After a non-hardware crash where the system had no chance to flush its buffers to disk (reset button, reboot -nq, reset from debugger) fsck reports a soft update inconsistency in preen mode during restart: Jan 2 10:18:01 bajor root: /dev/ad0s2g: LINK COUNT DIR I=253952 OWNER=root MODE=40775 Jan 2 10:18:01 bajor root: /dev/ad0s2g: SIZE=512 MTIME=Jan 2 10:14 2002 COUNT 2 SHOULD BE 3 Jan 2 10:18:01 bajor root: /dev/ad0s2g: LINK COUNT INCREASING Jan 2 10:18:01 bajor root: /dev/ad0s2g: UNEXPECTED SOFT UPDATE INCONSISTENCY; RUN fsck MANUALLY. From my understanding of soft updates (see http://www.mckusick.com/softdep/index.html) this should never happen except after a power failure with write-cache enabled on the disk or after media failures. How-To-Repeat: Start extracting ports.tgz from FreeBSD installation CDROM into a soft updates enabled partition and hit reset after a few seconds. The result is the same on three different machines with ATA-, SCSI- (Adaptec-2940) and SCSI-RAID-Controllers (ICP-Vortex). Changing the write-cache settings on the disks/controllers has no effect and there was definitly no power loss during the crash.
Responsible Changed From-To: freebsd-bugs->dillon Over to the resident soft updates man, since most of this stuff gets handled by Matt instead of Kirk these days.
Responsible Changed From-To: dillon->freebsd-bugs Back to the free pool.
Responsible Changed From-To: freebsd-bugs->freebsd-fs Over to maintainer(s).
State Changed From-To: open->feedback Is this still an issue on recent versions of FreeBSD?
State Changed From-To: feedback->closed as per previous comment