platform/kernel/linux-rpi.git
20 months agodm crypt: correct 'foo*' to 'foo *'
Heinz Mauelshagen [Tue, 31 Jan 2023 18:47:36 +0000 (19:47 +0100)]
dm crypt: correct 'foo*' to 'foo *'

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: fix trailing statements
Heinz Mauelshagen [Mon, 30 Jan 2023 21:13:54 +0000 (22:13 +0100)]
dm: fix trailing statements

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: fix undue/missing spaces
Heinz Mauelshagen [Mon, 30 Jan 2023 20:43:57 +0000 (21:43 +0100)]
dm: fix undue/missing spaces

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: correct block comments format.
Heinz Mauelshagen [Thu, 26 Jan 2023 14:48:30 +0000 (15:48 +0100)]
dm: correct block comments format.

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: address indent/space issues
Heinz Mauelshagen [Wed, 25 Jan 2023 22:31:55 +0000 (23:31 +0100)]
dm: address indent/space issues

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: address space issues relative to switch/while/for/...
Heinz Mauelshagen [Wed, 25 Jan 2023 21:57:42 +0000 (22:57 +0100)]
dm: address space issues relative to switch/while/for/...

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: avoid initializing static variables
Heinz Mauelshagen [Mon, 30 Jan 2023 20:28:24 +0000 (21:28 +0100)]
dm: avoid initializing static variables

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: enclose complex macros into parentheses where possible
Heinz Mauelshagen [Wed, 1 Feb 2023 20:51:04 +0000 (21:51 +0100)]
dm: enclose complex macros into parentheses where possible

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: avoid assignment in if conditions
Heinz Mauelshagen [Wed, 1 Feb 2023 20:17:44 +0000 (21:17 +0100)]
dm: avoid assignment in if conditions

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: change "unsigned" to "unsigned int"
Heinz Mauelshagen [Wed, 25 Jan 2023 20:14:58 +0000 (21:14 +0100)]
dm: change "unsigned" to "unsigned int"

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: use fsleep() instead of msleep() for deterministic sleep duration
Heinz Mauelshagen [Wed, 25 Jan 2023 21:44:39 +0000 (22:44 +0100)]
dm: use fsleep() instead of msleep() for deterministic sleep duration

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: prefer kmap_local_page() instead of deprecated kmap_atomic()
Heinz Mauelshagen [Tue, 7 Feb 2023 19:22:58 +0000 (20:22 +0100)]
dm: prefer kmap_local_page() instead of deprecated kmap_atomic()

Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: add missing SPDX-License-Indentifiers
Heinz Mauelshagen [Wed, 25 Jan 2023 20:00:44 +0000 (21:00 +0100)]
dm: add missing SPDX-License-Indentifiers

'GPL-2.0-only' is used instead of 'GPL-2.0' because SPDX has
deprecated its use.

Suggested-by: John Wiele <jwiele@redhat.com>
Signed-off-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm: send just one event on resize, not two
Mikulas Patocka [Tue, 7 Feb 2023 13:33:06 +0000 (08:33 -0500)]
dm: send just one event on resize, not two

Device mapper sends an uevent when the device is suspended, using the
function set_capacity_and_notify. However, this causes a race condition
with udev.

Udev skips scanning dm devices that are suspended. If we send an uevent
while we are suspended, udev will be racing with device mapper resume
code. If the device mapper resume code wins the race, udev will process
the uevent after the device is resumed and it will properly scan the
device.

However, if udev wins the race, it will receive the uevent, find out that
the dm device is suspended and skip scanning the device. This causes bugs
such as systemd unmounting the device - see
https://bugzilla.redhat.com/show_bug.cgi?id=2158628

This commit fixes this race.

We replace the function set_capacity_and_notify with set_capacity, so that
the uevent is not sent at this point. In do_resume, we detect if the
capacity has changed and we pass a boolean variable need_resize_uevent to
dm_kobject_uevent. dm_kobject_uevent adds "RESIZE=1" to the uevent if
need_resize_uevent is set.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Tested-by: Peter Rajnoha <prajnoha@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm table: check that a dm device doesn't reference itself
Benjamin Marzinski [Tue, 31 Jan 2023 21:22:57 +0000 (15:22 -0600)]
dm table: check that a dm device doesn't reference itself

If a DM device's table references itself, it will crash the kernel with an
infinite recursion.  Check for a self-reference in dm_get_device(). This
is a quick check, but it won't catch more complicated circular references.

Signed-off-by: Benjamin Marzinski <bmarzins@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
20 months agodm raid: fix some spelling mistakes in comments
Yu Zhe [Mon, 6 Feb 2023 03:27:39 +0000 (11:27 +0800)]
dm raid: fix some spelling mistakes in comments

Signed-off-by: Yu Zhe <yuzhe@nfschina.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm verity: stop using WQ_UNBOUND for verify_wq
Nathan Huckleberry [Thu, 2 Feb 2023 01:23:48 +0000 (17:23 -0800)]
dm verity: stop using WQ_UNBOUND for verify_wq

Setting WQ_UNBOUND increases scheduler latency on ARM64.  This is
likely due to the asymmetric architecture of ARM64 processors.

