-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Markus Koßmann wrote:
Am Montag, 7. April 2008 schrieb Jeff Mahoney:
Markus Koßmann wrote:
Am Sonntag, 6. April 2008 schrieb Jeff Mahoney:
Jeff Mahoney wrote:
Markus Koßmann wrote: > All KOTDs since 2.6.25-rc7-git2-HEAD_20080331133119 > -default are showing the following problem in the logs : > dentry_open called with NULL vfsmount I was wondering how long I'd be able to get away with that. Question answered. I'll come up with a fix today. Or not. I've fixed this problem, but run into others. I'll keep the
Jeff Mahoney wrote: list updated. I've just updated the reiserfs xattr patches to fix the problems. Tomorrow's KOTD will have the update.
It should contain the following in the changelog:
- patches.suse/reiserfs-kill-xattr-readdir.diff: Eliminated use of vfsmount-less dentry_open(). Thanks, that made "dentry_open called with NULL vfsmount" disappear.
But unfortnunately, there is another problem introduced with 2.6.25-rc7 that still exists. When I shutdown the system, there is a segmentation fault during the final umount and the system stops with extra sync but filesystems still mounted readonly . Nothing in the logs. Syslog seems to be allready stopped at that time. Can you switch to console 10 and see what the Oops says?
Sorry, don't have the time to write down the whole message
Kernel Bug at fs/dcache.c 637 Invalid Opcode:0000 [1] SMP
Backtrace begins with : shrink_dache_for_umount+0x37/0x47
RIP: shrink_dcache_for_umount_subtree 0x146/0x216
Ok, this means that there's a reference counting problem and a dentry isn't getting released before umount. I can't reproduce it locally, but I've been continuing the development of these patches. I'll update them as soon as I solve the ABBA deadlock I've been observing under load. What I'm seeing is that one thread is holding the xattr dir i_mutex and waiting on the journal while another thread is in the middle of a transaction and waiting on the xattr dir i_mutex. And then we're stuck. The solution is probably going to be defining a separate dir_action callback that is responsible for locking the dir in addition to avoiding the deadlock. My old version of these patches had a comment saying "Revisit the lock ordering here." Now I remember why. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.4-svn0 (GNU/Linux) Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org iD8DBQFIAFm1LPWxlyuTD7IRAuAmAJ9MU0tplnohEHWNxlla+khoq9YZcQCfRla2 jgjWjK0KFB5dRiqZqcc6E68= =cWn/ -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org