linux-2.6-block.git
8 years agoMerge branch 'for-4.12/block' into for-next
Jens Axboe [Sat, 8 Apr 2017 17:25:48 +0000 (11:25 -0600)]
Merge branch 'for-4.12/block' into for-next

8 years agoscsi: sd: Remove LBPRZ dependency for discards
Martin K. Petersen [Wed, 5 Apr 2017 17:21:25 +0000 (19:21 +0200)]
scsi: sd: Remove LBPRZ dependency for discards

Separating discards and zeroout operations allows us to remove the LBPRZ
block zeroing constraints from discards and honor the device preferences
for UNMAP commands.

If supported by the device, we'll also choose UNMAP over one of the
WRITE SAME variants for discards.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoscsi: sd: Separate zeroout and discard command choices
Martin K. Petersen [Wed, 5 Apr 2017 17:21:24 +0000 (19:21 +0200)]
scsi: sd: Separate zeroout and discard command choices

Now that zeroout and discards are distinct operations we need to
separate the policy of choosing the appropriate command. Create a
zeroing_mode which can be one of:

write: Zeroout assist not present, use regular WRITE
writesame: Allow WRITE SAME(10/16) with a zeroed payload
writesame_16_unmap: Allow WRITE SAME(16) with UNMAP
writesame_10_unmap: Allow WRITE SAME(10) with UNMAP

The last two are conditional on the device being thin provisioned with
LBPRZ=1 and LBPWS=1 or LBPWS10=1 respectively.

Whether to set the UNMAP bit or not depends on the REQ_NOUNMAP flag. And
if none of the _unmap variants are supported, regular WRITE SAME will be
used if the device supports it.

The zeroout_mode is exported in sysfs and the detected mode for a given
device can be overridden using the string constants above.

With this change in place we can now issue WRITE SAME(16) with UNMAP set
for block zeroing applications that require hard guarantees and
logical_block_size granularity. And at the same time use the UNMAP
command with the device's preferred granulary and alignment for discard
operations.

Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: remove the discard_zeroes_data flag
Christoph Hellwig [Wed, 5 Apr 2017 17:21:23 +0000 (19:21 +0200)]
block: remove the discard_zeroes_data flag

Now that we use the proper REQ_OP_WRITE_ZEROES operation everywhere we can
kill this hack.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodrbd: implement REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:22 +0000 (19:21 +0200)]
drbd: implement REQ_OP_WRITE_ZEROES

It seems like DRBD assumes its on the wire TRIM request always zeroes data.
Use that fact to implement REQ_OP_WRITE_ZEROES.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodrbd: make intelligent use of blkdev_issue_zeroout
Christoph Hellwig [Wed, 5 Apr 2017 17:21:21 +0000 (19:21 +0200)]
drbd: make intelligent use of blkdev_issue_zeroout

drbd always wants its discard wire operations to zero the blocks, so
use blkdev_issue_zeroout with the BLKDEV_ZERO_UNMAP flag instead of
reinventing it poorly.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: stop using discards for zeroing
Christoph Hellwig [Wed, 5 Apr 2017 17:21:20 +0000 (19:21 +0200)]
block: stop using discards for zeroing

Now that we have REQ_OP_WRITE_ZEROES implemented for all devices that
support efficient zeroing, we can remove the call to blkdev_issue_discard.
This means we only have two ways of zeroing left and can simplify the
code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agommc: remove the discard_zeroes_data flag
Christoph Hellwig [Wed, 5 Apr 2017 17:21:19 +0000 (19:21 +0200)]
mmc: remove the discard_zeroes_data flag

mmc only supports discarding on large alignments, so the zeroing code
would always fall back to explicit writings of zeroes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agorsxx: remove the discard_zeroes_data flag
Christoph Hellwig [Wed, 5 Apr 2017 17:21:18 +0000 (19:21 +0200)]
rsxx: remove the discard_zeroes_data flag

rsxx only supports discarding on large alignments, so the zeroing code
would always fall back to explicit writings of zeroes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agorbd: remove the discard_zeroes_data flag
Christoph Hellwig [Wed, 5 Apr 2017 17:21:17 +0000 (19:21 +0200)]
rbd: remove the discard_zeroes_data flag

rbd only supports discarding on large alignments, so the zeroing code
would always fall back to explicit writings of zeroes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agobrd: remove discard support
Christoph Hellwig [Wed, 5 Apr 2017 17:21:16 +0000 (19:21 +0200)]
brd: remove discard support

It's just a in-driver reimplementation of writing zeroes to the pages,
which fails if the discards aren't page aligned.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoloop: implement REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:15 +0000 (19:21 +0200)]
loop: implement REQ_OP_WRITE_ZEROES

It's identical to discard as hole punches will always leave us with
zeroes on reads.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agozram: implement REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:14 +0000 (19:21 +0200)]
zram: implement REQ_OP_WRITE_ZEROES

Just the same as discard if the block size equals the system page size.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme: implement REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:13 +0000 (19:21 +0200)]
nvme: implement REQ_OP_WRITE_ZEROES

But now for the real NVMe Write Zeroes yet, just to get rid of the
discard abuse for zeroing.  Also rename the quirk flag to be a bit
more self-explanatory.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agosd: implement unmapping Write Zeroes
Christoph Hellwig [Wed, 5 Apr 2017 17:21:12 +0000 (19:21 +0200)]
sd: implement unmapping Write Zeroes

Try to use a write same with unmap bit variant if the device supports it
and the caller allows for it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock_dev: use blkdev_issue_zerout for hole punches
Christoph Hellwig [Wed, 5 Apr 2017 17:21:11 +0000 (19:21 +0200)]
block_dev: use blkdev_issue_zerout for hole punches

This gets us support for non-discard efficient write of zeroes (e.g. NVMe)
and prepares for removing the discard_zeroes_data flag.

Also remove a pointless discard support check, which is done in
blkdev_issue_discard already.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: add a new BLKDEV_ZERO_NOFALLBACK flag
Christoph Hellwig [Wed, 5 Apr 2017 17:21:10 +0000 (19:21 +0200)]
block: add a new BLKDEV_ZERO_NOFALLBACK flag

This avoids fallbacks to explicit zeroing in (__)blkdev_issue_zeroout if
the caller doesn't want them.

Also clean up the convoluted check for the return condition that this
new flag is added to.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: add a REQ_NOUNMAP flag for REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:09 +0000 (19:21 +0200)]
block: add a REQ_NOUNMAP flag for REQ_OP_WRITE_ZEROES

If this flag is set logical provisioning capable device should
release space for the zeroed blocks if possible, if it is not set
devices should keep the blocks anchored.

Also remove an out of sync kerneldoc comment for a static function
that would have become even more out of data with this change.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: add a flags argument to (__)blkdev_issue_zeroout
Christoph Hellwig [Wed, 5 Apr 2017 17:21:08 +0000 (19:21 +0200)]
block: add a flags argument to (__)blkdev_issue_zeroout

Turn the existing discard flag into a new BLKDEV_ZERO_UNMAP flag with
similar semantics, but without referring to diŃ•card.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: stop using blkdev_issue_write_same for zeroing
Christoph Hellwig [Wed, 5 Apr 2017 17:21:07 +0000 (19:21 +0200)]
block: stop using blkdev_issue_write_same for zeroing

We'll always use the WRITE ZEROES code for zeroing now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodm kcopyd: switch to use REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:06 +0000 (19:21 +0200)]
dm kcopyd: switch to use REQ_OP_WRITE_ZEROES

It seems like the code currently passes whatever it was using for writes
to WRITE SAME.  Just switch it to WRITE ZEROES, although that doesn't
need any payload.