I've been unable to reproduce the results that claim WQ_UNBOUND gives
a performance boost on x86-64.

This flag is causing performance issues for multiple subsystems within
Android.  Notably, the same slowdown exists for decompression with
EROFS.

| open-prebuilt-camera  | WQ_UNBOUND | ~WQ_UNBOUND   |
|-----------------------|------------|---------------|
| verity wait time (us) | 11746      | 119 (-98%)    |
| erofs wait time (us)  | 357805     | 174205 (-51%) |

| sha256 ramdisk random read | WQ_UNBOUND    | ~WQ_UNBOUND |
|----------------------------|-----------=---|-------------|
| arm64 (accelerated)        | bw=42.4MiB/s  | bw=212MiB/s |
| arm64 (generic)            | bw=16.5MiB/s  | bw=48MiB/s  |
| x86_64 (generic)           | bw=233MiB/s   | bw=230MiB/s |

Using a alloc_workqueue() @max_active arg of num_online_cpus() only
made sense with WQ_UNBOUND. Switch the @max_active arg to 0 (aka
default, which is 256 per-cpu).

Also, eliminate 'wq_flags' since it really doesn't serve a purpose.

Cc: Sami Tolvanen <samitolvanen@google.com>
Cc: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Nathan Huckleberry <nhuck@google.com>
Reviewed-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm integrity: Remove bi_sector that's only used by commented debug code
Jiapeng Chong [Tue, 31 Jan 2023 06:09:41 +0000 (14:09 +0800)]
dm integrity: Remove bi_sector that's only used by commented debug code

drivers/md/dm-integrity.c:1738:13: warning: variable 'bi_sector' set but not used.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Link: https://bugzilla.openanolis.cn/show_bug.cgi?id=3895
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm crypt: Slightly simplify crypt_set_keyring_key()
Christophe JAILLET [Sat, 28 Jan 2023 15:43:00 +0000 (16:43 +0100)]
dm crypt: Slightly simplify crypt_set_keyring_key()

Use strchr() instead of strpbrk() when there is only 1 element in the set
of characters to look for.

This potentially saves a few cycles, but gcc does already account for
optimizing this pattern thanks to it's fold_builtin_strpbrk().

Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm ioctl: drop always-false condition
Sergey Shtylyov [Tue, 17 Jan 2023 20:59:55 +0000 (23:59 +0300)]
dm ioctl: drop always-false condition

The expression 'indata[3] > ULONG_MAX' always evaluates to false since
indata[] is declared as an array of *unsigned long* elements and #define
ULONG_MAX represents the max value of that exact type...

Note that gcc seems to be able to detect the dead code here and eliminate
this check anyway...

Found by Linux Verification Center (linuxtesting.org) with the SVACE static
analysis tool.

Signed-off-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Reviewed-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm flakey: fix logic when corrupting a bio
Mikulas Patocka [Sun, 22 Jan 2023 19:03:56 +0000 (14:03 -0500)]
dm flakey: fix logic when corrupting a bio

If "corrupt_bio_byte" is set to corrupt reads and corrupt_bio_flags is
used, dm-flakey would erroneously return all writes as errors. Likewise,
if "corrupt_bio_byte" is set to corrupt writes, dm-flakey would return
errors for all reads.

Fix the logic so that if fc->corrupt_bio_byte is non-zero, dm-flakey
will not abort reads on writes with an error.

Cc: stable@vger.kernel.org
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Reviewed-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm flakey: fix a bug with 32-bit highmem systems
Mikulas Patocka [Sun, 22 Jan 2023 19:03:31 +0000 (14:03 -0500)]
dm flakey: fix a bug with 32-bit highmem systems

The function page_address does not work with 32-bit systems with high
memory. Use bvec_kmap_local/kunmap_local instead.

Cc: stable@vger.kernel.org
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Reviewed-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm flakey: don't corrupt the zero page
Mikulas Patocka [Sun, 22 Jan 2023 19:02:57 +0000 (14:02 -0500)]
dm flakey: don't corrupt the zero page

When we need to zero some range on a block device, the function
__blkdev_issue_zero_pages submits a write bio with the bio vector pointing
to the zero page. If we use dm-flakey with corrupt bio writes option, it
will corrupt the content of the zero page which results in crashes of
various userspace programs. Glibc assumes that memory returned by mmap is
zeroed and it uses it for calloc implementation; if the newly mapped
memory is not zeroed, calloc will return non-zeroed memory.

Fix this bug by testing if the page is equal to ZERO_PAGE(0) and
avoiding the corruption in this case.

