> Here is my problem: we have here several Linux PC in hostile environment
> (students) and they are often uncleanly rebooted. The Sun Solaris
> stations nearly always recover automatically, but the Linux stations
> often require a manual root intervention. This upsets our sysadmin.
Gah. Don't rely on Solaris that much - it's *not* a s-u and it
is pretty able to corrupt fs silently. Example: sometime ago we had a
power outage. I did rsh <host> last -10 reboot for our boxen. To my
amazement two of them seemed to be unaffected, though they sit on the same
circuit as affected ones. WTF? Then I noticed the *dates* given by last on
those boxen. No way in hell could they stay up since the last summer
(damn, their kernels are newer then that!) OK, I telnetted there and said
last -10. Yeah, right. <foo> still logged in. Since Jun 1st. 1998.
Lots of swearing followed. After going through the thing with
truss and looking through the /var/adm/wtmpx it turned out that (a) on a
dirty reboot /var/adm/wtmpx got truncated to the block boundary;
(b) pututxline(3) happily kept adding new records to the end of file;
(c) last didn't like misaligned records (surprise, surprise); (d) neither
init(8) nor rc scripts cared to check the size of the thing; (e) size of
record is *not* a power of 2 (it's 372), so further dirty reboots (there
were ones) didn't restore the alignment. Overall: after the dirty reboot
last(1) has a chance to stop working. Forever (OK, until you truncate the
file by hands).
> So.. is there any plan/patch for integrating some softupdate feature to
> ext2 under Linux ? Or is there something I utterly missed ?
Yes, there is. It's 2.3 stuff, *NOT* a 2.2 one.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to firstname.lastname@example.org
Please read the FAQ at http://www.tux.org/lkml/