Untested, and confused by the code, maybe someone who understands it
better than me can help..

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodm: support REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:05 +0000 (19:21 +0200)]
dm: support REQ_OP_WRITE_ZEROES

Copy & paste from the REQ_OP_WRITE_SAME code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodm io: discards don't take a payload
Christoph Hellwig [Wed, 5 Apr 2017 17:21:04 +0000 (19:21 +0200)]
dm io: discards don't take a payload

Fix up do_region to not allocate a bio_vec for discards.  We've
got rid of the discard payload allocated by the caller years ago.

Obviously this wasn't actually harmful given how long it's been
there, but it's still good to avoid the pointless allocation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agomd: support REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:03 +0000 (19:21 +0200)]
md: support REQ_OP_WRITE_ZEROES

Copy & paste from the REQ_OP_WRITE_SAME code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agosd: implement REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:02 +0000 (19:21 +0200)]
sd: implement REQ_OP_WRITE_ZEROES

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: implement splitting of REQ_OP_WRITE_ZEROES bios
Christoph Hellwig [Wed, 5 Apr 2017 17:21:01 +0000 (19:21 +0200)]
block: implement splitting of REQ_OP_WRITE_ZEROES bios

Copy and past the REQ_OP_WRITE_SAME code to prepare to implementations
that limit the write zeroes size.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: renumber REQ_OP_WRITE_ZEROES
Christoph Hellwig [Wed, 5 Apr 2017 17:21:00 +0000 (19:21 +0200)]
block: renumber REQ_OP_WRITE_ZEROES

Make life easy for implementations that needs to send a data buffer
to the device (e.g. SCSI) by numbering it as a data out command.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agosd: split sd_setup_discard_cmnd
Christoph Hellwig [Wed, 5 Apr 2017 17:20:59 +0000 (19:20 +0200)]
sd: split sd_setup_discard_cmnd

Split sd_setup_discard_cmnd into one function per provisioning type.  While
this creates some very slight duplication of boilerplate code it keeps the
code modular for additions of new provisioning types, and for reusing the
write same functions for the upcoming scsi implementation of the Write Zeroes
operation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge branch 'for-4.12/block' into for-next
Jens Axboe [Fri, 7 Apr 2017 20:24:22 +0000 (14:24 -0600)]
Merge branch 'for-4.12/block' into for-next

8 years agoblock: sed-opal: Tone down all the pr_* to debugs
Scott Bauer [Fri, 7 Apr 2017 19:58:50 +0000 (13:58 -0600)]
block: sed-opal: Tone down all the pr_* to debugs

Lets not flood the kernel log with messages unless
the user requests so.

Signed-off-by: Scott Bauer <scott.bauer@intel.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge branch 'for-4.12/block' into for-next
Jens Axboe [Fri, 7 Apr 2017 18:46:11 +0000 (12:46 -0600)]
Merge branch 'for-4.12/block' into for-next

8 years agoMerge branch 'for-linus' into for-next
Jens Axboe [Fri, 7 Apr 2017 18:46:07 +0000 (12:46 -0600)]
Merge branch 'for-linus' into for-next

8 years agoblk-mq: Clarify comments in blk_mq_dispatch_rq_list()
Bart Van Assche [Fri, 7 Apr 2017 18:16:51 +0000 (11:16 -0700)]
blk-mq: Clarify comments in blk_mq_dispatch_rq_list()

The blk_mq_dispatch_rq_list() implementation got modified several
times but the comments in that function were not updated every
time. Since it is nontrivial what is going on, update the comments
in blk_mq_dispatch_rq_list().

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: Make it safe to use RCU to iterate over blk_mq_tag_set.tag_list
Bart Van Assche [Fri, 7 Apr 2017 18:16:49 +0000 (11:16 -0700)]
blk-mq: Make it safe to use RCU to iterate over blk_mq_tag_set.tag_list

Since the next patch in this series will use RCU to iterate over
tag_list, make this safe. Add lockdep_assert_held() statements
in functions that iterate over tag_list to make clear that using
list_for_each_entry() instead of list_for_each_entry_rcu() is
fine in these functions.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: use true instead of 1 for blk_mq_queue_data.last
Omar Sandoval [Wed, 5 Apr 2017 19:01:36 +0000 (12:01 -0700)]
blk-mq: use true instead of 1 for blk_mq_queue_data.last

Trivial cleanup.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: make driver tag failure path easier to follow
Omar Sandoval [Wed, 5 Apr 2017 19:01:35 +0000 (12:01 -0700)]
blk-mq: make driver tag failure path easier to follow

Minor cleanup that makes it easier to figure out what's going on in the
driver tag allocation failure path of blk_mq_dispatch_rq_list().

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq-sched: provide hooks for initializing hardware queue data
Omar Sandoval [Wed, 5 Apr 2017 19:01:34 +0000 (12:01 -0700)]
blk-mq-sched: provide hooks for initializing hardware queue data

Schedulers need to be informed when a hardware queue is added or removed
at runtime so they can allocate/free per-hardware queue data. So,
replace the blk_mq_sched_init_hctx_data() helper, which only makes sense
at init time, with .init_hctx() and .exit_hctx() hooks.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge branch 'for-linus' into for-4.12/block
Jens Axboe [Fri, 7 Apr 2017 18:45:20 +0000 (12:45 -0600)]
Merge branch 'for-linus' into for-4.12/block

We've added a considerable amount of fixes for stalls and issues
with the blk-mq scheduling in the 4.11 series since forking
off the for-4.12/block branch. We need to do improvements on
top of that for 4.12, so pull in the previous fixes to make
our lives easier going forward.

Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: Restart a single queue if tag sets are shared
Bart Van Assche [Fri, 7 Apr 2017 18:40:09 +0000 (12:40 -0600)]
blk-mq: Restart a single queue if tag sets are shared

To improve scalability, if hardware queues are shared, restart
a single hardware queue in round-robin fashion. Rename
blk_mq_sched_restart_queues() to reflect the new semantics.
Remove blk_mq_sched_mark_restart_queue() because this function
has no callers. Remove flag QUEUE_FLAG_RESTART because this
patch removes the code that uses this flag.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agodm rq: Avoid that request processing stalls sporadically
Bart Van Assche [Fri, 7 Apr 2017 18:16:54 +0000 (11:16 -0700)]
dm rq: Avoid that request processing stalls sporadically

While running the srp-test software I noticed that request
processing stalls sporadically at the beginning of a test, namely
when mkfs is run against a dm-mpath device. Every time when that
happened the following command was sufficient to resume request
processing:

    echo run >/sys/kernel/debug/block/dm-0/state

This patch avoids that such request processing stalls occur. The
test I ran is as follows:

    while srp-test/run_tests -d -r 30 -t 02-mq; do :; done

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: dm-devel@redhat.com
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoscsi: Avoid that SCSI queues get stuck
Bart Van Assche [Fri, 7 Apr 2017 18:16:53 +0000 (11:16 -0700)]
scsi: Avoid that SCSI queues get stuck

If a .queue_rq() function returns BLK_MQ_RQ_QUEUE_BUSY then the block
driver that implements that function is responsible for rerunning the
hardware queue once requests can be queued again successfully.

commit 52d7f1b5c2f3 ("blk-mq: Avoid that requeueing starts stopped
queues") removed the blk_mq_stop_hw_queue() call from scsi_queue_rq()
for the BLK_MQ_RQ_QUEUE_BUSY case. Hence change all calls to functions
that are intended to rerun a busy queue such that these examine all
hardware queues instead of only stopped queues.

