http://oss.sgi.com/bugzilla/show_bug.cgi?id=358
Summary: XFS 1.3 and 1.3.1 leave FS corrupted after crash/reboot
Product: Linux XFS
Version: unspecified
Platform: All
OS/Version: Linux
Status: NEW
Severity: normal
Priority: High
Component: XFS kernel code
AssignedTo: xfs-master@xxxxxxxxxxx
ReportedBy: mbellon@xxxxxxxxxx
I've got a problem that occurs on what, even after a lot of testing, appears to
be a single platform, an ATI 22[56] (MIPS with EIDE), which makes me uneasy that
this is a platform bug. However much analysis and test has been thrown at the
platform and we cannot find anything wrong with the processor, memory, DMA, EIDE
or disk drives.
XFS is running on MV PRO 3.1 (2.4.20++). Make a default file system (mkfs.xfs
2.5.11). Mount it and run a file system stress program (reads and writes files,
creates and deletes directories) - we can provide it. Reset machine in the
middle of the program running and reboot. Mount the file system; log replays and
file system is corrupted 100% of the time (hundreds of tests so far).
What we see is one of the test directories is corrupt - it has no "." or ".."
but you can cd into it and "cd .." out of it. Attempt to create a file via touch
and:
root@xxxxxxxxxxxx:/mnt/test6# touch foo
xfs_da_do_buf: bno 8388608
dir: inode 50331776
Filesystem "ide0(3,1)": XFS internal error xfs_da_do_buf(1) at line 2187 of file
xfs_da_btree.c. Caller 0x801b1404
This architecture does not implement dump_stack()
xfs_da_do_buf: bno 8388608
dir: inode 50331776
Filesystem "ide0(3,1)": XFS internal error xfs_da_do_buf(1) at line 2187 of file
xfs_da_btree.c. Caller 0x801b1404
This architecture does not implement dump_stack()
touch: creating `foo': Unknown error 990
No XFS parameters appear to affect the problem. Neither does the size of the
file system (10 GB, 40 GB and 160 GB). Problem does not show running same kernel
source on X86, PPC and other MIPS platforms. Every block device test passes on
all platforms including the ATI 22[56] - buffer layer seems OK. EXT3 passes this
test 100% of the time on all platforms.
Use xfs_repair and have it forget the log (-L) and the file system is OK, the
corrupted directory is gone and there are a few files in lost+found. The test
program deletes a great many files and then the directory quite often.
Since no problem is too trivial I report this hoping that there may be something
that can be shared or remembered. Is there a program to dump a log from user
space directly from a disk partition? Any debugging ideas?
sorry to bother you but I'm running out of things to try and learning XFS
internals is taking a lot of time.
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
|