First, when I e2fsck, a message dumps onto the console about how
there was an error with the disk (medium error, unrecovered read,
according to the SCSI messages); e2fsck reports "Error reading block
%d (Attempt to read block from filesystem resulted in short read) while
doing %s. Ignore error?" "%s" is sometimes "inode scan", among
other things. If I choose to ignore the error, nothing is done
by e2fsck about the fact that the block is unreadable; if I choose
to pay attention to it, this merely halts e2fsck. I'm using 1.09,
fwiw.
Second, despite being tune2fs'd so that the error behavior is
remount-ro, still I get kernel (2.1.29) panics when the kernel
can't read a block that's metadata, e.g. a block bitmap or
an inode. For example, (typed in:)
scsi0: MEDIUM ERROR on channel 0, id 1, lun 0, CDB: Read (6) 1c 4a d2 02 00
Current error sd08:13: sense key Medium Error
Additional sense indicates Unrecovered read error
scsidisk I/O error: dev 08:13, sector 753786
Kernel panic: EXT2-fs panic (device 08:13): ext2_read_inode: unable to read i-node block - inode=93545, block=376893
tune2fs -l confirms that the error behavior should be remount-ro; even
if blocks are getting corrupted faster than e2fsck can cope with them
(assuming it's coping properly at all), the kernel shouldn't be panicing
like that.
Keith (I can try things until I RMA the puppy back...)