Nicolai Stange [Sun, 4 Dec 2016 13:56:39 +0000 (14:56 +0100)]
block: fix unintended fallthrough in generic_make_request_checks()
Since commit e73c23ff736e ("block: add async variant of
blkdev_issue_zeroout") messages like the following show up:
EXT4-fs (dm-1): Delayed block allocation failed for inode 2368848 at
logical offset 0 with max blocks 1 with error 95
EXT4-fs (dm-1): This should not happen!! Data will be lost
Due to the following fallthrough introduced with
commit 2d253440b5af ("block: Define zoned block device operations"),
generic_make_request_checks() would accept a REQ_OP_WRITE_SAME bio only
if the block device supports "write same" *and* is a zoned one:
switch (bio_op(bio)) {
[...]
case REQ_OP_WRITE_SAME:
if (!bdev_write_same(bio->bi_bdev))
goto not_supported;
case REQ_OP_ZONE_REPORT:
case REQ_OP_ZONE_RESET:
if (!bdev_is_zoned(bio->bi_bdev))
goto not_supported;
break;
[...]
}
Thus, although the bio setup as done by __blkdev_issue_write_same() from
commit e73c23ff736e ("block: add async variant of blkdev_issue_zeroout")
would succeed, its actual submission would not, resulting in the
EOPNOTSUPP == 95.
Fix this by removing the fallthrough which, due to the lack of an explicit
comment, seems to be unintended anyway.
Josef Bacik [Fri, 2 Dec 2016 21:19:12 +0000 (16:19 -0500)]
nbd: use loff_t for blocksize and nbd_set_size args
If we have large devices (say like the 40t drive I was trying to test with) we
will end up overflowing the int arguments to nbd_set_size and not get the right
size for our device. Fix this by using loff_t everywhere so I don't have to
think about this again. Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Rabin Vincent [Thu, 1 Dec 2016 08:18:28 +0000 (09:18 +0100)]
block: protect iterate_bdevs() against concurrent close
If a block device is closed while iterate_bdevs() is handling it, the
following NULL pointer dereference occurs because bdev->b_disk is NULL
in bdev_get_queue(), which is called from blk_get_backing_dev_info() (in
turn called by the mapping_cap_writeback_dirty() call in
__filemap_fdatawrite_range()):
The crash is easily reproducible by running the following command, if an
msleep(100) is inserted before the call to func() in iterate_devs():
while :; do head -c1 /dev/nullb0; done > /dev/null & while :; do sync; done
Fix it by holding the bd_mutex across the func() call and only calling
func() if the bdev is opened.
Cc: stable@vger.kernel.org Fixes: 5c0d6b60a0ba ("vfs: Create function for iterating over block devices") Reported-and-tested-by: Wei Fang <fangwei1@huawei.com> Signed-off-by: Rabin Vincent <rabinv@axis.com> Signed-off-by: Jan Kara <jack@suse.cz> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Pan Bian [Thu, 1 Dec 2016 02:10:46 +0000 (10:10 +0800)]
block: mtip32xx: set error code on failure
Fix bug https://bugzilla.kernel.org/show_bug.cgi?id=188531. In function
mtip_block_initialize(), variable rv takes the return value, and its
value should be negative on errors. rv is initialized as 0 and is not
reset when the call to ida_pre_get() fails. So 0 may be returned.
The return value 0 indicates that there is no error, which may be
inconsistent with the execution status. This patch fixes the bug by
explicitly assigning -ENOMEM to rv on the branch that ida_pre_get()
fails.
Signed-off-by: Pan Bian <bianpan2016@163.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Add support for handling write zeroes command on target.
Call into __blkdev_issue_zeroout, which the block layer expands into the
best suitable variant of zeroing the LBAs. Allow write zeroes operation
to deallocate the LBAs when calling __blkdev_issue_zeroout.
Allow write zeroes operations (REQ_OP_WRITE_ZEROES) on the block
device, if the device supports optional command bit set for write
zeroes. Add support to setup write zeroes command. Set maximum possible
write zeroes sectors in one write zeroes command according to
nvme write zeroes command definition.
