xfs_db problem

Hello,

Recently we've noticed that xfs_admin (xfs_db ver. 2.10.2) causes segfault or FP traps while trying
to use following command on FAT partition:

# xfs_admin -u /dev/sda1
xfs_admin: unexpected XFS SB magic number 0xeb58906d
Floating point exception

and then makes error in dmesg:

xfs_db[10385] trap divide error ip:80b32b0 sp:fffbd900 error:0 in xfs_db[8048000+7f000]

We know that xfs_admin is dedicated tool for XFS only not for FAT or other filesystems,
but we would like to have xfs_db which will be able to cope with non-xfs filesystems also.

After some investigations we've found some place and put there exit(1) - in xfsprogs-2.10.2/db/init.c

in init() procedure:

    (...)
        sbp = &xmount.m_sb;
        if (sbp->sb_magicnum != XFS_SB_MAGIC) {
                fprintf(stderr, _("%s: unexpected XFS SB magic number
0x%08x\n"),
                        progname, sbp->sb_magicnum);
                exit(1);
        }
    (...)

Is it the right way to fix this problem?

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

The thing is, xfs_db could in theory be used to *fix* the bad magic number, so refusing to continue might be a problem. But perhaps when invoked as xfs_admin, it should just bail out. It would be interesting to find where the actual failure is, though...

FWIW, I see a slightly different failure on a 128M vfat fs:

[root@host tmp]# file fsfile
fsfile: x86 boot sector, mkdosfs boot message display [root@host tmp]# xfs_admin -u fsfile
xfs_admin: unexpected XFS SB magic number 0xeb3c906d
xfs_admin: size check failed
xfs_admin: failed to alloc 20875030464 bytes: Cannot allocate memory

It probably depends on the details of the (non-xfs) filesystem it's pointed at.

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

We have bugzilla open for this, too. I think it's reasonable to get a warning, but not the additional problems when calling xfs_db directly because that's a low-level tool for people that know what they do.

But when called through wrappers like xfs_admin we should first do the magic number check. I think the proper way to implement this is to add an option to xfs_db to perform the magic number check, and make sure all the wrapper scripts set it, similar to how they already use -p <progname>.

Any volunteer to implement this? It should be fairly simple.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs