When a sequence times out we have no idea what happened to the
frame. And we do not know if we will ever receive the frame.
Hence we cannot re-use the xid as we would risk data corruption
if the xid had been re-used and the timed out frame would be
received after that.
So we need to quarantine the xid until the lport is reset.
Yes, I know this will (eventually) deplete the xid pool.
But for now it's the safest method.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

The cached exchange index might be invalid, in which case
we should drop down to allocate a new one.
And we should not try to access an invalid exchange when
responding to a BA_ABTS.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When the queue depth is reduced we should print out the reason
for this; it might be due to a queue full condition.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

Occasionally it might happen that we receive a PRLI while we're still
waiting for our PLOGI response. In that case we should return
'busy' LS status instead of 'plogi required' LS status.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

PRLI is only required if the port is acting as an initiator; ports
which support target functionality only do not need to send PRLI.
At the same time the PRLI state is only used if the port initiated
a PRLI transfer; if we received a PRLI request we should _not_
change the state as this would cause our PRLI response to be dropped.
And when we receive a PRLI response we need to check if an image
pair has been established; if not the remote port cannot act as a
target for us and we need to disable target functionality.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

The libfc stack generates an RTV request, so we should be implementing
an RTV responder, too.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

We only ever use the 'fp' argument for fc_rport_error() to
encapsulate the error code, so we can as well do away with that
and pass the error directly.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When a command is aborted it might already have the DID_TIME_OUT
status set, so we shouldn't be overwriting that.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When setting the FCP timeout we need to ensure a lower boundary
for E_D_TOV and R_A_TOV, otherwise we'd be getting spurious I/O
issues due to the fcp timer firing too early.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

If fc_rport_error_retry() is attempting to retry the remote
port state we should be waiting for the configured e_d_tov
value rather than the default.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

We should be using the configured R_A_TOV value when sending the
exchange.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

If a command times out libfc is sending an REC, which also
might fail (due to frames being lost or something).
If no data has been transferred we can simply retry
the command, but the current code sets a state of FC_ERROR,
which then is being translated into DID_ERROR, resulting
in an I/O error.
So to handle this properly we need to set a separate
state FC_TRANS_RESET and mapping it onto DID_SOFT_RETRY.
Signed-off-by: NHannes Reinecke <hare@suse.de>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

This reverts commit 3e22760d.

This revert came about because of efforts by Ewan Milne, Curtis Taylor
and I.  In researching this issue, significant performance issues were
seen on large CPU count systems using the software FCOE stack.  Hannes
also weighed in.

The same was not apparent on much smaller low count CPU systems.  The
behavior introduced by commit 3e22760d
lands sup with large count CPU systems seeing continual
blk_requeue_request() calls due to ML_QUEUE_HOST_BUSY.

fc_exch_alloc() used to try all the available exchange managers in the
list for an available exchange id, but this was changed in 2010 so that
if the first matched exchange manager couldn't allocate one, it fails
and we end up returning host busy.  This was due to commit:

Setting the ddp_min module parameter to fcoe to 128MB prevents the
->match function from permitting the use of the offload exchange manager
for the frame, and we no longer see the problem with host busy status,
since it uses the larger non-offloaded pool.

Reverting commit 3e22760d was tested to
also prevent the host busy issue due to failing allocations.
Suggested-by: NEwan Milne <emilne@redhat.com>
Suggested-by: NCurtis Taylor <cjt@us.ibm.com>
Tested-by: NLaurence Oberman <loberman@redhat.com>
Signed-off-by: NLaurence Oberman <loberman@redhat.com>

When we receive an FLOGI but have already sent our own we should
not advance the state machine but rather wait for our FLOGI to
return before continuing with PLOGI.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When the port is already started we don't need to login; that
will only confuse the state machine.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When fc_rport_login() is called while the rport is not
in RPORT_ST_INIT, RPORT_ST_READY, or RPORT_ST_DELETE
login is already in progress and there's no need to
drop down to FLOGI; doing so will only confuse the
other side.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When an ELS response handler receives a -FC_EX_CLOSED, the rdata->rp_mutex is
already held which can lead to a deadlock condition like the following stack trace:

