openwrt/staging/blogic.git
6 years agoaoe: mark expected switch fall-through
Gustavo A. R. Silva [Wed, 1 Aug 2018 22:30:20 +0000 (17:30 -0500)]
aoe: mark expected switch fall-through

In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.

Addresses-Coverity-ID: 114722 ("Missing break in switch")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: make iolatency avg_lat exponentially decay
Dennis Zhou (Facebook) [Thu, 2 Aug 2018 06:15:41 +0000 (23:15 -0700)]
block: make iolatency avg_lat exponentially decay

Currently, avg_lat is calculated by accumulating the mean of every
window in a long running cumulative average. As time goes on, the metric
becomes less and less useful due to the accumulated history.

This patch reuses the same calculation done in load averages to make the
avg_lat metric more lively. Unlike load averages, the avg only advances
when a window elapses (due to an io). Idle periods extend the most
recent window. Bucketing is used to limit the history of avg_lat by
binding it to the window size. So, the window range for 1/exp (decay
rate) is [1 min, 2.5 min) when windows elapse immediately.

The current sample window size is exposed in the debug info to enable
calculation of the window range.

Signed-off-by: Dennis Zhou <dennisszhou@gmail.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-cgroup: clear the throttle queue on fork
Josef Bacik [Tue, 31 Jul 2018 16:39:04 +0000 (12:39 -0400)]
blk-cgroup: clear the throttle queue on fork

We were hitting a panic in production where we put too many times on the
request queue.  This is because we'd get the throttle_queue of the
parent if we fork()'ed while we needed to be throttled, but we didn't
have a reference on it.  Instead just clear these flags on fork so the
child doesn't pay for the sins of its father.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-cgroup: hold the queue ref during throttling
Josef Bacik [Tue, 31 Jul 2018 16:39:03 +0000 (12:39 -0400)]
blk-cgroup: hold the queue ref during throttling

The blkg lifetime is protected by the queue lifetime, so we need to put
the queue _after_ we're done using the blkg.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: fix blkg leak in timer_fn
Josef Bacik [Tue, 31 Jul 2018 16:39:02 +0000 (12:39 -0400)]
blk-iolatency: fix blkg leak in timer_fn

At this point we have a ref on the blkg, we need to drop it if we don't
have a iolat.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock/bsg-lib: use PTR_ERR_OR_ZERO to simplify the flow path
zhong jiang [Tue, 31 Jul 2018 16:13:14 +0000 (00:13 +0800)]
block/bsg-lib: use PTR_ERR_OR_ZERO to simplify the flow path

Simplify the code by using the PTR_ERR_OR_ZERO, instead of the
open code. It is better.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: zhong jiang <zhongjiang@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agot10-pi: provide empty t10_pi_complete() for !CONFIG_BLK_DEV_INTEGRITY
Jens Axboe [Tue, 31 Jul 2018 15:10:26 +0000 (09:10 -0600)]
t10-pi: provide empty t10_pi_complete() for !CONFIG_BLK_DEV_INTEGRITY

Fixes a link failure whtn BLK_DEV_INTEGRITY isn't defined.

Fixes: 10c41ddd6132 ("block: move dif_prepare/dif_complete functions to block layer")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: blk_init_allocated_queue() set q->fq as NULL in the fail case
xiao jin [Mon, 30 Jul 2018 06:11:12 +0000 (14:11 +0800)]
block: blk_init_allocated_queue() set q->fq as NULL in the fail case

We find the memory use-after-free issue in __blk_drain_queue()
on the kernel 4.14. After read the latest kernel 4.18-rc6 we
think it has the same problem.

Memory is allocated for q->fq in the blk_init_allocated_queue().
If the elevator init function called with error return, it will
run into the fail case to free the q->fq.

Then the __blk_drain_queue() uses the same memory after the free
of the q->fq, it will lead to the unpredictable event.

The patch is to set q->fq as NULL in the fail case of
blk_init_allocated_queue().

Fixes: commit 7c94e1c157a2 ("block: introduce blk_flush_queue to drive flush machinery")
Cc: <stable@vger.kernel.org>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: xiao jin <jin.xiao@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonvme: use blk API to remap ref tags for IOs with metadata
Max Gurtovoy [Sun, 29 Jul 2018 21:15:33 +0000 (00:15 +0300)]
nvme: use blk API to remap ref tags for IOs with metadata

Also moved the logic of the remapping to the nvme core driver instead
of implementing it in the nvme pci driver. This way all the other nvme
transport drivers will benefit from it (in case they'll implement metadata
support).

Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Acked-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: move dif_prepare/dif_complete functions to block layer
Max Gurtovoy [Sun, 29 Jul 2018 21:15:32 +0000 (00:15 +0300)]
block: move dif_prepare/dif_complete functions to block layer

Currently these functions are implemented in the scsi layer, but their
actual place should be the block layer since T10-PI is a general data
integrity feature that is used in the nvme protocol as well. Also, use
the tuple size from the integrity profile since it may vary between
integrity types.

Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: move ref_tag calculation func to the block layer
Max Gurtovoy [Sun, 29 Jul 2018 21:15:31 +0000 (00:15 +0300)]
block: move ref_tag calculation func to the block layer

Currently this function is implemented in the scsi layer, but it's
actual place should be the block layer since T10-PI is a general
data integrity feature that is used in the nvme protocol as well.

Suggested-by: Christoph Hellwig <hch@lst.de>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: don't account for split bio's size in cgroup stats
Josef Bacik [Mon, 30 Jul 2018 14:10:01 +0000 (10:10 -0400)]
block: don't account for split bio's size in cgroup stats

We need to check in blkcg_bio_issue_check if the bio is flagged as
QUEUE_ENTERED, because if it is then we've already accounted for the
size of the IO in the cgroup stats.  We can still however account for
the extra IO since it'll be another request.

Reported-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agopktcdvd: Fix possible Spectre-v1 for pkt_devs
Jinbum Park [Sat, 28 Jul 2018 04:20:44 +0000 (13:20 +0900)]
pktcdvd: Fix possible Spectre-v1 for pkt_devs

User controls @dev_minor which to be used as index of pkt_devs.
So, It can be exploited via Spectre-like attack. (speculative execution)

This kind of attack leaks address of pkt_devs, [1]
It leads an attacker to bypass security mechanism such as KASLR.

So sanitize @dev_minor before using it to prevent attack.

[1] https://github.com/jinb-park/linux-exploit/
tree/master/exploit-remaining-spectre-gadget/leak_pkt_devs.c

Signed-off-by: Jinbum Park <jinb.park7@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agopartitions/aix: append null character to print data from disk
Mauricio Faria de Oliveira [Thu, 26 Jul 2018 01:46:29 +0000 (22:46 -0300)]
partitions/aix: append null character to print data from disk

Even if properly initialized, the lvname array (i.e., strings)
is read from disk, and might contain corrupt data (e.g., lack
the null terminating character for strings).