Since no other functions than scsi_internal_device_block() and
scsi_internal_device_unblock() should ever stop or restart a SCSI
queue, change the blk_mq_delay_queue() call into a
blk_mq_delay_run_hw_queue() call.

Fixes: commit 52d7f1b5c2f3 ("blk-mq: Avoid that requeueing starts stopped queues")
Fixes: commit 7e79dadce222 ("blk-mq: stop hardware queue in blk_mq_delay_queue()")
Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Sagi Grimberg <sagi@grimberg.me>
Cc: Long Li <longli@microsoft.com>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: Introduce blk_mq_delay_run_hw_queue()
Bart Van Assche [Fri, 7 Apr 2017 18:16:52 +0000 (11:16 -0700)]
blk-mq: Introduce blk_mq_delay_run_hw_queue()

Introduce a function that runs a hardware queue unconditionally
after a delay. Note: there is already a function that stops and
restarts a hardware queue after a delay, namely blk_mq_delay_queue().

This function will be used in the next patch in this series.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Long Li <longli@microsoft.com>
Cc: K. Y. Srinivasan <kys@microsoft.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: trace completion of all bios.
NeilBrown [Fri, 7 Apr 2017 15:40:52 +0000 (09:40 -0600)]
block: trace completion of all bios.

Currently only dm and md/raid5 bios trigger
trace_block_bio_complete().  Now that we have bio_chain() and
bio_inc_remaining(), it is not possible, in general, for a driver to
know when the bio is really complete.  Only bio_endio() knows that.

So move the trace_block_bio_complete() call to bio_endio().

Now trace_block_bio_complete() pairs with trace_block_bio_queue().
Any bio for which a 'queue' event is traced, will subsequently
generate a 'complete' event.

There are a few cases where completion tracing is not wanted.
1/ If blk_update_request() has already generated a completion
   trace event at the 'request' level, there is no point generating
   one at the bio level too.  In this case the bi_sector and bi_size
   will have changed, so the bio level event would be wrong

2/ If the bio hasn't actually been queued yet, but is being aborted
   early, then a trace event could be confusing.  Some filesystems
   call bio_endio() but do not want tracing.

3/ The bio_integrity code interposes itself by replacing bi_end_io,
   then restoring it and calling bio_endio() again.  This would produce
   two identical trace events if left like that.

To handle these, we introduce a flag BIO_TRACE_COMPLETION and only
produce the trace event when this is set.
We address point 1 above by clearing the flag in blk_update_request().
We address point 2 above by only setting the flag when
generic_make_request() is called.
We address point 3 above by clearing the flag after generating a
completion event.

When bio_split() is used on a bio, particularly in blk_queue_split(),
there is an extra complication.  A new bio is split off the front, and
may be handle directly without going through generic_make_request().
The old bio, which has been advanced, is passed to
generic_make_request(), so it will trigger a trace event a second
time.
Probably the best result when a split happens is to see a single
'queue' event for the whole bio, then multiple 'complete' events - one
for each component.  To achieve this was can:
- copy the BIO_TRACE_COMPLETION flag to the new bio in bio_split()
- avoid generating a 'queue' event if BIO_TRACE_COMPLETION is already set.
This way, the split-off bio won't create a queue event, the original
won't either even if it re-submitted to generic_make_request(),
but both will produce completion events, each for their own range.

So if generic_make_request() is called (which generates a QUEUED
event), then bi_endio() will create a single COMPLETE event for each
range that the bio is split into, unless the driver has explicitly
requested it not to.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock: simple improvements for bio->flags
NeilBrown [Fri, 7 Apr 2017 01:10:44 +0000 (11:10 +1000)]
block: simple improvements for bio->flags

The comment for the 'flags' field of 'bio' mentions
"command" which is no longer stored there, and doesn't
mention the bvec pool number, which is.

BIO_RESET_BITS is set in such a way that it would need to be
updated if new bits were added, which is easy to miss.

BVEC_POOL_BITS is larger than needed.  The BVEC_POOL_IDX()
ranges from 0 to 6, so 3 bits are sufficient.

This patch make improvements in each of these areas.

Signed-off-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: remap queues when adding/removing hardware queues
Omar Sandoval [Fri, 7 Apr 2017 14:53:11 +0000 (08:53 -0600)]
blk-mq: remap queues when adding/removing hardware queues

blk_mq_update_nr_hw_queues() used to remap hardware queues, which is the
behavior that drivers expect. However, commit 4e68a011428a changed
blk_mq_queue_reinit() to not remap queues for the case of CPU
hotplugging, inadvertently making blk_mq_update_nr_hw_queues() not remap
queues as well. This breaks, for example, NBD's multi-connection mode,
leaving the added hardware queues unused. Fix it by making
blk_mq_update_nr_hw_queues() explicitly remap the queues.

Fixes: 4e68a011428a ("blk-mq: don't redistribute hardware queues on a CPU hotplug event")
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq-sched: fix crash in switch error path
Omar Sandoval [Fri, 7 Apr 2017 14:52:27 +0000 (08:52 -0600)]
blk-mq-sched: fix crash in switch error path

In elevator_switch(), if blk_mq_init_sched() fails, we attempt to fall
back to the original scheduler. However, at this point, we've already
torn down the original scheduler's tags, so this causes a crash. Doing
the fallback like the legacy elevator path is much harder for mq, so fix
it by just falling back to none, instead.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq-sched: set up scheduler tags when bringing up new queues
Omar Sandoval [Wed, 5 Apr 2017 19:01:31 +0000 (12:01 -0700)]
blk-mq-sched: set up scheduler tags when bringing up new queues

If a new hardware queue is added at runtime, we don't allocate scheduler
tags for it, leading to a crash. This hooks up the scheduler framework
to blk_mq_{init,exit}_hctx() to make sure everything gets properly
initialized/freed.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq-sched: refactor scheduler initialization
Omar Sandoval [Wed, 5 Apr 2017 19:01:30 +0000 (12:01 -0700)]
blk-mq-sched: refactor scheduler initialization

Preparation cleanup for the next couple of fixes, push
blk_mq_sched_setup() and e->ops.mq.init_sched() into a helper.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: use the right hctx when getting a driver tag fails
Omar Sandoval [Fri, 7 Apr 2017 14:56:26 +0000 (08:56 -0600)]
blk-mq: use the right hctx when getting a driver tag fails

While dispatching requests, if we fail to get a driver tag, we mark the
hardware queue as waiting for a tag and put the requests on a
hctx->dispatch list to be run later when a driver tag is freed. However,
blk_mq_dispatch_rq_list() may dispatch requests from multiple hardware
queues if using a single-queue scheduler with a multiqueue device. If
blk_mq_get_driver_tag() fails, it doesn't update the hardware queue we
are processing. This means we end up using the hardware queue of the
previous request, which may or may not be the same as that of the
current request. If it isn't, the wrong hardware queue will end up
waiting for a tag, and the requests will be on the wrong dispatch list,
leading to a hang.

