Currenly we doesn't assume that a server may break a lease
from RWH to RW which causes us setting a wrong lease state
on a file and thus mistakenly flushing data and byte-range
locks and purging cached data on the client. This leads to
performance degradation because subsequent IOs go directly
to the server.
Fix this by propagating new lease state and epoch values
to the oplock break handler through cifsFileInfo structure
and removing the use of cifsInodeInfo flags for that. It
allows to avoid some races of several lease/oplock breaks
using those flags in parallel.
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
After doing mount() successfully we call cifs_try_adding_channels()
which will open as many channels as it can.
Channels are closed when the master session is closed.
The master connection becomes the first channel.
,-------------> global cifs_tcp_ses_list <-------------------------.
| |
'- TCP_Server_Info <--> TCP_Server_Info <--> TCP_Server_Info <-'
(master con) (chan#1 con) (chan#2 con)
| ^ ^ ^
v '--------------------|--------------------'
cifs_ses |
- chan_count = 3 |
- chans[] ---------------------'
- smb3signingkey[]
(master signing key)
Note how channel connections don't have sessions. That's because
cifs_ses can only be part of one linked list (list_head are internal
to the elements).
For signing keys, each channel has its own signing key which must be
used only after the channel has been bound. While it's binding it must
use the master session signing key.
For encryption keys, since channel connections do not have sessions
attached we must now find matching session by looping over all sessions
in smb2_get_enc_key().
Each channel is opened like a regular server connection but at the
session setup request step it must set the
SMB2_SESSION_REQ_FLAG_BINDING flag and use the session id to bind to.
Finally, while sending in compound_send_recv() for requests that
aren't negprot, ses-setup or binding related, use a channel by cycling
through the available ones (round-robin).
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Even when mounting modern protocol version the server may be
configured without supporting SMB2.1 leases and the client
uses SMB2 oplock to optimize IO performance through local caching.
However there is a problem in oplock break handling that leads
to missing a break notification on the client who has a file
opened. It latter causes big latencies to other clients that
are trying to open the same file.
The problem reproduces when there are multiple shares from the
same server mounted on the client. The processing code tries to
match persistent and volatile file ids from the break notification
with an open file but it skips all share besides the first one.
Fix this by looking up in all shares belonging to the server that
issued the oplock break.
Cc: Stable <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Helps distinguish between an interrupted close and a truly
unmatched open.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
If Close command is interrupted before sending a request
to the server the client ends up leaking an open file
handle. This wastes server resources and can potentially
block applications that try to remove the file or any
directory containing this file.
Fix this by putting the close command into a worker queue,
so another thread retries it later.
Cc: Stable <stable@vger.kernel.org>
Tested-by: Frank Sorenson <sorenson@redhat.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
The smb2/smb3 message checking code was logging to dmesg when mounting
with encryption ("seal") for compounded SMB3 requests. When encrypted
the whole frame (including potentially multiple compounds) is read
so the length field is longer than in the case of non-encrypted
case (where length field will match the the calculated length for
the particular SMB3 request in the compound being validated).
Avoids the warning on mount (with "seal"):
"srv rsp padded more than expected. Length 384 not ..."
Signed-off-by: Steve French <stfrench@microsoft.com>
In the oplock break handler, writing pending changes from pages puts
the FileInfo handle. If the refcount reaches zero it closes the handle
and waits for any oplock break handler to return, thus causing a deadlock.
To prevent this situation:
* We add a wait flag to cifsFileInfo_put() to decide whether we should
wait for running/pending oplock break handlers
* We keep an additionnal reference of the SMB FileInfo handle so that
for the rest of the handler putting the handle won't close it.
- The ref is bumped everytime we queue the handler via the
cifs_queue_oplock_break() helper.
- The ref is decremented at the end of the handler
This bug was triggered by xfstest 464.
Also important fix to address the various reports of
oops in smb2_push_mandatory_locks
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
CC: Stable <stable@vger.kernel.org>
Currently on lease break the client sets a caching level twice:
when oplock is detected and when oplock is processed. While the
1st attempt sets the level to the value provided by the server,
the 2nd one resets the level to None unconditionally.