So, make sure the partition name string used in pr_warn() has
the null terminating character.

Fixes: 6ceea22bbbc8 ("partitions: add aix lvm partition support files")
Suggested-by: Daniel J. Axtens <daniel.axtens@canonical.com>
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agopartitions/aix: fix usage of uninitialized lv_info and lvname structures
Mauricio Faria de Oliveira [Thu, 26 Jul 2018 01:46:28 +0000 (22:46 -0300)]
partitions/aix: fix usage of uninitialized lv_info and lvname structures

The if-block that sets a successful return value in aix_partition()
uses 'lvip[].pps_per_lv' and 'n[].name' potentially uninitialized.

For example, if 'numlvs' is zero or alloc_lvn() fails, neither is
initialized, but are used anyway if alloc_pvd() succeeds after it.

So, make the alloc_pvd() call conditional on their initialization.

This has been hit when attaching an apparently corrupted/stressed
AIX LUN, misleading the kernel to pr_warn() invalid data and hang.

    [...] partition (null) (11 pp's found) is not contiguous
    [...] partition (null) (2 pp's found) is not contiguous
    [...] partition (null) (3 pp's found) is not contiguous
    [...] partition (null) (64 pp's found) is not contiguous

Fixes: 6ceea22bbbc8 ("partitions: add aix lvm partition support files")
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: stop using the deprecated get_seconds()
Arnd Bergmann [Thu, 26 Jul 2018 04:17:41 +0000 (12:17 +0800)]
bcache: stop using the deprecated get_seconds()

The get_seconds function is deprecated now since it returns a 32-bit
value that will eventually overflow, and we are replacing it throughout
the kernel with ktime_get_seconds() or ktime_get_real_seconds() that
return a time64_t.

bcache uses get_seconds() to read the current system time and store it in
the superblock as well as in uuid_entry structures that are user visible.

Unfortunately, the two structures in are still limited to 32 bits, so this
won't fix any real problems but will still overflow in year 2106. Let's
at least document that properly, in case we get an updated format in the
future it can be fixed. We still have a long time before the overflow
and checking the tools at https://github.com/koverstreet/bcache-tools
reveals no access to any of them.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: do not assign in if condition in bcache_device_init()
Florian Schmaus [Thu, 26 Jul 2018 04:17:40 +0000 (12:17 +0800)]
bcache: do not assign in if condition in bcache_device_init()

Fixes an error condition reported by checkpatch.pl which is caused by
assigning a variable in an if condition.

Signed-off-by: Florian Schmaus <flo@geekplace.eu>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: do not assign in if condition in bcache_init()
Florian Schmaus [Thu, 26 Jul 2018 04:17:39 +0000 (12:17 +0800)]
bcache: do not assign in if condition in bcache_init()

Fixes an error condition reported by checkpatch.pl which is caused by
assigning a variable in an if condition.

Signed-off-by: Florian Schmaus <flo@geekplace.eu>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: free heap cache_set->flush_btree in bch_journal_free
Shenghui Wang [Thu, 26 Jul 2018 04:17:38 +0000 (12:17 +0800)]
bcache: free heap cache_set->flush_btree in bch_journal_free

Free the cache_set->flush_bree heap memory on journal free.

Signed-off-by: Wang Sheng-Hui <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: do not assign in if condition register_bcache()
Florian Schmaus [Thu, 26 Jul 2018 04:17:37 +0000 (12:17 +0800)]
bcache: do not assign in if condition register_bcache()

Fixes an error condition reported by checkpatch.pl which is caused by
assigning a variable in an if condition.

Signed-off-by: Florian Schmaus <flo@geekplace.eu>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: fix I/O significant decline while backend devices registering
Tang Junhui [Thu, 26 Jul 2018 04:17:36 +0000 (12:17 +0800)]
bcache: fix I/O significant decline while backend devices registering

I attached several backend devices in the same cache set, and produced lots
of dirty data by running small rand I/O writes in a long time, then I
continue run I/O in the others cached devices, and stopped a cached device,
after a mean while, I register the stopped device again, I see the running
I/O in the others cached devices dropped significantly, sometimes even
jumps to zero.

In currently code, bcache would traverse each keys and btree node to count
the dirty data under read locker, and the writes threads can not get the
btree write locker, and when there is a lot of keys and btree node in the
registering device, it would last several seconds, so the write I/Os in
others cached device are blocked and declined significantly.

In this patch, when a device registering to a ache set, which exist others
cached devices with running I/Os, we get the amount of dirty data of the
device in an incremental way, and do not block other cached devices all the
time.

Patch v2: Rename some variables and macros name as Coly suggested.

Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: calculate the number of incremental GC nodes according to the total of btree...
Tang Junhui [Thu, 26 Jul 2018 04:17:35 +0000 (12:17 +0800)]
bcache: calculate the number of incremental GC nodes according to the total of btree nodes

This patch base on "[PATCH] bcache: finish incremental GC".

Since incremental GC would stop 100ms when front side I/O comes, so when
there are many btree nodes, if GC only processes constant (100) nodes each
time, GC would last a long time, and the front I/Os would run out of the
buckets (since no new bucket can be allocated during GC), and I/Os be
blocked again.

So GC should not process constant nodes, but varied nodes according to the
number of btree nodes. In this patch, GC is divided into constant (100)
times, so when there are many btree nodes, GC can process more nodes each
time, otherwise GC will process less nodes each time (but no less than
MIN_GC_NODES).

Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: finish incremental GC
Tang Junhui [Thu, 26 Jul 2018 04:17:34 +0000 (12:17 +0800)]
bcache: finish incremental GC

In GC thread, we record the latest GC key in gc_done, which is expected
to be used for incremental GC, but in currently code, we didn't realize
it. When GC runs, front side IO would be blocked until the GC over, it
would be a long time if there is a lot of btree nodes.

This patch realizes incremental GC, the main ideal is that, when there
are front side I/Os, after GC some nodes (100), we stop GC, release locker
of the btree node, and go to process the front side I/Os for some times
(100 ms), then go back to GC again.

By this patch, when we doing GC, I/Os are not blocked all the time, and
there is no obvious I/Os zero jump problem any more.

Patch v2: Rename some variables and macros name as Coly suggested.

Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: simplify the calculation of the total amount of flash dirty data
Tang Junhui [Thu, 26 Jul 2018 04:17:33 +0000 (12:17 +0800)]
bcache: simplify the calculation of the total amount of flash dirty data

Currently we calculate the total amount of flash only devices dirty data
by adding the dirty data of each flash only device under registering
locker. It is very inefficient.

In this patch, we add a member flash_dev_dirty_sectors in struct cache_set
to record the total amount of flash only devices dirty data in real time,
so we didn't need to calculate the total amount of dirty data any more.

Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoreadahead: stricter check for bdi io_pages
Markus Stockhausen [Fri, 27 Jul 2018 15:09:53 +0000 (09:09 -0600)]
readahead: stricter check for bdi io_pages

ondemand_readahead() checks bdi->io_pages to cap the maximum pages
that need to be processed. This works until the readit section. If
we would do an async only readahead (async size = sync size) and
target is at beginning of window we expand the pages by another
get_next_ra_size() pages. Btrace for large reads shows that kernel
always issues a doubled size read at the beginning of processing.
Add an additional check for io_pages in the lower part of the func.
The fix helps devices that hard limit bio pages and rely on proper
handling of max_hw_read_sectors (e.g. older FusionIO cards). For
that reason it could qualify for stable.

Fixes: 9491ae4a ("mm: don't cap request size based on read-ahead setting")
Cc: stable@vger.kernel.org
Signed-off-by: Markus Stockhausen stockhausen@collogia.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoscsi: virtio_scsi: fix pi_bytes{out,in} on 4 KiB block size devices
Greg Edwards [Thu, 26 Jul 2018 19:52:54 +0000 (15:52 -0400)]
scsi: virtio_scsi: fix pi_bytes{out,in} on 4 KiB block size devices

When the underlying device is a 4 KiB logical block size device with a
protection interval exponent of 0, i.e. 4096 bytes data + 8 bytes PI, the
driver miscalculates the pi_bytes{out,in} by a factor of 8x (64 bytes).

This leads to errors on all reads and writes on 4 KiB logical block size
devices when CONFIG_BLK_DEV_INTEGRITY is enabled and the
VIRTIO_SCSI_F_T10_PI feature bit has been negotiated.

Fixes: e6dc783a38ec0 ("virtio-scsi: Enable DIF/DIX modes in SCSI host LLD")
Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Edwards <gedwards@ddn.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: move bio_integrity_{intervals,bytes} into blkdev.h
Greg Edwards [Wed, 25 Jul 2018 14:22:58 +0000 (10:22 -0400)]
block: move bio_integrity_{intervals,bytes} into blkdev.h

This allows bio_integrity_bytes() to be called from drivers instead of
open coding it.

Acked-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Greg Edwards <gedwards@ddn.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoxen/blkfront: remove unused macros
Juergen Gross [Wed, 25 Jul 2018 07:42:07 +0000 (09:42 +0200)]
xen/blkfront: remove unused macros

Remove some macros not used anywhere.

Acked-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoMerge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-4.19/block
Jens Axboe [Wed, 25 Jul 2018 14:43:30 +0000 (08:43 -0600)]
Merge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-4.19/block

Pull NVMe updates from Christoph:

"Highlights:

 - massively improved tracepoints (Keith Busch)
 - support for larger inline data in the RDMA host and target
   (Steve Wise)
 - RDMA setup/teardown path fixes and refactor (Sagi Grimberg)
 - Command Supported and Effects log support for the NVMe target
   (Chaitanya Kulkarni)
 - buffered I/O support for the NVMe target (Chaitanya Kulkarni)

 plus the usual set of cleanups and small enhancements."

* 'nvme-4.19' of git://git.infradead.org/nvme:
  nvmet: don't use uuid_le type
  nvmet: check fileio lba range access boundaries
  nvmet: fix file discard return status
  nvme-rdma: centralize admin/io queue teardown sequence
  nvme-rdma: centralize controller setup sequence
  nvme-rdma: unquiesce queues when deleting the controller
  nvme-rdma: mark expected switch fall-through
  nvme: add disk name to trace events
  nvme: add controller name to trace events
  nvme: use hw qid in trace events
  nvme: cache struct nvme_ctrl reference to struct nvme_request
  nvmet-rdma: add an error flow for post_recv failures
  nvmet-rdma: add unlikely check in the fast path
  nvmet-rdma: support max(16KB, PAGE_SIZE) inline data
  nvme-rdma: support up to 4 segments of inline data
  nvmet: add buffered I/O support for file backed ns
  nvmet: add commands supported and effects log page
  nvme: move init of keep_alive work item to controller initialization
  nvme.h: resync with nvme-cli

6 years agoblock: allow max_discard_segments to be stacked
Mike Snitzer [Fri, 20 Jul 2018 18:57:38 +0000 (14:57 -0400)]
block: allow max_discard_segments to be stacked

Set max_discard_segments to USHRT_MAX in blk_set_stacking_limits() so
that blk_stack_limits() can stack up this limit for stacked devices.

before:

$ cat /sys/block/nvme0n1/queue/max_discard_segments
256
$ cat /sys/block/dm-0/queue/max_discard_segments
1

after:

$ cat /sys/block/nvme0n1/queue/max_discard_segments
256
$ cat /sys/block/dm-0/queue/max_discard_segments
256

Fixes: 1e739730c5b9e ("block: optionally merge discontiguous discard bios into a single request")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: unexport bio_clone_bioset
Christoph Hellwig [Tue, 24 Jul 2018 07:52:34 +0000 (09:52 +0200)]
block: unexport bio_clone_bioset

Now only used by the bounce code, so move it there and mark the function
static.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agomd: remove a bogus comment
Christoph Hellwig [Tue, 24 Jul 2018 07:52:33 +0000 (09:52 +0200)]
md: remove a bogus comment

The function name mentioned doesn't exist, and the code next to it
doesn't match the description either.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: remove bio_clone_kmalloc
Christoph Hellwig [Tue, 24 Jul 2018 07:52:32 +0000 (09:52 +0200)]
block: remove bio_clone_kmalloc

Unused now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoexofs: use bio_clone_fast in _write_mirror
Christoph Hellwig [Tue, 24 Jul 2018 07:52:31 +0000 (09:52 +0200)]
exofs: use bio_clone_fast in _write_mirror

The mirroring code never changes the bio data or biovecs.  This means
we can reuse the biovec allocation easily instead of duplicating it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by Boaz Harrosh <ooo@electrozaur.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: don't clone bio in bch_data_verify
Christoph Hellwig [Tue, 24 Jul 2018 07:52:30 +0000 (09:52 +0200)]
bcache: don't clone bio in bch_data_verify

We immediately overwrite the biovec array, so instead just allocate
a new bio and copy over the disk, setor and size.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: bio_set_pages_dirty can't see NULL bv_page in a valid bio_vec
Christoph Hellwig [Tue, 24 Jul 2018 12:04:13 +0000 (14:04 +0200)]
block: bio_set_pages_dirty can't see NULL bv_page in a valid bio_vec

So don't bother handling it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: simplify bio_check_pages_dirty
Christoph Hellwig [Tue, 24 Jul 2018 12:04:12 +0000 (14:04 +0200)]
block: simplify bio_check_pages_dirty

bio_check_pages_dirty currently inviolates the invariant that bv_page of
a bio_vec inside bi_vcnt shouldn't be zero, and that is going to become
really annoying with multpath biovecs.  Fortunately there isn't any
all that good reason for it - once we decide to defer freeing the bio
to a workqueue holding onto a few additional pages isn't really an
issue anymore.  So just check if there is a clean page that needs
dirtying in the first path, and do a second pass to free them if there
was none, while the cache is still hot.

