linux-2.6-microblaze.git
3 years agonvme: move nvme_validate_ns
Christoph Hellwig [Mon, 28 Sep 2020 11:55:22 +0000 (13:55 +0200)]
nvme: move nvme_validate_ns

Move nvme_validate_ns just above its only remaining caller.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: query namespace identifiers before adding the namespace
Christoph Hellwig [Mon, 28 Sep 2020 12:07:56 +0000 (14:07 +0200)]
nvme: query namespace identifiers before adding the namespace

Check the namespace identifier list first thing when scanning namespaces.
This keeps the code to query the CSI common between the alloc and validate
path, and helps to structure the code better for multiple command set
support.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: revalidate zone bitmaps in nvme_update_ns_info
Christoph Hellwig [Mon, 28 Sep 2020 10:30:16 +0000 (12:30 +0200)]
nvme: revalidate zone bitmaps in nvme_update_ns_info

Consolidate the two calls into a single place.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
3 years agonvme: remove nvme_update_formats
Christoph Hellwig [Mon, 28 Sep 2020 09:10:36 +0000 (11:10 +0200)]
nvme: remove nvme_update_formats

Now that the queue is frozen before updating ->lba_shift we can't hit the
invalid references mentioned in the comment any more.  More importantly
this code would not have helped us if the format was changed by another
controller or through implementation defined back channels.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: update the known admin effects
Christoph Hellwig [Fri, 25 Sep 2020 06:34:43 +0000 (08:34 +0200)]
nvme: update the known admin effects

A Format NVM command can change the capabilities of namespaces, while
Sanitize does change the Logical Block Content and must be serialized.

Also remove CSUPP bit for Format - it is not a mandatory command,
and we don't check for the bit anyway.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: set the queue limits in nvme_update_ns_info
Christoph Hellwig [Mon, 28 Sep 2020 10:05:28 +0000 (12:05 +0200)]
nvme: set the queue limits in nvme_update_ns_info

Only set the queue limits once we have the real block size.  This also
updates the limits on a rescan if needed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: remove the 0 lba_shift check in nvme_update_ns_info
Christoph Hellwig [Mon, 28 Sep 2020 10:12:02 +0000 (12:12 +0200)]
nvme: remove the 0 lba_shift check in nvme_update_ns_info

We can no longer reach this code if Identify Namespace failed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: clean up the check for too large logic block sizes
Christoph Hellwig [Mon, 28 Sep 2020 10:03:13 +0000 (12:03 +0200)]
nvme: clean up the check for too large logic block sizes

Use a single statement to set both the capacity and fake block size
instead of two.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
3 years agonvme: freeze the queue over ->lba_shift updates
Christoph Hellwig [Mon, 28 Sep 2020 10:11:42 +0000 (12:11 +0200)]
nvme: freeze the queue over ->lba_shift updates

Ensure that there can't be any I/O in flight went we change the disk
geometry in nvme_update_ns_info, most notable the LBA size by lifting
the queue free from nvme_update_disk_info into the caller

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: factor out a nvme_configure_metadata helper
Christoph Hellwig [Fri, 25 Sep 2020 05:19:13 +0000 (07:19 +0200)]
nvme: factor out a nvme_configure_metadata helper

Factor out a helper from nvme_update_ns_info that configures the
per-namespaces metadata and PI settings.  Also make sure the helpers
clear the flags explicitly instead of all of ->features to allow for
potentially reusing ->features for future non-metadata flags.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: call nvme_identify_ns as the first thing in nvme_alloc_ns_block
Christoph Hellwig [Mon, 28 Sep 2020 10:34:04 +0000 (12:34 +0200)]
nvme: call nvme_identify_ns as the first thing in nvme_alloc_ns_block

Check if the namespace actually exists as the very first thing and don't
bother with any extra work if not.  This should speed up and simplify
the sequential scanning for NVMe 1.0 devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: lift the check for an unallocated namespace into nvme_identify_ns
Christoph Hellwig [Mon, 28 Sep 2020 10:33:19 +0000 (12:33 +0200)]
nvme: lift the check for an unallocated namespace into nvme_identify_ns

Move the check from the two callers into the common helper.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
3 years agonvme: rename __nvme_revalidate_disk
Christoph Hellwig [Mon, 28 Sep 2020 10:14:20 +0000 (12:14 +0200)]
nvme: rename __nvme_revalidate_disk

Rename __nvme_revalidate_disk to nvme_update_ns_info and pass a
namespace instead of the gendisk.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: rename _nvme_revalidate_disk
Christoph Hellwig [Mon, 28 Sep 2020 08:25:54 +0000 (10:25 +0200)]
nvme: rename _nvme_revalidate_disk

Rename _nvme_revalidate_disk to nvme_validate_ns to better describe
what the function does, and pass the struct nvme_ns instead of the
gendisk to better match the call chain.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: rename nvme_validate_ns to nvme_validate_or_alloc_ns
Christoph Hellwig [Mon, 28 Sep 2020 07:42:17 +0000 (09:42 +0200)]
nvme: rename nvme_validate_ns to nvme_validate_or_alloc_ns

Use a slightly more descriptive name to enable reusing nvme_validate_ns
in the next patch for a lower level function.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: remove the disk argument to nvme_update_zone_info
Christoph Hellwig [Thu, 20 Aug 2020 12:02:18 +0000 (14:02 +0200)]
nvme: remove the disk argument to nvme_update_zone_info

The queue can trivially be derived from the nvme_ns structure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme: fix initialization of the zone bitmaps
Christoph Hellwig [Thu, 20 Aug 2020 07:31:36 +0000 (09:31 +0200)]
nvme: fix initialization of the zone bitmaps

The removal of the ->revalidate_disk method broke the initialization of
the zone bitmaps, as nvme_revalidate_disk now never gets called during
initialization.

Move the zone related code from nvme_revalidate_disk into a new helper in
zns.c, and call it from nvme_alloc_ns in addition to nvme_validate_ns to
ensure the zone bitmaps are initialized during probe.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agoblock: optimize blk_queue_zoned_model for !CONFIG_BLK_DEV_ZONED
Christoph Hellwig [Thu, 20 Aug 2020 07:56:58 +0000 (09:56 +0200)]
block: optimize blk_queue_zoned_model for !CONFIG_BLK_DEV_ZONED

Always return BLK_ZONED_NONE if zoned device support is not enabled.
This allows various compiler optimizations including the dead code
elimination that we so like for avoiding ifdefs.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvme-loop: don't put ctrl on nvme_init_ctrl error
Chaitanya Kulkarni [Wed, 30 Sep 2020 04:24:17 +0000 (21:24 -0700)]
nvme-loop: don't put ctrl on nvme_init_ctrl error

The function nvme_init_ctrl() gets the ctrl reference & when it fails it
does put the ctrl reference in the error unwind code.