This adds a new block layer operation to zero out a range of
LBAs. This allows to implement zeroing for devices that don't use
either discard with a predictable zero pattern or WRITE SAME of zeroes.
The prominent example of that is NVMe with the Write Zeroes command,
but in the future, this should also help with improving the way
zeroing discards work. For this operation, suitable entry is exported in
sysfs which indicate the number of maximum bytes allowed in one
write zeroes operation by the device.
Damien Le Moal [Thu, 1 Dec 2016 07:00:25 +0000 (16:00 +0900)]
block: Check partition alignment on zoned block devices
Both blkdev_report_zones and blkdev_reset_zones can operate on a partition of
a zoned block device. However, the first and last zones reported for a
partition make sense only if the partition start sector and size are aligned
on the device zone size. The same applies for zone reset. Resetting the first
or the last zone of a partition straddling zones may impact neighboring
partitions. Finally, if a partition start sector is not at the beginning of a
sequential zone, it will be impossible to write to the first sectors of the
partition on a host-managed device.
Avoid all these problems and incoherencies by ignoring partitions that are not
zone aligned.
Note: Even with CONFIG_BLK_DEV_ZONED disabled, bdev_is_zoned() will report the
correct disk zoning type (host-aware, host-managed or none) but
bdev_zone_size() will always return 0 for zoned block devices (i.e. the zone
size is unknown). So test this as a way to ensure that a zoned block device is
being handled as such. As a result, for a host-aware devices, unaligned zone
partitions will be accepted with CONFIG_BLK_DEV_ZONED disabled. That is, the
disk will be treated as a regular block device (as it should). If zoned block
device support is enabled, only aligned partitions will be accepted.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Javier González [Mon, 28 Nov 2016 21:39:14 +0000 (22:39 +0100)]
lightnvm: transform target get/set bad block
Since targets are given a virtual target device, it is necessary to
translate all communication between targets and the backend device.
Implement the translation layer for get/set bad block table.
Javier González [Mon, 28 Nov 2016 21:39:12 +0000 (22:39 +0100)]
lightnvm: introduce max_phys_sects helper function
Target devices do not have access to the device driver operations.
Introduce a helper function that exposes the max. number of physical
sectors supported by the underlying device.
Javier González [Mon, 28 Nov 2016 21:39:11 +0000 (22:39 +0100)]
lightnvm: introduce helpers for generic ops in rrpc
Avoid calling media manager and device-specific operations directly from
rrpc. Create helper functions on lightnvm's core instead.
Signed-off-by: Javier González <javier@cnexlabs.com>
Made it work with null_blk as well. Signed-off-by: Matias Bjørling <m@bjorling.me> Signed-off-by: Matias Bjørling <m@bjorling.me> Signed-off-by: Jens Axboe <axboe@fb.com>
Javier González [Mon, 28 Nov 2016 21:39:10 +0000 (22:39 +0100)]
lightnvm: eliminate nvm_lun abstraction in mm
In order to naturally support multi-target instances on an Open-Channel
SSD, targets should own the LUNs they get blocks from and manage
provisioning internally. This is done in several steps.
Since targets own the LUNs the are instantiated on top of and manage the
free block list internally, there is no need for a LUN abstraction in
the media manager. LUNs are intrinsically managed as in the physical
layout (ch:0,lun:0, ..., ch:0,lun:n, ch:1,lun:0, ch:1,lun:n, ...,
ch:m,lun:0, ch:m,lun:n) and given to the targets based on the target
creation ioctl. This simplifies LUN management and clears the path for a
partition manager to sit directly underneath LightNVM targets.
Javier González [Mon, 28 Nov 2016 21:39:09 +0000 (22:39 +0100)]
lightnvm: eliminate nvm_block abstraction on mm
In order to naturally support multi-target instances on an Open-Channel
SSD, targets should own the LUNs they get blocks from and manage
provisioning internally. This is done in several steps.
A part of this transformation is that targets manage their blocks
internally. This patch eliminates the nvm_block abstraction and moves
block management to the target logic. The rrpc target is transformed.