Also use the chance to micro-optimize bio_dirty_fn a bit by not saving
irq state - we know we are called from a workqueue.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Rename the null_blk_mod kernel module back into null_blk
Bart Van Assche [Mon, 23 Jul 2018 21:18:33 +0000 (14:18 -0700)]
block: Rename the null_blk_mod kernel module back into null_blk

Commit ca4b2a011948 ("null_blk: add zone support") breaks several
blktests scripts because it renamed the null_blk kernel module into
null_blk_mod. Hence rename null_blk_mod back into null_blk.

Fixes: ca4b2a011948 ("null_blk: add zone support")
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Matias Bjorling <matias.bjorling@wdc.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonvmet: don't use uuid_le type
Andy Shevchenko [Tue, 17 Jul 2018 14:17:36 +0000 (17:17 +0300)]
nvmet: don't use uuid_le type

Don't use sizeof(uuid_le) where none of the parameters is type of uuid_le.
Since both arguments are u8 [16], use size of destination there.

Moreover, uuid_le is a deprecated type, and nvmet is using uuid_t
already.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet: check fileio lba range access boundaries
Sagi Grimberg [Wed, 11 Jul 2018 13:13:04 +0000 (16:13 +0300)]
nvmet: check fileio lba range access boundaries

Fail out-of-bounds with a proper status code.

Fixes: d5eff33ee6f8 ("nvmet: add simple file backed ns support")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet: fix file discard return status
Sagi Grimberg [Wed, 11 Jul 2018 09:43:16 +0000 (12:43 +0300)]
nvmet: fix file discard return status

If nvmet_copy_from_sgl failed, we falsly return successful
completion status.

Fixes: d5eff33ee6f8 ("nvmet: add simple file backed ns support")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-rdma: centralize admin/io queue teardown sequence
Sagi Grimberg [Mon, 9 Jul 2018 09:49:07 +0000 (12:49 +0300)]
nvme-rdma: centralize admin/io queue teardown sequence

We follow the same queue teardown sequence in delete, reset and error
recovery. Centralize the logic.  This patch does not change any
functionality.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-rdma: centralize controller setup sequence
Sagi Grimberg [Mon, 9 Jul 2018 09:49:06 +0000 (12:49 +0300)]
nvme-rdma: centralize controller setup sequence

Centralize controller sequence to a single routine that correctly cleans
up after failures instead of having multiple apperances in several flows
(create, reset, reconnect).

One thing that we also gain here are the sanity/boundary checks also
when connecting back to a dynamic controller.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-rdma: unquiesce queues when deleting the controller
Sagi Grimberg [Mon, 9 Jul 2018 09:49:05 +0000 (12:49 +0300)]
nvme-rdma: unquiesce queues when deleting the controller

If the controller is going away, we need to unquiesce the IO queues so
that all pending request can fail gracefully before moving forward with
controller deletion. Do that before we destroy the IO queues so
blk_cleanup_queue won't block in freeze.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-rdma: mark expected switch fall-through
Gustavo A. R. Silva [Thu, 5 Jul 2018 13:14:00 +0000 (08:14 -0500)]
nvme-rdma: mark expected switch fall-through

In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: add disk name to trace events
Keith Busch [Fri, 29 Jun 2018 22:50:03 +0000 (16:50 -0600)]
nvme: add disk name to trace events

This will print the disk name to the nvme event trace for io requests so
a user can better distinguish traffic to different disks. This can be used
to  create disk based filters. For example, to see only nvme0n2 traffic:

  echo "disk == \"nvme0n2\"" > /sys/kernel/debug/tracing/events/nvme/filter

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: turned __assign_disk_name into an inline function]
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: add controller name to trace events
Keith Busch [Mon, 2 Jul 2018 15:15:03 +0000 (09:15 -0600)]
nvme: add controller name to trace events

This appends the controller instance to the nvme trace buffer to
distinguish which controller is dispatching and completing a command.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: use hw qid in trace events
Keith Busch [Fri, 29 Jun 2018 22:50:01 +0000 (16:50 -0600)]
nvme: use hw qid in trace events

We can not match a command to its completion based on the command
id alone. We need the submitting queue identifier to pair with the
completion, so this patch adds that to the trace buffer.

This patch is also collapsing the admin and IO submission traces into a
single one so we don't need to duplicate this and creating unnecessary
code branches: we know if the command is an admin vs IO based on the qid.

And since we're here, the patch fixes code formatting in the area.

Signed-off-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
[hch: move the qid helper to nvme.h and made it an inline function]
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: cache struct nvme_ctrl reference to struct nvme_request
Sagi Grimberg [Fri, 29 Jun 2018 22:50:00 +0000 (16:50 -0600)]
nvme: cache struct nvme_ctrl reference to struct nvme_request

We will need to reference the controller in the setup and completion
time for tracing and future traffic based keep alive support.

Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet-rdma: add an error flow for post_recv failures
Max Gurtovoy [Sun, 1 Jul 2018 09:20:24 +0000 (12:20 +0300)]
nvmet-rdma: add an error flow for post_recv failures

Posting receive buffer operation can fail, thus we should make
sure to have an error flow during initialization phase. While
we're here, add a debug print in case of a failure.

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet-rdma: add unlikely check in the fast path
Max Gurtovoy [Wed, 27 Jun 2018 11:58:02 +0000 (14:58 +0300)]
nvmet-rdma: add unlikely check in the fast path

ib_post_send operation should succeed unless something unusual
happened to the ib device.

Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet-rdma: support max(16KB, PAGE_SIZE) inline data
Steve Wise [Wed, 20 Jun 2018 14:15:10 +0000 (07:15 -0700)]
nvmet-rdma: support max(16KB, PAGE_SIZE) inline data

The patch enables inline data sizes using up to 4 recv sges, and capping
the size at 16KB or at least 1 page size.  So on a 4K page system, up to
16KB is supported, and for a 64K page system 1 page of 64KB is supported.

We avoid > 0 order page allocations for the inline buffers by using
multiple recv sges, one for each page.  If the device cannot support
the configured inline data size due to lack of enough recv sges, then
log a warning and reduce the inline size.

Add a new configfs port attribute, called param_inline_data_size,
to allow configuring the size of inline data for a given nvmf port.
The maximum size allowed is still enforced by nvmet-rdma with
NVMET_RDMA_MAX_INLINE_DATA_SIZE, which is now max(16KB, PAGE_SIZE).
And the default size, if not specified via configfs, is still PAGE_SIZE.
This preserves the existing behavior, but allows larger inline sizes
for small page systems.  If the configured inline data size exceeds
NVMET_RDMA_MAX_INLINE_DATA_SIZE, a warning is logged and the size is
reduced.  If param_inline_data_size is set to 0, then inline data is
disabled for that nvmf port.

