Wang ShaoBo [Thu, 24 Nov 2022 01:58:17 +0000 (09:58 +0800)]
drbd: destroy workqueue when drbd device was freed
A submitter workqueue is dynamically allocated by init_submitter()
called by drbd_create_device(), we should destroy it when this
device is not needed or destroyed.
Wang ShaoBo [Thu, 24 Nov 2022 01:58:16 +0000 (09:58 +0800)]
drbd: remove call to memset before free device/resource/connection
This revert c2258ffc56f2 ("drbd: poison free'd device, resource and
connection structs"), add memset is odd here for debugging, there are
some methods to accurately show what happened, such as kdump.
Damien Le Moal [Thu, 24 Nov 2022 02:12:08 +0000 (11:12 +0900)]
block: mq-deadline: Do not break sequential write streams to zoned HDDs
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:
Damien Le Moal [Thu, 24 Nov 2022 02:12:07 +0000 (11:12 +0900)]
block: mq-deadline: Fix dd_finish_request() for zoned devices
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: c807ab520fc3 ("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>
DRBD currently has a mix of GPL-2.0 and GPL-2.0-or-later SPDX license
identifiers. We have decided to stick with GPL 2.0 only, so consistently
use that identifier.
block: fix missing nr_hw_queues update in blk_mq_realloc_tag_set_tags
The commit ee9d55210c2f ("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.
blk-crypto: move internal only declarations to blk-crypto-internal.h
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.
blk-crypto: add a blk_crypto_config_supported_natively helper
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.
blk-crypto: don't use struct request_queue for public interfaces
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.
Waiman Long [Sat, 5 Nov 2022 00:59:02 +0000 (20:59 -0400)]
blk-cgroup: Flush stats at blkgs destruction path
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.
Waiman Long [Sat, 5 Nov 2022 00:59:01 +0000 (20:59 -0400)]
blk-cgroup: Optimize blkcg_rstat_flush()
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%.
Waiman Long [Sat, 5 Nov 2022 00:59:00 +0000 (20:59 -0400)]
blk-cgroup: Return -ENOMEM directly in blkcg_css_alloc() error path
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.
Yu Kuai [Tue, 15 Nov 2022 14:10:54 +0000 (22:10 +0800)]
block: don't allow a disk link holder to itself
After creating a dm device, then user can reload such dm with itself,
and dead loop will be triggered because dm keep looking up to itself.
Test procedures:
1) dmsetup create test --table "xxx sda", assume dm-0 is created
2) dmsetup suspend test
3) dmsetup reload test --table "xxx dm-0"
4) dmsetup resume test
Yu Kuai [Tue, 15 Nov 2022 14:10:53 +0000 (22:10 +0800)]
block: store the holder kobject in bd_holder_disk
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.
Yu Kuai [Tue, 15 Nov 2022 14:10:52 +0000 (22:10 +0800)]
block: fix use after free for bd_holder_dir
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.
dm is a bit special in that it opens the underlying devices. Commit 89f871af1b26 ("dm: delay registering the gendisk") tried to accommodate
that by allowing to add the holder to the list before add_gendisk and
then just add them to sysfs once add_disk is called. But that leads to
really odd lifetime problems and error handling problems as we can't
know the state of the kobjects and don't unwind properly. To fix this
switch to just registering all existing table_devices with the holder
code right after add_disk, and remove them before calling del_gendisk.
Fixes: 89f871af1b26 ("dm: delay registering the gendisk") Reported-by: Yu Kuai <yukuai3@huawei.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Yu Kuai <yukuai3@huawei.com> Reviewed-by: Mike Snitzer <snitzer@kernel.org> Link: https://lore.kernel.org/r/20221115141054.1051801-7-yukuai1@huaweicloud.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
Take the list unlink and free into close_table_device so that no half
torn down table_devices exist. Also remove the check for a NULL bdev
as that can't happen - open_table_device never adds a table_device to
the list that does not have a valid block_device.
Move all the logic for allocation the table_device and linking it into
the list into the open_table_device. This keeps the code tidy and
ensures that the table_devices only exist in fully initialized state.
free_table_devices just warns and frees all table_device structures when
the target removal did not remove them. This should never happen, but
if it did, just freeing the structure without deleting them from the
list or cleaning up the resources would not help at all. So just WARN on
a non-empty list instead.
block: clear ->slave_dir when dropping the main slave_dir reference
Zero out the pointer to ->slave_dir so that the holder code doesn't
incorrectly treat the object as alive when add_disk failed or after
del_gendisk was called.