Cc: stable@vger.kernel.org
Fixes: a00f5276e266 ("dm flakey: Properly corrupt multi-page bios.")
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Reviewed-by: Sweet Tea Dorminy <sweettea-kernel@dorminy.me>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm cache: Add some documentation to dm-cache-background-tracker.h
Joe Thornber [Thu, 26 Jan 2023 10:14:26 +0000 (10:14 +0000)]
dm cache: Add some documentation to dm-cache-background-tracker.h

Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm cache: free background tracker's queued work in btracker_destroy
Joe Thornber [Thu, 26 Jan 2023 09:59:10 +0000 (09:59 +0000)]
dm cache: free background tracker's queued work in btracker_destroy

Otherwise the kernel can BUG with:

[ 2245.426978] =============================================================================
[ 2245.435155] BUG bt_work (Tainted: G    B   W         ): Objects remaining in bt_work on __kmem_cache_shutdown()
[ 2245.445233] -----------------------------------------------------------------------------
[ 2245.445233]
[ 2245.454879] Slab 0x00000000b0ce2b30 objects=64 used=2 fp=0x000000000a3c6a4e flags=0x17ffffc0000200(slab|node=0|zone=2|lastcpupid=0x1fffff)
[ 2245.467300] CPU: 7 PID: 10805 Comm: lvm Kdump: loaded Tainted: G    B   W          6.0.0-rc2 #19
[ 2245.476078] Hardware name: Dell Inc. PowerEdge R7525/0590KW, BIOS 2.5.6 10/06/2021
[ 2245.483646] Call Trace:
[ 2245.486100]  <TASK>
[ 2245.488206]  dump_stack_lvl+0x34/0x48
[ 2245.491878]  slab_err+0x95/0xcd
[ 2245.495028]  __kmem_cache_shutdown.cold+0x31/0x136
[ 2245.499821]  kmem_cache_destroy+0x49/0x130
[ 2245.503928]  btracker_destroy+0x12/0x20 [dm_cache]
[ 2245.508728]  smq_destroy+0x15/0x60 [dm_cache_smq]
[ 2245.513435]  dm_cache_policy_destroy+0x12/0x20 [dm_cache]
[ 2245.518834]  destroy+0xc0/0x110 [dm_cache]
[ 2245.522933]  dm_table_destroy+0x5c/0x120 [dm_mod]
[ 2245.527649]  __dm_destroy+0x10e/0x1c0 [dm_mod]
[ 2245.532102]  dev_remove+0x117/0x190 [dm_mod]
[ 2245.536384]  ctl_ioctl+0x1a2/0x290 [dm_mod]
[ 2245.540579]  dm_ctl_ioctl+0xa/0x20 [dm_mod]
[ 2245.544773]  __x64_sys_ioctl+0x8a/0xc0
[ 2245.548524]  do_syscall_64+0x5c/0x90
[ 2245.552104]  ? syscall_exit_to_user_mode+0x12/0x30
[ 2245.556897]  ? do_syscall_64+0x69/0x90
[ 2245.560648]  ? do_syscall_64+0x69/0x90
[ 2245.564394]  entry_SYSCALL_64_after_hwframe+0x63/0xcd
[ 2245.569447] RIP: 0033:0x7fe52583ec6b
...
[ 2245.646771] ------------[ cut here ]------------
[ 2245.651395] kmem_cache_destroy bt_work: Slab cache still has objects when called from btracker_destroy+0x12/0x20 [dm_cache]
[ 2245.651408] WARNING: CPU: 7 PID: 10805 at mm/slab_common.c:478 kmem_cache_destroy+0x128/0x130

Found using: lvm2-testsuite --only "cache-single-split.sh"

Ben bisected and found that commit 0495e337b703 ("mm/slab_common:
Deleting kobject in kmem_cache_destroy() without holding
slab_mutex/cpu_hotplug_lock") first exposed dm-cache's incomplete
cleanup of its background tracker work objects.

Reported-by: Benjamin Marzinski <bmarzins@redhat.com>
Tested-by: Benjamin Marzinski <bmarzins@redhat.com>
Cc: stable@vger.kernel.org # 6.0+
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@kernel.org>
21 months agodm: improve shrinker debug names
Mike Snitzer [Wed, 25 Jan 2023 18:01:46 +0000 (13:01 -0500)]
dm: improve shrinker debug names

Commit e33c267ab70d ("mm: shrinkers: provide shrinkers with names")
chose some fairly bad names for DM's shrinkers.

Fixes: e33c267ab70d ("mm: shrinkers: provide shrinkers with names")
Signed-off-by : Mike Snitzer <snitzer@kernel.org>

21 months agoblock: Default to use cgroup support for BFQ
Ulf Hansson [Mon, 30 Jan 2023 12:12:40 +0000 (13:12 +0100)]
block: Default to use cgroup support for BFQ

Assuming that both Kconfig options, BLK_CGROUP and IOSCHED_BFQ are set, we
most likely want cgroup support for BFQ too (BFQ_GROUP_IOSCHED), so let's
make it default y.

Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Link: https://lore.kernel.org/r/20230130121240.159456-1-ulf.hansson@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: remove unused bfq_wr_max_time in struct bfq_data
Kemeng Shi [Mon, 16 Jan 2023 09:51:53 +0000 (17:51 +0800)]
block, bfq: remove unused bfq_wr_max_time in struct bfq_data

bfqd->bfq_wr_max_time is set to 0 in bfq_init_queue and is never changed.
It is only used in bfq_wr_duration when bfq_wr_max_time > 0 which never
meets, so bfqd->bfq_wr_max_time is not used actually. Just remove it.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-9-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: remove unnecessary goto tag in bfq_dispatch_rq_from_bfqq
Kemeng Shi [Mon, 16 Jan 2023 09:51:52 +0000 (17:51 +0800)]
block, bfq: remove unnecessary goto tag in bfq_dispatch_rq_from_bfqq

We jump to tag only for returning current rq. Return directly to
remove this tag.

Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116095153.3810101-8-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: remove redundant check in bfq_put_cooperator
Kemeng Shi [Mon, 16 Jan 2023 09:51:51 +0000 (17:51 +0800)]
block, bfq: remove redundant check in bfq_put_cooperator

We have already avoided a circular list in bfq_setup_merge (see comments
in bfq_setup_merge() for details), so bfq_queue will not appear in it's
new_bfqq list. Just remove this check.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-7-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: remove unnecessary dereference to get async_bfqq
Kemeng Shi [Mon, 16 Jan 2023 09:51:50 +0000 (17:51 +0800)]
block, bfq: remove unnecessary dereference to get async_bfqq

The async_bfqq is assigned with bfqq->bic->bfqq[0], use it directly.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-6-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: use helper macro RQ_BFQQ to get bfqq of request
Kemeng Shi [Mon, 16 Jan 2023 09:51:49 +0000 (17:51 +0800)]
block, bfq: use helper macro RQ_BFQQ to get bfqq of request

Use helper macro RQ_BFQQ to get bfqq of request.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-5-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: initialize bfqq->decrease_time_jif correctly
Kemeng Shi [Mon, 16 Jan 2023 09:51:48 +0000 (17:51 +0800)]
block, bfq: initialize bfqq->decrease_time_jif correctly

Inject limit is updated or reset when time_is_before_eq_jiffies(
decrease_time_jif + several msecs) or think-time state changes.
decrease_time_jif is initialized to 0 and will be set to current jiffies
when inject limit is updated or reset. If the jiffies is slightly greater
than LONG_MAX, time_is_after_eq_jiffies(0) will keep for a long time, so as
time_is_after_eq_jiffies(decrease_time_jif + several msecs). If the
think-time state never chages, then the injection will not work as expected
for long time.

To be more specific:
Function bfq_update_inject_limit maybe triggered when jiffies pasts
decrease_time_jif + msecs_to_jiffies(10) in bfq_add_request by setting
bfqd->wait_dispatch to true.
Function bfq_reset_inject_limit are called in two conditions:
1. jiffies pasts bfqq->decrease_time_jif + msecs_to_jiffies(1000) in
function bfq_add_request.
2. jiffies pasts bfqq->decrease_time_jif + msecs_to_jiffies(100) or
bfq think-time state change from short to long.

Fix this by initializing bfqq->decrease_time_jif to current jiffies
to trigger service injection soon when service injection conditions
are met.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-4-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: remove unsed parameter reason in bfq_bfqq_is_slow
Kemeng Shi [Mon, 16 Jan 2023 09:51:47 +0000 (17:51 +0800)]
block, bfq: remove unsed parameter reason in bfq_bfqq_is_slow

Parameter reason is never used, just remove it.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-3-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: correctly raise inject limit in bfq_choose_bfqq_for_injection
Kemeng Shi [Mon, 16 Jan 2023 09:51:46 +0000 (17:51 +0800)]
block, bfq: correctly raise inject limit in bfq_choose_bfqq_for_injection

Function bfq_choose_bfqq_for_injection may temporarily raise inject limit
to one request if current inject_limit is 0 before search of the source
queue for injection. However the search below will reset inject limit to
bfqd->in_service_queue which is zero for raised inject limit. Then the
temporarily raised inject limit never works as expected.
Assigment limit to bfqd->in_service_queue in search is needed as limit
maybe overwriten to min_t(unsigned int, 1, limit) for condition that
a large in-flight request is on non-rotational devices in found queue.
So we need to reset limit to bfqd->in_service_queue for normal case.