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-rdma: support up to 4 segments of inline data
Steve Wise [Wed, 20 Jun 2018 14:15:05 +0000 (07:15 -0700)]
nvme-rdma: support up to 4 segments of inline data

Allow up to 4 segments of inline data for NVMF WRITE operations. This
reduces latency for small WRITEs by removing the need for the target to
issue a READ WR for IB, or a REG_MR + READ WR chain for iWarp.

Also cap the inline segments used based on the limitations of the
device.

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet: add buffered I/O support for file backed ns
Chaitanya Kulkarni [Wed, 20 Jun 2018 04:01:41 +0000 (00:01 -0400)]
nvmet: add buffered I/O support for file backed ns

Add a new "buffered_io" attribute, which disabled direct I/O and thus
enables page cache based caching when enabled.   The attribute can only
be changed when the namespace is disabled as the file has to be reopend
for the change to take effect.

The possibly blocking read/write are deferred to a newly introduced
global workqueue.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet: add commands supported and effects log page
Chaitanya Kulkarni [Mon, 11 Jun 2018 17:40:07 +0000 (13:40 -0400)]
nvmet: add commands supported and effects log page

This patch adds support for Commands Supported and Effects log page
(Log Identifier 05h) for NVMeOF. This also makes it easier to find
which commands are supported, e.g. :-

subnqn    : testnqn1
Admin Command Set
ACS2     [Get Log Page                    ] 00000001
ACS6     [Identify                        ] 00000001
ACS8     [Abort                           ] 00000001
ACS9     [Set Features                    ] 00000001
ACS10    [Get Features                    ] 00000001
ACS12    [Asynchronous Event Request      ] 00000001
ACS24    [Keep Alive                      ] 00000001

NVM Command Set
IOCS0    [Flush                           ] 00000001
IOCS1    [Write                           ] 00000001
IOCS2    [Read                            ] 00000001
IOCS8    [Write Zeroes                    ] 00000001
IOCS9    [Dataset Management              ] 00000001

This partticular functionality can be used from the host side to examine
the NVMeOF ctrl commands supported.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: move init of keep_alive work item to controller initialization
James Smart [Tue, 12 Jun 2018 23:28:24 +0000 (16:28 -0700)]
nvme: move init of keep_alive work item to controller initialization

Currently, the code initializes the keep alive work item whenever
nvme_start_keep_alive() is called. However, this routine is called
several times while reconnecting, etc. Although it's hoped that keep
alive is always disabled and not scheduled when start is called,
re-initing if it were scheduled or completing can have very bad
side effects. There's no need for re-initialization.

Move the keep_alive work item and cmd struct initialization to
controller init.

Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme.h: resync with nvme-cli
Revanth Rajashekar [Fri, 15 Jun 2018 18:39:27 +0000 (12:39 -0600)]
nvme.h: resync with nvme-cli

Added some feature ids present in nvme-cli but not kernel.

Signed-off-by: Revanth Rajashekar <revanth.rajashekar@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agoblk-mq: fail the request in case issue failure
Ming Lei [Sun, 22 Jul 2018 06:10:15 +0000 (14:10 +0800)]
blk-mq: fail the request in case issue failure

Inside blk_mq_try_issue_list_directly(), if the request is issued as
failed, we shouldn't try to do it again, otherwise the warning in
blk_mq_start_request() will be triggered. This change is aligned to
behaviour of other ways of request issue & dispatch.

Fixes: 6ce3dd6eec1 ("blk-mq: issue directly if hw queue isn't busy in case of 'none'")
Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: Laurence Oberman <loberman@redhat.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: LKP <lkp@01.org>
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-rq-qos: make depth comparisons unsigned
Josef Bacik [Fri, 20 Jul 2018 01:42:13 +0000 (21:42 -0400)]
blk-rq-qos: make depth comparisons unsigned

With the change to use UINT_MAX I broke the depth check as any value of
inflight (ie 0) would be less than (int)UINT_MAX.  Fix this by changing
everything to unsigned int to match the depth.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblkcg: Track DISCARD statistics and output them in cgroup io.stat
Tejun Heo [Wed, 18 Jul 2018 11:47:41 +0000 (04:47 -0700)]
blkcg: Track DISCARD statistics and output them in cgroup io.stat

Add tracking of REQ_OP_DISCARD ios to the per-cgroup io.stat.  Two
fields, dbytes and dios, to respectively count the total bytes and
number of discards are added.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Cc: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Track DISCARD statistics and output them in stat and diskstat
Michael Callahan [Wed, 18 Jul 2018 11:47:40 +0000 (04:47 -0700)]
block: Track DISCARD statistics and output them in stat and diskstat

Add tracking of REQ_OP_DISCARD ios to the partition statistics and
append them to the various stat files in /sys as well as
/proc/diskstats.  These are tracked with the same four stats as reads
and writes:

Number of discard ios completed.
Number of discard ios merged
Number of discard sectors completed
Milliseconds spent on discard requests

This is done via adding a new STAT_DISCARD define to genhd.h and then
using it to index that stat field for discard requests.

tj: Refreshed on top of v4.17 and other previous updates.

Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Add and use op_stat_group() for indexing disk_stat fields.
Michael Callahan [Wed, 18 Jul 2018 11:47:39 +0000 (04:47 -0700)]
block: Add and use op_stat_group() for indexing disk_stat fields.

Add and use a new op_stat_group() function for indexing partition stat
fields rather than indexing them by rq_data_dir() or bio_data_dir().
This function works similarly to op_is_sync() in that it takes the
request::cmd_flags or bio::bi_opf flags and determines which stats
should et updated.

In addition, the second parameter to generic_start_io_acct() and
generic_end_io_acct() is now a REQ_OP rather than simply a read or
write bit and it uses op_stat_group() on the parameter to determine
the stat group.

Note that the partition in_flight counts are not part of the per-cpu
statistics and as such are not indexed via this function.  It's now
indexed by op_is_write().

tj: Refreshed on top of v4.17.  Updated to pass around REQ_OP.

Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Joshua Morris <josh.h.morris@us.ibm.com>
Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Matias Bjorling <mb@lightnvm.io>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Cc: Alasdair Kergon <agk@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Define and use STAT_READ and STAT_WRITE
Michael Callahan [Wed, 18 Jul 2018 11:47:38 +0000 (04:47 -0700)]
block: Define and use STAT_READ and STAT_WRITE

Add defines for STAT_READ and STAT_WRITE for indexing the partition
stat entries. This clarifies some fs/ code which has hardcoded 1 for
STAT_WRITE and will make it easier to extend the stats with additional
fields.

tj: Refreshed on top of v4.17.

Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Add part_stat_read_accum to read across field entries.
Michael Callahan [Wed, 18 Jul 2018 11:47:37 +0000 (04:47 -0700)]
block: Add part_stat_read_accum to read across field entries.