Fixes: 89f871af1b26 ("dm: delay registering the gendisk") Reported-by: Yu Kuai <yukuai3@huawei.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Yu Kuai <yukuai3@huawei.com> Reviewed-by: Mike Snitzer <snitzer@kernel.org> Link: https://lore.kernel.org/r/20221115141054.1051801-2-yukuai1@huaweicloud.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
sbitmap: Try each queue to wake up at least one waiter
Jan reported the new algorithm as merged might be problematic if the
queue being awaken becomes empty between the waitqueue_active inside
sbq_wake_ptr check and the wake up. If that happens, wake_up_nr will
not wake up any waiter and we loose too many wake ups. In order to
guarantee progress, we need to wake up at least one waiter here, if
there are any. This now requires trying to wake up from every queue.
Instead of walking through all the queues with sbq_wake_ptr, this call
moves the wake up inside that function. In a previous version of the
patch, I found that updating wake_index several times when walking
through queues had a measurable overhead. This ensures we only update
it once, at the end.
Fixes: 4f8126bb2308 ("sbitmap: Use single per-bitmap counting to wake up queued tags") Reported-by: Jan Kara <jack@suse.cz> Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de> Reviewed-by: Jan Kara <jack@suse.cz> Link: https://lore.kernel.org/r/20221115224553.23594-4-krisman@suse.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Sbitmap code will need to know how many waiters were actually woken for
its batched wakeups implementation. Return the number of woken
exclusive waiters from __wake_up() to facilitate that.
sbitmap: Advance the queue index before waking up a queue
When a queue is awaken, the wake_index written by sbq_wake_ptr currently
keeps pointing to the same queue. On the next wake up, it will thus
retry the same queue, which is unfair to other queues, and can lead to
starvation. This patch, moves the index update to happen before the
queue is returned, such that it will now try a different queue first on
the next wake up, improving fairness.
Fixes: 4f8126bb2308 ("sbitmap: Use single per-bitmap counting to wake up queued tags") Reported-by: Jan Kara <jack@suse.cz> Reviewed-by: Jan Kara <jack@suse.cz> Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de> Link: https://lore.kernel.org/r/20221115224553.23594-2-krisman@suse.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
While the block device code should switch to implementing
->writepages instead of ->writepage eventually, the current
implementation is entirely pointless as it does the same looping over
->writepage as the generic code if no ->writepages is present.
Remove blkdev_writepages so that we can eventually unexport
generic_writepages.
Pavel Begunkov [Wed, 2 Nov 2022 15:18:23 +0000 (15:18 +0000)]
bio: shrink max number of pcpu cached bios
The downside of the bio pcpu cache is that bios of a cpu will be never
freed unless there is new I/O issued from that cpu. We currently keep
max 512 bios, which feels too much, half it.
Pavel Begunkov [Wed, 2 Nov 2022 15:18:22 +0000 (15:18 +0000)]
bio: add pcpu caching for non-polling bio_put
This patch extends REQ_ALLOC_CACHE to IRQ completions, whenever
currently it's only limited to iopoll. Instead of guarding the list with
irq toggling on alloc, which is expensive, it keeps an additional
irq-safe list from which bios are spliced in batches to ammortise
overhead. On the put side it toggles irqs, but in many cases they're
already disabled and so cheap.
Pavel Begunkov [Wed, 2 Nov 2022 15:18:20 +0000 (15:18 +0000)]
bio: don't rob starving biosets of bios
Biosets keep a mempool, so as long as requests complete we can always
can allocate and have forward progress. Percpu bio caches break that
assumptions as we may complete into the cache of one CPU and after try
and fail to allocate with another CPU. We also can't grab from another
CPU's cache without tricky sync.
If we're allocating with a bio while the mempool is undersaturated,
remove REQ_ALLOC_CACHE flag, so on put it will go straight to mempool.
It might try to free into mempool more requests than required, but
assuming than there is no memory starvation in the system it'll
stabilise and never hit that path.
Pavel Begunkov [Wed, 2 Nov 2022 15:18:19 +0000 (15:18 +0000)]
mempool: introduce mempool_is_saturated
Introduce a helper mempool_is_saturated(), which tells if the mempool is
under-filled or not. We need it to figure out whether it should be
freed right into the mempool or could be cached with top level caches.