The fix is twofold:

1. Make sure we save which hardware queue we were trying to get a
   request for in blk_mq_get_driver_tag() regardless of whether it
   succeeds or not.
2. Make blk_mq_dispatch_rq_list() take a request_queue instead of a
   blk_mq_hw_queue to make it clear that it must handle multiple
   hardware queues, since I've already messed this up on a couple of
   occasions.

This didn't appear in testing with nvme and mq-deadline because nvme has
more driver tags than the default number of scheduler tags. However,
with the blk_mq_update_nr_hw_queues() fix, it showed up with nbd.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge tag 'xfs-4.11-fixes-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Linus Torvalds [Thu, 6 Apr 2017 21:42:05 +0000 (14:42 -0700)]
Merge tag 'xfs-4.11-fixes-3' of git://git./fs/xfs/xfs-linux

Pull XFS fixes from Darrick Wong:
 "Here are three more fixes for 4.11.

  The first one reworks the inline directory verifier to check the
  working copy of the directory metadata and to avoid triggering a
  periodic crash in xfs/348. The second patch fixes a regression in hole
  punching at EOF that corrupts files; and the third patch closes a
  kernel memory disclosure bug.

  Summary:

   - rework the inline directory verifier to avoid crashes on disk
     corruption

   - don't change file size when punching holes w/ KEEP_SIZE

   - close a kernel memory exposure bug"

* tag 'xfs-4.11-fixes-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  xfs: fix kernel memory exposure problems
  xfs: Honor FALLOC_FL_KEEP_SIZE when punching ends of files
  xfs: rework the inline directory verifiers

8 years agoMerge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus
Linus Torvalds [Thu, 6 Apr 2017 20:16:34 +0000 (13:16 -0700)]
Merge branch 'upstream' of git://git.linux-mips.org/ralf/upstream-linus

Pull MIPS fixes from Ralf Baechle:
 "Lantiq:
    - Fix adding xbar resoures causing a panic

  Loongson3:
    - Some Loongson 3A don't identify themselves as having an FTLB so
      hardwire that knowledge into CPU probing.
    - Handle Loongson 3 TLB peculiarities in the fast path of the RDHWR
      emulation.
    - Fix invalid FTLB entries with huge page on VTLB+FTLB platforms
    - Add missing calculation of S-cache and V-cache cache-way size

  Ralink:
    - Fix typos in rt3883 pinctrl data

  Generic:
    - Force o32 fp64 support on 32bit MIPS64r6 kernels
    - Yet another build fix after the linux/sched.h changes
    - Wire up statx system call
    - Fix stack unwinding after introduction of IRQ stack
    - Fix spinlock code to build even for microMIPS with recent binutils

  SMP-CPS:
    - Fix retrieval of VPE mask on big endian CPUs"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus:
  MIPS: IRQ Stack: Unwind IRQ stack onto task stack
  MIPS: c-r4k: Fix Loongson-3's vcache/scache waysize calculation
  MIPS: Flush wrong invalid FTLB entry for huge page
  MIPS: Check TLB before handle_ri_rdhwr() for Loongson-3
  MIPS: Add MIPS_CPU_FTLB for Loongson-3A R2
  MIPS: Lantiq: fix missing xbar kernel panic
  MIPS: smp-cps: Fix retrieval of VPE mask on big endian CPUs
  MIPS: Wire up statx system call
  MIPS: Include asm/ptrace.h now linux/sched.h doesn't
  MIPS: ralink: Fix typos in rt3883 pinctrl
  MIPS: End spinlocks with .insn
  MIPS: Force o32 fp64 support on 32bit MIPS64r6 kernels

8 years agoMerge tag 'trace-v4.11-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
Linus Torvalds [Thu, 6 Apr 2017 20:12:12 +0000 (13:12 -0700)]
Merge tag 'trace-v4.11-rc5' of git://git./linux/kernel/git/rostedt/linux-trace

Pull tracing fix from Steven Rostedt:
 "Wei Yongjun fixed a long standing bug in the ring buffer startup test.

  If for some unknown reason, the kthread that is created fails to be
  created, the return from kthread_create() is an PTR_ERR and not a
  NULL. The test incorrectly checks for NULL instead of an error"

* tag 'trace-v4.11-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  ring-buffer: Fix return value check in test_ringbuffer()

8 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 6 Apr 2017 03:17:38 +0000 (20:17 -0700)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Reject invalid updates to netfilter expectation policies, from Pablo
    Neira Ayuso.

 2) Fix memory leak in nfnl_cthelper, from Jeffy Chen.

 3) Don't do stupid things if we get a neigh_probe() on a neigh entry
    whose ops lack a solicit method. From Eric Dumazet.

 4) Don't transmit packets in r8152 driver when the carrier is off, from
    Hayes Wang.

 5) Fix ipv6 packet type detection in aquantia driver, from Pavel
    Belous.

 6) Don't write uninitialized data into hw registers in bna driver, from
    Arnd Bergmann.

 7) Fix locking in ping_unhash(), from Eric Dumazet.

 8) Make BPF verifier range checks able to understand certain sequences
    emitted by LLVM, from Alexei Starovoitov.

 9) Fix use after free in ipconfig, from Mark Rutland.

10) Fix refcount leak on force commit in openvswitch, from Jarno
    Rajahalme.

11) Fix various overflow checks in AF_PACKET, from Andrey Konovalov.

12) Fix endianness bug in be2net driver, from Suresh Reddy.

13) Don't forget to wake TX queues when processing a timeout, from
    Grygorii Strashko.

14) ARP header on-stack storage is wrong in flow dissector, from Simon
    Horman.

15) Lost retransmit and reordering SNMP stats in TCP can be
    underreported. From Yuchung Cheng.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (82 commits)
  nfp: fix potential use after free on xdp prog
  tcp: fix reordering SNMP under-counting
  tcp: fix lost retransmit SNMP under-counting
  sctp: get sock from transport in sctp_transport_update_pmtu
  net: ethernet: ti: cpsw: fix race condition during open()
  l2tp: fix PPP pseudo-wire auto-loading
  bnx2x: fix spelling mistake in macros HW_INTERRUT_ASSERT_SET_*
  l2tp: take reference on sessions being dumped
  tcp: minimize false-positives on TCP/GRO check
  sctp: check for dst and pathmtu update in sctp_packet_config
  flow dissector: correct size of storage for ARP
  net: ethernet: ti: cpsw: wake tx queues on ndo_tx_timeout
  l2tp: take a reference on sessions used in genetlink handlers
  l2tp: hold session while sending creation notifications
  l2tp: fix duplicate session creation
  l2tp: ensure session can't get removed during pppol2tp_session_ioctl()
  l2tp: fix race in l2tp_recv_common()
  sctp: use right in and out stream cnt
  bpf: add various verifier test cases for self-tests
  bpf, verifier: fix rejection of unaligned access checks for map_value_adj
  ...

