diff --git a/en/releases/5.3R/todo.sgml b/en/releases/5.3R/todo.sgml index 61a485dcaa..1e718ce480 100644 --- a/en/releases/5.3R/todo.sgml +++ b/en/releases/5.3R/todo.sgml @@ -1,387 +1,376 @@ - + %includes; %developers; N/A"> Done"> In progress"> Needs testing"> Not done"> Unknown"> ]> &header;

Open Issues

This is a list of open issues that need to be resolved for FreeBSD 5.3. If you have any updates for this list, please e-mail re@FreeBSD.org.

Show stopper defects for 5.3-RELEASE

- - - - - - -
IssueStatusResponsibleDescription
/dev/random stalls single user and sysinstall&status.wip;&a.markm;The kernel random number generator's requirement of needing to be - fed entropy early on has been disabled so that sysinstall and single-user - mode apps don't block with strange messages. Work is still needed in - rc.d to enable the higher-quality entropy before critical startup scripts - start and the system reaches multiuser mode.
General instability and lockups under high load &status.new; &a.scottl;, &a.julian; Problems persist with crashes and hangs under heavy load, especially - under SMP. The recent introduction of full-scale preemption exasperated + under SMP. The recent introduction of full-scale preemption exacerbated the problem, though preemption has been turned off temporarily while this problem is debugged. Speculation on the source of the problem seems to center around problems in the scheduler and appear to be common to both the 4BSD and ULE schedulers. This needs to be driven to root cause and fixed in order for 5.3 to be considered STABLE.
Source upgrade incompatibility &status.new; &a.das; Symbols that have moved between libc and libm have made source upgrades problematic. The 5.3 world sources must be buildable and installable from a 5.2.1 system.

Required features for 5.3-RELEASE

IssueStatusResponsibleDescription
GDB 6.1 kernel debugging support &status.wip &a.marcel; GDB 6.1.1 import does not include FreeBSD kernel debugging support.
BIND9 import into 5-CURRENT &status.wip; &a.dougb; BIND9 must be imported for 5.3-RELEASE.
KSE support for sparc64 -- -- Kernel bits implemented, userland not implemented. Required for 5.3-RELEASE.
Fine-grained network stack locking without Giant &status.wip; &a.rwatson; Significant parts of the network stack (especially IPv4, UNIX domain IPC, and sockets) now have fine-grained locking of their data structures. It's possible to run many common network subsystems and services without the Giant lock. However, a number of device drivers and less mainstream network subsystems are currently not MPSAFE. By 5.3-RELEASE, it is necessary to have the vast majority of network code running without Giant, including sockets, permitting complete local<->remote delivery without grabbing Giant.
kld support for amd64 &status.wip; &a.obrien; &a.peter; KLDs work when loaded from userland, but not from the loader. kldxref might also need work. Required for amd64 to be promoted to tier-1 status.
GDB thread support &status.wip; &a.davidxu;,&a.marcel; With improved support for threading primitives, support is now required to ease debugging of threaded applications. Ideally, this support will work for both libthr and libkse threading models.
Scheduler cleanup and resolution &status.wip; -- Currently, two schedulers are present: SCHED_ULE (default), an SMP-optimized scheduler created as part of SMPng, and SCHED_4BSD, an SMP-adapted version of the original 4BSD scheduler. They have quite different performance properties, with ULE providing strong interactivity characteristics, and performing quite well in a number of benchmarks, and 4BSD showing greater strength in IPC intensive user space benchmarks, such as databases. One of these schedulers must be the default for 5.3, and whichever one it is, it requires careful measurement, analysis, and optimization before the release in order to address its deficiencies.
Reports of UFS2 "large disk" problems &status.wip &a.scottl; There have been several reports that growfs(8) works improperly with large disk sizes, and other size-related nits in the current disk and label management tool set. These must be resolved for 5.3-RELEASE.

