[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: another quota related ext3fs crash...



On Sat, 9 Mar 2002, Andrew Morton wrote:

> Michael Renner wrote:
> >
> > this message didn't get parsed by ksymoops
> >
> > Mar  9 17:15:13 srck trottelkunde Assertion failure in journal_start() at transaction.c:226: "handle->h_transaction->t_journal == journal"
>
> This one is a bit of a red herring.  The real error is this one:
>
> > Dec  9 19:55:30 srck trottelkunde Kernel panic: EXT3-fs panic (device ide1(22,66)): load_block_bitmap: block_group >= groups_count - block_group = 131071, groups_count = 94
>
> Which called panic(), which called sys_sync(), which tried to sync
> some other filesystem while inside this filesystem's transaction.
> Really, panic() shouldn't be calling sys_sync().

i bet this behaviour isn't designed that way in the kernel sources? :)

> So the question is, why is truncate trying to access such a wild
> blockgroup number?  ext3_free_blocks() checks the block number so
> possibly some of the in-memory superblock information has been
> corrupted.

could this problem get tracked down if i had jbd debugging enabled or
where these errors completely outside of ext3/jbd's scope?

> Hard to say.  Are you running anything unusual on the machine? Anything
> which could help us to understand why you get this happening, but
> (to my knowledge) nobody else does?  Is it good quality hardware?


software:

well, this is more a "play around & test stuff"-box of mine, 4 big cpu
hogs (2 * lame, encoding a radio live stream from a tv/radio card; 2 *
halflife dedicated servers). the interesting stuff here is that the hlds
processes are reniced to -20 and -15, and the lame processes to -10. is it
possible that this renicing interferes with kernel-internals?

aside from that there's the more usual apache/mysql reniced to +5,
shoutcast servers for streaming the stuff encoded by lame, except that
there's nothing usual (imho)


kernel:

well, it's an -ac kernel, the main reason for using it is the vfsv0 quota
(and the somehow better vm during the major vanilla-vm wrapup from 2.4.7
till 2.4.15 or so)

loaded kernel modules include i2c (for lm-sensors and bttv), bttv (for the
radio card) and the lm-sensors module for my sensor chip

unusual from the default configs i've seen so far:

devfs kernel + userland daemon running, local APIC (uniprocessor system)
is enabled, video mode selection support is enabled (i hate 80*24
terminals)

i can mail you the .config of my current kernel if that would help


hardware:

i haven't got that "high quality" hardware in this server.

the motherboard is a epox 8kta3+ with via kt133a chipset. the drives are
attached to the via ide controller, the onboard HPT370 is deactivated.

cpu is a athlon 1400mhz, 3 256MB dimms (should be fine, infineon ram
with manufacturer warranty)

the drive in question is a seagate ST320430A (according to
/proc/hdd/model) which runs at udma66

except from that there's a tseng lebs ET6000, a 3com905C, the onboard via
soundcard (radio card out -> soundcard line in) and a hauppauge wintv pci
theater



i hope i've covered all sections, do you need further information on any
of these points?


best regards,
michael





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]