Jens Axboe [Mon, 14 Nov 2022 19:57:50 +0000 (12:57 -0700)]
Merge branch 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md into for-6.2/block
Pull MD fixes from Song.
* 'md-next' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
md/raid1: stop mdx_raid1 thread when raid1 array run failed
md/raid5: use bdev_write_cache instead of open coding it
md: fix a crash in mempool_free
md/raid0, raid10: Don't set discard sectors for request queue
md/bitmap: Fix bitmap chunk size overflow issues
md: introduce md_ro_state
md: factor out __md_set_array_info()
lib/raid6: drop RAID6_USE_EMPTY_ZERO_PAGE
raid5-cache: use try_cmpxchg in r5l_wake_reclaim
drivers/md/md-bitmap: check the return value of md_bitmap_get_counter()
Jiang Li [Mon, 7 Nov 2022 14:16:59 +0000 (22:16 +0800)]
md/raid1: stop mdx_raid1 thread when raid1 array run failed
fail run raid1 array when we assemble array with the inactive disk only,
but the mdx_raid1 thread were not stop, Even if the associated resources
have been released. it will caused a NULL dereference when we do poweroff.
Mikulas Patocka [Fri, 4 Nov 2022 13:53:38 +0000 (09:53 -0400)]
md: fix a crash in mempool_free
There's a crash in mempool_free when running the lvm test
shell/lvchange-rebuild-raid.sh.
The reason for the crash is this:
* super_written calls atomic_dec_and_test(&mddev->pending_writes) and
wake_up(&mddev->sb_wait). Then it calls rdev_dec_pending(rdev, mddev)
and bio_put(bio).
* so, the process that waited on sb_wait and that is woken up is racing
with bio_put(bio).
* if the process wins the race, it calls bioset_exit before bio_put(bio)
is executed.
* bio_put(bio) attempts to free a bio into a destroyed bio set - causing
a crash in mempool_free.
We fix this bug by moving bio_put before atomic_dec_and_test.
We also move rdev_dec_pending before atomic_dec_and_test as suggested by
Neil Brown.
The function md_end_flush has a similar bug - we must call bio_put before
we decrement the number of in-progress bios.
Xiao Ni [Wed, 2 Nov 2022 02:07:30 +0000 (10:07 +0800)]
md/raid0, raid10: Don't set discard sectors for request queue
It should use disk_stack_limits to get a proper max_discard_sectors
rather than setting a value by stack drivers.
And there is a bug. If all member disks are rotational devices,
raid0/raid10 set max_discard_sectors. So the member devices are
not ssd/nvme, but raid0/raid10 export the wrong value. It reports
warning messages in function __blkdev_issue_discard when mkfs.xfs
like this:
Signed-off-by: Xiao Ni <xni@redhat.com> Reported-by: Yi Zhang <yi.zhang@redhat.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Song Liu <song@kernel.org>
- limit bitmap chunk size internal u64 variable to values not overflowing
the u32 bitmap superblock structure variable stored on persistent media
- assign bitmap chunk size internal u64 variable from unsigned values to
avoid possible sign extension artifacts when assigning from a s32 value
The bug has been there since at least kernel 4.0.
Steps to reproduce it:
1: mdadm -C /dev/mdx -l 1 --bitmap=internal --bitmap-chunk=256M -e 1.2
-n2 /dev/rnbd1 /dev/rnbd2
2 resize member device rnbd1 and rnbd2 to 8 TB
3 mdadm --grow /dev/mdx --size=max
The bitmap_chunksize will overflow without patch.
Cc: stable@vger.kernel.org Signed-off-by: Florian-Ewald Mueller <florian-ewald.mueller@ionos.com> Signed-off-by: Jack Wang <jinpu.wang@ionos.com> Signed-off-by: Song Liu <song@kernel.org>
Giulio Benetti [Wed, 19 Oct 2022 16:04:07 +0000 (18:04 +0200)]
lib/raid6: drop RAID6_USE_EMPTY_ZERO_PAGE
RAID6_USE_EMPTY_ZERO_PAGE is unused and hardcoded to 0, so let's drop it.