This happens because the oplock/lease processing code was changed
to avoid races between page cache flushes and oplock breaks.
The commit c11f1df500 ("cifs: Wait for writebacks to complete
before attempting write.") fixed the races for oplocks but didn't
apply the same changes for leases resulting in overwriting the
server granted value to None. Fix this by properly processing
lease breaks.
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
CC: Stable <stable@vger.kernel.org>
We should add any credits granted to us from unmatched server responses.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Otherwise we gradually leak credits leading to potential
hung session.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
CC: Stable <stable@vger.kernel.org>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
I ran into some cases where server was returning the wrong length
on frames but I couldn't easily match them to the command in the
network trace (or server logs) since I need the command and/or
multiplex id to find the offending SMB2/SMB3 command. Add these
two fields to the log message. In the case of padding too much
it may not be a problem in all cases but might have correlated
to a network disconnect case in some problems we have been
looking at. In the case of frame too short is even more important.
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Some SMB2/3 servers, Win2016 but possibly others too, adds padding
not only between PDUs in a compound but also to the final PDU.
This padding extends the PDU to a multiple of 8 bytes.
Check if the unexpected length looks like this might be the case
and avoid triggering the log messages for :
"SMB2 server sent bad RFC1001 len %d not %d\n"
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
We really, really want to be encouraging use of secure dialects,
and SMB3.1.1 offers useful security features, and will soon
be the recommended dialect for many use cases. Simplify the code
by removing the CONFIG_CIFS_SMB311 ifdef so users don't disable
it in the build, and create compatibility and/or security issues
with modern servers - many of which have been supporting this
dialect for multiple years.
Also clarify some of the Kconfig text for cifs.ko about
SMB3.1.1 and current supported features in the module.
Signed-off-by: Steve French <stfrench@microsoft.com>
Acked-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Use a read lease for the cached root fid so that we can detect
when the content of the directory changes (via a break) at which time
we close the handle. On next access to the root the handle will be reopened
and cached again.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
cifs->master_tlink is NULL against Win Server 2016 (which is
strange.. not sure why) and is dereferenced in cifs_sb_master_tcon().
move master_tlink getter to cifsglob.h so it can be used from
smb2misc.c
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Acked-by: Pavel Shilovsky <pshilov@microsoft.com>
Since header_preamble_size is 0 for SMB2+ we can remove it in those
code paths that are only invoked from SMB2.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
struct smb2_hdr is now just a wrapper for smb2_sync_hdr.
We can thus get rid of smb2_hdr completely and access the sync header directly.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
smb2_hdr is just a wrapper around smb2_sync_hdr at this stage
and smb2_hdr is going away.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
The two structures smb2_oplock_breaq_req/rsp are now basically identical.
Replace this with a single definition of a smb2_oplock_break structure.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
and change the smb2 version to take heder_preamble_size into account
instead of hardcoding it as 4 bytes.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <smfrench@gmail.com>
More cleanup of use of hardcoded 4 byte RFC1001 field size
Signed-off-by: Steve French <smfrench@gmail.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
The length checking for SMB3.11 negotiate request includes
"negotiate contexts" which caused a buffer validation problem
and a confusing warning message on SMB3.11 mount e.g.:
SMB2 server sent bad RFC1001 len 236 not 170
Fix the length checking for SMB3.11 negotiate to account for
the new negotiate context so that we don't log a warning on
SMB3.11 mount by default but do log warnings if lengths returned
by the server are incorrect.
CC: Stable <stable@vger.kernel.org>
Signed-off-by: Steve French <smfrench@gmail.com>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
This variable is set to 4 for all protocol versions and replaces
the hardcoded constant 4 throughought the code.
This will later be updated to reflect whether a response packet
has a 4 byte length preamble or not once we start removing this
field from the SMB2+ dialects.
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
SMB3.11 clients must implement pre-authentification integrity.
* new mechanism to certify requests/responses happening before Tree
Connect.
