp);
	@@ -1574,9 +1566,6 @@
	vn_seqc_write_end(fdvp);
	want_seqc_end = false;
	vfs_ref(mp);
	- MPASS(checkpath_locked);
	- sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
	- checkpath_locked = false;
	VOP_UNLOCK(fdvp);
	VOP_UNLOCK(fvp);
	vref(tdvp);
	@@ -1763,9 +1752,6 @@
	vn_seqc_write_end(fdvp);
	}

	- if (checkpath_locked)
	- sx_xunlock(&VFSTOUFS(mp)->um_checkpath_lock);
	-
	vput(fdvp);
	vput(fvp);

	diff --git a/sys/ufs/ufs/ufsmount.h b/sys/ufs/ufs/ufsmount.h
	--- a/sys/ufs/ufs/ufsmount.h
	+++ b/sys/ufs/ufs/ufsmount.h
	@@ -97,8 +97,6 @@
	uint64_t um_maxsymlinklen; /* (c) max size of short
	symlink */
	struct mtx um_lock; /* (c) Protects ufsmount & fs */
	- struct sx um_checkpath_lock; /* (c) Protects ufs_checkpath()
	- result */
	struct mount_softdeps um_softdep; / (c) softdep mgmt structure */
	struct vnode um_quotas[MAXQUOTAS]; / (q) pointer to quota files */
	struct ucred um_cred[MAXQUOTAS]; / (q) quota file access cred */

File Metadata

Mime Type: text/plain
Expires: Fri, May 8, 5:59 PM (12 h, 21 m)
Storage Engine: blob
Storage Format: Raw Data
Storage Handle: 28611485
Default Alt Text: D50648.1778263195.diff (20 KB)

D50648.1778263195.diffNo OneTemporaryActions

D50648.1778263195.diffView Options

File Metadata

Event Timeline

D50648.1778263195.diff
No OneTemporary
Actions

D50648.1778263195.diff
View Options