Commit graph

64 commits

Author SHA1 Message Date
Jeff Layton
3bc303c254 cifs: convert oplock breaks to use slow_work facility (try #4)
This is the fourth respin of the patch to convert oplock breaks to
use the slow_work facility.

A customer of ours was testing a backport of one of the earlier
patchsets, and hit a "Busy inodes after umount..." problem. An oplock
break job had raced with a umount, and the superblock got torn down and
its memory reused. When the oplock break job tried to dereference the
inode->i_sb, the kernel oopsed.

This patchset has the oplock break job hold an inode and vfsmount
reference until the oplock break completes.  With this, there should be
no need to take a tcon reference (the vfsmount implicitly holds one
already).

Currently, when an oplock break comes in there's a chance that the
oplock break job won't occur if the allocation of the oplock_q_entry
fails. There are also some rather nasty races in the allocation and
handling these structs.

Rather than allocating oplock queue entries when an oplock break comes
in, add a few extra fields to the cifsFileInfo struct. Get rid of the
dedicated cifs_oplock_thread as well and queue the oplock break job to
the slow_work thread pool.

This approach also has the advantage that the oplock break jobs can
potentially run in parallel rather than be serialized like they are
today.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2009-09-24 18:33:18 +00:00
Jeff Layton
1b49c55661 cifs: protect GlobalOplock_Q with its own spinlock
Right now, the GlobalOplock_Q is protected by the GlobalMid_Lock. That
lock is also used for completely unrelated purposes (mostly for managing
the global mid queue). Give the list its own dedicated spinlock
(cifs_oplock_lock) and rename the list to cifs_oplock_list to
eliminate the camel-case.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2009-09-01 22:25:29 +00:00
Steve French
da505c386c [CIFS] Make socket retry timeouts consistent between blocking and nonblocking cases
We have used approximately 15 second timeouts on nonblocking sends in the past, and
also 15 second SMB timeout (waiting for server responses, for most request types).
Now that we can do blocking tcp sends,
make blocking send timeout approximately the same (15 seconds).

Signed-off-by: Steve French <sfrench@us.ibm.com>
2009-01-29 03:32:13 +00:00
Jeff Layton
0496e02d87 cifs: turn smb_send into a wrapper around smb_sendv
cifs: turn smb_send into a wrapper around smb_sendv

Rename smb_send2 to smb_sendv to make it consistent with kernel naming
conventions for functions that take a vector.

There's no need to have 2 functions to handle sending SMB calls. Turn
smb_send into a wrapper around smb_sendv. This also allows us to
properly mark the socket as needing to be reconnected when there's a
partial send from smb_send.

Also, in practice we always use the address and noblocksnd flag
that's attached to the TCP_Server_Info. There's no need to pass
them in as separate args to smb_sendv.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2009-01-29 03:32:12 +00:00
Volker Lendecke
ac6a3ef405 Remove an already-checked error condition in SendReceiveBlockingLock
Remove an already-checked error condition in SendReceiveBlockingLock

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:13 +00:00
Volker Lendecke
698e96a826 Streamline SendReceiveBlockingLock: Use "goto out:" in an error condition
Streamline SendReceiveBlockingLock: Use "goto out:" in an error condition

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:13 +00:00
Volker Lendecke
17c8bfed8a Streamline SendReceiveBlockingLock: Use "goto out:" in an error condition
Streamline SendReceiveBlockingLock: Use "goto out:" in an error condition

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Steve French
2b2bdfba7a [CIFS] Streamline SendReceive[2] by using "goto out:" in an error condition
Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Volker Lendecke
8e4f2e8a1e Slightly streamline SendReceive[2]
Slightly streamline SendReceive[2]

Remove an else branch by naming the error condition what it is

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Volker Lendecke
829049cbb1 Check the return value of cifs_sign_smb[2]
Check the return value of cifs_sign_smb[2]

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Volker Lendecke
27a97a613b [CIFS] Slightly simplify wait_for_free_request(), remove an unnecessary "else" branch
This is no functional change, because in the "if" branch we do an early
"return 0;".

