With the pktcdvdv removal, bio_copy_data_iter is unused now. Fold the
logic into bio_copy_data and remove the separate lower level function.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221206144407.722049-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no need to update tg->slice_start[rw] to start when they are
equal already. So remove "eq" part of check before update slice_start.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-10-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no need to check elapsed time from last upgrade for each node in
hierarchy. Move this check before traversing as throtl_can_upgrade do
to remove repeat check.
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-9-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Function tg_last_low_overflow_time is called with intermediate node as
following:
throtl_hierarchy_can_downgrade
throtl_tg_can_downgrade
tg_last_low_overflow_time
throtl_hierarchy_can_upgrade
throtl_tg_can_upgrade
tg_last_low_overflow_time
throtl_hierarchy_can_downgrade/throtl_hierarchy_can_upgrade will traverse
from leaf node to sub-root node and pass traversed intermediate node
to tg_last_low_overflow_time.
No such limit could be found from context and implementation of
tg_last_low_overflow_time, so remove this limit in comment.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-8-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit c79892c557 ("blk-throttle: add upgrade logic for LIMIT_LOW
state") added upgrade logic for low limit and methioned that
1. "To determine if a cgroup exceeds its limitation, we check if the cgroup
has pending request. Since cgroup is throttled according to the limit,
pending request means the cgroup reaches the limit."
2. "If a cgroup has limit set for both read and write, we consider the
combination of them for upgrade. The reason is read IO and write IO can
interfere with each other. If we do the upgrade based in one direction IO,
the other direction IO could be severly harmed."
Besides, we also determine that cgroup reaches low limit if low limit is 0,
see comment in throtl_tg_can_upgrade.
Collect the information above, the desgin of upgrade check is as following:
1.The low limit is reached if limit is zero or io is already queued.
2.Cgroup will pass upgrade check if low limits of READ and WRITE are both
reached.
Simpfy the check code described above to removce repeat check and improve
readability. There is no functional change.
Detail equivalence proof is as following:
All replaced conditions to return true are as following:
condition 1
(!read_limit && !write_limit)
condition 2
read_limit && sq->nr_queued[READ] &&
(!write_limit || sq->nr_queued[WRITE])
condition 3
write_limit && sq->nr_queued[WRITE] &&
(!read_limit || sq->nr_queued[READ])
Transferring condition 2 as following:
(read_limit && sq->nr_queued[READ]) &&
(!write_limit || sq->nr_queued[WRITE])
is equivalent to
(read_limit && sq->nr_queued[READ]) &&
(!write_limit || (write_limit && sq->nr_queued[WRITE]))
is equivalent to
condition 2.1
(read_limit && sq->nr_queued[READ] &&
!write_limit) ||
condition 2.2
(read_limit && sq->nr_queued[READ] &&
(write_limit && sq->nr_queued[WRITE]))
Transferring condition 3 as following:
write_limit && sq->nr_queued[WRITE] &&
(!read_limit || sq->nr_queued[READ])
is equivalent to
(write_limit && sq->nr_queued[WRITE]) &&
(!read_limit || (read_limit && sq->nr_queued[READ]))
is equivalent to
condition 3.1
((write_limit && sq->nr_queued[WRITE]) &&
!read_limit) ||
condition 3.2
((write_limit && sq->nr_queued[WRITE]) &&
(read_limit && sq->nr_queued[READ]))
Condition 3.2 is the same as condition 2.2, so all conditions we get to
return are as following:
(!read_limit && !write_limit) (1)
(!read_limit && (write_limit && sq->nr_queued[WRITE])) (3.1)
((read_limit && sq->nr_queued[READ]) && !write_limit) (2.1)
((write_limit && sq->nr_queued[WRITE]) &&
(read_limit && sq->nr_queued[READ])) (2.2)
As we can extract conditions "(a1 || a2) && (b1 || b2)" to:
a1 && b1
a1 && b2
a2 && b1
ab && b2
Considering that:
a1 = !read_limit
a2 = read_limit && sq->nr_queued[READ]
b1 = !write_limit
b2 = write_limit && sq->nr_queued[WRITE]
We can pack replaced conditions to
(!read_limit || (read_limit && sq->nr_queued[READ])) &&
(!write_limit || (write_limit && sq->nr_queued[WRITE]))
which is equivalent to
(!read_limit || sq->nr_queued[READ]) &&
(!write_limit || sq->nr_queued[WRITE])
Reported-by: kernel test robot <lkp@intel.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-6-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In C language, When executing "if (expression1 && expression2)" and
expression1 return false, the expression2 may not be executed.