Actually, we have already make sure bfqd->rq_in_driver is < limit before
search, then
 -Limit is >= 1 as bfqd->rq_in_driver is >= 0. Then min_t(unsigned int,
1, limit) is always 1. So we can simply check bfqd->rq_in_driver with
1 instead of result of min_t(unsigned int, 1, limit) for larget request in
non-rotational device case to avoid overwritting limit and the bug is gone.
 -For normal case, we have already check bfqd->rq_in_driver is < limit,
so we can return found bfqq unconditionally to remove unncessary check.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20230116095153.3810101-2-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agosbitmap: correct wake_batch recalculation to avoid potential IO hung
Kemeng Shi [Mon, 16 Jan 2023 20:50:59 +0000 (04:50 +0800)]
sbitmap: correct wake_batch recalculation to avoid potential IO hung

Commit 180dccb0dba4f ("blk-mq: fix tag_get wait task can't be awakened")
mentioned that in case of shared tags, there could be just one real
active hctx(queue) because of lazy detection of tag idle. Then driver tag
allocation may wait forever on this real active hctx(queue) if wake_batch
is > hctx_max_depth where hctx_max_depth is available tags depth for the
actve hctx(queue). However, the condition wake_batch > hctx_max_depth is
not strong enough to avoid IO hung as the sbitmap_queue_wake_up will only
wake up one wait queue for each wake_batch even though there is only one
waiter in the woken wait queue. After this, there is only one tag to free
and wake_batch may not be reached anymore. Commit 180dccb0dba4f ("blk-mq:
fix tag_get wait task can't be awakened") methioned that driver tag
allocation may wait forever. Actually, the inactive hctx(queue) will be
truely idle after at most 30 seconds and will call blk_mq_tag_wakeup_all
to wake one waiter per wait queue to break the hung. But IO hung for 30
seconds is also not acceptable. Set batch size to small enough that depth
of the shared hctx(queue) is enough to wake up all of the queues like
sbq_calc_wake_batch do to fix this potential IO hung.

Although hctx_max_depth will be clamped to at least 4 while wake_batch
recalculation does not do the clamp, the wake_batch will be always
recalculated to 1 when hctx_max_depth <= 4.

Fixes: 180dccb0dba4 ("blk-mq: fix tag_get wait task can't be awakened")
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116205059.3821738-6-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agosbitmap: add sbitmap_find_bit to remove repeat code in __sbitmap_get/__sbitmap_get_sh...
Kemeng Shi [Mon, 16 Jan 2023 20:50:58 +0000 (04:50 +0800)]
sbitmap: add sbitmap_find_bit to remove repeat code in __sbitmap_get/__sbitmap_get_shallow

There are three differences between __sbitmap_get and
__sbitmap_get_shallow when searching free bit:
1. __sbitmap_get_shallow limit number of bit to search per word.
__sbitmap_get has no such limit.
2. __sbitmap_get_shallow always searches with wrap set. __sbitmap_get set
wrap according to round_robin.
3. __sbitmap_get_shallow always searches from first bit in first word.
__sbitmap_get searches from first bit when round_robin is not set
otherwise searches from SB_NR_TO_BIT(sb, alloc_hint).

Add helper function sbitmap_find_bit function to do common search while
accept "limit depth per word", "wrap flag" and "first bit to
search" from caller to support the need of both __sbitmap_get and
__sbitmap_get_shallow.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116205059.3821738-5-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agosbitmap: rewrite sbitmap_find_bit_in_index to reduce repeat code
Kemeng Shi [Mon, 16 Jan 2023 20:50:57 +0000 (04:50 +0800)]
sbitmap: rewrite sbitmap_find_bit_in_index to reduce repeat code

Rewrite sbitmap_find_bit_in_index as following:
1. Rename sbitmap_find_bit_in_index to sbitmap_find_bit_in_word
2. Accept "struct sbitmap_word *" directly instead of accepting
"struct sbitmap *" and "int index" to get "struct sbitmap_word *".
3. Accept depth/shallow_depth and wrap for __sbitmap_get_word from caller
to support need of both __sbitmap_get_shallow and __sbitmap_get.

With helper function sbitmap_find_bit_in_word, we can remove repeat
code in __sbitmap_get_shallow to find bit considring deferred clear.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116205059.3821738-4-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agosbitmap: remove redundant check in __sbitmap_queue_get_batch
Kemeng Shi [Mon, 16 Jan 2023 20:50:56 +0000 (04:50 +0800)]
sbitmap: remove redundant check in __sbitmap_queue_get_batch

Commit fbb564a557809 ("lib/sbitmap: Fix invalid loop in
__sbitmap_queue_get_batch()") mentioned that "Checking free bits when
setting the target bits. Otherwise, it may reuse the busying bits."
This commit add check to make sure all masked bits in word before
cmpxchg is zero. Then the existing check after cmpxchg to check any
zero bit is existing in masked bits in word is redundant.

Actually, old value of word before cmpxchg is stored in val and we
will filter out busy bits in val by "(get_mask & ~val)" after cmpxchg.
So we will not reuse busy bits methioned in commit fbb564a557809
("lib/sbitmap: Fix invalid loop in __sbitmap_queue_get_batch()"). Revert
new-added check to remove redundant check.