Signed-off-by: Giulio Benetti <giulio.benetti@benettiengineering.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Song Liu <song@kernel.org>
Uros Bizjak [Thu, 20 Oct 2022 15:51:04 +0000 (17:51 +0200)]
raid5-cache: use try_cmpxchg in r5l_wake_reclaim
Use try_cmpxchg instead of cmpxchg (*ptr, old, new) == old in
r5l_wake_reclaim. 86 CMPXCHG instruction returns success in ZF flag, so
this change saves a compare after cmpxchg (and related move instruction in
front of cmpxchg).
Also, try_cmpxchg implicitly assigns old *ptr value to "old" when cmpxchg
fails. There is no need to re-read the value in the loop.
Note that the value from *ptr should be read using READ_ONCE to prevent
the compiler from merging, refetching or reordering the read.
No functional change intended.
Cc: Song Liu <song@kernel.org> Signed-off-by: Uros Bizjak <ubizjak@gmail.com> Signed-off-by: Song Liu <song@kernel.org>
sbitmap: Use single per-bitmap counting to wake up queued tags
sbitmap suffers from code complexity, as demonstrated by recent fixes,
and eventual lost wake ups on nested I/O completion. The later happens,
from what I understand, due to the non-atomic nature of the updates to
wait_cnt, which needs to be subtracted and eventually reset when equal
to zero. This two step process can eventually miss an update when a
nested completion happens to interrupt the CPU in between the wait_cnt
updates. This is very hard to fix, as shown by the recent changes to
this code.
The code complexity arises mostly from the corner cases to avoid missed
wakes in this scenario. In addition, the handling of wake_batch
recalculation plus the synchronization with sbq_queue_wake_up is
non-trivial.
This patchset implements the idea originally proposed by Jan [1], which
removes the need for the two-step updates of wait_cnt. This is done by
tracking the number of completions and wakeups in always increasing,
per-bitmap counters. Instead of having to reset the wait_cnt when it
reaches zero, we simply keep counting, and attempt to wake up N threads
in a single wait queue whenever there is enough space for a batch.
Waking up less than batch_wake shouldn't be a problem, because we
haven't changed the conditions for wake up, and the existing batch
calculation guarantees at least enough remaining completions to wake up
a batch for each queue at any time.
Performance-wise, one should expect very similar performance to the
original algorithm for the case where there is no queueing. In both the
old algorithm and this implementation, the first thing is to check
ws_active, which bails out if there is no queueing to be managed. In the
new code, we took care to avoid accounting completions and wakeups when
there is no queueing, to not pay the cost of atomic operations
unnecessarily, since it doesn't skew the numbers.
For more interesting cases, where there is queueing, we need to take
into account the cross-communication of the atomic operations. I've
been benchmarking by running parallel fio jobs against a single hctx
nullb in different hardware queue depth scenarios, and verifying both
IOPS and queueing.
Each experiment was repeated 5 times on a 20-CPU box, with 20 parallel
jobs. fio was issuing fixed-size randwrites with qd=64 against nullb,
varying only the hardware queue length per test.
The following is a similar experiment, ran against a nullb with a single
bitmap shared by 20 hctx spread across 2 NUMA nodes. This has 40
parallel fio jobs operating on the same device
It has also survived blktests and a 12h-stress run against nullb. I also
ran the code against nvme and a scsi SSD, and I didn't observe
performance regression in those. If there are other tests you think I
should run, please let me know and I will follow up with results.
Cc: Hugh Dickins <hughd@google.com> Cc: Keith Busch <kbusch@kernel.org> Cc: Liu Song <liusong@linux.alibaba.com> Suggested-by: Jan Kara <jack@suse.cz> Signed-off-by: Gabriel Krisman Bertazi <krisman@suse.de> Link: https://lore.kernel.org/r/20221105231055.25953-1-krisman@suse.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
This fixes crashes in bfq_add_bfqq_busy due to waker_bfqq being NULL,
but woken_list_node still being hashed. This would happen when
bfq_init_rq() expects a brand new allocated queue to be returned from
bfq_get_bfqq_handle_split() and unconditionally updates waker_bfqq
without resetting woken_list_node. Since we can always return oom_bfqq
when attempting to allocate, we cannot assume waker_bfqq starts as NULL.
Avoid setting woken_bfqq for oom_bfqq entirely, as it's not useful.
(Sort of) cherry-picked from the out-of-tree drbd9 branch. Original
commit message by Joel Colledge:
This simplifies drbd_submit_peer_request by removing most of the
arguments. It also makes the treatment of the op better aligned with
that in struct bio.
Determine fault_type dynamically using information which is already
available instead of passing it in as a parameter.