For "tg_within_bps_limit(tg, bio, bps_limit, &bps_wait) &&
tg_within_iops_limit(tg, bio, iops_limit, &iops_wait))", if bps is
limited, tg_within_bps_limit will return false and
tg_within_iops_limit will not be called. So even bps and iops are
both limited, iops_wait will not be calculated and is always zero.
So wait time of iops is always ignored.
Fix this by always calling tg_within_bps_limit and tg_within_iops_limit
to get wait time for both bps and iops.
Observed that:
1. Wait time in tg_within_iops_limit/tg_within_bps_limit need always
be stored as wait argument is always passed.
2. wait time is stored to zero if iops/bps is limited otherwise non-zero
is stored.
Simpfy tg_within_iops_limit/tg_within_bps_limit by removing wait argument
and return wait time directly. Caller tg_may_dispatch checks if wait time
is zero to find if iops/bps is limited.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-5-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Ignore cgroup without io queued in blk_throtl_cancel_bios for two
reasons:
1. Save cpu cycle for trying to dispatch cgroup which is no io queued.
2. Avoid non-consistent state that cgroup is inserted to service queue
without THROTL_TG_PENDING set as tg_update_disptime will unconditional
re-insert cgroup to service queue. If we are on the default hierarchy,
IO dispatched from child in tg_dispatch_one_bio will trigger inserting
cgroup to service queue without erase first and ruin the tree.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-4-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Consider situation as following (on the default hierarchy):
HDD
|
root (bps limit: 4k)
|
child (bps limit :8k)
|
fio bs=8k
Rate of fio is supposed to be 4k, but result is 8k. Reason is as
following:
Size of single IO from fio is larger than bytes allowed in one
throtl_slice in child, so IOs are always queued in child group first.
When queued IOs in child are dispatched to parent group, BIO_BPS_THROTTLED
is set and these IOs will not be limited by tg_within_bps_limit anymore.
Fix this by only set BIO_BPS_THROTTLED when the bio traversed the entire
tree.
There patch has no influence on situation which is not on the default
hierarchy as each group is a single root group without parent.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-3-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On the default hierarchy (cgroup2), the throttle interface files don't
exist in the root cgroup, so the ablity to limit the whole system
by configuring root group is not existing anymore. In general, cgroup
doesn't wanna be in the business of restricting resources at the
system level, so correct the stale comment that we can limit whole
system to we can only limit subtree.
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Link: https://lore.kernel.org/r/20221205115709.251489-2-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With the removal of the pktcdvd driver, there are no in-kernel users of
the devnode callback in struct block_device_operations, so it can be
safely removed. If it is needed for new block drivers in the future, it
can be brought back.
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Link: https://lore.kernel.org/r/20221203140747.1942969-1-gregkh@linuxfoundation.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make the description of @gendisk to @disk in blkcg_schedule_throttle()
to clear the below warnings:
block/blk-cgroup.c:1850: warning: Function parameter or member 'disk' not described in 'blkcg_schedule_throttle'
block/blk-cgroup.c:1850: warning: Excess function parameter 'gendisk' description in 'blkcg_schedule_throttle'
Fixes: de185b56e8 ("blk-cgroup: pass a gendisk to blkcg_schedule_throttle")
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3338
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Link: https://lore.kernel.org/r/20221202011713.14834-1-yang.lee@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use only one hyphen in kernel-doc notation between the function name
and its short description.
The is the documented kerenl-doc format. It also fixes the HTML
presentation to be consistent with other functions.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Link: https://lore.kernel.org/r/20221201070331.25685-1-rdunlap@infradead.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no iocg_pd_init function. The pd_alloc_fn function pointer of
iocost policy is set with ioc_pd_init. Just correct it.
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221018121932.10792-6-shikemeng@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we trace vtime_base_rate instead of vtime_rate, there is nowhere
which accesses now->vrate except function ioc_now using now->vrate locally.