Fixes: fbb564a55780 ("lib/sbitmap: Fix invalid loop in __sbitmap_queue_get_batch()")
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116205059.3821738-3-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agosbitmap: remove unnecessary calculation of alloc_hint in __sbitmap_get_shallow
Kemeng Shi [Mon, 16 Jan 2023 20:50:55 +0000 (04:50 +0800)]
sbitmap: remove unnecessary calculation of alloc_hint in __sbitmap_get_shallow

Updates to alloc_hint in the loop in __sbitmap_get_shallow() are mostly
pointless and equivalent to setting alloc_hint to zero (because
SB_NR_TO_BIT() considers only low sb->shift bits from alloc_hint). So
simplify the logic.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Link: https://lore.kernel.org/r/20230116205059.3821738-2-shikemeng@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-cgroup: synchronize pd_free_fn() from blkg_free_workfn() and blkcg_deactivate_pol...
Yu Kuai [Thu, 19 Jan 2023 11:03:50 +0000 (19:03 +0800)]
blk-cgroup: synchronize pd_free_fn() from blkg_free_workfn() and blkcg_deactivate_policy()

Currently parent pd can be freed before child pd:

t1: remove cgroup C1
blkcg_destroy_blkgs
 blkg_destroy
  list_del_init(&blkg->q_node)
  // remove blkg from queue list
  percpu_ref_kill(&blkg->refcnt)
   blkg_release
    call_rcu

t2: from t1
__blkg_release
 blkg_free
  schedule_work
t4: deactivate policy
blkcg_deactivate_policy
 pd_free_fn
 // parent of C1 is freed first
t3: from t2
 blkg_free_workfn
  pd_free_fn

If policy(for example, ioc_timer_fn() from iocost) access parent pd from
child pd after pd_offline_fn(), then UAF can be triggered.

Fix the problem by delaying 'list_del_init(&blkg->q_node)' from
blkg_destroy() to blkg_free_workfn(), and using a new disk level mutex to
synchronize blkg_free_workfn() and blkcg_deactivate_policy().

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230119110350.2287325-4-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-cgroup: support to track if policy is online
Yu Kuai [Thu, 19 Jan 2023 11:03:49 +0000 (19:03 +0800)]
blk-cgroup: support to track if policy is online

A new field 'online' is added to blkg_policy_data to fix following
2 problem:

1) In blkcg_activate_policy(), if pd_alloc_fn() with 'GFP_NOWAIT'
   failed, 'queue_lock' will be dropped and pd_alloc_fn() will try again
   without 'GFP_NOWAIT'. In the meantime, remove cgroup can race with
   it, and pd_offline_fn() will be called without pd_init_fn() and
   pd_online_fn(). This way null-ptr-deference can be triggered.

2) In order to synchronize pd_free_fn() from blkg_free_workfn() and
   blkcg_deactivate_policy(), 'list_del_init(&blkg->q_node)' will be
   delayed to blkg_free_workfn(), hence pd_offline_fn() can be called
   first in blkg_destroy(), and then blkcg_deactivate_policy() will
   call it again, we must prevent it.

The new field 'online' will be set after pd_online_fn() and will be
cleared after pd_offline_fn(), in the meantime pd_offline_fn() will only
be called if 'online' is set.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230119110350.2287325-3-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-cgroup: dropping parent refcount after pd_free_fn() is done
Yu Kuai [Thu, 19 Jan 2023 11:03:48 +0000 (19:03 +0800)]
blk-cgroup: dropping parent refcount after pd_free_fn() is done

Some cgroup policies will access parent pd through child pd even
after pd_offline_fn() is done. If pd_free_fn() for parent is called
before child, then UAF can be triggered. Hence it's better to guarantee
the order of pd_free_fn().

Currently refcount of parent blkg is dropped in __blkg_release(), which
is before pd_free_fn() is called in blkg_free_work_fn() while
blkg_free_work_fn() is called asynchronously.

This patch make sure pd_free_fn() called from removing cgroup is ordered
by delaying dropping parent refcount after calling pd_free_fn() for
child.

BTW, pd_free_fn() will also be called from blkcg_deactivate_policy()
from deleting device, and following patches will guarantee the order.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230119110350.2287325-2-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-mq: cleanup unused methods: blk_mq_hw_sysfs_store
Zhong Jinghua [Sat, 28 Jan 2023 03:04:19 +0000 (11:04 +0800)]
blk-mq: cleanup unused methods: blk_mq_hw_sysfs_store

We found that the blk_mq_hw_sysfs_store interface has no place to use.
The object default_hw_ctx_attrs using blk_mq_hw_sysfs_ops only uses
the show method and does not use the store method.

Since this patch:
4a46f05ebf99 ("blk-mq: move hctx and ctx counters from sysfs to debugfs")
moved the store method to debugfs, the store method is not used anymore.

So let me do some tiny work to clean up unused code.

