FreeBSD The Power to Serve

FreeBSD 6.0 Open Issues

This is a list of open issues that need to be resolved for FreeBSD 6.0. If you have any updates for this list, please e-mail re@FreeBSD.org.

Show stopper defects for 6.0-RELEASE

Issue Status Responsible Description

Required features for 6.0-RELEASE

Issue Status Responsible Description

Desired features for 6.0-RELEASE

Issue Status Responsible Description
devfs umount panic Not done   There is a race condition between device removal and devfs umounts that causes "Memory modified after free" panics. Can be reproduced by doing 'mdconfig -u' concurrently with unmounting a devfs instance.
/dev/kmem panic Not done   Kris has noticed panics on SMP machines when there was ABI breakage of libkvm and world was not rebuilt and utilities like fstat were used. This suggests panics can be caused by incorrect accesses to /dev/kmem.
KLDs on sparc64 Not done   On sparc64 machines with more than 4Gb memory KLDs are not usable and will panic the system. The problem is reportedly with how the KLDs are compiled, it only works if the code ends up below 4G.
Max RAM on sparc64 Not done   Maximum RAM on sparc64 appears to be limited to 16Gb.
make -jN Not done   Doing 'make -jN', then suspending/resuming it may result in make reporting it lost child process(es).
OpenBSM Deferred for future release Robert Watson The integration of OpenBSM is waiting on some final licensing hurdles. Once those are cleared, it will be a very desirable feature for 6.0.
update sysinstall disk labeling In progress Craig Rodrigues Sysinstall could use the same fixes recently made to fdisk so it plays nice with GEOM and disk labeling. This does not cause problems during install because nothing on the disk is mounted when its label is being manipulated but it can cause problems if sysinstall gets used on a live system to adjust labels on existing disks which sys-admins tend to do.

Documentation items that must be resolved for 6.0

Issue Status Responsible Description
dhclient Not done freebsd-doc The ISC DHCP client v3.x has been removed in favor of the OpenBSD DHCP client which was based on ISC DHCP v2.X. Minimal updates to the FreeBSD Handbook may be required.
WPA / hostap In progress loader@freebsdmall.com, Sam Leffler, Murray Stokely Significant new wireless functionality will be available with FreeBSD 6.0. The wireless section of the FreeBSD Handbook should be updated to describe how to connect to networks with WPA keys, setup host access points, etc. A howto in progress is available here.