Javier González [Mon, 28 Nov 2016 21:39:08 +0000 (22:39 +0100)]
lightnvm: remove debug lun statistics from gennvm
Since LUNs are managed internally on targets, the media manager has no
access to the free LUN lists. Thus, debug functions that show LUN
information on the device should not be implemented on the media
manager, but rather on the target in itself.
Javier González [Mon, 28 Nov 2016 21:39:06 +0000 (22:39 +0100)]
lightnvm: move block provisioning to targets
In order to naturally support multi-target instances on an Open-Channel
SSD, targets should own the LUNs they get blocks from and manage
provisioning internally. This is done in several steps.
This patch moves the block provisioning inside of the target and removes
the get/put block interface from the media manager.
Javier González [Mon, 28 Nov 2016 21:39:05 +0000 (22:39 +0100)]
lightnvm: manage lun partitions internally in mm
LUNs are exclusively owned by targets implementing a block device FTL.
Doing this reservation requires at the moment a 2-way callback gennvm
<-> target. The reason behind this is that LUNs were not assumed to
always be exclusively owned by targets. However, this design decision
goes against I/O determinism QoS (two targets would mix I/O on the same
parallel unit in the device).
This patch makes LUN reservation as part of the target creation on the
media manager. This makes that LUNs are always exclusively owned by the
target instantiated on top of them. LUN stripping and/or sharing should
be implemented on the target itself or the layers on top.
Javier González [Mon, 28 Nov 2016 21:39:04 +0000 (22:39 +0100)]
lightnvm: remove gen_lun abstraction
The gen_lun abstraction in the generic media manager was conceived on
the assumption that a single target would instantiated on top of it.
This has complicated target design to implement multi-instances. Remove
this abstraction and move its logic to nvm_lun, which manages physical
lun geometry and operations.
Javier González [Mon, 28 Nov 2016 21:39:02 +0000 (22:39 +0100)]
lightnvm: remove unnecessary variables in rrpc
Before vectored I/Os were supported on rrpc, the physical address was
stored as part of the nvm_rqd request. This variable become obsolete
when the ppa_list was introduced. Cleanup this variable.
Javier González [Mon, 28 Nov 2016 21:39:01 +0000 (22:39 +0100)]
lightnvm: make address conversion functions global
Targets are assumed to used the same generic ppa format, where the
address is partitioned on ch:lun:block:pg:pl:sec. Thus, make the
function in charge of transforming the ppa address from a linear format
to the generic one available to all targets.
This function will be needed by the media manager in order to do target
mapping translations when targets are divided on different physical
partitions.
Javier González [Mon, 28 Nov 2016 21:38:59 +0000 (22:38 +0100)]
lightnvm: remove sysfs configuration interface
LightNVM used to be managed and configured through sysfs. Since the
introduction of management ioctls this interface is redundant and
outdated. Get rid of it.
Javier González [Mon, 28 Nov 2016 21:38:56 +0000 (22:38 +0100)]
lightnvm: export set bad block table
Bad blocks should be managed by block owners. This would be either
targets for data blocks or sysblk for system blocks.
In order to support this, export two functions: One to mark a block as
an specific type (e.g., bad block) and another to update the bad block
table on the device.
Javier González [Mon, 28 Nov 2016 21:38:55 +0000 (22:38 +0100)]
lightnvm: do not protect block 0
Device blocks should be marked by the device and considered as bad
blocks by the media manager. Thus, do not make assumptions on which
blocks are going to be used by the device. In doing so we might lose
valid blocks from the free list.
Matias Bjørling [Mon, 28 Nov 2016 21:38:53 +0000 (22:38 +0100)]
nvme: lightnvm: attach lightnvm sysfs to nvme block device
Previously, LBA read and write were not supported in the lightnvm
specification. Now that it supports it, lets use the traditional
NVMe gendisk, and attach the lightnvm sysfs geometry export.
Matias Bjørling [Mon, 28 Nov 2016 21:38:52 +0000 (22:38 +0100)]
nvme: lightnvm: frees wrong cmd structure
When struct nvme_request was introduced, the nvme_nvm_submit_io was
converted to the new interface. The interface moves nvme_nvm_command
data structure into the struct request pdu. On io completion, rq->cmd is
freed, which should have been the dereferenced pdu nvme_request->cmd.