When creating loop ctrl in nvme_loop_create_ctrl() if nvme_init_ctrl()
returns non zero (i.e. error) value it jumps to the "out_put_ctrl" label
which calls nvme_put_ctrl(), that will lead to douple ctrl put in error
unwind path.

Update nvme_loop_create_ctrl() such that this patch removes the
"out_put_ctrl" label, add a new "out" label after nvme_put_ctrl() in
error unwind path and jump to newly added label when nvme_init_ctrl()
call retuns an error.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agoscsi: megaraid_sas: Added support for shared host tagset for cpuhotplug
Kashyap Desai [Wed, 19 Aug 2020 15:20:35 +0000 (23:20 +0800)]
scsi: megaraid_sas: Added support for shared host tagset for cpuhotplug

Fusion adapters can steer completions to individual queues, and
we now have support for shared host-wide tags.
So we can enable multiqueue support for fusion adapters.

Once driver enable shared host-wide tags, cpu hotplug feature is also
supported as it was enabled using below patchsets -
commit bf0beec0607d ("blk-mq: drain I/O when all CPUs in a hctx are
offline")

Currently driver has provision to disable host-wide tags using
"host_tagset_enable" module parameter.

Once we do not have any major performance regression using host-wide
tags, we will drop the hand-crafted interrupt affinity settings.

Performance is also meeting the expecatation - (used both none and
mq-deadline scheduler)
24 Drive SSD on Aero with/without this patch can get 3.1M IOPs
3 VDs consist of 8 SAS SSD on Aero with/without this patch can get 3.1M
IOPs.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoscsi: scsi_debug: Support host tagset
John Garry [Wed, 19 Aug 2020 15:20:33 +0000 (23:20 +0800)]
scsi: scsi_debug: Support host tagset

When host_max_queue is set (> 0), set the Scsi_Host.host_tagset such that
blk-mq will use a hostwide tagset over all SCSI host submission queues.

This means that we may expose all submission queues and always use the hwq
chosen by blk-mq.

And since if sdebug_host_max_queue is set, sdebug_max_queue is fixed to the
same value, we can simplify how sdebug_driver_template.can_queue is set.

Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoscsi: hisi_sas: Switch v3 hw to MQ
John Garry [Wed, 19 Aug 2020 15:20:32 +0000 (23:20 +0800)]
scsi: hisi_sas: Switch v3 hw to MQ

Now that the block layer provides a shared tag, we can switch the driver
to expose all HW queues.

Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoscsi: core: Show nr_hw_queues in sysfs
John Garry [Wed, 19 Aug 2020 15:20:31 +0000 (23:20 +0800)]
scsi: core: Show nr_hw_queues in sysfs

So that we don't use a value of 0 for when Scsi_Host.nr_hw_queues is unset,
use the tag_set->nr_hw_queues (which holds 1 for this case).

Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoscsi: Add host and host template flag 'host_tagset'
Hannes Reinecke [Wed, 19 Aug 2020 15:20:30 +0000 (23:20 +0800)]
scsi: Add host and host template flag 'host_tagset'

Add Host and host template flag 'host_tagset' so hostwide tagset can be
shared on multiple reply queues after the SCSI device's reply queue is
converted to blk-mq hw queue.

[jpg: Update comment on .can_queue and add Scsi_Host.host_tagset]

Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Don Brace<don.brace@microsemi.com> #SCSI resv cmds patches used
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: scsi_ioctl: Avoid the use of one-element arrays
Gustavo A. R. Silva [Fri, 2 Oct 2020 23:10:33 +0000 (18:10 -0500)]
block: scsi_ioctl: Avoid the use of one-element arrays

One-element arrays are being deprecated[1]. Replace the one-element array
with a simple object of type compat_caddr_t: 'compat_caddr_t unused'[2],
once it seems this field is actually never used.

Also, update struct cdrom_generic_command in UAPI by adding an
anonimous union to avoid using the one-element array _reserved_.

[1] https://www.kernel.org/doc/html/v5.9-rc1/process/deprecated.html#zero-length-and-one-element-arrays
[2] https://github.com/KSPP/linux/issues/86

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://lore.kernel.org/lkml/5f76f5d0.qJ4t%2FHWuRzSW7bTa%25lkp@intel.com/
Build-tested-by: kernel test robot <lkp@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agorsxx: Use fallthrough pseudo-keyword
Gustavo A. R. Silva [Fri, 2 Oct 2020 23:59:28 +0000 (18:59 -0500)]
rsxx: Use fallthrough pseudo-keyword

Replace /* Fall through. */ comment with the new pseudo-keyword macro
fallthrough[1].

[1] https://www.kernel.org/doc/html/v5.7/process/deprecated.html?highlight=fallthrough#implicit-switch-case-fall-through

Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove embedded struct cache_sb from struct cache_set
Coly Li [Thu, 1 Oct 2020 06:50:56 +0000 (14:50 +0800)]
bcache: remove embedded struct cache_sb from struct cache_set

Since bcache code was merged into mainline kerrnel, each cache set only
as one single cache in it. The multiple caches framework is here but the
code is far from completed. Considering the multiple copies of cached
data can also be stored on e.g. md raid1 devices, it is unnecessary to
support multiple caches in one cache set indeed.

The previous preparation patches fix the dependencies of explicitly
making a cache set only have single cache. Now we don't have to maintain
an embedded partial super block in struct cache_set, the in-memory super
block can be directly referenced from struct cache.

This patch removes the embedded struct cache_sb from struct cache_set,
and fixes all locations where the superb lock was referenced from this
removed super block by referencing the in-memory super block of struct
cache.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: check and set sync status on cache's in-memory super block
Coly Li [Thu, 1 Oct 2020 06:50:55 +0000 (14:50 +0800)]
bcache: check and set sync status on cache's in-memory super block

Currently the cache's sync status is checked and set on cache set's in-
memory partial super block. After removing the embedded struct cache_sb
from cache set and reference cache's in-memory super block from struct
cache_set, the sync status can set and check directly on cache's super
block.

This patch checks and sets the cache sync status directly on cache's
in-memory super block. This is a preparation for later removing embedded
struct cache_sb from struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove can_attach_cache()
Coly Li [Thu, 1 Oct 2020 06:50:54 +0000 (14:50 +0800)]
bcache: remove can_attach_cache()

After removing the embedded struct cache_sb from struct cache_set, cache
set will directly reference the in-memory super block of struct cache.
It is unnecessary to compare block_size, bucket_size and nr_in_set from
the identical in-memory super block in can_attach_cache().