Signed-off-by: Volker Lendecke <vl@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Volker Lendecke
8fbbd365cc Simplify allocate_mid() slightly: Remove some unnecessary "else" branches
Simplify allocate_mid() slightly: Remove some unnecessary "else" branches

Signed-off-by: Volker Lendecke <vl@samba.org>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:12 +00:00
Volker Lendecke
6d9c6d5431 [CIFS] In SendReceive, move consistency check out of the mutexed region
inbuf->smb_buf_length does not change in in wait_for_free_request() or in
allocate_mid(), so we can check it early.

Signed-off-by: Volker Lendecke <vl@samba.org>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:11 +00:00
Jeff Layton
8570552425 cifs: fix wait_for_response to time out sleeping processes correctly
cifs: fix wait_for_response to time out sleeping processes correctly

The current scheme that CIFS uses to sleep and wait for a response is
not quite what we want. After sending a request, wait_for_response puts
the task to sleep with wait_event(). One of the conditions for
wait_event is a timeout (using time_after()).

The problem with this is that there is no guarantee that the process
will ever be woken back up. If the server stops sending data, then
cifs_demultiplex_thread will leave its response queue sleeping.

I think the only thing that saves us here is the fact that
cifs_dnotify_thread periodically (every 15s) wakes up sleeping processes
on all response_q's that have calls in flight.  This makes for
unnecessary wakeups of some processes. It also means large variability
in the timeouts since they're all woken up at once.

Instead of this, put the tasks to sleep with wait_event_timeout. This
makes them wake up on their own if they time out. With this change,
cifs_dnotify_thread should no longer be needed.

I've been testing this in conjunction with some other patches that I'm
working on. It doesn't seem to affect performance at all with with heavy
I/O. Identical iozone -ac runs complete in almost exactly the same time
(<1% difference in times).

Thanks to Wasrshi Nimara for initially pointing this out. Wasrshi, it
would be nice to know whether this patch also helps your testcase.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Cc: Wasrshi Nimara <warshinimara@gmail.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:11 +00:00
Jeff Layton
72ca545b2d cifs: convert tcpSem to a mutex
Mutexes are preferred for single-holder semaphores...

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:09 +00:00
Jeff Layton
24b9b06ba7 cifs: remove unused SMB session pointer from struct mid_q_entry
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-12-26 02:29:09 +00:00
Steve French
61de800d33 [CIFS] fix error in smb_send2
smb_send2 exit logic was strange, and with the previous change
could cause us to fail large
smb writes when all of the smb was not sent as one chunk.

Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-10-30 20:15:22 +00:00
Steve French
edf1ae4038 [CIFS] Reduce number of socket retries in large write path
CIFS in some heavy stress conditions cifs could get EAGAIN
repeatedly in smb_send2 which led to repeated retries and eventually
failure of large writes which could lead to data corruption.

There are three changes that were suggested by various network
developers:

1) convert cifs from non-blocking to blocking tcp sendmsg
(we left in the retry on failure)
2) change cifs to not set sendbuf and rcvbuf size for the socket
(let tcp autotune the buffer sizes since that works much better
in the TCP stack now)
3) if we have a partial frame sent in smb_send2, mark the tcp
session as invalid (close the socket and reconnect) so we do
not corrupt the remaining part of the SMB with the beginning
of the next SMB.

This does not appear to hurt performance measurably and has
been run in various scenarios, but it definately removes
a corruption that we were seeing in some high stress
test cases.

Acked-by: Shirish Pargaonkar <shirishp@us.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-10-29 00:47:57 +00:00
Pekka Enberg
232087cb73 cifs: don't use GFP_KERNEL with GFP_NOFS
GFP_KERNEL and GFP_NOFS are mutually exclusive. If you combine them, you end up
with plain GFP_KERNEL which can deadlock in cases where you really want
GFP_NOFS.

Cc: Steve French <sfrench@samba.org>
Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-09-22 22:23:56 +00:00
Shirish Pargaonkar
176803562b [CIFS] cifs send2 not retrying enough in some cases on full socket
There are cases in which, on a full socket which requires retry on
sending data by the app (cifs in this case), that we were not
retrying since we did not reinitialize a counter.

This fixes the retry logic to retry up to 15 seconds on stuck
sockets.