* supersedes VALIDATE_NEGOTIATE
* fixes signing for SMB3.11
Signed-off-by: Aurelien Aptel <aaptel@suse.com>
Signed-off-by: Steve French <smfrench@gmail.com>
CC: Stable <stable@vger.kernel.org>
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com>
Signed-off-by: Steve French <smfrench@gmail.com>
Acked-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
When the final cifsFileInfo_put() is called from cifsiod and an oplock
break work is queued, lockdep complains loudly:
=============================================
[ INFO: possible recursive locking detected ]
4.11.0+ #21 Not tainted
---------------------------------------------
kworker/0:2/78 is trying to acquire lock:
("cifsiod"){++++.+}, at: flush_work+0x215/0x350
but task is already holding lock:
("cifsiod"){++++.+}, at: process_one_work+0x255/0x8e0
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock("cifsiod");
lock("cifsiod");
*** DEADLOCK ***
May be due to missing lock nesting notation
2 locks held by kworker/0:2/78:
#0: ("cifsiod"){++++.+}, at: process_one_work+0x255/0x8e0
#1: ((&wdata->work)){+.+...}, at: process_one_work+0x255/0x8e0
stack backtrace:
CPU: 0 PID: 78 Comm: kworker/0:2 Not tainted 4.11.0+ #21
Workqueue: cifsiod cifs_writev_complete
Call Trace:
dump_stack+0x85/0xc2
__lock_acquire+0x17dd/0x2260
? match_held_lock+0x20/0x2b0
? trace_hardirqs_off_caller+0x86/0x130
? mark_lock+0xa6/0x920
lock_acquire+0xcc/0x260
? lock_acquire+0xcc/0x260
? flush_work+0x215/0x350
flush_work+0x236/0x350
? flush_work+0x215/0x350
? destroy_worker+0x170/0x170
__cancel_work_timer+0x17d/0x210
? ___preempt_schedule+0x16/0x18
cancel_work_sync+0x10/0x20
cifsFileInfo_put+0x338/0x7f0
cifs_writedata_release+0x2a/0x40
? cifs_writedata_release+0x2a/0x40
cifs_writev_complete+0x29d/0x850
? preempt_count_sub+0x18/0xd0
process_one_work+0x304/0x8e0
worker_thread+0x9b/0x6a0
kthread+0x1b2/0x200
? process_one_work+0x8e0/0x8e0
? kthread_create_on_node+0x40/0x40
ret_from_fork+0x31/0x40
This is a real warning. Since the oplock is queued on the same
workqueue this can deadlock if there is only one worker thread active
for the workqueue (which will be the case during memory pressure when
the rescuer thread is handling it).
Furthermore, there is at least one other kind of hang possible due to
the oplock break handling if there is only worker. (This can be
reproduced without introducing memory pressure by having passing 1 for
the max_active parameter of cifsiod.) cifs_oplock_break() can wait
indefintely in the filemap_fdatawait() while the cifs_writev_complete()
work is blocked:
sysrq: SysRq : Show Blocked State
task PC stack pid father
kworker/0:1 D 0 16 2 0x00000000
Workqueue: cifsiod cifs_oplock_break
Call Trace:
__schedule+0x562/0xf40
? mark_held_locks+0x4a/0xb0
schedule+0x57/0xe0
io_schedule+0x21/0x50
wait_on_page_bit+0x143/0x190
? add_to_page_cache_lru+0x150/0x150
__filemap_fdatawait_range+0x134/0x190
? do_writepages+0x51/0x70
filemap_fdatawait_range+0x14/0x30
filemap_fdatawait+0x3b/0x40
cifs_oplock_break+0x651/0x710
? preempt_count_sub+0x18/0xd0
process_one_work+0x304/0x8e0
worker_thread+0x9b/0x6a0
kthread+0x1b2/0x200
? process_one_work+0x8e0/0x8e0
? kthread_create_on_node+0x40/0x40
ret_from_fork+0x31/0x40
dd D 0 683 171 0x00000000
Call Trace:
__schedule+0x562/0xf40
? mark_held_locks+0x29/0xb0
schedule+0x57/0xe0
io_schedule+0x21/0x50
wait_on_page_bit+0x143/0x190
? add_to_page_cache_lru+0x150/0x150
__filemap_fdatawait_range+0x134/0x190
? do_writepages+0x51/0x70
filemap_fdatawait_range+0x14/0x30
filemap_fdatawait+0x3b/0x40
filemap_write_and_wait+0x4e/0x70