This is a preparation patch for latter removing cache_set->sb from
struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: don't check seq numbers in register_cache_set()
Coly Li [Thu, 1 Oct 2020 06:50:53 +0000 (14:50 +0800)]
bcache: don't check seq numbers in register_cache_set()

In order to update the partial super block of cache set, the seq numbers
of cache and cache set are checked in register_cache_set(). If cache's
seq number is larger than cache set's seq number, cache set must update
its partial super block from cache's super block. It is unncessary when
the embedded struct cache_sb is removed from struct cache set.

This patch removed the seq numbers checking from register_cache_set(),
because later there will be no such partial super block in struct cache
set, the cache set will directly reference in-memory super block from
struct cache. This is a preparation patch for removing embedded struct
cache_sb from struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: only use bucket_bytes() on struct cache
Coly Li [Thu, 1 Oct 2020 06:50:52 +0000 (14:50 +0800)]
bcache: only use bucket_bytes() on struct cache

Because struct cache_set and struct cache both have struct cache_sb,
macro bucket_bytes() currently are used on both of them. When removing
the embedded struct cache_sb from struct cache_set, this macro won't be
used on struct cache_set anymore.

This patch unifies all bucket_bytes() usage only on struct cache, this is
one of the preparation to remove the embedded struct cache_sb from
struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove useless bucket_pages()
Coly Li [Thu, 1 Oct 2020 06:50:51 +0000 (14:50 +0800)]
bcache: remove useless bucket_pages()

It seems alloc_bucket_pages() is the only user of bucket_pages().
Considering alloc_bucket_pages() is removed from bcache code, it is safe
to remove the useless macro bucket_pages() now.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove useless alloc_bucket_pages()
Coly Li [Thu, 1 Oct 2020 06:50:50 +0000 (14:50 +0800)]
bcache: remove useless alloc_bucket_pages()

Now no one uses alloc_bucket_pages() anymore, remove it from bcache.h.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: only use block_bytes() on struct cache
Coly Li [Thu, 1 Oct 2020 06:50:49 +0000 (14:50 +0800)]
bcache: only use block_bytes() on struct cache

Because struct cache_set and struct cache both have struct cache_sb,
therefore macro block_bytes() can be used on both of them. When removing
the embedded struct cache_sb from struct cache_set, this macro won't be
used on struct cache_set anymore.

This patch unifies all block_bytes() usage only on struct cache, this is
one of the preparation to remove the embedded struct cache_sb from
struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: add set_uuid in struct cache_set
Coly Li [Thu, 1 Oct 2020 06:50:48 +0000 (14:50 +0800)]
bcache: add set_uuid in struct cache_set

This patch adds a separated set_uuid[16] in struct cache_set, to store
the uuid of the cache set. This is the preparation to remove the
embedded struct cache_sb from struct cache_set.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove for_each_cache()
Coly Li [Thu, 1 Oct 2020 06:50:47 +0000 (14:50 +0800)]
bcache: remove for_each_cache()

Since now each cache_set explicitly has single cache, for_each_cache()
is unnecessary. This patch removes this macro, and update all locations
where it is used, and makes sure all code logic still being consistent.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: explicitly make cache_set only have single cache
Coly Li [Thu, 1 Oct 2020 06:50:46 +0000 (14:50 +0800)]
bcache: explicitly make cache_set only have single cache

Currently although the bcache code has a framework for multiple caches
in a cache set, but indeed the multiple caches never completed and users
use md raid1 for multiple copies of the cached data.

This patch does the following change in struct cache_set, to explicitly
make a cache_set only have single cache,
- Change pointer array "*cache[MAX_CACHES_PER_SET]" to a single pointer
  "*cache".
- Remove pointer array "*cache_by_alloc[MAX_CACHES_PER_SET]".
- Remove "caches_loaded".

Now the code looks as exactly what it does in practic: only one cache is
used in the cache set.

Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: remove 'int n' from parameter list of bch_bucket_alloc_set()
Coly Li [Thu, 1 Oct 2020 06:50:45 +0000 (14:50 +0800)]
bcache: remove 'int n' from parameter list of bch_bucket_alloc_set()

The parameter 'int n' from bch_bucket_alloc_set() is not cleared
defined. From the code comments n is the number of buckets to alloc, but
from the code itself 'n' is the maximum cache to iterate. Indeed all the
locations where bch_bucket_alloc_set() is called, 'n' is alwasy 1.

This patch removes the confused and unnecessary 'int n' from parameter
list of  bch_bucket_alloc_set(), and explicitly allocates only 1 bucket
for its caller.

Signed-off-by: Coly Li <colyli@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: Convert to DEFINE_SHOW_ATTRIBUTE
Qinglang Miao [Thu, 1 Oct 2020 06:50:44 +0000 (14:50 +0800)]
bcache: Convert to DEFINE_SHOW_ATTRIBUTE

Use DEFINE_SHOW_ATTRIBUTE macro to simplify the code.

As inode->iprivate equals to third parameter of
debugfs_create_file() which is NULL. So it's equivalent
to original code logic.

Signed-off-by: Qinglang Miao <miaoqinglang@huawei.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: check c->root with IS_ERR_OR_NULL() in mca_reserve()
Dongsheng Yang [Thu, 1 Oct 2020 06:50:43 +0000 (14:50 +0800)]
bcache: check c->root with IS_ERR_OR_NULL() in mca_reserve()

In mca_reserve(c) macro, we are checking root whether is NULL or not.
But that's not enough, when we read the root node in run_cache_set(),
if we got an error in bch_btree_node_read_done(), we will return
ERR_PTR(-EIO) to c->root.

And then we will go continue to unregister, but before calling
unregister_shrinker(&c->shrink), there is a possibility to call
bch_mca_count(), and we would get a crash with call trace like that:

[ 2149.876008] Unable to handle kernel NULL pointer dereference at virtual address 00000000000000b5
... ...
[ 2150.598931] Call trace:
[ 2150.606439]  bch_mca_count+0x58/0x98 [escache]
[ 2150.615866]  do_shrink_slab+0x54/0x310
[ 2150.624429]  shrink_slab+0x248/0x2d0
[ 2150.632633]  drop_slab_node+0x54/0x88
[ 2150.640746]  drop_slab+0x50/0x88
[ 2150.648228]  drop_caches_sysctl_handler+0xf0/0x118
[ 2150.657219]  proc_sys_call_handler.isra.18+0xb8/0x110
[ 2150.666342]  proc_sys_write+0x40/0x50
[ 2150.673889]  __vfs_write+0x48/0x90
[ 2150.681095]  vfs_write+0xac/0x1b8
[ 2150.688145]  ksys_write+0x6c/0xd0
[ 2150.695127]  __arm64_sys_write+0x24/0x30
[ 2150.702749]  el0_svc_handler+0xa0/0x128
[ 2150.710296]  el0_svc+0x8/0xc

Signed-off-by: Dongsheng Yang <dongsheng.yang@easystack.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: share register sysfs with async register
Coly Li [Thu, 1 Oct 2020 06:50:42 +0000 (14:50 +0800)]
bcache: share register sysfs with async register

Previously the experimental async registration uses a separate sysfs
file register_async. Now the async registration code seems working well
for a while, we can do furtuher testing with it now.