Just remove it.
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221018121932.10792-5-shikemeng@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit ac33e91e2d ("blk-iocost: implement vtime loss
compensation") rename original vtime_rate to vtime_base_rate
and current vtime_rate is original vtime_rate with compensation.
The current rate showed in tracepoint is mixed with vtime_rate
and vtime_base_rate:
1) In function ioc_adjust_base_vrate, the first trace_iocost_ioc_vrate_adj
shows vtime_rate, the second trace_iocost_ioc_vrate_adj shows
vtime_base_rate.
2) In function iocg_activate shows vtime_rate by calling
TRACE_IOCG_PATH(iocg_activate...
3) In function ioc_check_iocgs shows vtime_rate by calling
TRACE_IOCG_PATH(iocg_idle...
Trace vtime_base_rate instead of vtime_rate as:
1) Before commit ac33e91e2d ("blk-iocost: implement vtime loss
compensation"), the traced rate is without compensation, so still
show rate without compensation.
2) The vtime_base_rate is more stable while vtime_rate heavily depends on
excess budeget on current period which may change abruptly in next period.
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221018121932.10792-4-shikemeng@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit ac33e91e2daca("blk-iocost: implement vtime loss compensation")
split vtime_rate into vtime_rate and vtime_base_rate, we need reset both
vtime_base_rate and vtime_rate when device parameters are refreshed.
If vtime_base_rate is no reset here, vtime_rate will be overwritten with
old vtime_base_rate soon in ioc_refresh_vrate.
Signed-off-by: Kemeng Shi <shikemeng@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221018121932.10792-3-shikemeng@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit 10c70d95c0 ("block: remove the bd_openers checks in
blk_drop_partitions") we allow rereading of partition table although
there are users of the block device. This has an undesirable consequence
that e.g. if sda and sdb are assembled to a RAID1 device md0 with
partitions, BLKRRPART ioctl on sda will rescan partition table and
create sda1 device. This partition device under a raid device confuses
some programs (such as libstorage-ng used for initial partitioning for
distribution installation) leading to failures.
Fix the problem refusing to rescan partitions if there is another user
that has the block device exclusively open.
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20221130135344.2ul4cyfstfs3znxg@quack3
Fixes: 10c70d95c0 ("block: remove the bd_openers checks in blk_drop_partitions")
Signed-off-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20221130175653.24299-1-jack@suse.cz
[axboe: fold in followup fix]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We can't just say that the last reference release may block, as any
reference dropped could be the last one. So move the might_sleep() from
blk_free_queue to blk_put_queue and update the documentation.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221114042637.1009333-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The kobject embedded into the request_queue is used for the queue
directory in sysfs, but that is a child of the gendisks directory and is
intimately tied to it. Move this kobject to the gendisk and use a
refcount_t in the request_queue for the actual request_queue refcounting
that is completely unrelated to the device model.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221114042637.1009333-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_register_queue fails to handle errors from blk_mq_sysfs_register,
leaks various resources on errors and accidentally sets queue refs percpu
refcount to percpu mode on kobject_add failure. Fix all that by
properly unwinding on errors.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221114042637.1009333-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split the debugfs removal from blk_unregister_queue into a helper so that
the it can be reused for blk_register_queue error handling.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221114042637.1009333-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Prepare for changes to the block layer sysfs handling by passing the
readily available gendisk to blk_crypto_sysfs_{,un}register.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221114042637.1009333-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Rename deadline_is_seq_writes() to deadline_is_seq_write() (remove the
"s" plural) to more correctly reflect the fact that this function tests
a single request, not multiple requests.
Fixes: 015d02f485 ("block: mq-deadline: Do not break sequential write streams to zoned HDDs")
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Link: https://lore.kernel.org/r/20221126025550.967914-2-damien.lemoal@opensource.wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
mq-deadline ensures an in order dispatching of write requests to zoned
block devices using a per zone lock (a bit). This implies that for any
purely sequential write workload, the drive is exercised most of the
time at a maximum queue depth of one.
However, when such sequential write workload crosses a zone boundary
(when sequentially writing multiple contiguous zones), zone write
locking may prevent the last write to one zone to be issued (as the
previous write is still being executed) but allow the first write to the
following zone to be issued (as that zone is not yet being writen and
not locked). This result in an out of order delivery of the sequential
write commands to the device every time a zone boundary is crossed.