8 years agonfp: fix potential use after free on xdp prog
Jakub Kicinski [Tue, 4 Apr 2017 22:56:55 +0000 (15:56 -0700)]
nfp: fix potential use after free on xdp prog

We should unregister the net_device first, before we give back
our reference on xdp_prog.  Otherwise xdp_prog may be freed
before .ndo_stop() disabled the datapath.  Found by code inspection.

Fixes: ecd63a0217d5 ("nfp: add XDP support in the driver")
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: fix reordering SNMP under-counting
Yuchung Cheng [Tue, 4 Apr 2017 21:15:40 +0000 (14:15 -0700)]
tcp: fix reordering SNMP under-counting

Currently the reordering SNMP counters only increase if a connection
sees a higher degree then it has previously seen. It ignores if the
reordering degree is not greater than the default system threshold.
This significantly under-counts the number of reordering events
and falsely convey that reordering is rare on the network.

This patch properly and faithfully records the number of reordering
events detected by the TCP stack, just like the comment says "this
exciting event is worth to be remembered". Note that even so TCP
still under-estimate the actual reordering events because TCP
requires TS options or certain packet sequences to detect reordering
(i.e. ACKing never-retransmitted sequence in recovery or disordered
 state).

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agotcp: fix lost retransmit SNMP under-counting
Yuchung Cheng [Tue, 4 Apr 2017 21:15:39 +0000 (14:15 -0700)]
tcp: fix lost retransmit SNMP under-counting

The lost retransmit SNMP stat is under-counting retransmission
that uses segment offloading. This patch fixes that so all
retransmission related SNMP counters are consistent.

Fixes: 10d3be569243 ("tcp-tso: do not split TSO packets at retransmit time")
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoblock: move timeout field in struct request to pack better
Jens Axboe [Wed, 5 Apr 2017 18:16:38 +0000 (12:16 -0600)]
block: move timeout field in struct request to pack better

After commit 64c7f1d1572c, we went from 1 to 2 holes in my
test setup. If we move the timeout field a bit, we remove
both of those holes and shrink struct request by 8 bytes.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblock, scsi: move the retries field to struct scsi_request
Christoph Hellwig [Wed, 5 Apr 2017 17:18:12 +0000 (19:18 +0200)]
block, scsi: move the retries field to struct scsi_request

Instead of bloating the generic struct request with it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme: move the retries count to struct nvme_request
Christoph Hellwig [Wed, 5 Apr 2017 17:18:11 +0000 (19:18 +0200)]
nvme: move the retries count to struct nvme_request

The way NVMe uses this field is entirely different from the older
SCSI/BLOCK_PC usage, so move it into struct nvme_request.

Also reduce the size of the file to a unsigned char so that we leave
space for additional smaller fields that will appear soon.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme: mark nvme_max_retries static
Christoph Hellwig [Wed, 5 Apr 2017 17:18:10 +0000 (19:18 +0200)]
nvme: mark nvme_max_retries static

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme: cleanup nvme_req_needs_retry
Christoph Hellwig [Wed, 5 Apr 2017 17:18:09 +0000 (19:18 +0200)]
nvme: cleanup nvme_req_needs_retry

Don't pass the status explicitly but derive it from the requeust,
and unwind the complex condition to be more readable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme: move ->retries setup to nvme_setup_cmd
Christoph Hellwig [Wed, 5 Apr 2017 17:18:08 +0000 (19:18 +0200)]
nvme: move ->retries setup to nvme_setup_cmd

->retries is counting the number of times a command is resubmitted, and
be cleared on the first time we see the command.  We currently don't do
that for non-PCIe command, which is easily fixed by moving the setup
to common code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge tag 'mfd-fixes-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd
Linus Torvalds [Wed, 5 Apr 2017 16:04:26 +0000 (09:04 -0700)]
Merge tag 'mfd-fixes-4.11' of git://git./linux/kernel/git/lee/mfd

Pull MFD bug fix from Lee Jones:
 "Increase buffer size om cros-ec to allow for SPI messages"

* tag 'mfd-fixes-4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/lee/mfd:
  mfd: cros-ec: Fix host command buffer size

8 years agoremove the obsolete hd driver
Christoph Hellwig [Wed, 5 Apr 2017 13:59:23 +0000 (15:59 +0200)]
remove the obsolete hd driver

This driver is for pre-IDE hardisk that are only found in PC from the
stoneage of personal computing, and which we don't support elsewhere
in the kernel these days.

It's also been marked broken forever.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoblk-mq: Remove blk_mq_queue_data.list
Bart Van Assche [Wed, 5 Apr 2017 15:39:18 +0000 (08:39 -0700)]
blk-mq: Remove blk_mq_queue_data.list

The block layer core sets blk_mq_queue_data.list but no block
drivers read that member. Hence remove it and also the code that
is used to set this member.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge tag 'kbuild-fixes-v4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/masah...
Linus Torvalds [Wed, 5 Apr 2017 15:37:28 +0000 (08:37 -0700)]
Merge tag 'kbuild-fixes-v4.11' of git://git./linux/kernel/git/masahiroy/linux-kbuild

Pull Kbuild fixes from Masahiro Yamada:

 - hand-off primary maintainership of Kbuild

 - fix build warnings

 - fix build error when GCOV is enabled with old compiler

 - fix HAVE_ASM_GOTO check when GCC plugin is enabled

* tag 'kbuild-fixes-v4.11' of git://git.kernel.org/pub/scm/linux/kernel/git/masahiroy/linux-kbuild:
  gconfig: remove misleading parentheses around a condition
  jump label: fix passing kbuild_cflags when checking for asm goto support
  Kbuild: use cc-disable-warning consistently for maybe-uninitialized
  kbuild: external module build warnings when KBUILD_OUTPUT set and W=1
  MAINTAINERS: add Masahiro Yamada as a Kbuild maintainer

8 years agosctp: get sock from transport in sctp_transport_update_pmtu
Xin Long [Tue, 4 Apr 2017 05:39:55 +0000 (13:39 +0800)]
sctp: get sock from transport in sctp_transport_update_pmtu

This patch is almost to revert commit 02f3d4ce9e81 ("sctp: Adjust PMTU
updates to accomodate route invalidation."). As t->asoc can't be NULL
in sctp_transport_update_pmtu, it could get sk from asoc, and no need
to pass sk into that function.

It is also to remove some duplicated codes from that function.

Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agocfq: Disable writeback throttling by default
Jan Kara [Tue, 4 Apr 2017 12:31:30 +0000 (14:31 +0200)]
cfq: Disable writeback throttling by default

Writeback throttling does not play well with CFQ since that also tries
to throttle async writes. As a result async writeback can get starved in
presence of readers. As an example take a benchmark simulating
postgreSQL database running over a standard rotating SATA drive. There
are 16 processes doing random reads from a huge file (2*machine memory),
1 process doing random writes to the huge file and calling fsync once
per 50000 writes and 1 process doing sequential 8k writes to a
relatively small file wrapping around at the end of the file and calling
fsync every 5 writes. Under this load read latency easily exceeds the
target latency of 75 ms (just because there are so many reads happening
against a relatively slow disk) and thus writeback is throttled to a
point where only 1 write request is allowed at a time. Blktrace data
then looks like:

  8,0    1        0     8.347751764     0  m   N cfq workload slice:40000000
  8,0    1        0     8.347755256     0  m   N cfq293A  / set_active wl_class: 0 wl_type:0
  8,0    1        0     8.347784100     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    1     3814     8.347763916  5839 UT   N [kworker/u9:2] 1
  8,0    0        0     8.347777605     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    1        0     8.347784100     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    3     1596     8.354364057     0  C   R 156109528 + 8 (6906954) [0]
  8,0    3        0     8.354383193     0  m   N cfq6196SN / complete rqnoidle 0
  8,0    3        0     8.354386476     0  m   N cfq schedule dispatch
  8,0    3        0     8.354399397     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    3        0     8.354404705     0  m   N cfq293A  / dispatch_insert
  8,0    3        0     8.354409454     0  m   N cfq293A  / dispatched a request
  8,0    3        0     8.354412527     0  m   N cfq293A  / activate rq, drv=1
  8,0    3     1597     8.354414692     0  D   W 145961400 + 24 (6718452) [swapper/0]
  8,0    3        0     8.354484184     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    3        0     8.354487536     0  m   N cfq293A  / slice expired t=0
  8,0    3        0     8.354498013     0  m   N / served: vt=5888102466265088 min_vt=5888074869387264
  8,0    3        0     8.354502692     0  m   N cfq293A  / sl_used=6737519 disp=1 charge=6737519 iops=0 sect=24
  8,0    3        0     8.354505695     0  m   N cfq293A  / del_from_rr
...
  8,0    0     1810     8.354728768     0  C   W 145961400 + 24 (314076) [0]
  8,0    0        0     8.354746927     0  m   N cfq293A  / complete rqnoidle 0
...
  8,0    1     3829     8.389886102  5839  G   W 145962968 + 24 [kworker/u9:2]
  8,0    1     3830     8.389888127  5839  P   N [kworker/u9:2]
  8,0    1     3831     8.389908102  5839  A   W 145978336 + 24 <- (8,4) 44000
  8,0    1     3832     8.389910477  5839  Q   W 145978336 + 24 [kworker/u9:2]
  8,0    1     3833     8.389914248  5839  I   W 145962968 + 24 (28146) [kworker/u9:2]
  8,0    1        0     8.389919137     0  m   N cfq293A  / insert_request
  8,0    1        0     8.389924305     0  m   N cfq293A  / add_to_rr
  8,0    1     3834     8.389933175  5839 UT   N [kworker/u9:2] 1
...
  8,0    0        0     9.455290997     0  m   N cfq workload slice:40000000
  8,0    0        0     9.455294769     0  m   N cfq293A  / set_active wl_class:0 wl_type:0
  8,0    0        0     9.455303499     0  m   N cfq293A  / fifo=ffff880003166090
  8,0    0        0     9.455306851     0  m   N cfq293A  / dispatch_insert
  8,0    0        0     9.455311251     0  m   N cfq293A  / dispatched a request
  8,0    0        0     9.455314324     0  m   N cfq293A  / activate rq, drv=1
  8,0    0     2043     9.455316210  6204  D   W 145962968 + 24 (1065401962) [pgioperf]
  8,0    0        0     9.455392407     0  m   N cfq293A  / Not idling.  st->count:1
  8,0    0        0     9.455395969     0  m   N cfq293A  / slice expired t=0
  8,0    0        0     9.455404210     0  m   N / served: vt=5888958194597888 min_vt=5888941810597888
  8,0    0        0     9.455410077     0  m   N cfq293A  / sl_used=4000000 disp=1 charge=4000000 iops=0 sect=24
  8,0    0        0     9.455416851     0  m   N cfq293A  / del_from_rr
...
  8,0    0     2045     9.455648515     0  C   W 145962968 + 24 (332305) [0]
  8,0    0        0     9.455668350     0  m   N cfq293A  / complete rqnoidle 0
...
  8,0    1     4371     9.455710115  5839  G   W 145978336 + 24 [kworker/u9:2]
  8,0    1     4372     9.455712350  5839  P   N [kworker/u9:2]
  8,0    1     4373     9.455730159  5839  A   W 145986616 + 24 <- (8,4) 52280
  8,0    1     4374     9.455732674  5839  Q   W 145986616 + 24 [kworker/u9:2]
  8,0    1     4375     9.455737563  5839  I   W 145978336 + 24 (27448) [kworker/u9:2]
  8,0    1        0     9.455742871     0  m   N cfq293A  / insert_request
  8,0    1        0     9.455747550     0  m   N cfq293A  / add_to_rr
  8,0    1     4376     9.455756629  5839 UT   N [kworker/u9:2] 1

So we can see a Q event for a write request, then IO is blocked by
writeback throttling and G and I events for the request happen only once
other writeback IO is completed. Thus CFQ always sees only one write
request. When it sees it, it queues the async queue behind all the read
queues and the async queue gets scheduled after about one second. When
it is scheduled, that one request gets dispatched and async queue is
expired as it has no more requests to submit. Overall we submit about
one write request per second.

Although this scheduling is beneficial for read latency, writes are
heavily starved and this causes large delays all over the system (due to
processes blocking on page lock, transaction starts, etc.). When
writeback throttling is disabled, write throughput is about one fifth of
a read throughput which roughly matches readers/writers ratio and
overall the system stalls are much shorter.

Mixing writeback throttling logic with CFQ throttling logic is always a
recipe for surprises as CFQ assumes it sees the big part of the picture
which is not necessarily true when writeback throttling is blocking
requests. So disable writeback throttling logic by default when CFQ is
used as an IO scheduler.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoring-buffer: Fix return value check in test_ringbuffer()
Wei Yongjun [Fri, 17 Jun 2016 17:33:59 +0000 (17:33 +0000)]
ring-buffer: Fix return value check in test_ringbuffer()

In case of error, the function kthread_run() returns ERR_PTR()
and never returns NULL. The NULL test in the return value check
should be replaced with IS_ERR().

Link: http://lkml.kernel.org/r/1466184839-14927-1-git-send-email-weiyj_lk@163.com
Cc: stable@vger.kernel.org
Fixes: 6c43e554a ("ring-buffer: Add ring buffer startup selftest")
Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
8 years agomfd: cros-ec: Fix host command buffer size
Vic Yang [Fri, 24 Mar 2017 17:44:01 +0000 (18:44 +0100)]
mfd: cros-ec: Fix host command buffer size

For SPI, we can get up to 32 additional bytes for response preamble.
The current overhead (2 bytes) may cause problems when we try to receive
a big response. Update it to 32 bytes.

Without this fix we could see a kernel BUG when we receive a big response
from the Chrome EC when is connected via SPI.