This patch changes the async bcache registration shares the same sysfs
file /sys/fs/bcache/register (and register_quiet). Async registration
will be default behavior if BCACHE_ASYNC_REGISTRATION is set in kernel
configure. By default, BCACHE_ASYNC_REGISTRATION is not configured yet.

Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agonull_blk: add support for max open/active zone limit for zoned devices
Niklas Cassel [Fri, 28 Aug 2020 10:54:00 +0000 (12:54 +0200)]
null_blk: add support for max open/active zone limit for zoned devices

Add support for user space to set a max open zone and a max active zone
limit via configfs. By default, the default values are 0 == no limit.

Call the block layer API functions used for exposing the configured
limits to sysfs.

Add accounting in null_blk_zoned so that these new limits are respected.
Performing an operation that would exceed these limits results in a
standard I/O error.

A max open zone limit exists in the ZBC standard.
While null_blk_zoned is used to test the Zoned Block Device model in
Linux, when it comes to differences between ZBC and ZNS, null_blk_zoned
mostly follows ZBC.

Therefore, implement the manage open zone resources function from ZBC,
but additionally add support for max active zones.
This enables user space not only to test against a device with an open
zone limit, but also to test against a device with an active zone limit.

Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge tag 'nvme-5.10-2020-09-27' of git://git.infradead.org/nvme into for-5.10/drivers
Jens Axboe [Mon, 28 Sep 2020 15:02:48 +0000 (09:02 -0600)]
Merge tag 'nvme-5.10-2020-09-27' of git://git.infradead.org/nvme into for-5.10/drivers

Pull NVMe updates from Christoph:

"nvme updates for 5.10

 - fix keep alive timer modification (Amit Engel)
 - order the PCI ID list more sensibly (Andy Shevchenko)
 - cleanup the open by controller helper (Chaitanya Kulkarni)
 - use an xarray for th CSE log lookup (Chaitanya Kulkarni)
 - support ZNS in nvmet passthrough mode (Chaitanya Kulkarni)
 - fix nvme_ns_report_zones (me)
 - add a sanity check to nvmet-fc (James Smart)
 - fix interrupt allocation when too many polled queues are specified
   (Jeffle Xu)
 - small nvmet-tcp optimization (Mark Wunderlich)"

* tag 'nvme-5.10-2020-09-27' of git://git.infradead.org/nvme:
  nvme-pci: allocate separate interrupt for the reserved non-polled I/O queue
  nvme: fix error handling in nvme_ns_report_zones
  nvmet-fc: fix missing check for no hostport struct
  nvmet: add passthru ZNS support
  nvmet: handle keep-alive timer when kato is modified by a set features cmd
  nvmet-tcp: have queue io_work context run on sock incoming cpu
  nvme-pci: Move enumeration by class to be last in the table
  nvme: use an xarray to lookup the Commands Supported and Effects log
  nvme: lift the file open code from nvme_ctrl_get_by_path

3 years agonvme-pci: allocate separate interrupt for the reserved non-polled I/O queue
Jeffle Xu [Thu, 24 Sep 2020 07:01:22 +0000 (09:01 +0200)]
nvme-pci: allocate separate interrupt for the reserved non-polled I/O queue

One queue will be reserved for non-polled IO when nvme.poll_queues is
greater or equal than the number of IO queues that the nvme controller
can provide. Currently the reserved queue for non-polled IO will reuse
the interrupt used by admin queue in this case, e.g, vector 0.

This can work and the performance may not be an issue since the admin
queue is used unfrequently. However this behaviour may be inconsistent
with that when nvme.poll_queues is smaller than the number of IO
queues available.

Thus allocate separate interrupt for this reserved queue, and thus make
the behaviour consistent.

Signed-off-by: Jeffle Xu <jefflexu@linux.alibaba.com>
[hch: minor cleanups, mostly to the pre-existing surrounding code]
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme: fix error handling in nvme_ns_report_zones
Christoph Hellwig [Sun, 30 Aug 2020 10:00:10 +0000 (12:00 +0200)]
nvme: fix error handling in nvme_ns_report_zones

nvme_submit_sync_cmd can return positive NVMe error codes in addition to
the negative Linux error code, which are currently ignored.  Fix this
by removing __nvme_ns_report_zones and handling the errors from
nvme_submit_sync_cmd in the caller instead of multiplexing the return
value and the number of zones reported into a single return value.

Fixes: 240e6ee272c0 ("nvme: support for zoned namespaces")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com>
3 years agonvmet-fc: fix missing check for no hostport struct
James Smart [Tue, 22 Sep 2020 21:54:12 +0000 (14:54 -0700)]
nvmet-fc: fix missing check for no hostport struct

A hostport port pointer is allowed to be NULL as it is not allocated if
the lldd does not support the new interfaces for NVME LS request support.
The hostport free routine validates the handle but forgot to validate the
hostport pointer.

Validate the hostport pointer before using it to validate the handle.

Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvmet: add passthru ZNS support
Chaitanya Kulkarni [Tue, 22 Sep 2020 21:17:18 +0000 (14:17 -0700)]
nvmet: add passthru ZNS support

In the default passthru implementation NVMeOF target passthru ctrl is
not capable of handling Zoned Namespaces (ZNS).

Update the nvmet_parse_pasthru_admin_cmd() to allow
NVME_ID_CNS_CS_CTRL/NVME_CIS_ZNS and NVME_ID_CNS_CS_NS/NVME_CIS_ZNS.

With this addition NVMeOF Passthru allows Zoned Namespaces.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvmet: handle keep-alive timer when kato is modified by a set features cmd
Amit Engel [Wed, 16 Sep 2020 17:47:20 +0000 (20:47 +0300)]
nvmet: handle keep-alive timer when kato is modified by a set features cmd

A user may modify the kato by a set features cmd.  To properly deal
with races or a kato value of 0 (no keep alive enabled) change
nvmet_set_feat_kato to first disable the timer, then set the value
and then re-enable the timer.

Signed-off-by: Amit Engel <amit.engel@dell.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvmet-tcp: have queue io_work context run on sock incoming cpu
Mark Wunderlich [Fri, 28 Aug 2020 01:00:53 +0000 (01:00 +0000)]
nvmet-tcp: have queue io_work context run on sock incoming cpu

No real good need to spread queues artificially. Usually the
target will serve multiple hosts, and it's better to run on the socket
incoming cpu for better affinitization rather than spread queues on all
online cpus.

