linux-cifs-client Digest, Vol 39, Issue 19
Quoted 83 lines · [email protected] wrote:
Send linux-cifs-client mailing list submissions to
[email protected]To subscribe or unsubscribe via the World Wide Web, visit
https://lists.samba.org/mailman/listinfo/linux-cifs-client
or, via email, send a message with subject or body 'help' to
[email protected]You can reach the person managing the list at
[email protected]When replying, please edit your Subject line so it is more specific than "Re: Contents of linux-cifs-client digest..."
Today's Topics:
1.
i_mutexand deadlock (Steve French (smfltc))
2. Re:i_mutexand deadlock (Dave Kleikamp)----------------------------------------------------------------------
Message: 1
Date: Fri, 23 Feb 2007 10:02:16 -0600
From: "Steve French (smfltc)" <[email protected]>
Subject: [linux-cifs-client]i_mutexand deadlock
To: [email protected]
Cc: [email protected]
Message-ID: <[email protected]>
Content-Type: text/plain; charset=ISO-8859-1; format=flowedA field in
i_size_write(i_size_seqcount) must be protected against simultaneous update otherwise we risk looping ini_size_read.The suggestion in
fs.his to usei_mutexwhich seems too dangerous due to the possibility of deadlock.There are 65 places in the fs directory which lock an
i_mutex, and seven more in the mm directory. The vfs does clearly lock file inodes in some paths before calling into a particular filesystem (nfs, ext3, cifs etc.) - in particular for fsync but probably for others that are harder to trace. This seems to introduce the possibility of deadlock if a filesystem also usesi_mutexto protect file size updates
Documentation/filesystems/Lockingdescribes the use ofi_mutex(was "i_sem" previously) and indicates that it is held by the vfs on three additional calls on file inodes which concern me (for deadlock possibility), setattr, truncate and unlink.nfs seems to limit its use of
i_mutexto llseek andinvalidate_mapping, and does not appear to grab thei_mutex(or any sem for that matter) to protecti_size_write(nfs callsi_size_writeinnfs_grow_file) - and for the case ofnfs_fhget(in which they bypassi_size_writeand seti_sizedirectly) does not seem to grabi_mutexeither.ext3 also does not use
i_mutexfor this purpose (protectingi_size_write) - ony to protect a journalling ioctl.I am concerned about using
i_mutexto protect the cifs calls toi_size_write(although it seems to fix a problem reported ini_size_readunder stress) because of the following:1) no one else calls
i_size_writeAFAIK (on our file inodes)
2) we don't block insidei_size_writedo we ... (so why in the world do they take a slow mutex instead of a fast spinlock)
3) we don't really know what happens inside fsync (the paths through the page cache code seem complex and we don't want to reenter writepage in low memory conditions and deadlock updating the file size), and there is some concern that the vfs takes thei_mutexin other paths on file inodes before entering our code and could deadlock.Any reason, why an fs shouldn't simply use something else (a spinlock) other than
i_mutexto protect thei_size_writecall?------------------------------
Message: 2
Date: Fri, 23 Feb 2007 10:29:53 -0600
From: Dave Kleikamp <[email protected]>
Subject: Re: [linux-cifs-client]i_mutexand deadlock
To: "Steve French (smfltc)" <[email protected]>
Cc: [email protected], [email protected]
Message-ID: <[email protected]>
Content-Type: text/plainQuoted 5 lines · Steve French (smfltc)
A field in
i_size_write(i_size_seqcount) must be protected against simultaneous update otherwise we risk looping ini_size_read.The suggestion in
fs.his to usei_mutexwhich seems too dangerous due to the possibility of deadlock.I'm not sure if it's as much a suggestion as a way of documenting the locking that exists (or existed when the comment was written).
"...
i_size_write()does need locking around it (normallyi_mutex) ..."Quoted 2 lines
There are 65 places in the fs directory which lock an
i_mutex, and seven more in the mm directory. The vfs does clearly lock file inodes in some paths before calling into a particular filesystem (nfs, ext3, cifs etc.) - in particular for fsync but probably for others that are harder to trace. This seems to introduce the possibility of deadlock if a filesystem also usesi_mutexto protect file size updates
Quoted 10 lines
Quoted 5 lines
I am concerned about using
i_mutexto protect the cifs calls toi_size_write(although it seems to fix a problem reported ini_size_readunder stress) because of the following:1) no one else calls
i_size_writeAFAIK (on our file inodes)I think you're right.
I produced a patch (attached to http://bugzilla.kernel.org/show_bug.cgi?id=7903) which
fixes this, has tested out fine, has no sideeffects/changes outside of cifs and simply uses inode->i_lock (spinlock) to protect i_size_write calls on cifs inodes which seemed faster/safer especially as there was one other path outside the cifs code which callsi_size_write (vmtruncate in mm/memory.c) which was easy to work around since cifs
could call its own version of this small function. This is also a lot easier to prove/
understand as there were 72 places which called i_mutex outside of cifs (ie in mm and fs)
many of which were potential deadlocks if cifs used i_mutex to protect its i_size_writes
(network filesystems do updates of the file size in more places than a local filesystem
of course ... so have to be careful about grabbing the same sem twice ...)
I don't want to push that over akpm's objections as he knows the mm well, but this seems the simplest/safest approach.