After commit 287922eb0b18 ("block: defer timeouts to a workqueue"),
deleting the timeout work after freezing the queue shouldn't be
necessary, since the synchronization is already enforced by the
acquisition of a q_usage_counter reference in blk_mq_timeout_work.
Signed-off-by: Gabriel Krisman Bertazi <krisman@linux.vnet.ibm.com> Reviewed-by: Ming Lei <ming.lei@canonical.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Mon, 28 Nov 2016 16:40:34 +0000 (09:40 -0700)]
blk-wbt: allow wbt to be enabled always through sysfs
Currently there's no way to enable wbt if it's not enabled in the
kernel config by default for a device. Allow a write to the
'wbt_lat_usec' queue sysfs file to enable wbt.
This is useful for both the kernel config case, but also if the
device is CFQ managed and it was turned off by default.
Jens Axboe [Mon, 28 Nov 2016 16:25:50 +0000 (09:25 -0700)]
blk-wbt: cleanup disable-by-default for CFQ
Make it clear that we are disabling wbt for the specified queued,
if it was enabled by default. This is in preparation for allowing
users to re-enable wbt, and not have it disabled automatically
again.
Jens Axboe [Mon, 28 Nov 2016 16:22:47 +0000 (09:22 -0700)]
blk-wbt: allow reset of default latency through sysfs
Allow a write of '-1' to reset the default latency target for
a given device. This removes knowledge of the different default
settings for rotational vs non-rotational from user space.
Josef Bacik [Tue, 22 Nov 2016 19:04:40 +0000 (14:04 -0500)]
nbd: add multi-connection support
NBD can become contended on its single connection. We have to serialize all
writes and we can only process one read response at a time. Fix this by
allowing userspace to provide multiple connections to a single nbd device. This
coupled with block-mq drastically increases performance in multi-process cases.
Thanks,
Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Tejun Heo [Mon, 21 Nov 2016 23:03:32 +0000 (18:03 -0500)]
block,blkcg: use __GFP_NOWARN for best-effort allocations in blkcg
blkcg allocates some per-cgroup data structures with GFP_NOWAIT and
when that fails falls back to operations which aren't specific to the
cgroup. Occassional failures are expected under pressure and falling
back to non-cgroup operation is the right thing to do.
Unfortunately, I forgot to add __GFP_NOWARN to these allocations and
these expected failures end up creating a lot of noise. Add
__GFP_NOWARN.
Ming Lei [Fri, 11 Nov 2016 12:05:30 +0000 (20:05 +0800)]
block: drbd: remove impossible failure handling
For a non-cloned bio, bio_add_page() only returns failure when
the io vec table is full, but in that case, bio->bi_vcnt can't
be zero at all.
So remove the impossible failure handling.
Reviewed-by: Christoph Hellwig <hch@lst.de> Acked-by: Lars Ellenberg <lars.ellenberg@linbit.com> Signed-off-by: Ming Lei <tom.leiming@gmail.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Ming Lei [Tue, 22 Nov 2016 15:57:21 +0000 (08:57 -0700)]
block: bio: pass bvec table to bio_init()
Some drivers often use external bvec table, so introduce
this helper for this case. It is always safe to access the
bio->bi_io_vec in this way for this case.
After converting to this usage, it will becomes a bit easier
to evaluate the remaining direct access to bio->bi_io_vec,
so it can help to prepare for the following multipage bvec
support.
Signed-off-by: Ming Lei <tom.leiming@gmail.com> Reviewed-by: Christoph Hellwig <hch@lst.de>
Fixed up the new O_DIRECT cases.
Jens Axboe [Tue, 22 Nov 2016 15:12:39 +0000 (08:12 -0700)]
block_dev: get rid of blksize bits calculation
We store the bits in the bdev sector size locally, but we don't use
the calculation anymore. All we do with it is shift it back up to
the bdev sector size. So let's just use that directly and kill the
variable and bits calculation.