Signed-off-by: Shirish Pargaonkar <shirishp@us.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-07-29 21:26:13 +00:00
Steve French
5d941ca628 [CIFS] Fix oops when slow oplock process races with unmount
If a tcon is being freed in call tconInfoFree, clean up any entries that may
exist in global oplock queue as the tcon structure hanging off of those entries
will be invalid and can cause oops while accesing any elements in the
tcon structure.

Signed-off-by: Shirish Pargaonkar <shirishp@us.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-04-15 18:40:48 +00:00
Steve French
90c81e0b0e [CIFS] clean up some hard to read ifdefs
Christoph had noticed too many ifdefs in the CIFS code making it
hard to read.  This patch removes about a quarter of them from
the C files in cifs by improving a few key ifdefs in the .h files.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-02-12 20:32:36 +00:00
Steve French
ad7a2926b9 [CIFS] reduce checkpatch warnings
Signed-off-by: Steve French <sfrench@us.ibm.com>
2008-02-07 23:25:02 +00:00
Steve French
133672efbc [CIFS] Fix buffer overflow if server sends corrupt response to small
request

In SendReceive() function in transport.c - it memcpy's
message payload into a buffer passed via out_buf param. The function
assumes that all buffers are of size (CIFSMaxBufSize +
MAX_CIFS_HDR_SIZE) , unfortunately it is also called with smaller
(MAX_CIFS_SMALL_BUFFER_SIZE) buffers.  There are eight callers
(SMB worker functions) which are primarily affected by this change:

TreeDisconnect, uLogoff, Close, findClose, SetFileSize, SetFileTimes,
Lock and PosixLock

CC: Dave Kleikamp <shaggy@austin.ibm.com>
CC: Przemyslaw Wegrzyn <czajnik@czajsoft.pl>
Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-11-13 22:41:37 +00:00
Steve French
a761ac579b [CIFS] log better errors on failed mounts
Also returns more accurate errors to mount for the cases of
account expired and password expired

Acked-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-10-18 21:45:27 +00:00
Steve French
26f57364d7 [CIFS] formatting cleanup found by checkpatch
Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-08-30 22:09:15 +00:00
Steve French
50c2f75388 [CIFS] whitespace/formatting fixes
This should be the last big batch of whitespace/formatting fixes.
checkpatch warnings for the cifs directory are down about 90% and
many of the remaining ones are harder to remove or make the code
harder to read.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-07-13 00:33:32 +00:00
Steve French
b609f06ac4 [CIFS] Fix packet signatures for NTLMv2 case
Signed-off-by: Yehuda Sadeh Weinraub <Yehuda.Sadeh@expand.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-07-09 07:55:14 +00:00
Steve French
79a58d1f60 [CIFS] whitespace cleanup
checkpatch.pl redux

Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-07-06 22:44:50 +00:00
Steve French
8a236264f7 [CIFS] cifs_prepare_write was incorrectly rereading page in some cases
Noticed by Shaggy.

Signed-off-by: Shaggy <shaggy@us.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2007-03-06 00:31:00 +00:00
Christoph Lameter
e18b890bb0 [PATCH] slab: remove kmem_cache_t
Replace all uses of kmem_cache_t with struct kmem_cache.

The patch was generated using the following script:

	#!/bin/sh
	#
	# Replace one string by another in all the kernel sources.
	#

	set -e

	for file in `find * -name "*.c" -o -name "*.h"|xargs grep -l $1`; do
		quilt add $file
		sed -e "1,\$s/$1/$2/g" $file >/tmp/$$
		mv /tmp/$$ $file
		quilt refresh
	done

The script was run like this

	sh replace kmem_cache_t "struct kmem_cache"

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 08:39:25 -08:00
Christoph Lameter
e94b176609 [PATCH] slab: remove SLAB_KERNEL
SLAB_KERNEL is an alias of GFP_KERNEL.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 08:39:24 -08:00
Christoph Lameter
e6b4f8da3a [PATCH] slab: remove SLAB_NOFS
SLAB_NOFS is an alias of GFP_NOFS.

Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-12-07 08:39:23 -08:00
Jeremy Allison
7ee1af765d [CIFS]
Allow Windows blocking locks to be cancelled via a
CANCEL_LOCK call. TODO - restrict this to servers
that support NT_STATUS codes (Win9x will probably
not support this call).

