The following scenario shows some locking problems with nfsd under
2.6.0:
- export filesystem
- lazy unmount filesystem
- unexport filesystem
The following call trace is produced:
bad: scheduling while atomic!
Call Trace:
[schedule+1301/1456] schedule+0x585/0x590
[__filemap_fdatawrite+164/208] __remove_from_page_cache+0x24/0x60
[__pagevec_lru_add+121/256] __pagevec_release+0x29/0x40
[truncate_inode_pages+498/656] truncate_inode_pages+0xc2/0x290
[__iget+1/96] destroy_inode+0x51/0x60
[__iget+1/96] destroy_inode+0x51/0x60
[invalidate_inodes+138/208] dispose_list+0xaa/0xb0
[prune_icache+100/496] invalidate_inodes+0xa4/0xd0
[sget+75/368] generic_shutdown_super+0x7b/0x190
[get_sb_bdev+183/352] kill_anon_super+0x17/0x50
[generic_shutdown_super+14/400] deactivate_super+0x5e/0xc0
[_end+274119914/1069647496] svc_export_put+0x82/0x90 [nfsd]
[_end+273198636/1069647496] cache_clean+0x164/0x250 [sunrpc]
[_end+273198973/1069647496] cache_flush+0x15/0x50 [sunrpc]
[_end+274126880/1069647496] nfsd_export_flush+0x38/0x50 [nfsd]
[_end+274095171/1069647496] nfsd+0x32b/0x3b0 [nfsd]
[_end+274094360/1069647496] nfsd+0x0/0x3b0 [nfsd]
[kernel_thread_helper+5/12] kernel_thread_helper+0x5/0xc
The heart of the problem seems to be that cache_clean is holding the
cache_list_lock spinlock while calling cache_put (svc_export_put),
which calls mntput, which may sleep.
Miklos
-------------------------------------------------------
The SF.Net email is sponsored by EclipseCon 2004
Premiere Conference on Open Tools Development and Integration
See the breadth of Eclipse activity. February 3-5 in Anaheim, CA.
http://www.eclipsecon.org/osdn
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs