"[bgmilne]" <[EMAIL PROTECTED]> writes: > Danny, it seems much better on 9.2b2 than on 9.1, I had an automatic reboot > (instead of seeing the REBOOT LINUX message) on my new 9.2b2 install, and I > can't get it to do the old bad behaviour by hard resets, with root on ext3. The > ext3 root partition it's journal recovered now before init even starts. > > Also, if left alone during boot after an unclean shutdown, it does the right > thing, and the user doesn't come back to a "Do you want to run fsck?"-type question. > > With bootsplash in silent mode by default, the user probably won't see any > tempting prompts anyway ;-). Even if fsck is run by the user, it's now after a > successful journal recovery, so shouldn't result in corruption of files. > > (I couldn't find any changelog claiming credit for this apparent > improvement, so maybe I haven't been testing enough, maybe I should > hard reset while installing packages, running rsync, playing music, > and compiling a kernel?).
for fsck, i suggest to frederic lepied that fsck should result in rebooting if exit code is 2 because fsck (sanely of course) directly fix the fs, thus bypassing kernel's vfs[1]. actually frederic lepied is the one to be credited for the implementation. fred noticed that 3 should resulted in immediate reboot too. we could prevent "rebooting because of fsck fixes bypassed the vfs" if we add the fsck for the root fs in the initrd so that root fs is fixed even before being mounted ro by initcripts. drawback: memory usage at bootstrapping time, especially for xfs. another improvement would be to alter fsck so that it claimed the partition (using /dev/raw access) so that a fs driver would not mount it while being checked (that would need some kernel-2.6.x backport through). but this is not as sexy as checking fs from initrd because it does not provide any more safety since the root fs is: - either already ro mounted and checked from initscripts - or not yet mounted and checked from initrd