While such behavior does not break the sequential write constraint of
zoned block devices (and does not generate any write error), some zoned
hard-disks react badly to seeing these out of order writes, resulting in
lower write throughput.
This problem can be addressed by always dispatching the first request
of a stream of sequential write requests, regardless of the zones
targeted by these sequential writes. To do so, the function
deadline_skip_seq_writes() is introduced and used in
deadline_next_request() to select the next write command to issue if the
target device is an HDD (blk_queue_nonrot() being false).
deadline_fifo_request() is modified using the new
deadline_earlier_request() and deadline_is_seq_write() helpers to ignore
requests in the fifo list that have a preceding request in lba order
that is sequential.
With this fix, a sequential write workload executed with the following
fio command:
fio --name=seq-write --filename=/dev/sda --zonemode=zbd --direct=1 \
--size=68719476736 --ioengine=libaio --iodepth=32 --rw=write \
--bs=65536
results in an increase from 225 MB/s to 250 MB/s of the write throughput
of an SMR HDD (11% increase).
Cc: <stable@vger.kernel.org>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20221124021208.242541-3-damien.lemoal@opensource.wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
dd_finish_request() tests if the per prio fifo_list is not empty to
determine if request dispatching must be restarted for handling blocked
write requests to zoned devices with a call to
blk_mq_sched_mark_restart_hctx(). While simple, this implementation has
2 problems:
1) Only the priority level of the completed request is considered.
However, writes to a zone may be blocked due to other writes to the
same zone using a different priority level. While this is unlikely to
happen in practice, as writing a zone with different IO priorirites
does not make sense, nothing in the code prevents this from
happening.
2) The use of list_empty() is dangerous as dd_finish_request() does not
take dd->lock and may run concurrently with the insert and dispatch
code.
Fix these 2 problems by testing the write fifo list of all priority
levels using the new helper dd_has_write_work(), and by testing each
fifo list using list_empty_careful().
Fixes: c807ab520f ("block/mq-deadline: Add I/O priority support")
Cc: <stable@vger.kernel.org>
Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20221124021208.242541-2-damien.lemoal@opensource.wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Allow the compiler to verify consistency of function declarations and
function definitions. This patch fixes the following sparse errors:
block/blk-crypto-profile.c:241:14: error: no previous prototype for ‘blk_crypto_get_keyslot’ [-Werror=missing-prototypes]
241 | blk_status_t blk_crypto_get_keyslot(struct blk_crypto_profile *profile,
| ^~~~~~~~~~~~~~~~~~~~~~
block/blk-crypto-profile.c:318:6: error: no previous prototype for ‘blk_crypto_put_keyslot’ [-Werror=missing-prototypes]
318 | void blk_crypto_put_keyslot(struct blk_crypto_keyslot *slot)
| ^~~~~~~~~~~~~~~~~~~~~~
block/blk-crypto-profile.c:344:6: error: no previous prototype for ‘__blk_crypto_cfg_supported’ [-Werror=missing-prototypes]
344 | bool __blk_crypto_cfg_supported(struct blk_crypto_profile *profile,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
block/blk-crypto-profile.c:373:5: error: no previous prototype for ‘__blk_crypto_evict_key’ [-Werror=missing-prototypes]
373 | int __blk_crypto_evict_key(struct blk_crypto_profile *profile,
| ^~~~~~~~~~~~~~~~~~~~~~
Cc: Eric Biggers <ebiggers@google.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20221123172923.434339-1-bvanassche@acm.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The commit ee9d55210c ("blk-mq: simplify blk_mq_realloc_tag_set_tags")
cleaned up the function blk_mq_realloc_tag_set_tags. After this change,
the function does not update nr_hw_queues of struct blk_mq_tag_set when
new nr_hw_queues value is smaller than original. This results in failure
of queue number change of block devices. To avoid the failure, add the
missing nr_hw_queues update.