[<ffffffffa04d8f18>] fc_rport_plogi_resp+0x28/0x200 [libfc]
[<ffffffffa04cfa1a>] fc_invoke_resp+0x6a/0xe0 [libfc]
[<ffffffffa04d0c08>] fc_exch_mgr_reset+0x1b8/0x280 [libfc]
[<ffffffffa04d87b3>] fc_rport_logoff+0x43/0xd0 [libfc]
[<ffffffffa04ce73d>] fc_disc_stop+0x6d/0xf0 [libfc]
[<ffffffffa04ce7ce>] fc_disc_stop_final+0xe/0x20 [libfc]
[<ffffffffa04d55f7>] fc_fabric_logoff+0x17/0x70 [libfc]

The other ELS handlers need to follow the FLOGI response handler and simply do
a kref_put against the fc_rport_priv struct and exit when receving a
-FC_EX_CLOSED response.
Signed-off-by: NChad Dupuis <chad.dupuis@cavium.com>
Reviewed-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

The list of attached 'rdata' remote port structures is RCU
protected, so there is no need to take the 'disc_mutex' when
traversing it.
Rather we should be using rcu_read_lock() and kref_get_unless_zero()
to validate the entries.
We need, however, take the disc_mutex when deleting an entry;
otherwise we risk clashes with list_add.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

The kref handling in fc_rport is a mess. This patch updates
the kref handling according to the following rules:

- Take a reference whenever scheduling a workqueue
- Take a reference whenever an ELS command is send
- Drop the reference at the end of the workqueue function
- Drop the reference at the end of handling ELS replies
- Take a reference when allocating an rport
- Drop the reference when removing an rport
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When all exchanges are reset the upper layers have already logged out of
the remote port, so the exchanges can be reset without sending any ABTS.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NChad Dupuis <chad.dupuis@qlogic.com>
Tested-by: NChad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

FC-LS mandates that we should invalidate all sequences before sending a
LOGO. And we should set the event to RPORT_EV_STOP when a LOGO request
has been received to signal that all exchanges are terminated.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NChad Dupuis <chad.dupuis@qlogic.com>
Tested-by: NChad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When running in point-to-multipoint mode PLOGI is done after FLOGI
completed. So when the PLOGI fails we should be sending a LOGO to the
remote port.

[mkp: Applied by hand]
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NChad Dupuis <chad.dupuis@qlogic.com>
Tested-by: NChad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

When receiving a PRLO it just means that the operating parameters have
changed, it does _not_ mean that the port doesn't want to communicate
with us.  So instead of implicitly logging out we should be issueing a
PRLI to figure out the new operating parameters.  We can always recover
once PRLI fails.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Reviewed-by: NChad Dupuis <chad.dupuis@qlogic.com>
Tested-by: NChad Dupuis <chad.dupuis@qlogic.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

In the receive path libfc extracts a cpu number from the ox_id in the
fiber channel header and uses that to do a per_cpu_ptr conversion.  If,
for some reason, a frame is received with an invalid ox_id, per_cpu_ptr
will return an invalid pointer and the libfc receive path will panic the
system trying to use it.

I'm currently looking at such a case, and I don't yet know why a cpu
number > nr_cpu_ids is appearing in an exchange id.  But adding a sanity
check in libfc prevents a system panic, and seems like good idea when
dealing with frames coming in from the network.
Signed-off-by: NChris Leech <cleech@redhat.com>
Acked-by: NJohannes Thumshirn <jth@kernel.org>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

The fc_get_host_stats() function contains a complex conversion from
jiffies to timespec to seconds. As we try to get rid of uses of struct
timespec, we can clean this up and replace it with a simpler
computation.