Signed-off-by: Vic Yang <victoryang@google.com>
Tested-by: Enric Balletbo i Serra <enric.balletbo.collabora.com>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
8 years agoblock: fix inheriting request priority from bio
Adam Manzanares [Tue, 4 Apr 2017 15:25:14 +0000 (08:25 -0700)]
block: fix inheriting request priority from bio

In 4.10 I introduced a patch that associates the ioc priority with
each request in the block layer. This work was done in the single queue
block layer code. This patch unifies ioc priority to request mapping across
the single/multi queue block layers.

I have tested this patch with the null block device driver with the following
parameters.

null_blk queue_mode=2 irqmode=0 use_per_node_hctx=1 nr_devices=1

I have not seen a performance regression with this patch and I would appreciate
any feedback or additional testing.

I have also verified that io priorities are passed to the device when using
the SQ and MQ path to a SATA HDD that supports io priorities.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Adam Manzanares <adam.manzanares@wdc.com>
Reviewed-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agoMerge tag 'gpio-v4.11-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Linus Torvalds [Tue, 4 Apr 2017 18:40:20 +0000 (11:40 -0700)]
Merge tag 'gpio-v4.11-3' of git://git./linux/kernel/git/linusw/linux-gpio

Pull late GPIO fixes from Linus Walleij:
 "Some late coming ACPI fixes for GPIO.

  We're dealing with ACPI issues here. The first is related to wake IRQs
  on Bay Trail/Cherry Trail CPUs which are common in laptops. The second
  is about proper probe deferral when reading _CRS properties.

  For my untrained eye it seems there was some quarrel between the BIOS
  and the kernel about who is supposed to deal with wakeups from GPIO
  lines"

* tag 'gpio-v4.11-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-gpio:
  ACPI / gpio: do not fall back to parsing _CRS when we get a deferral
  gpio: acpi: Call enable_irq_wake for _IAE GpioInts with Wake set

8 years agoMerge tag 'wireless-drivers-for-davem-2017-04-03' of git://git.kernel.org/pub/scm...
David S. Miller [Tue, 4 Apr 2017 18:36:54 +0000 (11:36 -0700)]
Merge tag 'wireless-drivers-for-davem-2017-04-03' of git://git./linux/kernel/git/kvalo/wireless-drivers

Kalle Valo says:

====================
wireless-drivers fixes for 4.11

iwlwifi

* an RCU fix
* a fix for a potential out-of-bounds access crash
* a fix for IBSS which has been broken since DQA was enabled

rtlwifi

* fix scheduling while atomic regression

brcmfmac

* fix use-after-free bug found by KASAN
====================

Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'nios2-v4.11-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan...
Linus Torvalds [Tue, 4 Apr 2017 18:16:52 +0000 (11:16 -0700)]
Merge tag 'nios2-v4.11-fix' of git://git./linux/kernel/git/lftan/nios2

Pull nios2 fix from Ley Foon Tan:

 - nios2: reserve boot memory for device tree

* tag 'nios2-v4.11-fix' of git://git.kernel.org/pub/scm/linux/kernel/git/lftan/nios2:
  nios2: reserve boot memory for device tree

8 years agonet: ethernet: ti: cpsw: fix race condition during open()
Sekhar Nori [Mon, 3 Apr 2017 12:04:28 +0000 (17:34 +0530)]
net: ethernet: ti: cpsw: fix race condition during open()

TI's cpsw driver handles both OF and non-OF case for phy
connect. Unfortunately of_phy_connect() returns NULL on
error while phy_connect() returns ERR_PTR().

To handle this, cpsw_slave_open() overrides the return value
from phy_connect() to make it NULL or error.

This leaves a small window, where cpsw_adjust_link() may be
invoked for a slave while slave->phy pointer is temporarily
set to -ENODEV (or some other error) before it is finally set
to NULL.

_cpsw_adjust_link() only handles the NULL case, and an oops
results when ERR_PTR() is seen by it.

Note that cpsw_adjust_link() checks PHY status for each
slave whenever it is invoked. It can so happen that even
though phy_connect() for a given slave returns error,
_cpsw_adjust_link() is still called for that slave because
the link status of another slave changed.

Fix this by using a temporary pointer to store return value
of {of_}phy_connect() and do a one-time write to slave->phy.

Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Reported-by: Yan Liu <yan-liu@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agoMerge tag 'drm-fixes-for-v4.11-rc6' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Tue, 4 Apr 2017 17:12:15 +0000 (10:12 -0700)]
Merge tag 'drm-fixes-for-v4.11-rc6' of git://people.freedesktop.org/~airlied/linux

Pull drm fixes from Dave Airlie:
 "This is just mostly stuff that missed rc5, from vmwgfx and msm
  drivers"

* tag 'drm-fixes-for-v4.11-rc6' of git://people.freedesktop.org/~airlied/linux:
  drm/msm: Make sure to detach the MMU during GPU cleanup
  drm/msm/hdmi: redefinitions of macros not required
  drm/msm/mdp5: Update SSPP_MAX value
  drm/msm/dsi: Fix bug in dsi_mgr_phy_enable
  drm/msm: Don't allow zero sized buffer objects
  drm/msm: Fix wrong pointer check in a5xx_destroy
  drm/msm: adreno: fix build error without debugfs
  drm/vmwgfx: fix integer overflow in vmw_surface_define_ioctl()
  drm/vmwgfx: Remove getparam error message
  drm/ttm: Avoid calling drm_ht_remove from atomic context
  drm/ttm, drm/vmwgfx: Relax permission checking when opening surfaces
  drm/vmwgfx: avoid calling vzalloc with a 0 size in vmw_get_cap_3d_ioctl()
  drm/vmwgfx: NULL pointer dereference in vmw_surface_define_ioctl()
  drm/vmwgfx: Type-check lookups of fence objects

8 years agol2tp: fix PPP pseudo-wire auto-loading
Guillaume Nault [Mon, 3 Apr 2017 11:23:15 +0000 (13:23 +0200)]
l2tp: fix PPP pseudo-wire auto-loading

PPP pseudo-wire type is 7 (11 is L2TP_PWTYPE_IP).

Fixes: f1f39f911027 ("l2tp: auto load type modules")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agobnx2x: fix spelling mistake in macros HW_INTERRUT_ASSERT_SET_*
Colin Ian King [Mon, 3 Apr 2017 10:19:10 +0000 (11:19 +0100)]
bnx2x: fix spelling mistake in macros HW_INTERRUT_ASSERT_SET_*

Trival fix, rename HW_INTERRUT_ASSERT_SET_* to HW_INTERRUPT_ASSERT_SET_*

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: Yuval Mintz <Yuval.Mintz@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agol2tp: take reference on sessions being dumped
Guillaume Nault [Mon, 3 Apr 2017 10:03:13 +0000 (12:03 +0200)]
l2tp: take reference on sessions being dumped

Take a reference on the sessions returned by l2tp_session_find_nth()
(and rename it l2tp_session_get_nth() to reflect this change), so that
caller is assured that the session isn't going to disappear while
processing it.

For procfs and debugfs handlers, the session is held in the .start()
callback and dropped in .show(). Given that pppol2tp_seq_session_show()
dereferences the associated PPPoL2TP socket and that
l2tp_dfs_seq_session_show() might call pppol2tp_show(), we also need to
call the session's .ref() callback to prevent the socket from going
away from under us.

