2010/11/19 Jeff Layton <jlayton@xxxxxxxxxx>: > On Fri, 19 Nov 2010 14:55:57 +0300 > I don't think we should apply any patches to fix a problem that we > don't really understand. If it mostly goes away with cifsFYI turned on > then it sounds like a race condition of some sort. It looks like > invalidate_inode_pages2 invalidates locked pages whereas > invalidate_remote_inode doesn't, so that may be a hint, or it could > just change the timing of the race. > > You'll need to ascertain the nature of the race to ensure that the fix > is really fixing the problem and not just papering over the bug. > I checked all the lock/unlock calls for pages in CIFS code and didn't find any problem places. I added debug message that shows how many pages invalidate_remote_inode invalidates and for wrong case it is 0: /home/piastry/git/cifs-2.6/fs/cifs/file.c: CIFS VFS: in cifs_close as Xid: 1922 with uid: 500 /home/piastry/git/cifs-2.6/fs/cifs/cifssmb.c: In CIFSSMBClose /home/piastry/git/cifs-2.6/fs/cifs/transport.c: For smb_command 4 /home/piastry/git/cifs-2.6/fs/cifs/transport.c: Sending smb: total_len 45 /home/piastry/git/cifs-2.6/fs/cifs/connect.c: rfc1002 length 0x27 /home/piastry/git/cifs-2.6/fs/cifs/file.c: closing last open instance for inode e2402828 /home/piastry/git/cifs-2.6/fs/cifs/file.c: CIFS VFS: leaving cifs_close (xid = 1922) rc = 0 /home/piastry/git/cifs-2.6/fs/cifs/inode.c: CIFS VFS: in cifs_revalidate as Xid: 1923 with uid: 500 /home/piastry/git/cifs-2.6/fs/cifs/inode.c: Revalidate: /_test4321_ inode 0xe2402690 count 1 dentry: 0xf6f0b550 d_time 1962130 jiffies 2313421 /home/piastry/git/cifs-2.6/fs/cifs/file.c: CIFS VFS: in cifs_writepages as Xid: 1924 with uid: 500 /home/piastry/git/cifs-2.6/fs/cifs/file.c: CIFS VFS: leaving cifs_writepages (xid = 1924) rc = 0 /home/piastry/git/cifs-2.6/fs/cifs/inode.c: invalidating remote inode since open detected it changed /home/piastry/git/cifs-2.6/fs/cifs/inode.c: invalidate returned: 0 ^^^ /home/piastry/git/cifs-2.6/fs/cifs/inode.c: CIFS VFS: leaving cifs_revalidate (xid = 1923) rc = 0 /home/piastry/git/cifs-2.6/fs/cifs/file.c: CIFS VFS: in cifs_open as Xid: 1925 with uid: 500 /home/piastry/git/cifs-2.6/fs/cifs/file.c: inode = 0xe2402690 file flags are 0x8000 for /_test4321_ For right cases it is "invalidate returned: 1". This message was taken from dmesg after a run of my python script I posted before. So, the page can be locked and then successfully unlocked by VFS (because invalidate_inode_pages2 locks a page before processing it). Anyway I think we shouldn't use invalidate_remote_inode at all - we can miss pages that are under some VFS process. In this case invalidate_inode_pages2 is ok - I don't think it can break any things. -- Best regards, Pavel Shilovsky. -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
