non-critical ext3-fs errors?

I notice these errors every now and then on my 100GB WD drive (partitioned into bits) and particularly on this partition that I compile everything on.

EXT3-fs error (device ide0(3,7)): ext3_readdir: bad entry in directory #17821: directory entry across blocks - offset=24, inode=17822, rec_len=4076, name_len=3

Now these errors dont cause the system to panic like other ext3 errors
do, so i'm wondering what the significance of these errors are. I'm
running 2.4.19-rc3 and the drive is on VIA vt82c686b (rev 40) IDE
UDMA100 controller with dma enabled.

This is definitely a bad entry - offset + rec_len (4100) != blocksize (4096). Strange. You need to run e2fsck -f on this partition. It sounds like you are getting some corruption or bit flips happening.

Well, this is an error, and the next time the computer is restarted it should force a full fsck on the partition. The other "panic" (oops) situations are when things are totally shot and it is better to stop doing anything than try and continue. In this case, it is possible to continue, but that doesn't mean things are OK.

Cheers, Andreas
--
Andreas Dilger
http://www-mddsp.enel.ucalgary.ca/People/adilger/
http://sourceforge.net/projects/ext2resize/

I rebooted after setting max mount count to 1 so it would force and fsck, I got no error messages. So, I dropped to shell in single mode and attempted to unmount /usr (the partition with the issues) but umount
insisted /usr was busy. I looked at ps, nothing executed from usr. I
looked at my shell, nothing executed on /usr. I looked at lsof, nothing showing /usr files being accessed. It didn't make any sense at all but then there are other surprises with 2.4.19-rc3 that come later. I fsck -f'd the partition again with dma disabled and still, no errors reported. Either it fixed them behind the scenes or didn't detect
them. I then went and looked at my dmesg to notice this interesting
message

PDC20262: chipset revision 1
ide: Skipping Promise RAID controller.

Now this is funny. I dont remember telling the kernel to skip my other ide controller. I double checked the config, nope, no ignoring or skipping of the promise ide controller. This did not occur in previous
kernels, it seems to be brand new in rc3. I confess to having used jam
kernels with rc2 and rc1 though which had other ide patches in it. So
what gives with this ? Why is rc3 skipping the promise ide
controller? And why is fsck not reporting any fixes to errors that
dmesg shows is being detected during use?

That is not the right thing to do, FYI. That will mean that each time you boot (even after clean shutdown) you will get an fsck. Instead, you should change the mount count (-C), or the time it was last checked (-T on more recent tune2fs) to force it to fsck. That will make it a one-time event.

Yes, you could also touch /forcefsck or whatever, but that would force it for all filesystems, again probably not what you want.

Cheers, Andreas
--
Andreas Dilger
http://www-mddsp.enel.ucalgary.ca/People/adilger/
http://sourceforge.net/projects/ext2resize/