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

Re: ext3 corruption with vanilla kernel 2.4.19?



Hi,

On Fri, 2003-01-10 at 10:56, Nicola Ragozzino wrote:

> After compiling and booting a RH 8.0 box with a 2.4.19 vanilla kernel
> i've started to experience some problems on my /home partition, (i/o
> errors on 2 files).

Well, I/O errors can occassionally be due to a filesystem fault
(especially if a filesystem corruption has resulted in a file trying to
exist beyond the end of the device.)  But that is very very rare: 99% of
the time, the IO errors are hardware or, sometimes, device-driver
related.

> After rebooting the machine failed the to mount /home and dropped me in
> the recovery shell.
> The fsck is still running after 9 hours.

How far has it got?  What indications of progress has it made?

> The system is a dual athlon MP  with 1Gb ram and ASUS MB.
> the /home patrition is mounted on 3 120 Gb maxtor disks on ATA 100 bus
> in RAID 5 (software).
> The point is: i've heard same problems from another sysadmin switching
> on vanilla kernel on is RH 8.0 box with ext3.

The biggest danger I've found is that recent 2.4 kernels enable higher
levels of UDMA on more chipsets, and sometimes that exposes borderline
hardware which could work correctly at slower speeds but which fails at
the higher UDMA speeds.  I've got several examples of such hardware
myself, where the disk has to be forced to a lower DMA rate or the
hardware corrupts data silently.

> Question one: Is really normal that fsck takes so long or i can just
> consider the partition dead and reboot?

It depends,  240GB is a lot of data for it to fix, and software raid5 is
relatively slow on writes.  In addition, you'll probably have a
background raid5 reconstruction trying to proceed (cat /proc/mdstat to
check), so all in all, it's not inconceivable (though it does sound slow
to me.)  Check whether the disk is still in DMA mode, though --- if it
has ended up in PIO, it will be _dead_ slow.

> Question two: It was just an unluckly coincidence or there is some know
> incompatibility between RH 8.0 and vanilla kernels on ext3?

There are no known ext3 incompatibilities, but there have been a lot of
IDE driver changes in recent kernels (not just in Red Hat's --- the
upstream drivers have been getting a lot of work for recent chipsets),
so that's the area I'd suspect first.

Cheers,
 Stephen





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