Signed-off-by: Zhong Jinghua <zhongjinghua@huawei.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230128030419.2780298-1-zhongjinghua@huawei.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agos390/dcssblk:: don't call bio_split_to_limits
Christoph Hellwig [Mon, 23 Jan 2023 07:53:56 +0000 (08:53 +0100)]
s390/dcssblk:: don't call bio_split_to_limits

s390 iterates over the bio using bio_for_each_segment and doesn't need
any bio splitting.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Alexander Gordeev <agordeev@linux.ibm.com>
Link: https://lore.kernel.org/r/20230123075356.60847-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agops3vram: remove bio splitting
Christoph Hellwig [Mon, 23 Jan 2023 07:47:18 +0000 (08:47 +0100)]
ps3vram: remove bio splitting

ps3vram iterates over the bio one segment, that is page aligned and max
page sized chunk, a time.  Because of that there is no point in
calling bio_split_to_limits, or explicitly setting the default limits
that are only used by bio_split_to_limits.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Geoff Levand <geoff@infradead.org>
Link: https://lore.kernel.org/r/20230123074718.57951-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: treat poll queue enter similarly to timeouts
Jens Axboe [Fri, 20 Jan 2023 14:51:07 +0000 (07:51 -0700)]
block: treat poll queue enter similarly to timeouts

We ran into an issue where a production workload would randomly grind to
a halt and not continue until the pending IO had timed out. This turned
out to be a complicated interaction between queue freezing and polled
IO:

1) You have an application that does polled IO. At any point in time,
   there may be polled IO pending.

2) You have a monitoring application that issues a passthrough command,
   which is marked with side effects such that it needs to freeze the
   queue.

3) Passthrough command is started, which calls blk_freeze_queue_start()
   on the device. At this point the queue is marked frozen, and any
   attempt to enter the queue will fail (for non-blocking) or block.

4) Now the driver calls blk_mq_freeze_queue_wait(), which will return
   when the queue is quiesced and pending IO has completed.

5) The pending IO is polled IO, but any attempt to poll IO through the
   normal iocb_bio_iopoll() -> bio_poll() will fail when it gets to
   bio_queue_enter() as the queue is frozen. Rather than poll and
   complete IO, the polling threads will sit in a tight loop attempting
   to poll, but failing to enter the queue to do so.

The end result is that progress for either application will be stalled
until all pending polled IO has timed out. This causes obvious huge
latency issues for the application doing polled IO, but also long delays
for passthrough command.

Fix this by treating queue enter for polled IO just like we do for
timeouts. This allows quick quiesce of the queue as we still poll and
complete this IO, while still disallowing queueing up new IO.

Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: change div64_u64 to DIV64_U64_ROUND_UP in ioc_refresh_params()
Li Nan [Tue, 17 Jan 2023 07:08:06 +0000 (15:08 +0800)]
blk-iocost: change div64_u64 to DIV64_U64_ROUND_UP in ioc_refresh_params()

vrate_min is calculated by DIV64_U64_ROUND_UP, but vrate_max is calculated
by div64_u64. Vrate_min may be 1 greater than vrate_max if the input
values min and max of cost.qos are equal.

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230117070806.3857142-6-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: fix divide by 0 error in calc_lcoefs()
Li Nan [Tue, 17 Jan 2023 07:08:05 +0000 (15:08 +0800)]
blk-iocost: fix divide by 0 error in calc_lcoefs()

echo max of u64 to cost.model can cause divide by 0 error.

  # echo 8:0 rbps=18446744073709551615 > /sys/fs/cgroup/io.cost.model

  divide error: 0000 [#1] PREEMPT SMP
  RIP: 0010:calc_lcoefs+0x4c/0xc0
  Call Trace:
   <TASK>
   ioc_refresh_params+0x2b3/0x4f0
   ioc_cost_model_write+0x3cb/0x4c0
   ? _copy_from_iter+0x6d/0x6c0
   ? kernfs_fop_write_iter+0xfc/0x270
   cgroup_file_write+0xa0/0x200
   kernfs_fop_write_iter+0x17d/0x270
   vfs_write+0x414/0x620
   ksys_write+0x73/0x160
   __x64_sys_write+0x1e/0x30
   do_syscall_64+0x35/0x80
   entry_SYSCALL_64_after_hwframe+0x63/0xcd

calc_lcoefs() uses the input value of cost.model in DIV_ROUND_UP_ULL,
overflow would happen if bps plus IOC_PAGE_SIZE is greater than
ULLONG_MAX, it can cause divide by 0 error.

Fix the problem by setting basecost

Signed-off-by: Li Nan <linan122@huawei.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230117070806.3857142-5-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: read params inside lock in sysfs apis
Yu Kuai [Tue, 17 Jan 2023 07:08:04 +0000 (15:08 +0800)]
blk-iocost: read params inside lock in sysfs apis

Otherwise, user might get abnormal values if params is updated
concurrently.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230117070806.3857142-4-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: don't allow to configure bio based device
Yu Kuai [Tue, 17 Jan 2023 07:08:03 +0000 (15:08 +0800)]
blk-iocost: don't allow to configure bio based device

iocost is based on rq_qos, which can only work for request based device,
thus it doesn't make sense to configure iocost for bio based device.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230117070806.3857142-3-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: check return value of match_u64()
Yu Kuai [Tue, 17 Jan 2023 07:08:02 +0000 (15:08 +0800)]
blk-iocost: check return value of match_u64()