Simply dividing the difference in jiffies by HZ is not only much more
efficient, it also avoids a problem that causes the
seconds_since_last_reset value to be incorrect if jiffies has overrun
since the 'boot_time' value was recorded.
Signed-off-by: NArnd Bergmann <arnd@arndb.de>
Reviewed-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

Originally libfc would just be initializing the refcount to '1', and
using the disc_mutex to synchronize if and when the final put should be
happening.  This has a race condition as the mutex might be delayed,
causing other threads to access an invalid structure.  This patch
updates the rport reference counting to increase the reference every
time 'rport_lookup' is called, and decreases the reference
correspondingly.  This removes the need to hold 'disc_mutex' when
removing the structure, and avoids the above race condition.
Signed-off-by: NHannes Reinecke <hare@suse.com>
Acked-by: NVasu Dev <vasu.dev@intel.com>
Reviewed-by: NJohannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>

Signed-off-by: NSebastian Herbszt <herbszt@gmx.de>
Acked-by: NVasu Dev <vasu.dev@intel.com>
Reviewed-by: NHannes Reinecke <hare@suse.de>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Since fc_fcp_cleanup_cmd() can sleep this function must not
be called while holding a spinlock. This patch avoids that
fc_fcp_cleanup_each_cmd() triggers the following bug:

BUG: scheduling while atomic: sg_reset/1512/0x00000202
1 lock held by sg_reset/1512:
 #0:  (&(&fsp->scsi_pkt_lock)->rlock){+.-...}, at: [<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Preemption disabled at:[<ffffffffc0225cd5>] fc_fcp_cleanup_each_cmd.isra.21+0xa5/0x150 [libfc]
Call Trace:
 [<ffffffff816c612c>] dump_stack+0x4f/0x7b
 [<ffffffff810828bc>] __schedule_bug+0x6c/0xd0
 [<ffffffff816c87aa>] __schedule+0x71a/0xa10
 [<ffffffff816c8ad2>] schedule+0x32/0x80
 [<ffffffffc0217eac>] fc_seq_set_resp+0xac/0x100 [libfc]
 [<ffffffffc0218b11>] fc_exch_done+0x41/0x60 [libfc]
 [<ffffffffc0225cff>] fc_fcp_cleanup_each_cmd.isra.21+0xcf/0x150 [libfc]
 [<ffffffffc0225f43>] fc_eh_device_reset+0x1c3/0x270 [libfc]
 [<ffffffff814a2cc9>] scsi_try_bus_device_reset+0x29/0x60
 [<ffffffff814a3908>] scsi_ioctl_reset+0x258/0x2d0
 [<ffffffff814a2650>] scsi_ioctl+0x150/0x440
 [<ffffffff814b3a9d>] sd_ioctl+0xad/0x120
 [<ffffffff8132f266>] blkdev_ioctl+0x1b6/0x810
 [<ffffffff811da608>] block_ioctl+0x38/0x40
 [<ffffffff811b4e08>] do_vfs_ioctl+0x2f8/0x530
 [<ffffffff811b50c1>] SyS_ioctl+0x81/0xa0
 [<ffffffff816cf8b2>] system_call_fastpath+0x16/0x7a
Signed-off-by: NBart Van Assche <bart.vanassche@sandisk.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Due to patch "libfc: Do not invoke the response handler after
fc_exch_done()" (commit ID 7030fd62) the lport_recv() call
in fc_exch_recv_req() is passed a dangling pointer. Avoid this
by moving the fc_frame_free() call from fc_invoke_resp() to its
callers. This patch fixes the following crash:

general protection fault: 0000 [#3] PREEMPT SMP
RIP: fc_lport_recv_req+0x72/0x280 [libfc]
Call Trace:
 fc_exch_recv+0x642/0xde0 [libfc]
 fcoe_percpu_receive_thread+0x46a/0x5ed [fcoe]
 kthread+0x10a/0x120
 ret_from_fork+0x42/0x70
Signed-off-by: NBart Van Assche <bart.vanassche@sandisk.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Signed-off-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NJames Bottomley <JBottomley@Odin.com>

Drop the now unused reason argument from the ->change_queue_depth method.
Also add a return value to scsi_adjust_queue_depth, and rename it to
scsi_change_queue_depth now that it can be used as the default
->change_queue_depth implementation.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NHannes Reinecke <hare@suse.de>

All drivers use the implementation for ramping the queue up and down, so
instead of overloading the change_queue_depth method call the
implementation diretly if the driver opts into it by setting the
track_queue_depth flag in the host template.