cifs_flush+0x6a/0xb0
filp_close+0x52/0xa0
__close_fd+0xdc/0x150
SyS_close+0x33/0x60
entry_SYSCALL_64_fastpath+0x1f/0xbe
Showing all locks held in the system:
2 locks held by kworker/0:1/16:
#0: ("cifsiod"){.+.+.+}, at: process_one_work+0x255/0x8e0
#1: ((&cfile->oplock_break)){+.+.+.}, at: process_one_work+0x255/0x8e0
Showing busy workqueues and worker pools:
workqueue cifsiod: flags=0xc
pwq 0: cpus=0 node=0 flags=0x0 nice=0 active=1/1
in-flight: 16:cifs_oplock_break
delayed: cifs_writev_complete, cifs_echo_request
pool 0: cpus=0 node=0 flags=0x0 nice=0 hung=0s workers=3 idle: 750 3
Fix these problems by creating a a new workqueue (with a rescuer) for
the oplock break work.
Signed-off-by: Rabin Vincent <rabinv@axis.com>
Signed-off-by: Steve French <smfrench@gmail.com>
CC: Stable <stable@vger.kernel.org>
A signal can interrupt a SendReceive call which result in incoming
responses to the call being ignored. This is a problem for calls such as
open which results in the successful response being ignored. This
results in an open file resource on the server.
The patch looks into responses which were cancelled after being sent and
in case of successful open closes the open fids.
For this patch, the check is only done in SendReceive2()
RH-bz: 1403319
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Pavel Shilovsky <pshilov@microsoft.com>
Cc: Stable <stable@vger.kernel.org>
In order to support compounding and encryption we need to separate
RFC1001 length field and SMB2 header structure because the protocol
treats them differently. This change will allow to simplify parsing
of such complex SMB2 packets further.
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Remove the global file_list_lock to simplify cifs/smb3 locking and
have spinlocks that more closely match the information they are
protecting.
Add new tcon->open_file_lock and file->file_info_lock spinlocks.
Locks continue to follow a heirachy,
cifs_socket --> cifs_ses --> cifs_tcon --> cifs_file
where global tcp_ses_lock still protects socket and cifs_ses, while the
the newer locks protect the lower level structure's information
(tcon and cifs_file respectively).
CC: Stable <stable@vger.kernel.org>
Signed-off-by: Steve French <steve.french@primarydata.com>
Signed-off-by: Pavel Shilovsky <pshilov@microsoft.com>
Reviewed-by: Aurelien Aptel <aaptel@suse.com>
Reviewed-by: Germano Percossi <germano.percossi@citrix.com>
that's the bulk of filesystem drivers dealing with inodes of their own
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Coverity reports a warning for referencing the beginning of the
SMB2/SMB3 frame using the ProtocolId field as an array. Although
it works the same either way, this patch should quiet the warning
and might be a little clearer.
Reported by Coverity (CID 741269)
Signed-off-by: Steve French <smfrench@gmail.com>
Acked-by: Shirish Pargaonkar <shirishpargaonkar@gmail.com>
Acked-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Jeff Layton <jlayton@poochiereds.net>
We have encountered failures when When testing smb2 mounts on ppc64
machines when using both Samba as well as Windows 2012.
On poking around, the problem was determined to be caused by the
high endian MessageID passed in the header for smb2. On checking the
corresponding MID for smb1 is converted to LE before being sent on the
wire.
We have tested this patch successfully on a ppc64 machine.
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Replace all __constant_foo to foo() except in smb2status.h (1700 lines to
update).