Add a part_stat_read_accum macro to genhd.h to read and sum across
field entries.  For example to sum up the number read and write
sectors completed.  In addition to being ar reasonable cleanup by
itself this will make it easier to add new stat fields in the future.

tj: Refreshed on top of v4.17.

Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: make bdev_ops->rw_page() take a REQ_OP instead of bool
Tejun Heo [Wed, 18 Jul 2018 11:47:36 +0000 (04:47 -0700)]
block: make bdev_ops->rw_page() take a REQ_OP instead of bool

c11f0c0b5bb9 ("block/mm: make bdev_ops->rw_page() take a bool for
read/write") replaced @op with boolean @is_write, which limited the
amount of information going into ->rw_page() and more importantly
page_endio(), which removed the need to expose block internals to mm.

Unfortunately, we want to track discards separately and @is_write
isn't enough information.  This patch updates bdev_ops->rw_page() to
take REQ_OP instead but leaves page_endio() to take bool @is_write.
This allows the block part of operations to have enough information
while not leaking it to mm.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Mike Christie <mchristi@redhat.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agopktcdvd: remove assignment in if condition
RAGHU Halharvi [Tue, 17 Jul 2018 17:02:12 +0000 (22:32 +0530)]
pktcdvd: remove assignment in if condition

* Remove checkpatch errors caused due to assignment operation in if
condition

Signed-off-by: RAGHU Halharvi <raghuhack78@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-mq: issue directly if hw queue isn't busy in case of 'none'
Ming Lei [Tue, 10 Jul 2018 01:03:31 +0000 (09:03 +0800)]
blk-mq: issue directly if hw queue isn't busy in case of 'none'

In case of 'none' io scheduler, when hw queue isn't busy, it isn't
necessary to enqueue request to sw queue and dequeue it from
sw queue because request may be submitted to hw queue asap without
extra cost, meantime there shouldn't be much request in sw queue,
and we don't need to worry about effect on IO merge.

There are still some single hw queue SCSI HBAs(HPSA, megaraid_sas, ...)
which may connect high performance devices, so 'none' is often required
for obtaining good performance.

This patch improves IOPS and decreases CPU unilization on megaraid_sas,
per Kashyap's test.

Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: Laurence Oberman <loberman@redhat.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Reported-by: Kashyap Desai <kashyap.desai@broadcom.com>
Tested-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: truncate our current time
Josef Bacik [Mon, 16 Jul 2018 16:12:23 +0000 (12:12 -0400)]
blk-iolatency: truncate our current time

In our longer tests we noticed that some boxes would degrade to the
point of uselessness.  This is because we truncate the current time when
saving it in our bio, but I was using the raw current time to subtract
from.  So once the box had been up a certain amount of time it would
appear as if our IO's were taking several years to complete.  Fix this
by truncating the current time so it matches the issue time.  Verified
this worked by running with this patch for a week on our test tier.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: don't change the latency window
Josef Bacik [Mon, 16 Jul 2018 16:12:22 +0000 (12:12 -0400)]
blk-iolatency: don't change the latency window

Early versions of these patches had us waiting for seconds at a time
during submission, so we had to adjust the timing window we monitored
for latency.  Now we don't do things like that so this is unnecessary
code.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: assume that chunks are closed on 1.2 devices
Hans Holmberg [Fri, 13 Jul 2018 08:48:45 +0000 (10:48 +0200)]
lightnvm: pblk: assume that chunks are closed on 1.2 devices

We can't know if a block is closed or not on 1.2 devices, so assume
closed state to make sure that blocks are erased before writing.

Fixes: 32ef9412c114 ("lightnvm: pblk: implement get log report chunk")
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add asynchronous partial read
Heiner Litz [Fri, 13 Jul 2018 08:48:44 +0000 (10:48 +0200)]
lightnvm: pblk: add asynchronous partial read

In the read path, partial reads are currently performed synchronously
which affects performance for workloads that generate many partial
reads. This patch adds an asynchronous partial read path as well as
the required partial read ctx.

Signed-off-by: Heiner Litz <hlitz@ucsc.edu>
Reviewed-by: Igor Konopko <igor.j.konopko@intel.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: mark expected switch fall-through
Gustavo A. R. Silva [Fri, 13 Jul 2018 08:48:43 +0000 (10:48 +0200)]
lightnvm: pblk: mark expected switch fall-through

In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: expose generic disk name on pr_* msgs
Matias Bjørling [Fri, 13 Jul 2018 08:48:42 +0000 (10:48 +0200)]
lightnvm: pblk: expose generic disk name on pr_* msgs

The error messages in pblk does not say which pblk instance that
a message occurred from. Update each error message to reflect the
instance it belongs to, and also prefix it with pblk, so we know
the message comes from the pblk module.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: limit get chunk meta request size
Matias Bjørling [Fri, 13 Jul 2018 08:48:41 +0000 (10:48 +0200)]
lightnvm: limit get chunk meta request size

For devices that does not specify a limit on its transfer size, the
get_chk_meta command may send down a single I/O retrieving the full
chunk metadata table. Resulting in large 2-4MB I/O requests. Instead,
split up the I/Os to a maximum of 256KB and issue them separately to
reduce memory requirements.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix read_bitmap for 32bit archs
Matias Bjørling [Fri, 13 Jul 2018 08:48:40 +0000 (10:48 +0200)]
lightnvm: pblk: fix read_bitmap for 32bit archs

If using pblk on a 32bit architecture, and there is a need to
perform a partial read, the partial read bitmap will only have
allocated 32 entries, where as 64 are needed.

Make sure that the read_bitmap is initialized to 64bits on 32bit
architectures as well.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Igor Konopko <igor.j.konopko@intel.com>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: Remove redundant rq->__data_len initialization
Bart Van Assche [Fri, 13 Jul 2018 08:48:39 +0000 (10:48 +0200)]
lightnvm: Remove redundant rq->__data_len initialization

Since both blk_old_get_request() and blk_mq_alloc_request() initialize
rq->__data_len to zero, it is not necessary to initialize that member
in nvme_nvm_alloc_request(). Hence remove the rq->__data_len
initialization from nvme_nvm_alloc_request().

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: enable line minor version detection
Matias Bjørling [Fri, 13 Jul 2018 08:48:38 +0000 (10:48 +0200)]
lightnvm: pblk: enable line minor version detection

When recovering a line, an extra check was added when debugging was
active, such that minor version where also checked. Unfortunately,
this used the ifdef NVM_DEBUG, which is not correct.

Instead use the proper DEBUG def, and now that it compiles, also fix
the variable.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Fixes: d0ab0b1ab991f ("lightnvm: pblk: check data lines version on recovery")
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: move NVM_DEBUG to pblk
Matias Bjørling [Fri, 13 Jul 2018 08:48:37 +0000 (10:48 +0200)]
lightnvm: move NVM_DEBUG to pblk

There is no users of CONFIG_NVM_DEBUG in the LightNVM subsystem. All
users are in pblk. Rename NVM_DEBUG to NVM_PBLK_DEBUG and enable
only for pblk.