Damien Le Moal [Tue, 22 Nov 2016 06:38:49 +0000 (15:38 +0900)]
block_dev: Fixed direct I/O bio sector calculation
A direct I/O alignment must be always checked against the device blocks size,
but the I/O offset (bio->bi_iter.bi_sector must always use 512B sector unit, and
not the actual logical block size.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Shaun Tancheff [Mon, 21 Nov 2016 21:52:23 +0000 (15:52 -0600)]
block: apply blk_partition_remap to REQ_OP_ZONE_RESET
If a ZBC device is partitioned and operations are performed on the partition
the zone information is rebased to the partition, however the zone reset
is not mapped from the partition to device as are other operations.
This causes the API (report zones / reset zone) to be unbalanced in this
regard. Checking for the zone reset op code explicitly will balance the
API.
Since commit 87374179 ("block: add a proper block layer data direction
encoding") we only or the new op and flags into bi_opf in bio_set_op_attrs
instead of clearing the old value. I've not seen any breakage with the
new behavior, but it seems dangerous.
Also convert it to an inline function to make the argument passing
safer.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Tobias Klauser [Fri, 18 Nov 2016 14:16:06 +0000 (15:16 +0100)]
block: Change extern inline to static inline
With compilers which follow the C99 standard (like modern versions of
gcc and clang), "extern inline" does the opposite thing from older
versions of gcc (emits code for an externally linkable version of the
inline function).
"static inline" does the intended behavior in all cases instead.
Description taken from commit 6d91857d4826 ("staging, rtl8192e,
LLVMLinux: Change extern inline to static inline").
This also fixes the following GCC warning when building with CONFIG_PM
disabled:
./include/linux/blkdev.h:1143:20: warning: no previous prototype for 'blk_set_runtime_active' [-Wmissing-prototypes]
Similar to the simple fast path, but we now need a dio structure to
track multiple-bio completions. It's basically a cut-down version
of the new iomap-based direct I/O code for filesystems, but without
all the logic to call into the filesystem for extent lookup or
allocation, and without the complex I/O completion workqueue handler
for AIO - instead we just use the FUA bit on the bios to ensure
data is flushed to stable storage.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Mon, 14 Nov 2016 20:03:03 +0000 (13:03 -0700)]
blk-mq: make the polling code adaptive
The previous commit introduced the hybrid sleep/poll mode. Take
that one step further, and use the completion latencies to
automatically sleep for half the mean completion time. This is
a good approximation.
This changes the 'io_poll_delay' sysfs file a bit to expose the
various options. Depending on the value, the polling code will
behave differently:
-1 Never enter hybrid sleep mode
0 Use half of the completion mean for the sleep delay
>0 Use this specific value as the sleep delay
Signed-off-by: Jens Axboe <axboe@fb.com> Tested-By: Stephen Bates <sbates@raithlin.com> Reviewed-By: Stephen Bates <sbates@raithlin.com>
Jens Axboe [Mon, 14 Nov 2016 20:01:59 +0000 (13:01 -0700)]
blk-mq: implement hybrid poll mode for sync O_DIRECT
This patch enables a hybrid polling mode. Instead of polling after IO
submission, we can induce an artificial delay, and then poll after that.
For example, if the IO is presumed to complete in 8 usecs from now, we
can sleep for 4 usecs, wake up, and then do our polling. This still puts
a sleep/wakeup cycle in the IO path, but instead of the wakeup happening
after the IO has completed, it'll happen before. With this hybrid
scheme, we can achieve big latency reductions while still using the same
(or less) amount of CPU.
Signed-off-by: Jens Axboe <axboe@fb.com> Tested-By: Stephen Bates <sbates@raithlin.com> Reviewed-By: Stephen Bates <sbates@raithlin.com>
block: fast-path for small and simple direct I/O requests
This patch adds a small and simple fast patch for small direct I/O
requests on block devices that don't use AIO. Between the neat
bio_iov_iter_get_pages helper that avoids allocating a page array
for get_user_pages and the on-stack bio and biovec this avoid memory
allocations and atomic operations entirely in the direct I/O code
(lower levels might still do memory allocations and will usually
have at least some atomic operations, though).
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com> Tested-By: Stephen Bates <sbates@raithlin.com> Reviewed-By: Stephen Bates <sbates@raithlin.com>
Jens Axboe [Thu, 17 Nov 2016 19:30:37 +0000 (12:30 -0700)]
nbd: fix use-after-free of rq/bio in the xmit path
For writes, we can get a completion in while we're still iterating
the request and bio chain. If that happens, we're reading freed
memory and we can crash.