Fixes: ee9d55210c ("blk-mq: simplify blk_mq_realloc_tag_set_tags")
Reported-by: Chaitanya Kulkarni <chaitanyak@nvidia.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/linux-block/20221118140640.featvt3fxktfquwh@shindev/
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221122084917.2034220-1-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_crypto_get_keyslot, blk_crypto_put_keyslot, __blk_crypto_evict_key
and __blk_crypto_cfg_supported are only used internally by the
blk-crypto code, so move the out of blk-crypto-profile.h, which is
included by drivers that supply blk-crypto functionality.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221114042944.1009870-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a blk_crypto_config_supported_natively helper that wraps
__blk_crypto_cfg_supported to retrieve the crypto_profile from the
request queue. With this fscrypt can stop including
blk-crypto-profile.h and rely on the public consumer interface in
blk-crypto.h.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221114042944.1009870-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Switch all public blk-crypto interfaces to use struct block_device
arguments to specify the device they operate on instead of th
request_queue, which is a block layer implementation detail.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20221114042944.1009870-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
As noted by Michal, the blkg_iostat_set's in the lockless list
hold reference to blkg's to protect against their removal. Those
blkg's hold reference to blkcg. When a cgroup is being destroyed,
cgroup_rstat_flush() is only called at css_release_work_fn() which is
called when the blkcg reference count reaches 0. This circular dependency
will prevent blkcg from being freed until some other events cause
cgroup_rstat_flush() to be called to flush out the pending blkcg stats.
To prevent this delayed blkcg removal, add a new cgroup_rstat_css_flush()
function to flush stats for a given css and cpu and call it at the blkgs
destruction path, blkcg_destroy_blkgs(), whenever there are still some
pending stats to be flushed. This will ensure that blkcg reference
count can reach 0 ASAP.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221105005902.407297-4-longman@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For a system with many CPUs and block devices, the time to do
blkcg_rstat_flush() from cgroup_rstat_flush() can be rather long. It
can be especially problematic as interrupt is disabled during the flush.
It was reported that it might take seconds to complete in some extreme
cases leading to hard lockup messages.
As it is likely that not all the percpu blkg_iostat_set's has been
updated since the last flush, those stale blkg_iostat_set's don't need
to be flushed in this case. This patch optimizes blkcg_rstat_flush()
by keeping a lockless list of recently updated blkg_iostat_set's in a
newly added percpu blkcg->lhead pointer.
The blkg_iostat_set is added to a lockless list on the update side
in blk_cgroup_bio_start(). It is removed from the lockless list when
flushed in blkcg_rstat_flush(). Due to racing, it is possible that
blk_iostat_set's in the lockless list may have no new IO stats to be
flushed, but that is OK.
To protect against destruction of blkg, a percpu reference is gotten
when putting into the lockless list and put back when removed.
When booting up an instrumented test kernel with this patch on a
2-socket 96-thread system with cgroup v2, out of the 2051 calls to
cgroup_rstat_flush() after bootup, 1788 of the calls were exited
immediately because of empty lockless list. After an all-cpu kernel
build, the ratio became 6295424/6340513. That was more than 99%.
Signed-off-by: Waiman Long <longman@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221105005902.407297-3-longman@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For blkcg_css_alloc(), the only error that will be returned is -ENOMEM.
Simplify error handling code by returning this error directly instead
of setting an intermediate "ret" variable.
Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20221105005902.407297-2-longman@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We hold a reference to the holder kobject for each bd_holder_disk,
so to make the code a bit more robust, use a reference to it instead
of the block_device. As long as no one clears ->bd_holder_dir in
before freeing the disk, this isn't strictly required, but it does
make the code more clear and more robust.
Orignally-From: Christoph Hellwig <hch@lst.de>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221115141054.1051801-10-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, the caller of bd_link_disk_holer() get 'bdev' by
blkdev_get_by_dev(), which will look up 'bdev' by inode number 'dev'.
Howerver, it's possible that del_gendisk() can be called currently, and
'bd_holder_dir' can be freed before bd_link_disk_holer() access it, thus
use after free is triggered.
t1: t2:
bdev = blkdev_get_by_dev
del_gendisk
kobject_put(bd_holder_dir)
kobject_free()
bd_link_disk_holder
Fix the problem by checking disk is still live and grabbing a reference
to 'bd_holder_dir' first in bd_link_disk_holder().
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20221115141054.1051801-9-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