Note that a few drivers validated the new queue depth in their
change_queue_depth method, but as we never go over the queue depth
set during slave_configure or the sysfs file this isn't nessecary
and can safely be removed.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NHannes Reinecke <hare@suse.de>
Reviewed-by: NVenkatesh Srinivas <venkateshs@google.com>

Remove the tagged argument from scsi_adjust_queue_depth, and just let it
handle the queue depth.  For most drivers those two are fairly separate,
given that most modern drivers don't care about the SCSI "tagged" status
of a command at all, and many old drivers allow queuing of multiple
untagged commands in the driver.

Instead we start out with the ->simple_tags flag set before calling
->slave_configure, which is how all drivers actually looking at
->simple_tags except for one worke anyway.  The one other case looks
broken, but I've kept the behavior as-is for now.

Except for that we only change ->simple_tags from the ->change_queue_type,
and when rejecting a tag message in a single driver, so keeping this
churn out of scsi_adjust_queue_depth is a clear win.

Now that the usage of scsi_adjust_queue_depth is more obvious we can
also remove all the trivial instances in ->slave_alloc or ->slave_configure
that just set it to the cmd_per_lun default.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NHannes Reinecke <hare@suse.de>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>

Allow a driver to ask for block layer tags by setting .use_blk_tags in the
host template, in which case it will always see a valid value in
request->tag, similar to the behavior when using blk-mq.  This means even
SCSI "untagged" commands will now have a tag, which is especially useful
when using a host-wide tag map.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: NHannes Reinecke <hare@suse.de>

Most drivers use exactly the same implementation, so provide it as a
library function.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NBart Van Assche <bvanassche@acm.org>
Reviewed-by: NMike Christie <michaelc@cs.wisc.edu>
Reviewed-by: NMartin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: NHannes Reinecke <hare@suse.de>

The uses of "rcu_assign_pointer()" are NULLing out the pointers.
According to RCU_INIT_POINTER()'s block comment:
"1.   This use of RCU_INIT_POINTER() is NULLing out the pointer"
it is better to use it instead of rcu_assign_pointer() because it has a
smaller overhead.

The following Coccinelle semantic patch was used:
@@
@@

- rcu_assign_pointer
+ RCU_INIT_POINTER
  (..., NULL)
Signed-off-by: NAndreea-Cristina Bernat <bernat.ada@gmail.com>
Acked-by: NVasu Dev <vasu.dev@intel.com>
Signed-off-by: NChristoph Hellwig <hch@lst.de>

This patch avoids that the FCoE initiator sends a REC message after
having received a SCSI response with non-zero status and non-zero
DATA IN buffer length.
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Cc: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>

While the FCoE initiator driver invokes fc_exch_done() from inside
the libfc response handler, FCoE target drivers typically invoke
fc_exch_done() from outside the libfc response handler. The object
fc_exch.arg points at may disappear as soon as fc_exch_done() has
finished. So it's important not to invoke the response handler
function after fc_exch_done() has finished. Modify libfc such that
this guarantee is provided if fc_exch_done() is invoked from
outside a response handler. This patch fixes a sporadic crash in
FCoE target implementations after a command has been aborted.
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Cc: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>

Reduce the time during which the exchange lock is held by allocating
a frame before obtaining the exchange lock.
Signed-off-by: NBart Van Assche <bvanassche@acm.org>
Cc: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: NRobert Love <robert.w.love@intel.com>