Fixes: fd558d186df2 ("l2tp: Split pppol2tp patch into separate l2tp and ppp parts")
Fixes: 0ad6614048cf ("l2tp: Add debugfs files for dumping l2tp debug info")
Fixes: 309795f4bec2 ("l2tp: Add netlink control API for L2TP")
Signed-off-by: Guillaume Nault <g.nault@alphalink.fr>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 years agonvme: factor request completion code into a common helper
Christoph Hellwig [Thu, 30 Mar 2017 11:41:32 +0000 (13:41 +0200)]
nvme: factor request completion code into a common helper

This avoids duplicating the logic four times, and it also allows to keep
some helpers static in core.c or just opencode them.

Note that this loses printing the aborted status on completions in the
PCI driver as that uses a data structure not available any more.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-fc: drop ctrl for all command completions
Christoph Hellwig [Thu, 30 Mar 2017 11:41:31 +0000 (13:41 +0200)]
nvme-fc: drop ctrl for all command completions

A requeue means we go through nvme_fc_start_fcp_op again and get
another controller reference.  To make sure the refcount doesn't
leak we also need to drop it for every completion that came from
the LLDD.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-fc: increment request retries counter before requeuing
Sagi Grimberg [Wed, 29 Mar 2017 17:54:46 +0000 (20:54 +0300)]
nvme-fc: increment request retries counter before requeuing

This way our max retry limit holds as well.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-loop: increment request retries counter before requeuing
Sagi Grimberg [Wed, 29 Mar 2017 17:54:15 +0000 (20:54 +0300)]
nvme-loop: increment request retries counter before requeuing

This way our max retry limit holds as well.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-rdma: increment request retries counter before requeuing
Sagi Grimberg [Wed, 29 Mar 2017 17:51:10 +0000 (20:51 +0300)]
nvme-rdma: increment request retries counter before requeuing

This way our max retry limit holds as well.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme_fc: Clean up host fcpio done status handling
James Smart [Fri, 24 Mar 2017 03:41:27 +0000 (20:41 -0700)]
nvme_fc: Clean up host fcpio done status handling

As Dan Carpenter pointed out: mixing 16-bit nvme status with 32-bit
error status from driver. Corrected comment on fcp request struct
status field, and converted done routine to explicitly set nvme status
codes for nvme status.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvmet_fc: Clear SG list to avoid double frees
James Smart [Fri, 24 Mar 2017 03:41:26 +0000 (20:41 -0700)]
nvmet_fc: Clear SG list to avoid double frees

Clear SG list to avoid double frees of payload page list

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme_fc: correct LS validation
James Smart [Fri, 24 Mar 2017 03:41:25 +0000 (20:41 -0700)]
nvme_fc: correct LS validation

LS validations shouldn't have been independent checks.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvmet_fc: Sync NVME LS reject reasons with spec
James Smart [Fri, 24 Mar 2017 03:41:24 +0000 (20:41 -0700)]
nvmet_fc: Sync NVME LS reject reasons with spec

nvmet_fc: Sync NVME LS reject reasons with spec

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme_fc: Add check of status_code in ERSP_IU
James Smart [Fri, 24 Mar 2017 03:41:23 +0000 (20:41 -0700)]
nvme_fc: Add check of status_code in ERSP_IU

Add check of status_code in ERSP_IU

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme_fc: Sync FC-NVME header with standard
James Smart [Fri, 24 Mar 2017 03:41:22 +0000 (20:41 -0700)]
nvme_fc: Sync FC-NVME header with standard

Update FC-NVME definitions to match FC-NVME r1.14 (16-020vB) plus
change voted in by 2/22 FC-NVME Adhoc (see HOSTID below).

Includes the following:
- Addition of "status_code" field to ERSP IU
- Addition of FC-NVME LS RJT reason_codes and reason_explanations
- CreateAssociation payload, HostID field shortened to 16 bytes

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-rdma: Support ctrl_loss_tmo
Sagi Grimberg [Sat, 18 Mar 2017 18:58:29 +0000 (20:58 +0200)]
nvme-rdma: Support ctrl_loss_tmo

Before scheduling a reconnect attempt, check
nr_reconnects against max_reconnects, if not
exhausted (or max_reconnects is not -1), schedule
a reconnect attempts, otherwise schedule ctrl
removal.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-fabrics: Allow ctrl loss timeout configuration
Sagi Grimberg [Sat, 18 Mar 2017 18:52:36 +0000 (20:52 +0200)]
nvme-fabrics: Allow ctrl loss timeout configuration

When a host sense that its controller session is damaged,
it tries to re-establish it periodically (reconnect every
reconnect_delay). It may very well be that the controller
is gone and never coming back, in this case the host will
try to reconnect forever.

Add a ctrl_loss_tmo to bound the number of reconnect attempts
to a specific controller (default to a reasonable 10 minutes).
The timeout configuration is actually translated into number of
reconnect attempts and not a schedule on its own but rather
divided with reconnect_delay. This is useful to prevent
racing flows of remove and reconnect, and it doesn't really
matter if we remove slightly sooner than what the user requested.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-rdma: get rid of local reconnect_delay
Sagi Grimberg [Sat, 18 Mar 2017 21:47:22 +0000 (23:47 +0200)]
nvme-rdma: get rid of local reconnect_delay

we already have it in opts.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-loop: retrieve iod from the cqe command_id
Sagi Grimberg [Mon, 27 Feb 2017 16:28:25 +0000 (18:28 +0200)]
nvme-loop: retrieve iod from the cqe command_id

useful to validate that the we didn't mess up
the command_id.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-loop: remove unneeded includes
Sagi Grimberg [Sun, 19 Mar 2017 04:32:09 +0000 (06:32 +0200)]
nvme-loop: remove unneeded includes

Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-fc: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 12:16:05 +0000 (14:16 +0200)]
nvme-fc: fix module_init (theoretical) error path

If nvmf_register_transport happened to fail
(it can't, but theoretically) we leak memory.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-loop: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 04:26:28 +0000 (06:26 +0200)]
nvme-loop: fix module_init (theoretical) error path

if nvmf_register_transport happend to fail, we
need to nvmet_unregister_transport as well.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvme-rdma: fix module_init (theoretical) error path
Sagi Grimberg [Sun, 19 Mar 2017 04:21:42 +0000 (06:21 +0200)]
nvme-rdma: fix module_init (theoretical) error path

If nvmf_register_transport happened to fail
(it can't, but theoretically) we leak memory.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvmet: use symbolic constants for log identifiers
Max Gurtovoy [Sun, 5 Mar 2017 22:30:38 +0000 (00:30 +0200)]
nvmet: use symbolic constants for log identifiers

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>
8 years agonvmet: Introduced helper routine for controller status check.
Parav Pandit [Tue, 28 Feb 2017 05:21:33 +0000 (23:21 -0600)]
nvmet: Introduced helper routine for controller status check.

This patch introduces helper function for checking controller
status during admin and io command processing which returns u16
status. As to bring consistency on returning status, other
friend functions also now return u16 status instead of int
to match the spec.

As part of the theseerror log prints in also prints qid on
which command error occured.

Signed-off-by: Parav Pandit <parav@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@fb.com>