Break out after the last segment and avoid having the iterator
read freed memory.
Reviewed-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Arnd Bergmann [Wed, 16 Nov 2016 15:29:57 +0000 (16:29 +0100)]
blk-wbt: fix old-style function declaration
The newly added driver causes a harmless warning in some configurations:
block/blk-wbt.c:250:1: error: ‘inline’ is not at beginning of declaration [-Werror=old-style-declaration]
static bool inline stat_sample_valid(struct blk_rq_stat *stat)
This makes it use the expected format for the declaration.
Ming Lei [Wed, 16 Nov 2016 10:07:05 +0000 (18:07 +0800)]
block: deal with stale req count of plug list
In both legacy and mq path, req count of plug list is computed
before allocating request, so the number can be stale when falling
back to slept allocation, also the new introduced wbt can sleep
too.
This patch deals with the case by checking if plug list becomes
empty, and fixes the KASAN report of 'BUG: KASAN: stack-out-of-bounds'
which is introduced by Shaohua's patches of dispatching big request.
Fixes: 600271d900002(blk-mq: immediately dispatch big size request) Fixes: 50d24c34403c6(block: immediately dispatch big size request) Cc: Shaohua Li <shli@fb.com> Signed-off-by: Ming Lei <ming.lei@canonical.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Damien Le Moal [Fri, 11 Nov 2016 05:53:26 +0000 (14:53 +0900)]
sd_zbc: Force use of READ16/WRITE16
Normally, sd_read_capacity sets sdp->use_16_for_rw to 1 based on the
disk capacity so that READ16/WRITE16 are used for large drives.
However, for a zoned disk with RC_BASIS set to 0, the capacity reported
through READ_CAPACITY may be very small, leading to use_16_for_rw not being
set and READ10/WRITE10 commands being used, even after the actual zoned disk
capacity is corrected in sd_zbc_read_zones. This causes LBA offset overflow for
accesses beyond 2TB.
As the ZBC standard makes it mandatory for ZBC drives to support
the READ16/WRITE16 commands anyway, make sure that use_16_for_rw is set.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
eviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Fri, 11 Nov 2016 19:24:46 +0000 (12:24 -0700)]
blk-mq: blk_mq_try_issue_directly() should lookup hardware queue
A previous commit changed this to pass in the hardware queue, but
it was using the wrong hardware queue. Hence a request that was
allocated on one hardware queue ended up being issued on another
one, and that caused IO timeouts and oopses on some drivers. Since
the request holds hardware queue private resources, like a tag,
we can't just issue it on a different hardware queue.
Fixes: 2253efc850c4 ("blk-mq: Move more code into blk_mq_direct_issue_request()") Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Wed, 9 Nov 2016 19:38:14 +0000 (12:38 -0700)]
block: hook up writeback throttling
Enable throttling of buffered writeback to make it a lot
more smooth, and has way less impact on other system activity.
Background writeback should be, by definition, background
activity. The fact that we flush huge bundles of it at the time
means that it potentially has heavy impacts on foreground workloads,
which isn't ideal. We can't easily limit the sizes of writes that
we do, since that would impact file system layout in the presence
of delayed allocation. So just throttle back buffered writeback,
unless someone is waiting for it.
The algorithm for when to throttle takes its inspiration in the
CoDel networking scheduling algorithm. Like CoDel, blk-wb monitors
the minimum latencies of requests over a window of time. In that
window of time, if the minimum latency of any request exceeds a
given target, then a scale count is incremented and the queue depth
is shrunk. The next monitoring window is shrunk accordingly. Unlike
CoDel, if we hit a window that exhibits good behavior, then we
simply increment the scale count and re-calculate the limits for that
scale value. This prevents us from oscillating between a
close-to-ideal value and max all the time, instead remaining in the
windows where we get good behavior.
Unlike CoDel, blk-wb allows the scale count to to negative. This
happens if we primarily have writes going on. Unlike positive
scale counts, this doesn't change the size of the monitoring window.
