First, we've been running many raid5+XFS, some over 1TB in size:
It's great! Thanks for adding this filesystem to Linux.
We'd like to update a couple of our systems which are running
vanilla 2.4.16+XFS to 2.4.19 (with XFS August 18th snapshot).
We're recovering from a disk failure, so I'm currently mounting
the fs read-only to be safe. 2.4.16 does this, and replays the
log just fine. 2.4.19, however, oopses with a NULL pointer
deref in xlog_do_log_recovery(0xf7d8679c, 0x36c58, 0x0).
Also of interest, xfs_logprint (version 2.1.2) lists a bunch of
entries in cycle 190, and then eventually prints a garbage entry
... bogus version, bogus cycle, etc. ... The 2.4.16 kernel apparently
does not find problem here. xfs_logprint's memory usage blows up
and eventually it gets oom-killed.
Should I send you just the log? (How do I do that?) Thanks again!
Be seeing you,