If dm-mpath encounters an reservation conflict it should not fail the
path (as communication with the target is not affected) but should
rather retry on another path.  However, in doing so we might be inducing
a ping-pong between paths, with no guarantee of any forward progress.
And arguably a reservation conflict is an unexpected error, so we should
be passing it upwards to allow the application to take appropriate
steps.

This change resolves a show-stopper problem seen with the pNFS SCSI
layout because it is trivial to hit reservation conflict based failover
loops without it.

Doubts were raised about the implications of this change relative to
products like IBM's SVC.  But there is little point withholding a fix
for Linux because a proprietary product may or may not have some issues
in its implementation of how it interfaces with Linux.  In the future,
if there is glaring evidence that this change is certainly problematic
we can revisit it.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Acked-by: Christoph Hellwig <hch@lst.de>
Tested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mike Snitzer <snitzer@redhat.com> # tweaked header

Return DM_MAPIO_DELAY_REQUEUE from .clone_and_map_rq.  Also, return
false from .busy, if all paths are down, so that blk-mq requests get
mapped via .clone_and_map_rq -- which results in DM_MAPIO_DELAY_REQUEUE
being returned to dm-rq.

This change allows for a noticeable reduction in cpu utilization
(reduced kworker load) while all paths are down, e.g.:

system CPU idleness (as measured by fio's --idle-prof=system):
before: system: 86.58%
after:  system: 98.60%

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>

When reinstating a path the blk-mq request_queue's requeue_list should
get kicked.  It makes sense to kick the requeue_list as part of the
existing hook (previously only used by bio-based support).

Rename process_queued_bios_list to process_queued_io_list.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>

Use autoremove_wake_function() instead of default_wake_function()
to make the dm wait loops more similar to other wait loops in the
kernel.  This patch does not change any functionality.

Signed-off-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>

If pg_init_retries is set and a request is queued against a multipath
device with all underlying block device request_queues in the "dying"
state then an infinite loop is triggered because activate_path() never
succeeds and hence never calls pg_init_done().

This change avoids that device removal triggers an infinite loop by
failing the activate_path() which causes the "dying" path to be failed.

Reported-by: Bart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org

Since commit 63a4cc24867d, bio->bi_rw contains flags in the lower
portion and the op code in the higher portions. This means that
old code that relies on manually setting bi_rw is most likely
going to be broken. Instead of letting that brokeness linger,
rename the member, to force old and out-of-tree code to break
at compile time instead of at runtime.

No intended functional changes in this commit.

Signed-off-by: Jens Axboe <axboe@fb.com>

Multiple flags were being tested without locking.  Protect against
non-atomic bit changes in m->flags by holding m->lock (while testing or
setting the queue_if_no_path related flags).

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

Allow a user to specify an optional feature 'queue_mode <mode>' where
<mode> may be "bio", "rq" or "mq" -- which corresponds to bio-based,
request_fn rq-based, and blk-mq rq-based respectively.

If the queue_mode feature isn't specified the default for the
"multipath" target is still "rq" but if dm_mod.use_blk_mq is set to Y
it'll default to mode "mq".

This new queue_mode feature introduces the ability for each multipath
device to have its own queue_mode (whereas before this feature all
multipath devices effectively had to have the same queue_mode).

This commit also goes a long way to eliminate the awkward (ab)use of
DM_TYPE_*, the associated filter_md_type() and other relatively fragile
and difficult to maintain code.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

Signed-off-by: Mike Snitzer <snitzer@redhat.com>

Add "multipath-bio" target that offers a bio-based multipath target as
an alternative to the request-based "multipath" target -- but in a
following commit "multipath-bio" will immediately be replaced by a new
"queue_mode" feature for the "multipath" target which will allow
bio-based mode to be selected.

When DM multipath was originally converted from bio-based to
request-based the motivation for the change was better dynamic load
balancing (by leveraging block core's request-based IO schedulers, for
merging and sorting, _before_ DM multipath would make the decision on
where to steer the IO -- based on path load and/or availability).

More background is available in this "Request-based Device-mapper
multipath and Dynamic load balancing" paper:
https://www.kernel.org/doc/ols/2007/ols2007v2-pages-235-244.pdf

But we've now come full circle where significantly faster storage
devices no longer need IOs to be made larger to drive optimal IO
performance.  And even if they do there have been changes to the block
and filesystem layers that help ensure upper layers are constructing
larger IOs.  In addition, SCSI's differentiated IO errors will propagate
through to bio-based IO completion hooks -- so that eliminates another
historic justiciation for request-based DM multipath.  Lastly, the block
layer's immutable biovec changes have made bio cloning cheaper than it
has ever been; whereas request cloning is still relatively expensive
(both on a CPU usage and memory footprint level).

As such, bio-based DM multipath offers the promise of a more efficient
IO path for high IOPs devices that are, or will be, emerging.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>