Signed-off-by: Jeremy Allison <jra@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
(cherry picked from 570d4d2d895569825d0d017d4e76b51138f68864 commit)
2006-08-11 21:28:47 +00:00
Steve French
5da07b0208 [CIFS] Make midState usage more consistent
Although harmless, we were sometimes treating midState like it contained
flags but they are exclusive states, and this makes that more clear.

Signed-off-by: Jeremy Allison <jra@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
(cherry picked from 586c057c3a68dd6ae0f3ba94fbf76798b1558074 commit)
2006-08-11 21:28:02 +00:00
Steve French
14a441a2b4 [CIFS] spinlock protect read of last srv response time in timeout path
Signed-off-by: Jeremy Allison <jra@samba.org>
Signed-off-by: Steve French <sfrench@us.ibm.com>
(cherry picked from b33a3f55e54fd210fc043eafcf83728b03bc9e02 commit)
2006-08-11 21:27:41 +00:00
Steve French
3a5ff61c18 [CIFS] Do not time out posix brl requests when using new posix setfileinfo
request and do not time out slow requests to a server that is still responding
well to other threads

Suggested by jra of Samba team

Signed-off-by: Steve French <sfrench@us.ibm.com>
(cherry picked from 89b57148115479eef074b8d3f86c4c86c96ac969 commit)
2006-08-11 21:27:07 +00:00
Steve French
26a21b980b [CIFS] Cleanup extra whitespace in dmesg logging. Update cifs change log 2006-05-31 18:05:34 +00:00
Steve French
4b8f930ff8 [CIFS] Free small buffers earlier so we exceed the cifs
small req buffer pool less often.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2006-02-26 16:41:18 +00:00
Steve French
0820e15a35 [CIFS] Do not zero non-existent iovec in SendReceive response processing.
Could cause memory leak in some readpaths depending on what junk followed it in the stack.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2006-01-23 12:50:04 -08:00
Steve French
0a4b92c05e [CIFS] Add worker function for Get ACL cifs style
Signed-off-by: Steve French <sfrench@us.ibm.com>
2006-01-12 15:44:21 -08:00
Steve French
ec637e3ffb [CIFS] Avoid extra large buffer allocation (and memcpy) in cifs_readpages
Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-12-12 20:53:18 -08:00
Steve French
84afc29b18 [CIFS] Readpages and readir performance improvements - eliminate extra
memcpy.  Part 1

Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-12-02 13:32:45 -08:00
Steve French
6ab16d2495 [CIFS] Fix umount --force to wake up the pending response queue, not just
the request queue. Also periodically wakeup response_q so threads can
check if stuck requests have timed out. Workaround Windows server illegal smb
length on transact2 findfirst response.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-11-29 20:55:11 -08:00
Steve French
0f2b27c438 [CIFS] Fix sparse warnings on smb bcc (byte count)
Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-11-16 14:25:50 -08:00
Steve French
1047abc159 [CIFS] CIFS Stats improvements
New cifs_writepages routine was not updated bytes written in cifs stats.
Also added ability to clear /proc/fs/cifs/Stats by writing (0 or 1) to it.
Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-10-11 19:58:06 -07:00
Steve French
5e1253b501 [CIFS] Correct cifs tcp retry when some data sent before getting EAGAIN.
Continue implementation of cifs umount begin to allow force unmounts of
cifs mounts.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-10-10 14:06:37 -07:00
Steve French
68058e7575 [CIFS] Reduce CIFS tcp congestion timeout (it was too long) and backoff
ever longer amounts (up to 15 seconds).  This improves performance
especially when using large wsize.

Signed-off-by: Steve French (sfrench@us.ibm.com)
2005-10-10 10:34:22 -07:00
Steve French
131afd0b74 [CIFS] /proc/fs/cifs debug code cleanup and new stats2
These changes to debug code and new stats are helpful in
debugging potential tcp performance/configuration problems under cifs.

Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-10-07 09:51:05 -07:00
Steve French
4a77118cd5 CIFS: Allow wsize to exceed CIFSMaxBufSize
This allows cifs_writepages to send data in larger chunks from the page
cache, without requiring larger memory allocations in other cases.

Signed-off-by: Dave Kleikamp <shaggy@austin.ibm.com>
Signed-off-by: Steve French <sfrench@us.ibm.com>
2005-10-05 15:14:33 -07:00