When the heavy writers finish, blk-bw quickly snaps back to it's
stable state of a zero scale count.
The patch registers a sysfs entry, 'wb_lat_usec'. This sets the latency
target to me met. It defaults to 2 msec for non-rotational storage, and
75 msec for rotational storage. Setting this value to '0' disables
blk-wb. Generally, a user would not have to touch this setting.
We don't enable WBT on devices that are managed with CFQ, and have
a non-root block cgroup attached. If we have a proportional share setup
on this particular disk, then the wbt throttling will interfere with
that. We don't have a strong need for wbt for that case, since we will
rely on CFQ doing that for us.
This shows a sync issue event (wbt_lat) that exceeded it's time. wbt_stat
dumps the current read/write stats for that window, and wbt_step shows a
step down event where we now scale back writes. Each trace includes the
device, 259:0 in this case.
Jens Axboe [Tue, 8 Nov 2016 04:32:37 +0000 (21:32 -0700)]
block: add scalable completion tracking of requests
For legacy block, we simply track them in the request queue. For
blk-mq, we track them on a per-sw queue basis, which we can then
sum up through the hardware queues and finally to a per device
state.
The stats are tracked in, roughly, 0.1s interval windows.
Add sysfs files to display the stats.
The feature is off by default, to avoid any extra overhead. In-kernel
users of it can turn it on by setting QUEUE_FLAG_STATS in the queue
flags. We currently don't turn it on if someone just reads any of
the stats files, that is something we could add as well.
Tejun Heo [Thu, 10 Nov 2016 16:16:37 +0000 (11:16 -0500)]
block: cfq_cpd_alloc() should use @gfp
cfq_cpd_alloc() which is the cpd_alloc_fn implementation for cfq was
incorrectly hard coding GFP_KERNEL instead of using the mask specified
through the @gfp parameter. This currently doesn't cause any actual
issues because all current callers specify GFP_KERNEL. Fix it.
Signed-off-by: Tejun Heo <tj@kernel.org> Reported-by: Dan Carpenter <dan.carpenter@oracle.com> Fixes: e4a9bde9589f ("blkcg: replace blkcg_policy->cpd_size with ->cpd_alloc/free_fn() methods") Signed-off-by: Jens Axboe <axboe@fb.com>
nvme: don't pass the full CQE to nvme_complete_async_event
We only need the status and result fields, and passing them explicitly
makes life a lot easier for the Fibre Channel transport which doesn't
have a full CQE for the fast path case.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
This adds a shared per-request structure for all NVMe I/O. This structure
is embedded as the first member in all NVMe transport drivers request
private data and allows to implement common functionality between the
drivers.
The first use is to replace the current abuse of the SCSI command
passthrough fields in struct request for the NVMe command passthrough,
but it will grow a field more fields to allow implementing things
like common abort handlers in the future.
The passthrough commands are handled by having a pointer to the SQE
(struct nvme_command) in struct nvme_request, and the union of the
possible result fields, which had to be turned from an anonymous
into a named union for that purpose. This avoids having to pass
a reference to a full CQE around and thus makes checking the result
a lot more lightweight.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <keith.busch@intel.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Arnd Bergmann [Wed, 9 Nov 2016 12:55:34 +0000 (13:55 +0100)]
skd: fix msix error handling
As reported by gcc -Wmaybe-uninitialized, the cleanup path for
skd_acquire_msix tries to free the already allocated msi-x vectors
in reverse order, but the index variable may not have been
used yet:
drivers/block/skd_main.c: In function ‘skd_acquire_irq’:
drivers/block/skd_main.c:3890:8: error: ‘i’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
This changes the failure path to skip releasing the interrupts
if we have not started requesting them yet.
Fixes: 180b0ae77d49 ("skd: use pci_alloc_irq_vectors") Signed-off-by: Arnd Bergmann <arnd@arndb.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Wed, 9 Nov 2016 02:39:28 +0000 (19:39 -0700)]
block: set REQ_SYNC if we clear REQ_FUA|REQ_PREFLUSH
If we insert a flush request, we clear REQ_PREFLUSH and/or REQ_FUA,
depending on flush settings. Since op_is_sync() factors those flags
in for deciding whether this request is sync or not, we should
set REQ_SYNC to avoid screwing up this accounting.