Also fix up the CONFIG_NVM_PBLK entry to follow the code style for
Kconfig files.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: handle case when mw_cunits equals to 0
Marcin Dziegielewski [Fri, 13 Jul 2018 08:48:36 +0000 (10:48 +0200)]
lightnvm: pblk: handle case when mw_cunits equals to 0

Some devices can expose mw_cunits equal to 0, it can cause the
creation of too small write buffer and cause performance to drop
on write workloads.

Additionally, write buffer size must cover write data requirements,
such as WS_MIN and MW_CUNITS - it must be greater than or equal to
the larger one multiplied by the number of PUs. However, for
performance reasons, use the WS_OPT value to calculation instead of
WS_MIN.

Because the place where buffer size is calculated was changed, this
patch also removes pgs_in_buffer filed in pblk structure.

Signed-off-by: Marcin Dziegielewski <marcin.dziegielewski@intel.com>
Signed-off-by: Igor Konopko <igor.j.konopko@intel.com>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: remove blkdev_entry_to_request() macro
Vladimir Zapolskiy [Fri, 13 Jul 2018 14:07:26 +0000 (17:07 +0300)]
block: remove blkdev_entry_to_request() macro

Remove blkdev_entry_to_request() macro, which remained unused through
the observable history, also note that it repeats list_entry_rq() macro
verbatim.

Signed-off-by: Vladimir Zapolskiy <vz@mleia.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: skd: Use %pad printk format for dma_addr_t values
Helge Deller [Thu, 12 Jul 2018 20:29:16 +0000 (22:29 +0200)]
block: skd: Use %pad printk format for dma_addr_t values

Use the existing %pad printk format to print dma_addr_t values.
This avoids the following warnings when compiling on the parisc64 platform:

drivers/block/skd_main.c: In function 'skd_preop_sg_list':
drivers/block/skd_main.c:660:4: warning: format '%llx' expects argument of type 'long long unsigned int', but argument 6 has type 'dma_addr_t {aka unsigned int}' [-Wformat=]

Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobsg: remove read/write support
Christoph Hellwig [Thu, 12 Jul 2018 08:09:59 +0000 (10:09 +0200)]
bsg: remove read/write support

The code poses a security risk due to user memory access in ->release
and had an API that can't be used reliably.  As far as we know it was
never used for real, but if that turns out wrong we'll have to revert
this commit and come up with a band aid.

Jann Horn did look software archives for users of this interface,
and the only users found were example code in sg3_utils, and optional
support in an optional module of the tgt user space iscsi target,
which looks like a proof of concept extension of the /dev/sg
read/write support.

Tony Battersby chimes in that the code is basically unsafe to use in
general:

  The read/write interface on /dev/bsg is impossible to use safely
  because the list of completed commands is per-device (bd->done_list)
  rather than per-fd like it is with /dev/sg.  So if program A and
  program B are both using the write/read interface on the same bsg
  device, then their command responses will get mixed up, and program
  A will read() some command results from program B and vice versa.
  So no, I don't use read/write on /dev/bsg.  From a security standpoint,
  it should definitely be fixed or removed.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: fix max_depth comparisons
Josef Bacik [Wed, 11 Jul 2018 14:34:42 +0000 (10:34 -0400)]
blk-iolatency: fix max_depth comparisons

max_depth used to be a u64, but I changed it to a unsigned int but
didn't convert my comparisons over everywhere.  Fix by using UINT_MAX
everywhere instead of (u64)-1.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: iolatency: avoid 64-bit division
Arnd Bergmann [Tue, 10 Jul 2018 15:21:34 +0000 (17:21 +0200)]
block: iolatency: avoid 64-bit division

On 32-bit architectures, dividing a 64-bit number needs to use the
do_div() function or something like it to avoid a link failure:

block/blk-iolatency.o: In function `iolatency_prfill_limit':
blk-iolatency.c:(.text+0x8cc): undefined reference to `__aeabi_uldivmod'

Using div_u64() gives us the best output and avoids the need for an
explicit cast.

Fixes: d70675121546 ("block: introduce blk-iolatency io controller")
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock/DAC960.c: fix defined but not used build warnings
Randy Dunlap [Sat, 7 Jul 2018 03:49:19 +0000 (20:49 -0700)]
block/DAC960.c: fix defined but not used build warnings

Fix build warnings in DAC960.c when CONFIG_PROC_FS is not enabled
by marking the unused functions as __maybe_unused.

../drivers/block/DAC960.c:6429:12: warning: 'dac960_proc_show' defined but not used [-Wunused-function]
../drivers/block/DAC960.c:6449:12: warning: 'dac960_initial_status_proc_show' defined but not used [-Wunused-function]
../drivers/block/DAC960.c:6456:12: warning: 'dac960_current_status_proc_show' defined but not used [-Wunused-function]

Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonull_blk: add zone support
Matias Bjørling [Fri, 6 Jul 2018 17:38:39 +0000 (19:38 +0200)]
null_blk: add zone support

Adds support for exposing a null_blk device through the zone device
interface.

The interface is managed with the parameters zoned and zone_size.
If zoned is set, the null_blk instance registers as a zoned block
device. The zone_size parameter defines how big each zone will be.

Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonull_blk: move shared definitions to header file
Matias Bjørling [Fri, 6 Jul 2018 17:38:38 +0000 (19:38 +0200)]
null_blk: move shared definitions to header file

Split the null_blk device driver, such that it can prepare for
zoned block interface support.

Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: Add default switch case to blk_pm_allow_request() to kill warning
Geert Uytterhoeven [Fri, 6 Jul 2018 08:49:35 +0000 (10:49 +0200)]
block: Add default switch case to blk_pm_allow_request() to kill warning

With gcc 4.9.0 and 7.3.0:

    block/blk-core.c: In function 'blk_pm_allow_request':
    block/blk-core.c:2747:2: warning: enumeration value 'RPM_ACTIVE' not handled in switch [-Wswitch]
      switch (rq->q->rpm_status) {
      ^

Convert the return statement below the switch() block into a default
case to fix this.

Fixes: e4f36b249b4d4e75 ("block: fix peeking requests during PM")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: fix infinite loop if the device loses discard capability
Mikulas Patocka [Tue, 3 Jul 2018 17:34:22 +0000 (13:34 -0400)]
block: fix infinite loop if the device loses discard capability

If __blkdev_issue_discard is in progress and a device mapper device is
reloaded with a table that doesn't support discard,
q->limits.max_discard_sectors is set to zero. This results in infinite
loop in __blkdev_issue_discard.