Note: The opf in receive_rs_deallocated was changed from
REQ_OP_WRITE_ZEROES to REQ_OP_DISCARD. This was required in the
out-of-tree module, and does not matter in-tree. The opf is ignored
anyway in drbd_submit_peer_request, since the discard/zero-out is
decided by the EE_TRIM flag.
Philipp Reisner [Wed, 9 Nov 2022 13:34:52 +0000 (14:34 +0100)]
drbd: disable discard support if granularity > max
The discard_granularity describes the minimum unit of a discard.
If that is larger than the maximal discard size, we need to disable
discards completely.
We currently only set q->limits.max_discard_sectors, but that is not
enough. Another field, max_hw_discard_sectors, was introduced in
commit 0034af036554 ("block: make /sys/block/<dev>/queue/discard_max_bytes
writeable").
The difference is that max_discard_sectors can be changed from user
space via sysfs, while max_hw_discard_sectors is the "hardware" upper
limit.
So use this helper, which sets both.
This is also a fixup for commit 998e9cbcd615 ("drbd: cleanup
decide_on_discard_support"): if discards are not supported, that does
not necessarily mean we also want to disable write_zeroes.
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:16 +0000 (11:41 -0600)]
ABI: sysfs-bus-pci: add documentation for p2pmem allocate
Add documentation for the p2pmem/allocate binary file which allows
for allocating p2pmem buffers in userspace for passing to drivers
that support them. (Currently only O_DIRECT to NVMe devices.)
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:15 +0000 (11:41 -0600)]
PCI/P2PDMA: Allow userspace VMA allocations through sysfs
Create a sysfs bin attribute called "allocate" under the existing
"p2pmem" group. The only allowable operation on this file is the mmap()
call.
When mmap() is called on this attribute, the kernel allocates a chunk of
memory from the genalloc and inserts the pages into the VMA. The
dev_pagemap .page_free callback will indicate when these pages are no
longer used and they will be put back into the genalloc.
On device unbind, remove the sysfs file before the memremap_pages are
cleaned up. This ensures unmap_mapping_range() is called on the files
inode and no new mappings can be created.
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:14 +0000 (11:41 -0600)]
block: set FOLL_PCI_P2PDMA in bio_map_user_iov()
When a bio's queue supports PCI P2PDMA, set FOLL_PCI_P2PDMA for
iov_iter_get_pages_flags(). This allows PCI P2PDMA pages to be
passed from userspace and enables the NVMe passthru requests to
use P2PDMA pages.
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:13 +0000 (11:41 -0600)]
block: set FOLL_PCI_P2PDMA in __bio_iov_iter_get_pages()
When a bio's queue supports PCI P2PDMA, set FOLL_PCI_P2PDMA for
iov_iter_get_pages_flags(). This allows PCI P2PDMA pages to be passed
from userspace and enables the O_DIRECT path in iomap based filesystems
and direct to block devices.
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:12 +0000 (11:41 -0600)]
lib/scatterlist: add check when merging zone device pages
Consecutive zone device pages should not be merged into the same sgl
or bvec segment with other types of pages or if they belong to different
pgmaps. Otherwise getting the pgmap of a given segment is not possible
without scanning the entire segment. This helper returns true either if
both pages are not zone device pages or both pages are zone device
pages with the same pgmap.
Factor out the check for page mergability into a pages_are_mergable()
helper and add a check with zone_device_pages_are_mergeable().
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:11 +0000 (11:41 -0600)]
block: add check when merging zone device pages
Consecutive zone device pages should not be merged into the same sgl
or bvec segment with other types of pages or if they belong to different
pgmaps. Otherwise getting the pgmap of a given segment is not possible
without scanning the entire segment. This helper returns true either if
both pages are not zone device pages or both pages are zone device
pages with the same pgmap.
Add a helper to determine if zone device pages are mergeable and use
this helper in page_is_mergeable().
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:09 +0000 (11:41 -0600)]
mm: introduce FOLL_PCI_P2PDMA to gate getting PCI P2PDMA pages
GUP Callers that expect PCI P2PDMA pages can now set FOLL_PCI_P2PDMA to
allow obtaining P2PDMA pages. If GUP is called without the flag and a
P2PDMA page is found, it will return an error in try_grab_page() or
try_grab_folio().