This should be less fragile.
Reported-by: Logan Gunthorpe <logang@deltatee.com> Fixes: b685d3d65ac ("block: treat REQ_FUA and REQ_PREFLUSH as synchronous") Signed-off-by: Jens Axboe <axboe@fb.com>
writeback: track if we're sleeping on progress in balance_dirty_pages()
Note in the bdi_writeback structure whenever a task ends up sleeping
waiting for progress. We can use that information in the lower layers
to increase the priority of writes.
Signed-off-by: Jens Axboe <axboe@fb.com> Reviewed-by: Jan Kara <jack@suse.cz>
Switch the skd driver to use pci_alloc_irq_vectors. We need to two calls to
pci_alloc_irq_vectors as skd only supports multiple MSI-X vectors, but not
multiple MSI vectors.
Otherwise this cleans up a lot of cruft and allows to a lot more common code.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
I've been looking over the multi page bio vec work again recently, and
one of the stumbling blocks is raw biovec access in the pktcdvd.
The first issue is that it directly sets up the page and offset pointers
in the biovec just before calling bio_add_page. As bio_add_page already
does the setup it's trivial to just switch it to stack variables for the
arguments.
The second issue is the copy code in pkt_make_local_copy, which
effectively is an opencoded version of bio_copy_data except that it
skips pages that already are the same in the ѕource and destination.
But we look at the only calleer we just set up the bio using bio_add_page
to point exactly to the page array that pkt_make_local_copy compares,
so the pages will always be the same and we can just remove this function.
Note that all this is done based on code inspection, I don't have any
packet writing hardware myself.
Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@fb.com>
Commit 0e87e58bf60e ("blk-mq: improve warning for running a queue on the
wrong CPU") attempts to avoid triggering the WARN_ON in
__blk_mq_run_hw_queue when the expected CPU is dead. Problem is, in the
last batch execution before round robin, blk_mq_hctx_next_cpu can
schedule a dead CPU and also update next_cpu to the next alive CPU in
the mask, which will trigger the WARN_ON despite the previous
workaround.
The following patch fixes this scenario by always scheduling the value
in hctx->next_cpu. This changes the moment when we round-robin the CPU
running the hctx, but it really doesn't matter, since it still executes
BLK_MQ_CPU_WORK_BATCH times in a row before switching to another CPU.
Fixes: 0e87e58bf60e ("blk-mq: improve warning for running a queue on the wrong CPU") Signed-off-by: Gabriel Krisman Bertazi <krisman@linux.vnet.ibm.com> Signed-off-by: Jens Axboe <axboe@fb.com>
Jens Axboe [Wed, 30 Mar 2016 16:21:08 +0000 (10:21 -0600)]
block: add code to track actual device queue depth
For blk-mq, ->nr_requests does track queue depth, at least at init
time. But for the older queue paths, it's simply a soft setting.
On top of that, it's generally larger than the hardware setting
on purpose, to allow backup of requests for merging.
Fill a hole in struct request with a 'queue_depth' member, that
drivers can call to more closely inform the block layer of the
real queue depth.
Signed-off-by: Jens Axboe <axboe@fb.com> Reviewed-by: Jan Kara <jack@suse.cz>
Shaohua Li [Fri, 4 Nov 2016 00:03:53 +0000 (17:03 -0700)]
block: immediately dispatch big size request
Currently block plug holds up to 16 non-mergeable requests. This makes
sense if the request size is small, eg, reduce lock contention. But if
request size is big enough, we don't need to worry about lock
contention. Holding such request makes no sense and it lows the disk
utilization.
In practice, this improves 10% throughput for my raid5 sequential write
workload.
The size (128k) is arbitrary right now, but it makes sure lock
contention is small. This probably could be more intelligent, eg, check
average request size holded. Since this is mainly for sequential IO,
probably not worthy.
V2: check the last request instead of the first request, so as long as
there is one big size request we flush the plug.
Signed-off-by: Shaohua Li <shli@fb.com> Signed-off-by: Jens Axboe <axboe@fb.com>