We rely on RSS to spread the work around sufficiently.

Signed-off-by: Mark Wunderlich <mark.wunderlich@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme-pci: Move enumeration by class to be last in the table
Andy Shevchenko [Tue, 18 Aug 2020 08:35:30 +0000 (11:35 +0300)]
nvme-pci: Move enumeration by class to be last in the table

It's unusual that we have enumeration by class in the middle of the table.
It might potentially be problematic in the future if we add another entry
after it.

So, move class matching entry to be the last in the ID table.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme: use an xarray to lookup the Commands Supported and Effects log
Chaitanya Kulkarni [Tue, 22 Sep 2020 21:05:29 +0000 (14:05 -0700)]
nvme: use an xarray to lookup the Commands Supported and Effects log

When using linked list we have to open code the locking, search, and
destroy operations with the loops even if data structure doesn't fall
into the fast path.

One of the main advantage of having XArray to store, search, and remove
items is that it handles all the locking by itself, avoids the loops
when using linked lists, provides clear API to replace the linked list's
search and destroy loops.

This patch replaces the ctrl->cel list with XArray and removes :-

a. Extra code needed for the linked list for ctrl->cel item management
   such as nvme_find_cel().
b. Destroy loop in the nvme_free_ctrl().
c. Explicit insertion locking in the nvme_get_effects_log().

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
3 years agonvme: lift the file open code from nvme_ctrl_get_by_path
Chaitanya Kulkarni [Thu, 17 Sep 2020 01:11:02 +0000 (18:11 -0700)]
nvme: lift the file open code from nvme_ctrl_get_by_path

Lift opening the file open/close code from nvme_ctrl_get_by_path into
the caller, just keeping a simple nvme_ctrl_from_file() helper.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
[hch: refactored a bit, split the bug fixes into a separate prep patch]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Logan Gunthorpe <logang@deltatee.com>
3 years agoMerge branch 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md...
Jens Axboe [Fri, 25 Sep 2020 13:48:20 +0000 (07:48 -0600)]
Merge branch 'md-next' of https://git./linux/kernel/git/song/md into for-5.10/drivers

Pull MD updates from Song.

* 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
  md/raid10: improve discard request for far layout
  md/raid10: improve raid10 discard request
  md/raid10: pull codes that wait for blocked dev into one function
  md/raid10: extend r10bio devs to raid disks
  md: add md_submit_discard_bio() for submitting discard bio
  md: Simplify code with existing definition RESYNC_SECTORS in raid10.c
  md/raid5: reallocate page array after setting new stripe_size
  md/raid5: resize stripe_head when reshape array
  md/raid5: let multiple devices of stripe_head share page
  md/raid6: let async recovery function support different page offset
  md/raid6: let syndrome computor support different page offset
  md/raid5: convert to new xor compution interface
  md/raid5: add new xor function to support different page offset
  md/raid5: make async_copy_data() to support different page offset
  md/raid5: add a new member of offset into r5dev
  md: only calculate blocksize once and use i_blocksize()

3 years agomd/raid10: improve discard request for far layout
Xiao Ni [Wed, 2 Sep 2020 12:00:23 +0000 (20:00 +0800)]
md/raid10: improve discard request for far layout

For far layout, the discard region is not continuous on disks. So it needs
far copies r10bio to cover all regions. It needs a way to know all r10bios
have finish or not. Similar with raid10_sync_request, only the first r10bio
master_bio records the discard bio. Other r10bios master_bio record the
first r10bio. The first r10bio can finish after other r10bios finish and
then return the discard bio.

Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid10: improve raid10 discard request
Xiao Ni [Wed, 2 Sep 2020 12:00:22 +0000 (20:00 +0800)]
md/raid10: improve raid10 discard request

Now the discard request is split by chunk size. So it takes a long time
to finish mkfs on disks which support discard function. This patch improve
handling raid10 discard request. It uses the similar way with patch
29efc390b (md/md0: optimize raid0 discard handling).

But it's a little complex than raid0. Because raid10 has different layout.
If raid10 is offset layout and the discard request is smaller than stripe
size. There are some holes when we submit discard bio to underlayer disks.

For example: five disks (disk1 - disk5)
D01 D02 D03 D04 D05
D05 D01 D02 D03 D04
D06 D07 D08 D09 D10
D10 D06 D07 D08 D09
The discard bio just wants to discard from D03 to D10. For disk3, there is
a hole between D03 and D08. For disk4, there is a hole between D04 and D09.
D03 is a chunk, raid10_write_request can handle one chunk perfectly. So
the part that is not aligned with stripe size is still handled by
raid10_write_request.

If reshape is running when discard bio comes and the discard bio spans the
reshape position, raid10_write_request is responsible to handle this
discard bio.

I did a test with this patch set.
Without patch:
time mkfs.xfs /dev/md0
real4m39.775s
user0m0.000s
sys0m0.298s

With patch:
time mkfs.xfs /dev/md0
real0m0.105s
user0m0.000s
sys0m0.007s

nvme3n1           259:1    0   477G  0 disk
└─nvme3n1p1       259:10   0    50G  0 part
nvme4n1           259:2    0   477G  0 disk
└─nvme4n1p1       259:11   0    50G  0 part
nvme5n1           259:6    0   477G  0 disk
└─nvme5n1p1       259:12   0    50G  0 part
nvme2n1           259:9    0   477G  0 disk
└─nvme2n1p1       259:15   0    50G  0 part
nvme0n1           259:13   0   477G  0 disk
└─nvme0n1p1       259:14   0    50G  0 part

Reviewed-by: Coly Li <colyli@suse.de>
Reviewed-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid10: pull codes that wait for blocked dev into one function
Xiao Ni [Tue, 25 Aug 2020 05:43:01 +0000 (13:43 +0800)]
md/raid10: pull codes that wait for blocked dev into one function

The following patch will reuse these logics, so pull the same codes into
one function.

Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid10: extend r10bio devs to raid disks
Xiao Ni [Tue, 25 Aug 2020 05:43:00 +0000 (13:43 +0800)]
md/raid10: extend r10bio devs to raid disks

Now it allocs r10bio->devs[conf->copies]. Discard bio needs to submit
to all member disks and it needs to use r10bio. So extend to
r10bio->devs[geo.raid_disks].

Reviewed-by: Coly Li <colyli@suse.de>
Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd: add md_submit_discard_bio() for submitting discard bio
Xiao Ni [Tue, 25 Aug 2020 05:42:59 +0000 (13:42 +0800)]
md: add md_submit_discard_bio() for submitting discard bio

Move these logic from raid0.c to md.c, so that we can also use it in
raid10.c.