Testing foci for 6.0-RELEASE

Issue Status Responsible Description
New dhclient Needs testing Brooks Davis The ISC DHCP client v3.x has been removed in favor of the OpenBSD DHCP client which was based on ISC DHCP v2.X. More testing is needed to make sure there are no major regressions in functionality. We are particularly interested in tests involving non-standard environments.
WPA / hostap Needs testing Sam Leffler Significant new wireless functionality is available in FreeBSD 6.0. As this functionality is rather new, further testing would be appreciated.
if_bridge Needs testing   A new bridge implementation has been added from NetBSD. As this has not seen a wider audience yet, further testing would be appreciated. See if_bridge(4) for configuration details.
sysinstall dhclient problems Needs testing Scott Long, Sam Leffler sysinstall has been updated to deal with the new DHCP client in 6.0, more testing is needed.
ULE Needs testing David Xu The ULE scheduler has been fixed on both HEAD and RELENG_6. More testing is needed.
aac(4) regression Needs testing Scott Long There had been a regression in the aac device driver on Dell 2550 machines. This should be fixed but more testing would be good.
vnode_pager_init nswbuf initialization bug Needs testing Alexander Kabaev There had been a serious performance regression due to a very low bound on the number of swap buffers per vnode (1). This should be fixed but more testing would be good.
fdisk & boot0cfg Needs testing Poul-Henning Kamp The boot0cfg and fdisk utilities need to be updated to use the proper GEOM model for updating the boot sector of a mounted disk.
Alarming devfs shutdown error needs to be quenched Done Robert Watson During system shutdown, devfs generates a spurious error message due to /dev being non-unmountable: "unmount of /dev failed (BUSY)". While this is not a product of serious breakage, it is a potentially alarming error, whose source should either be fixed, or the message should be forceably quenched, in order to avoid many false positive bug reports and user concern.
page fault at if_ethersubr.c: 284 Done Gleb Smirnoff Found by stress tests at http://www.holm.cc/stress/log/cons141.html.
Livelock Needs testing Tor Egge Found by stress tests at http://www.holm.cc/stress/log/cons139.html. This is the second most frequent panic reported by Peter Holm.
panic: softdep_setup_inomapdep: found inode Needs testing Tor Egge Found by stress tests at http://www.holm.cc/stress/log/cons138.html.
M_SAVEDINO Needs testing Tor Egge A 'memory modified after free' bug affecting M_SAVEDINO. Possibly a softupdates bug.
kgdb Needs testing Marcel Moolenaar kgdb has been modified to handle trap frames on all tier 1 and 2 architectures. This allows tracebacks to provide useful info.
panic: wrong b_bufobj Needs testing Tor Egge Found by stress tests at http://www.holm.cc/stress/log/cons140.html. This is high priority and can be provoked within minutes of testing!
Panic when filesystem fills Done Robert Watson Inadequate locking causes panics when calling kernel printf functions. This is most often seen when a filesystem fills up and uprintf() is called to report it to the console, but it can happen in many other places also. Properly locking the upper and lower parts of the tty subsystem likely cannot happen for 6.0, but temporary fixes must be developed and committed. A patch has now been committed that is believed to fix this problem by acquiring Giant in uprintf() and tprintf(), as well as a regression test. This work-around has now been merged to RELENG_6 and is ready for wide-spread testing.
panic on multicast socket close after interface removal Done Robert Watson Due to the addition of multicast locking, a long-present memory corruption bug now generates an immediate panic if an interface is removed while a multicast group is in use on the interface. This can manifest if running routed and using vlans. A regression test has now been committed as msocket_ifnet_remove. A short-term fix will be to GC references to the interface in the IPv4/IPv6 multicast socket options hung off of the inpcb; a long term solution will require determining the correct application semantics for interface removal as relates to multicast sockets. This problem is described in detail in PR kern/77665. A fix has now been committed to 7-CURRENT in the CVS HEAD, and merged to RELENG_6.
panic: handle_written_inodeblock: live inodedep Needs testing Don “Truck” Lewis, Tor Egge Found by stress tests at http://www.holm.cc/stress/log/cons142.html.
"snaplk" livelock Needs testing Don “Truck” Lewis Found by stress tests at http://www.holm.cc/stress/log/cons143.html.
fpudna: fpcurthread == curthread 1 times Needs testing Stephan Uphoff This warning is appearing sporadically on a dual AMD64 system. It appears to be warning against a legitimate problem, and thus should be investigated and fixed.
race condition in POSIX named fifos Done Robert Watson There have been multiple reports of panics when running with make -j on SMP systems, which appear to be associated with a race condition in the POSIX named fifo support. A number of regression tests have been written to confirm that any fix does not break fifos, and a number of other unrelated bugs were fixed in the process. The cause of the symptoms has been identified, and a workaround has been committed to 7-CURRENT for testing; it has been merged to RELENG_6 for testing.
UFS_EXTATTR_AUTOSTART Done Robert Watson Changes in VFS locking as part of the SMPng work have resulted in a new panic during the auto-starting of UFS1 extended attributes. A patch to fix this has been committed (ufs_extattr.c:1.82), and now requires further testing before it is merged to RELENG_6 and RELENG_6_0.

Stress Test Panics

The system is continuously being subjected to Peter Holm's Kernel Stress Test Suite. The following issues have recently been discovered from this test suite.



Last modified on: May 15, 2021 by Allan Jude