The check is safe to do before taking the reference to the page in both
cases seeing the page should be protected by either the appropriate
ptl or mmap_lock; or the gup fast guarantees preventing TLB flushes.
try_grab_folio() has one call site that WARNs on failure and cannot
actually deal with the failure of this function (it seems it will
get into an infinite loop). Expand the comment there to document a
couple more conditions on why it will not fail.
FOLL_PCI_P2PDMA cannot be set if FOLL_LONGTERM is set. This is to copy
fsdax until pgmap refcounts are fixed (see the link below for more
information).
Logan Gunthorpe [Fri, 21 Oct 2022 17:41:08 +0000 (11:41 -0600)]
mm: allow multiple error returns in try_grab_page()
In order to add checks for P2PDMA memory into try_grab_page(), expand
the error return from a bool to an int/error code. Update all the
callsites handle change in usage.
Also remove the WARN_ON_ONCE() call at the callsites seeing there
already is a WARN_ON_ONCE() inside the function if it fails.
Signed-off-by: Logan Gunthorpe <logang@deltatee.com> Reviewed-by: Dan Williams <dan.j.williams@intel.com> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Link: https://lore.kernel.org/r/20221021174116.7200-2-logang@deltatee.com Signed-off-by: Jens Axboe <axboe@kernel.dk>
Chao Leng [Tue, 1 Nov 2022 15:00:50 +0000 (16:00 +0100)]
nvme: use blk_mq_[un]quiesce_tagset
All controller namespaces share the same tagset, so we can use this
interface which does the optimal operation for parallel quiesce based on
the tagset type(e.g. blocking tagsets and non-blocking tagsets).
nvme connect_q should not be quiesced when quiesce tagset, so set the
QUEUE_FLAG_SKIP_TAGSET_QUIESCE to skip it when init connect_q.
Currently we use NVME_NS_STOPPED to ensure pairing quiescing and
unquiescing. If use blk_mq_[un]quiesce_tagset, NVME_NS_STOPPED will be
invalided, so introduce NVME_CTRL_STOPPED to replace NVME_NS_STOPPED.
In addition, we never really quiesce a single namespace. It is a better
choice to move the flag from ns to ctrl.
Signed-off-by: Chao Leng <lengchao@huawei.com>
[hch: rebased on top of prep patches] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Chao Leng <lengchao@huawei.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Link: https://lore.kernel.org/r/20221101150050.3510-15-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Chao Leng [Tue, 1 Nov 2022 15:00:49 +0000 (16:00 +0100)]
blk-mq: add tagset quiesce interface
Drivers that have shared tagsets may need to quiesce potentially a lot
of request queues that all share a single tagset (e.g. nvme). Add an
interface to quiesce all the queues on a given tagset. This interface is
useful because it can speedup the quiesce by doing it in parallel.
Because some queues should not need to be quiesced (e.g. the nvme
connect_q) when quiescing the tagset, introduce a
QUEUE_FLAG_SKIP_TAGSET_QUIESCE flag to allow this new interface to
ski quiescing a particular queue.
Signed-off-by: Chao Leng <lengchao@huawei.com>
[hch: simplify for the per-tag_set srcu_struct] Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Chao Leng <lengchao@huawei.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Link: https://lore.kernel.org/r/20221101150050.3510-14-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Nothing in blk_mq_wait_quiesce_done needs the request_queue now, so just
pass the tagset, and move the non-mq check into the only caller that
needs it.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Chao Leng <lengchao@huawei.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Link: https://lore.kernel.org/r/20221101150050.3510-13-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk-mq: skip non-mq queues in blk_mq_quiesce_queue
For submit_bio based queues there is no (S)RCU critical section during
I/O submission and thus nothing to wait for in blk_mq_wait_quiesce_done,
so skip doing any synchronization. No non-mq driver should be calling
this, but for now we have core callers that unconditionally call into it.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Link: https://lore.kernel.org/r/20221101150050.3510-11-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
1) mark the gendisk of all namespaces dead
2) unquiesce all I/O queues
These used to be be intertwined due to block layer issues, but aren't
any more. So move the unquiscing of the I/O queues into the callers,
and rename the rest of the function to the now more descriptive
nvme_mark_namespaces_dead.
nvme: don't unquiesce the admin queue in nvme_kill_queues
None of the callers of nvme_kill_queues needs it to unquiesce the
admin queues, as all of them already do it themselves:
1) nvme_reset_work explicit call nvme_start_admin_queue toward the
beginning of the function. The extra call to nvme_start_admin_queue
in nvme_reset_work this won't do anything as
NVME_CTRL_ADMIN_Q_STOPPED will already be cleared.