Reviewed-by: Coly Li <colyli@suse.de>
Reviewed-by: Guoqing Jiang <guoqing.jiang@cloud.ionos.com>
Signed-off-by: Xiao Ni <xni@redhat.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd: Simplify code with existing definition RESYNC_SECTORS in raid10.c
Zhen Lei [Fri, 21 Aug 2020 01:29:18 +0000 (09:29 +0800)]
md: Simplify code with existing definition RESYNC_SECTORS in raid10.c

#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)

"RESYNC_BLOCK_SIZE/512" is equal to "RESYNC_BLOCK_SIZE >> 9", replace it
with RESYNC_SECTORS.

Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: reallocate page array after setting new stripe_size
Yufen Yu [Thu, 20 Aug 2020 13:22:14 +0000 (09:22 -0400)]
md/raid5: reallocate page array after setting new stripe_size

When try to resize stripe_size, we also need to free old
shared page array and allocate new.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: resize stripe_head when reshape array
Yufen Yu [Thu, 20 Aug 2020 13:22:13 +0000 (09:22 -0400)]
md/raid5: resize stripe_head when reshape array

When reshape array, we try to reuse shared pages of old stripe_head,
and allocate more for the new one if needed.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: let multiple devices of stripe_head share page
Yufen Yu [Thu, 20 Aug 2020 13:22:12 +0000 (09:22 -0400)]
md/raid5: let multiple devices of stripe_head share page

In current implementation, grow_buffers() uses alloc_page() to
allocate the buffers for each stripe_head, i.e. allocate a page
for each dev[i] in stripe_head.

After setting stripe_size as a configurable value by writing
sysfs entry, it means that we always allocate 64K buffers, but
just use 4K of them when stripe_size is 4K in 64KB arm64.

To avoid wasting memory, we try to let multiple sh->dev share
one real page. That means, multiple sh->dev[i].page will point
to the only page with different offset. Example of 64K PAGE_SIZE
and 4K stripe_size as following:

                    64K PAGE_SIZE
          +---+---+---+---+------------------------------+
          |   |   |   |   |
          |   |   |   |   |
          +-+-+-+-+-+-+-+-+------------------------------+
            ^   ^   ^   ^
            |   |   |   +----------------------------+
            |   |   |                                |
            |   |   +-------------------+            |
            |   |                       |            |
            |   +----------+            |            |
            |              |            |            |
            +-+            |            |            |
              |            |            |            |
        +-----+-----+------+-----+------+-----+------+------+
sh      | offset(0) | offset(4K) | offset(8K) | offset(12K) |
 +      +-----------+------------+------------+-------------+
 +----> dev[0].page  dev[1].page  dev[2].page  dev[3].page

A new 'pages' array will be added into stripe_head to record shared
page used by this stripe_head. Allocate them when grow_buffers()
and free them when shrink_buffers().

After trying to share page, the users of sh->dev[i].page need to take
care of the related page offset: page of issued bio and page passed
to xor compution functions. But thanks for previous different page offset
supported. Here, we just need to set correct dev[i].offset.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid6: let async recovery function support different page offset
Yufen Yu [Thu, 20 Aug 2020 13:22:11 +0000 (09:22 -0400)]
md/raid6: let async recovery function support different page offset

For now, asynchronous raid6 recovery calculate functions are require
common offset for pages. But, we expect them to support different page
offset after introducing stripe shared page. Do that by simplily adding
page offset where each page address are referred. Then, replace the
old interface with the new ones in raid6 and raid6test.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid6: let syndrome computor support different page offset
Yufen Yu [Thu, 20 Aug 2020 13:22:10 +0000 (09:22 -0400)]
md/raid6: let syndrome computor support different page offset

For now, syndrome compute functions require common offset in the pages
array. However, we expect them to support different offset when try to
use shared page in the following. Simplily covert them by adding page
offset where each page address are referred.

Since the only caller of async_gen_syndrome() and async_syndrome_val()
are in raid6, we don't want to reserve the old interface but modify the
interface directly. After that, replacing old interfaces with new ones
for raid6 and raid6test.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: convert to new xor compution interface
Yufen Yu [Thu, 20 Aug 2020 13:22:09 +0000 (09:22 -0400)]
md/raid5: convert to new xor compution interface

We try to replace async_xor() and async_xor_val() with the new
introduced interface async_xor_offs() and async_xor_val_offs()
for raid456.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: add new xor function to support different page offset
Yufen Yu [Thu, 20 Aug 2020 13:22:08 +0000 (09:22 -0400)]
md/raid5: add new xor function to support different page offset

raid5 will call async_xor() and async_xor_val() to compute xor.
For now, both of them require the common src/dst page offset. But,
we want them to support different src/dst page offset for following
shared page.

Here, adding two new function async_xor_offs() and async_xor_val_offs()
respectively for async_xor() and async_xor_val().

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: make async_copy_data() to support different page offset
Yufen Yu [Thu, 20 Aug 2020 13:22:07 +0000 (09:22 -0400)]
md/raid5: make async_copy_data() to support different page offset

ops_run_biofill() and ops_run_biodrain() will call async_copy_data()
to copy sh->dev[i].page from or to bio page. For now, it implies the
offset of dev[i].page is 0. But we want to support different page offset
in the following.

Thus, pass page offset to these functions and replace 'page_offset'
with 'page_offset + poff'.

No functional change.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd/raid5: add a new member of offset into r5dev
Yufen Yu [Thu, 20 Aug 2020 13:22:06 +0000 (09:22 -0400)]
md/raid5: add a new member of offset into r5dev

Add a new member of offset into struct r5dev. It indicates the
offset of related dev[i].page. For now, since each device have a
privated page, the value is always 0. Thus, we set offset as 0
when allcate page in grow_buffers() and resize_stripes().

To support following different page offset, we try to use the page
offset rather than '0' directly for async_memcpy() and ops_run_io().

We try to support different page offset for xor compution functions
in the following. To avoid repeatly allocate a new array each time,
we add a memory region into scribble buffer to record offset.

No functional change.

Signed-off-by: Yufen Yu <yuyufen@huawei.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agomd: only calculate blocksize once and use i_blocksize()
Xianting Tian [Tue, 18 Aug 2020 05:42:06 +0000 (13:42 +0800)]
md: only calculate blocksize once and use i_blocksize()

We alreday has the interface i_blocksize(), which can be used
to get blocksize, so use it.
Only calculate blocksize once and use it within read_page().

Signed-off-by: Xianting Tian <tian.xianting@h3c.com>
Signed-off-by: Song Liu <songliubraving@fb.com>
3 years agonull_blk: Support shared tag bitmap
John Garry [Wed, 19 Aug 2020 15:20:29 +0000 (23:20 +0800)]
null_blk: Support shared tag bitmap

Support a shared tag bitmap, whereby request tags are unique over all
submission queues, and not just per submission queue.