This patch checks if max_discard_sectors is zero and aborts with
-EOPNOTSUPP.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Tested-by: Zdenek Kabelac <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock, mm: remove unnecessary __GFP_HIGH flag
Shakeel Butt [Tue, 3 Jul 2018 17:14:46 +0000 (10:14 -0700)]
block, mm: remove unnecessary __GFP_HIGH flag

The flag GFP_ATOMIC already contains __GFP_HIGH. There is no need to
explicitly or __GFP_HIGH again. So, just remove unnecessary __GFP_HIGH.

Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonull_blk: remove NULLB_DEV_FL_CONFIGURED on turning off nullb device
Liu Bo [Thu, 5 Jul 2018 19:07:13 +0000 (03:07 +0800)]
null_blk: remove NULLB_DEV_FL_CONFIGURED on turning off nullb device

Currently mbps knob could only be set once before switching power knob to
on, after power knob has been set at least once, there is no way to set
mbps knob again due to -EBUSY.

As nullb is mainly used for testing, in order to make it flexible, this
removes the flag NULLB_DEV_FL_CONFIGURED so that mbps knob can be reset
when power knob is off, e.g.

echo 0 > /config/nullb/a/power
echo 40 > /config/nullb/a/mbps
echo 1 > /config/nullb/a/power

So does other knobs under /config/nullb/a.

Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agomm: skip readahead if the cgroup is congested
Josef Bacik [Tue, 3 Jul 2018 15:15:03 +0000 (11:15 -0400)]
mm: skip readahead if the cgroup is congested

We noticed in testing we'd get pretty bad latency stalls under heavy
pressure because read ahead would try to do its thing while the cgroup
was under severe pressure.  If we're under this much pressure we want to
do as little IO as possible so we can still make progress on real work
if we're a throttled cgroup, so just skip readahead if our group is
under pressure.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoDocumentation: add a doc for blk-iolatency
Josef Bacik [Tue, 3 Jul 2018 15:15:02 +0000 (11:15 -0400)]
Documentation: add a doc for blk-iolatency

A basic documentation to describe the interface, statistics, and
behavior of io.latency.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: introduce blk-iolatency io controller
Josef Bacik [Tue, 3 Jul 2018 15:15:01 +0000 (11:15 -0400)]
block: introduce blk-iolatency io controller

Current IO controllers for the block layer are less than ideal for our
use case.  The io.max controller is great at hard limiting, but it is
not work conserving.  This patch introduces io.latency.  You provide a
latency target for your group and we monitor the io in short windows to
make sure we are not exceeding those latency targets.  This makes use of
the rq-qos infrastructure and works much like the wbt stuff.  There are
a few differences from wbt

 - It's bio based, so the latency covers the whole block layer in addition to
   the actual io.
 - We will throttle all IO types that comes in here if we need to.
 - We use the mean latency over the 100ms window.  This is because writes can
   be particularly fast, which could give us a false sense of the impact of
   other workloads on our protected workload.
 - By default there's no throttling, we set the queue_depth to INT_MAX so that
   we can have as many outstanding bio's as we're allowed to.  Only at
   throttle time do we pay attention to the actual queue depth.
 - We backcharge cgroups for root cg issued IO and induce artificial
   delays in order to deal with cases like metadata only or swap heavy
   workloads.

In testing this has worked out relatively well.  Protected workloads
will throttle noisy workloads down to 1 io at time if they are doing
normal IO on their own, or induce up to a 1 second delay per syscall if
they are doing a lot of root issued IO (metadata/swap IO).

Our testing has revolved mostly around our production web servers where
we have hhvm (the web server application) in a protected group and
everything else in another group.  We see slightly higher requests per
second (RPS) on the test tier vs the control tier, and much more stable
RPS across all machines in the test tier vs the control tier.

Another test we run is a slow memory allocator in the unprotected group.
Before this would eventually push us into swap and cause the whole box
to die and not recover at all.  With these patches we see slight RPS
drops (usually 10-15%) before the memory consumer is properly killed and
things recover within seconds.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agorq-qos: introduce dio_bio callback
Josef Bacik [Tue, 3 Jul 2018 15:15:00 +0000 (11:15 -0400)]
rq-qos: introduce dio_bio callback

wbt cares only about request completion time, but controllers may need
information that is on the bio itself, so add a done_bio callback for
rq-qos so things like blk-iolatency can use it to have the bio when it
completes.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: remove external dependency on wbt_flags
Josef Bacik [Tue, 3 Jul 2018 15:14:59 +0000 (11:14 -0400)]
block: remove external dependency on wbt_flags

We don't really need to save this stuff in the core block code, we can
just pass the bio back into the helpers later on to derive the same
flags and update the rq->wbt_flags appropriately.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-rq-qos: refactor out common elements of blk-wbt
Josef Bacik [Tue, 3 Jul 2018 15:32:35 +0000 (09:32 -0600)]
blk-rq-qos: refactor out common elements of blk-wbt

blkcg-qos is going to do essentially what wbt does, only on a cgroup
basis.  Break out the common code that will be shared between blkcg-qos
and wbt into blk-rq-qos.* so they can both utilize the same
infrastructure.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-stat: export helpers for modifying blk_rq_stat
Josef Bacik [Tue, 3 Jul 2018 15:14:57 +0000 (11:14 -0400)]
blk-stat: export helpers for modifying blk_rq_stat

We need to use blk_rq_stat in the blkcg qos stuff, so export some of
these helpers so they can be used by other things.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agomemcontrol: schedule throttling if we are congested
Tejun Heo [Tue, 3 Jul 2018 15:14:56 +0000 (11:14 -0400)]
memcontrol: schedule throttling if we are congested

Memory allocations can induce swapping via kswapd or direct reclaim.  If
we are having IO done for us by kswapd and don't actually go into direct
reclaim we may never get scheduled for throttling.  So instead check to
see if our cgroup is congested, and if so schedule the throttling.
Before we return to user space the throttling stuff will only throttle
if we actually required it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblkcg: add generic throttling mechanism
Josef Bacik [Tue, 3 Jul 2018 15:14:55 +0000 (11:14 -0400)]
blkcg: add generic throttling mechanism

Since IO can be issued from literally anywhere it's almost impossible to
do throttling without having some sort of adverse effect somewhere else
in the system because of locking or other dependencies.  The best way to
solve this is to do the throttling when we know we aren't holding any
other kernel resources.  Do this by tracking throttling in a per-blkg
basis, and if we require throttling flag the task that it needs to check
before it returns to user space and possibly sleep there.

This is to address the case where a process is doing work that is
generating IO that can't be throttled, whether that is directly with a
lot of REQ_META IO, or indirectly by allocating so much memory that it
is swamping the disk with REQ_SWAP.  We can't use task_add_work as we
don't want to induce a memory allocation in the IO path, so simply
saving the request queue in the task and flagging it to do the
notify_resume thing achieves the same result without the overhead of a
memory allocation.

Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>