2) nvme_remove calls nvme_dev_disable with shutdown flag set to true at
the very beginning of the function if the PCIe device was not present,
which is the precondition for the call to nvme_kill_queues.
nvme_dev_disable already calls nvme_start_admin_queue toward the
end of the function when the shutdown flag is set to true, so the
admin queue is already enabled at this point.
3) nvme_remove_dead_ctrl schedules a workqueue to unbind the driver,
which will end up in nvme_remove, which calls nvme_dev_disable with
the shutdown flag. This case will call nvme_start_admin_queue a bit
later than before.
4) apple_nvme_remove uses the same sequence as nvme_remove_dead_ctrl
above.
5) nvme_remove_namespaces only calls nvme_kill_queues when the
controller is in the DEAD state. That can only happen in the PCIe
driver, and only from nvme_remove. See item 2) above for the
conditions there.
So it is safe to just remove the call to nvme_start_admin_queue in
nvme_kill_queues without replacement.
nvme: remove the NVME_NS_DEAD check in nvme_remove_invalid_namespaces
The NVME_NS_DEAD check only made sense when we revalidated namespaces
in nvme_passthrough_end for commands that affected the namespace inventory.
These days NVME_NS_DEAD is only set during reset or when tearing down
namespaces, and we always remove all namespaces right after that.
nvme: don't remove namespaces in nvme_passthru_end
The call to nvme_remove_invalid_namespaces made sense when
nvme_passthru_end revalidated all namespaces and had to remove those that
didn't exist any more. Since we don't revalidate from nvme_passthru_end
now, this call is entirely spurious.
nvme-pci: refactor the tagset handling in nvme_reset_work
The code to create, update or delete a tagset and namespaces in
nvme_reset_work is a bit convoluted. Refactor it with a two high-level
conditionals for first probe vs reset and I/O queues vs no I/O queues
to make the code flow more clear.
block: set the disk capacity to 0 in blk_mark_disk_dead
nvme and xen-blkfront are already doing this to stop buffered writes from
creating dirty pages that can't be written out later. Move it to the
common code.
This also removes the comment about the ordering from nvme, as bd_mutex
not only is gone entirely, but also hasn't been used for locking updates
to the disk size long before that, and thus the ordering requirement
documented there doesn't apply any more.
Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Chao Leng <lengchao@huawei.com> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Link: https://lore.kernel.org/r/20221101150050.3510-2-hch@lst.de Signed-off-by: Jens Axboe <axboe@kernel.dk>
Kemeng Shi [Tue, 18 Oct 2022 11:12:40 +0000 (19:12 +0800)]
block: Replace struct rq_depth with unsigned int in struct iolatency_grp
We only need a max queue depth for every iolatency to limit the inflight io
number. Replace struct rq_depth with unsigned int to simplfy "struct
iolatency_grp" and save memory.
Kemeng Shi [Tue, 18 Oct 2022 11:12:39 +0000 (19:12 +0800)]
block: Correct comment for scale_cookie_change
Default queue depth of iolatency_grp is unlimited, so we scale down
quickly(once by half) in scale_cookie_change. Remove the "subtract
1/16th" part which is not the truth and add the actual way we
scale down.
Kemeng Shi [Tue, 18 Oct 2022 11:12:38 +0000 (19:12 +0800)]
block: Remove redundant parent blkcg_gp check in check_scale_change
Function blkcg_iolatency_throttle will make sure blkg->parent is not
NULL before calls check_scale_change. And function check_scale_change
is only called in blkcg_iolatency_throttle.
Split an elevator_disable helper from elevator_switch for the case where
we want to switch to no scheduler at all. This includes removing the
pointless elevator_switch_mq helper and removing the switch to no
schedule logic from blk_mq_init_sched.
block: cleanup the variable naming in elv_iosched_store
Use eq for the elevator_queue as done elsewhere. This frees e to be used
for the loop iterator instead of the odd __ prefix. In addition rename
elv to cur to make it more clear it is the currently selected elevator.
Do the request_module and repeated lookup in the only caller that cares,
pick a saner name that explains where are actually doing a lookup and
use a sane calling conventions that passes the queue first.
projects / mirror_ubuntu-kernels.git / log