As such, per device total queue depth is normally hw_queue_depth *
submit_queues, but hw_queue_depth when set. And a similar story for when
shared_tags is set, where that is the queue depth over all null blk
devices.

Signed-off-by: John Garry <john.garry@huawei.com>
Tested-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoMerge branch 'for-5.10/block' into for-5.10/drivers
Jens Axboe [Thu, 24 Sep 2020 19:44:39 +0000 (13:44 -0600)]
Merge branch 'for-5.10/block' into for-5.10/drivers

* for-5.10/block: (140 commits)
  bdi: replace BDI_CAP_NO_{WRITEBACK,ACCT_DIRTY} with a single flag
  bdi: invert BDI_CAP_NO_ACCT_WB
  bdi: replace BDI_CAP_STABLE_WRITES with a queue and a sb flag
  mm: use SWP_SYNCHRONOUS_IO more intelligently
  bdi: remove BDI_CAP_SYNCHRONOUS_IO
  bdi: remove BDI_CAP_CGROUP_WRITEBACK
  block: lift setting the readahead size into the block layer
  md: update the optimal I/O size on reshape
  bdi: initialize ->ra_pages and ->io_pages in bdi_init
  aoe: set an optimal I/O size
  bcache: inherit the optimal I/O size
  drbd: remove dead code in device_to_statistics
  fs: remove the unused SB_I_MULTIROOT flag
  block: mark blkdev_get static
  PM: mm: cleanup swsusp_swap_check
  mm: split swap_type_of
  PM: rewrite is_hibernate_resume_dev to not require an inode
  mm: cleanup claim_swapfile
  ocfs2: cleanup o2hb_region_dev_store
  dasd: cleanup dasd_scan_partitions
  ...

3 years agobdi: replace BDI_CAP_NO_{WRITEBACK,ACCT_DIRTY} with a single flag
Christoph Hellwig [Thu, 24 Sep 2020 06:51:40 +0000 (08:51 +0200)]
bdi: replace BDI_CAP_NO_{WRITEBACK,ACCT_DIRTY} with a single flag

Replace the two negative flags that are always used together with a
single positive flag that indicates the writeback capability instead
of two related non-capabilities.  Also remove the pointless wrappers
to just check the flag.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobdi: invert BDI_CAP_NO_ACCT_WB
Christoph Hellwig [Thu, 24 Sep 2020 06:51:39 +0000 (08:51 +0200)]
bdi: invert BDI_CAP_NO_ACCT_WB

Replace BDI_CAP_NO_ACCT_WB with a positive BDI_CAP_WRITEBACK_ACCT to
make the checks more obvious.  Also remove the pointless
bdi_cap_account_writeback wrapper that just obsfucates the check.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobdi: replace BDI_CAP_STABLE_WRITES with a queue and a sb flag
Christoph Hellwig [Thu, 24 Sep 2020 06:51:38 +0000 (08:51 +0200)]
bdi: replace BDI_CAP_STABLE_WRITES with a queue and a sb flag

The BDI_CAP_STABLE_WRITES is one of the few bits of information in the
backing_dev_info shared between the block drivers and the writeback code.
To help untangling the dependency replace it with a queue flag and a
superblock flag derived from it.  This also helps with the case of e.g.
a file system requiring stable writes due to its own checksumming, but
not forcing it on other users of the block device like the swap code.

One downside is that we an't support the stable_pages_required bdi
attribute in sysfs anymore.  It is replaced with a queue attribute which
also is writable for easier testing.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agomm: use SWP_SYNCHRONOUS_IO more intelligently
Christoph Hellwig [Thu, 24 Sep 2020 06:51:37 +0000 (08:51 +0200)]
mm: use SWP_SYNCHRONOUS_IO more intelligently

There is no point in trying to call bdev_read_page if SWP_SYNCHRONOUS_IO
is not set, as the device won't support it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobdi: remove BDI_CAP_SYNCHRONOUS_IO
Christoph Hellwig [Thu, 24 Sep 2020 06:51:36 +0000 (08:51 +0200)]
bdi: remove BDI_CAP_SYNCHRONOUS_IO

BDI_CAP_SYNCHRONOUS_IO is only checked in the swap code, and used to
decided if ->rw_page can be used on a block device.  Just check up for
the method instead.  The only complication is that zram needs a second
set of block_device_operations as it can switch between modes that
actually support ->rw_page and those who don't.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobdi: remove BDI_CAP_CGROUP_WRITEBACK
Christoph Hellwig [Thu, 24 Sep 2020 06:51:35 +0000 (08:51 +0200)]
bdi: remove BDI_CAP_CGROUP_WRITEBACK

Just checking SB_I_CGROUPWB for cgroup writeback support is enough.
Either the file system allocates its own bdi (e.g. btrfs), in which case
it is known to support cgroup writeback, or the bdi comes from the block
layer, which always supports cgroup writeback.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: lift setting the readahead size into the block layer
Christoph Hellwig [Thu, 24 Sep 2020 06:51:34 +0000 (08:51 +0200)]
block: lift setting the readahead size into the block layer

Drivers shouldn't really mess with the readahead size, as that is a VM
concept.  Instead set it based on the optimal I/O size by lifting the
algorithm from the md driver when registering the disk.  Also set
bdi->io_pages there as well by applying the same scheme based on
max_sectors.  To ensure the limits work well for stacking drivers a
new helper is added to update the readahead limits from the block
limits, which is also called from disk_stack_limits.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agomd: update the optimal I/O size on reshape
Christoph Hellwig [Thu, 24 Sep 2020 06:51:33 +0000 (08:51 +0200)]
md: update the optimal I/O size on reshape

The raid5 and raid10 drivers currently update the read-ahead size,
but not the optimal I/O size on reshape.  To prepare for deriving the
read-ahead size from the optimal I/O size make sure it is updated
as well.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Song Liu <song@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobdi: initialize ->ra_pages and ->io_pages in bdi_init
Christoph Hellwig [Thu, 24 Sep 2020 06:51:32 +0000 (08:51 +0200)]
bdi: initialize ->ra_pages and ->io_pages in bdi_init

Set up a readahead size by default, as very few users have a good
reason to change it.  This means code, ecryptfs, and orangefs now
set up the values while they were previously missing it, while ubifs,
mtd and vboxsf manually set it to 0 to avoid readahead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: David Sterba <dsterba@suse.com> [btrfs]
Acked-by: Richard Weinberger <richard@nod.at> [ubifs, mtd]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoaoe: set an optimal I/O size
Christoph Hellwig [Thu, 24 Sep 2020 06:51:31 +0000 (08:51 +0200)]
aoe: set an optimal I/O size

aoe forces a larger readahead size, but any reason to do larger I/O
is not limited to readahead.  Also set the optimal I/O size, and
remove the local constants in favor of just using SZ_2G.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agobcache: inherit the optimal I/O size
Christoph Hellwig [Thu, 24 Sep 2020 06:51:30 +0000 (08:51 +0200)]
bcache: inherit the optimal I/O size