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Cc: Steve French <sfrench@samba.org>
Cc: Jeff Layton <jlayton@poochiereds.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The previous patch allowed remapping reserved characters from directory
listenings, this patch adds conversion the other direction, allowing
opening of files with any of the seven reserved characters.
Signed-off-by: Steve French <smfrench@gmail.com>
Reviewed-by: Shirish Pargaonkar <shirishpargaonkar@gmail.com>
response
Writes fail to Mac servers with SMB2.1 mounts (works with cifs though) due
to them sending an incorrect RFC1001 length for the SMB2.1 Write response.
Workaround this problem. MacOS server sends a write response with 3 bytes
of pad beyond the end of the SMB itself. The RFC1001 length is 3 bytes
more than the sum of the SMB2.1 header length + the write reponse.
Incorporate feedback from Jeff and JRA to allow servers to send
a tcp frame that is even more than three bytes too long
(ie much longer than the SMB2/SMB3 request that it contains) but
we do log it once now. In the earlier version of the patch I had
limited how far off the length field could be before we fail the request.
Signed-off-by: Steve French <smfrench@gmail.com>
Joe Perches and Hans Wennborg noticed that various places in the
kernel were printing decimal numbers with 0x prefix.
printk("0x%d") or equivalent
This fixes the instances of this in the cifs driver.
CC: Hans Wennborg <hans@hanshq.net>
CC: Joe Perches <joe@perches.com>
Signed-off-by: Steve French <smfrench@gmail.com>
Problem reported in Red Hat bz 1040329 for strict writes where we cache
only when we hold oplock and write direct to the server when we don't.
When we receive an oplock break, we first change the oplock value for
the inode in cifsInodeInfo->oplock to indicate that we no longer hold
the oplock before we enqueue a task to flush changes to the backing
device. Once we have completed flushing the changes, we return the
oplock to the server.
There are 2 ways here where we can have data corruption
1) While we flush changes to the backing device as part of the oplock
break, we can have processes write to the file. These writes check for
the oplock, find none and attempt to write directly to the server.
These direct writes made while we are flushing from cache could be
overwritten by data being flushed from the cache causing data
corruption.
2) While a thread runs in cifs_strict_writev, the machine could receive
and process an oplock break after the thread has checked the oplock and
found that it allows us to cache and before we have made changes to the
cache. In that case, we end up with a dirty page in cache when we
shouldn't have any. This will be flushed later and will overwrite all
subsequent writes to the part of the file represented by this page.
Before making any writes to the server, we need to confirm that we are
not in the process of flushing data to the server and if we are, we
should wait until the process is complete before we attempt the write.
We should also wait for existing writes to complete before we process
an oplock break request which changes oplock values.
We add a version specific downgrade_oplock() operation to allow for
differences in the oplock values set for the different smb versions.
Cc: stable@vger.kernel.org
Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Reviewed-by: Pavel Shilovsky <piastry@etersoft.ru>
Signed-off-by: Steve French <smfrench@gmail.com>
that force a client to purge cache pages when a server requests it.
Signed-off-by: Pavel Shilovsky <pshilovsky@samba.org>
Signed-off-by: Steve French <smfrench@gmail.com>
that prepare the code to handle different types of SMB2 leases.
Signed-off-by: Pavel Shilovsky <pshilovsky@samba.org>
Signed-off-by: Steve French <smfrench@gmail.com>
If a server sends a lease break to a connection that doesn't have
opens with a lease key specified in the server response, we can't
find an open file to send an ack. Fix this by walking through
all connections we have.
Cc: <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilovsky@samba.org>
Signed-off-by: Steve French <smfrench@gmail.com>
This happens when we receive a lease break from a server, then
find an appropriate lease key in opened files and schedule the
oplock_break slow work. lw pointer isn't freed in this case.
Cc: <stable@vger.kernel.org>
Signed-off-by: Pavel Shilovsky <pshilovsky@samba.org>
Signed-off-by: Steve French <smfrench@gmail.com>