Desired features for 5.3-RELEASE

IssueStatusResponsibleDescription
truss support for ptrace -- -- Almost all process debugging tools have been updated to use non-procfs kernel primitives, with the exception of truss(1). As procfs is considered deprecated due to its inherent security risks, it is highly desirable to update truss to operate in a post-procfs world. &a.des; had prototype patches; &a.robert; is developing and testing patches now. Support for system call tracing has been added to ptrace().
Per-platform Thread-Local Storage &status.wip; &a.dfr;, &a.marcel; To complete support for thread-local storage on FreeBSD, per-architecture changes must be made. Currently pending platforms are amd64, alpha, ia64, i386, sparc64, and powerpc.
FAST_IPSEC and KAME compatibility &status.new; -- FAST_IPSEC currently cannot be used directly with the KAME IPv6 implementation, requiring an additional level of IP tunnel indirection to protect IPv6 packets when using hardware crypto acceleration. This issue must be resolved so that the two services may more easily be used together. Among other things, this will require a careful review of the handling of mbuf header copying and m_tag support in the KAME IPv6 code.
rpc.lockd(8) stability -- -- A process cannot be interrupted while waiting on a lock. Fixing this requires that the RPC code be taught how to deal with lock cancellation and interruption events.
Revised kld build infrastructure &status.new; &a.peter; Kernel modules are currently built independently from a kernel configuration, and independently from one another, resulting in substantially redundant compilation of objects, as well as the inability to easily manage compile-time options for kernel objects (such as MAC, PAE, etc) that may require conditional compilation in the kernel modules. In order to improve build performance and better support options of this sort, the KLD build infrastructure needs to be revamped. &a.peter; has done some initial prototyping, and should be contacted before starting on this work.
Merge of Darwin msdosfs, other fixes &status.new; -- Apple's Darwin operating system has fairly extensive improvements to msdosfs and other kernel services; these fixes must be reviewed and merged to the FreeBSD tree.
Race conditions in truss Errata candidate &a.robert; Truss appears to contain a race condition during the start-up of debugging, which can result in truss failing to attach to the process before it exits. The symptom is that truss reports that it cannot open the procfs node supporting the process being debugged. A bug also appears to exist where in truss will hang if execve() returns ENOENT. A further race appears to exist in which truss will return "PIOCWAIT: Input/output error" occasionally on startup. The fix for this sufficiently changes process execution handling that we will defer the fix to post-5.0 and consider this errata.
ACL_MASK override of umask support in UFS &status.new; &a.rwatson; Many systems supporting POSIX.1e ACLs permit a minor violation to that specification, in which the ACL_MASK entry overrides the umask, rather than being intersected with it. The resulting semantics can be useful in group-oriented environments, and as such would be very helpful on FreeBSD.
filedesc LOR &status.new; -- The LOR reported in PR kern/55175 needs to be fixed. Filedesc locking needs to be heavily reviewed in general.
MAC support for NFS Server &status.new; &a.rwatson; Currently, MAC protections are enforced only on locally originated file system operations (VOPs), and not on RPCs generated via the NFS server. Improvements in NFS server credential handling are required to correct this problem, as well as the introduction of new entry points to properly label NFS credentials and perform enforcement properly.
busdma in all PCI drivers &status.wip; -- All PCI drivers must use busdma for DMA; no use of vtophys() will be permitted for any recent device driver. ISA drivers may be exempt.
KSE support for alpha &status.wip; &a.marcel; Userland bits implemented, kernel bits not implemented.
CAM locking &status.wip; &a.scottl;, &a.gibbs; For kernel API/ABI compatibility reasons, it would be desirable to have the CAM locking strategy determined and loosely implemented for 5.3.

Documentation items that must be resolved for 5.3

IssueStatusResponsibleDescription
i386 Floppy Installation Docs &status.new; -- The installation documentation doesn't take into account the new floppy images (with a full kernel split across multiple disks). This should be updated.
Finish hardware notes trimming &status.wip; &a.simon; Finish removing mention of individual devices in the hardware notes and use auto-generated lists, based on driver manual pages, instead.

Testing focuses for 5.3-RELEASE

IssueStatusResponsibleDescription
SCHED_ULE as the default scheduler &status.untested; &a.jeff; SCHED_ULE provides better interactivity, higher performance, and the ability to support pinning and affinity. Basic HTT scheduling policies should be in place for 5.3 also.
make -DUSE_KQUEUE causes lockup with buildworld -jBIGNUM &status.untested; &a.green;, &a.jmg; Attempts to use make(1) with KQueues appears to result in a kernel hang under "heavy load". It would be desirable to fix this both from the perspective of building FreeBSD quickly as a developer, but also because it's an instability that could show up under other high load and heavy use of KQueues. See PR kern/57945 for a proposed patch and details. This appear to be the product of a locking problem, and must be fixed for 5.3.
KSE as the default threads library &status.untested; &a.davidxu;, &a.deischen; KSE has matured to the point of being more stable and POSIX-compliant than the traditional libc_r. All Tier-1 platforms MUST have stable KSE support for 5.3 in order to support a consistent transition. Additionally, all ports that depend on the pthreads API must be modified to properly detect and support the default threading library.
Updated binutils for all platforms &status.untested; &a.obrien; Binutils needs updating in order to support new platforms, newer GDB versions, and Thread Local Storage.
gcc 3.3 floating point alignment regression &status.untested;   The previous GCC 3.3 snapshot included regressions in alignment of floating point arguments, resulting in a substantial performance degradation. The recent GCC 3.4.2 import should fix this, but more testing is needed.
&footer;