Inherit the optimal I/O size setting just like the readahead window,
as any reason to do larger I/O does not apply to just readahead.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agodrbd: remove dead code in device_to_statistics
Christoph Hellwig [Thu, 24 Sep 2020 06:51:29 +0000 (08:51 +0200)]
drbd: remove dead code in device_to_statistics

Ever since the switch to blk-mq, a lower device not used for VM
writeback will not be marked congested, so the check will never
trigger.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agofs: remove the unused SB_I_MULTIROOT flag
Christoph Hellwig [Thu, 24 Sep 2020 06:51:28 +0000 (08:51 +0200)]
fs: remove the unused SB_I_MULTIROOT flag

The last user of SB_I_MULTIROOT is disappeared with commit f2aedb713c28
("NFS: Add fs_context support.")

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: mark blkdev_get static
Christoph Hellwig [Mon, 21 Sep 2020 07:19:58 +0000 (09:19 +0200)]
block: mark blkdev_get static

There are no users outside the core block code left now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoPM: mm: cleanup swsusp_swap_check
Christoph Hellwig [Mon, 21 Sep 2020 07:19:57 +0000 (09:19 +0200)]
PM: mm: cleanup swsusp_swap_check

Use blkdev_get_by_dev instead of bdget + blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agomm: split swap_type_of
Christoph Hellwig [Mon, 21 Sep 2020 07:19:56 +0000 (09:19 +0200)]
mm: split swap_type_of

swap_type_of is used for two entirely different purposes:

 (1) check what swap type a given device/offset corresponds to
 (2) find the first available swap device that can be written to

Mixing both in a single function creates an unreadable mess.  Create two
separate functions instead, and switch both to pass a dev_t instead of
a struct block_device to further simplify the code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoPM: rewrite is_hibernate_resume_dev to not require an inode
Christoph Hellwig [Mon, 21 Sep 2020 07:19:55 +0000 (09:19 +0200)]
PM: rewrite is_hibernate_resume_dev to not require an inode

Just check the dev_t to help simplifying the code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agomm: cleanup claim_swapfile
Christoph Hellwig [Mon, 21 Sep 2020 07:19:54 +0000 (09:19 +0200)]
mm: cleanup claim_swapfile

Use blkdev_get_by_dev instead of bdgrab + blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoocfs2: cleanup o2hb_region_dev_store
Christoph Hellwig [Mon, 21 Sep 2020 07:19:53 +0000 (09:19 +0200)]
ocfs2: cleanup o2hb_region_dev_store

Use blkdev_get_by_dev instead of igrab (aka open coded bdgrab) +
blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agodasd: cleanup dasd_scan_partitions
Christoph Hellwig [Mon, 21 Sep 2020 07:19:52 +0000 (09:19 +0200)]
dasd: cleanup dasd_scan_partitions

Use blkdev_get_by_dev instead of bdget_disk + blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Stefan Haberland <sth@linux.ibm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoraw: don't keep unopened block device around
Christoph Hellwig [Mon, 21 Sep 2020 07:19:51 +0000 (09:19 +0200)]
raw: don't keep unopened block device around

Turn binding into a normal dev_t as the struct block device doesn't
buy us anything and use blkdev_open_by_dev to actually open it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agozram: cleanup backing_dev_store
Christoph Hellwig [Mon, 21 Sep 2020 07:19:50 +0000 (09:19 +0200)]
zram: cleanup backing_dev_store

Use blkdev_get_by_dev instead of bdgrab + blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agopktcdvd: use blkdev_get_by_dev instead of open coding it
Christoph Hellwig [Mon, 21 Sep 2020 07:19:49 +0000 (09:19 +0200)]
pktcdvd: use blkdev_get_by_dev instead of open coding it

Replace bdget + blkdev_get by blkdev_get_by_dev.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agopktcdvd: remove the if 0'ed pkt_start_recovery function
Christoph Hellwig [Mon, 21 Sep 2020 07:19:48 +0000 (09:19 +0200)]
pktcdvd: remove the if 0'ed pkt_start_recovery function

Remove code which has been dead since the initial commit.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: cleanup blkdev_bszset
Christoph Hellwig [Mon, 21 Sep 2020 07:19:47 +0000 (09:19 +0200)]
block: cleanup blkdev_bszset

Use blkdev_get_by_dev instead of bdgrab + blkdev_get.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: cleanup partition scanning in register_disk
Christoph Hellwig [Mon, 21 Sep 2020 07:19:46 +0000 (09:19 +0200)]
block: cleanup partition scanning in register_disk

Use blkdev_get_by_dev instead of open coding it using bdget_disk +
blkdev_get, and split the code to read the partition table into a
separate helper to make it a little more obvious.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: move the NEED_PART_SCAN flag to struct gendisk
Christoph Hellwig [Mon, 21 Sep 2020 07:19:45 +0000 (09:19 +0200)]
block: move the NEED_PART_SCAN flag to struct gendisk

We can only scan for partitions on the whole disk, so move the flag
from struct block_device to struct gendisk.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: allow 'chunk_sectors' to be non-power-of-2
Mike Snitzer [Tue, 22 Sep 2020 02:32:49 +0000 (22:32 -0400)]
block: allow 'chunk_sectors' to be non-power-of-2

It is possible, albeit more unlikely, for a block device to have a non
power-of-2 for chunk_sectors (e.g. 10+2 RAID6 with 128K chunk_sectors,
which results in a full-stripe size of 1280K. This causes the RAID6's
io_opt to be advertised as 1280K, and a stacked device _could_ then be
made to use a blocksize, aka chunk_sectors, that matches non power-of-2
io_opt of underlying RAID6 -- resulting in stacked device's
chunk_sectors being a non power-of-2).

Update blk_queue_chunk_sectors() and blk_max_size_offset() to
accommodate drivers that need a non power-of-2 chunk_sectors.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 years agoblock: use lcm_not_zero() when stacking chunk_sectors
Mike Snitzer [Tue, 22 Sep 2020 02:32:48 +0000 (22:32 -0400)]
block: use lcm_not_zero() when stacking chunk_sectors

Like 'io_opt', blk_stack_limits() should stack 'chunk_sectors' using
lcm_not_zero() rather than min_not_zero() -- otherwise the final
'chunk_sectors' could result in sub-optimal alignment of IO to
component devices in the IO stack.

Also, if 'chunk_sectors' isn't a multiple of 'physical_block_size'
then it is a bug in the driver and the device should be flagged as
'misaligned'.

Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>