This patch fixs that the return value of match_u64() from ioc_qos_write()
is not checked,

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230117070806.3857142-2-yukuai1@huaweicloud.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblk-iocost: avoid 64-bit division in ioc_timer_fn
Arnd Bergmann [Wed, 18 Jan 2023 08:07:01 +0000 (09:07 +0100)]
blk-iocost: avoid 64-bit division in ioc_timer_fn

The behavior of 'enum' types has changed in gcc-13, so now the
UNBUSY_THR_PCT constant is interpreted as a 64-bit number because
it is defined as part of the same enum definition as some other
constants that do not fit within a 32-bit integer. This in turn
leads to some inefficient code on 32-bit architectures as well
as a link error:

arm-linux-gnueabi/bin/arm-linux-gnueabi-ld: block/blk-iocost.o: in function `ioc_timer_fn':
blk-iocost.c:(.text+0x68e8): undefined reference to `__aeabi_uldivmod'
arm-linux-gnueabi-ld: blk-iocost.c:(.text+0x6908): undefined reference to `__aeabi_uldivmod'

Split the enum definition to keep the 64-bit timing constants in
a separate enum type from those constants that can clearly fit
within a smaller type.

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Tejun Heo <tj@kernel.org>
Link: https://lore.kernel.org/r/20230118080706.3303186-1-arnd@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: ublk: fix doc build warning
Ming Lei [Wed, 18 Jan 2023 04:23:18 +0000 (12:23 +0800)]
block: ublk: fix doc build warning

Fix the following warning:

Documentation/block/ublk.rst:157: WARNING: Enumerated list ends without a blank line; unexpected unindent.
Documentation/block/ublk.rst:171: WARNING: Enumerated list ends without a blank line; unexpected unindent.

Fixes: 56f5160bc1b8 ("ublk_drv: add mechanism for supporting unprivileged ublk device")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230118042318.127900-1-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: introduce bdev_zone_no helper
Pankaj Raghav [Tue, 10 Jan 2023 14:36:35 +0000 (15:36 +0100)]
block: introduce bdev_zone_no helper

Add a generic bdev_zone_no() helper to calculate zone number for a
given sector in a block device. This helper internally uses disk_zone_no()
to find the zone number.

Use the helper bdev_zone_no() to calculate nr of zones. This lets us
make modifications to the math if needed in one place.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Link: https://lore.kernel.org/r/20230110143635.77300-4-p.raghav@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: add a new helper bdev_{is_zone_start, offset_from_zone_start}
Pankaj Raghav [Tue, 10 Jan 2023 14:36:34 +0000 (15:36 +0100)]
block: add a new helper bdev_{is_zone_start, offset_from_zone_start}

Instead of open coding to check for zone start, add a helper to improve
readability and store the logic in one place.

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20230110143635.77300-3-p.raghav@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: remove superfluous check for request queue in bdev_is_zoned()
Pankaj Raghav [Tue, 10 Jan 2023 14:36:33 +0000 (15:36 +0100)]
block: remove superfluous check for request queue in bdev_is_zoned()

Remove the superfluous request queue check in bdev_is_zoned() as
bdev_get_queue() can never return NULL.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Link: https://lore.kernel.org/r/20230110143635.77300-2-p.raghav@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: extend bio-cache for non-polled requests
Anuj Gupta [Tue, 17 Jan 2023 12:06:38 +0000 (17:36 +0530)]
block: extend bio-cache for non-polled requests

This patch modifies the present check, so that bio-cache is not limited
to iopoll.

Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Link: https://lore.kernel.org/r/20230117120638.72254-3-anuj20.g@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agonvme: set REQ_ALLOC_CACHE for uring-passthru request
Anuj Gupta [Tue, 17 Jan 2023 12:06:37 +0000 (17:36 +0530)]
nvme: set REQ_ALLOC_CACHE for uring-passthru request

This patch sets REQ_ALLOC_CACHE flag for uring-passthru requests.
This is a prep-patch so that normal / IRQ-driven uring-passthru
I/Os can also leverage bio-cache.

Signed-off-by: Anuj Gupta <anuj20.g@samsung.com>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
Link: https://lore.kernel.org/r/20230117120638.72254-2-anuj20.g@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: add mechanism for supporting unprivileged ublk device
Ming Lei [Fri, 6 Jan 2023 04:17:11 +0000 (12:17 +0800)]
ublk_drv: add mechanism for supporting unprivileged ublk device

unprivileged ublk device is helpful for container use case, such
as: ublk device created in one unprivileged container can be controlled
and accessed by this container only.

Implement this feature by adding flag of UBLK_F_UNPRIVILEGED_DEV, and if
this flag isn't set, any control command has been run from privileged
user. Otherwise, any control command can be sent from any unprivileged
user, but the user has to be permitted to access the ublk char device
to be controlled.

In case of UBLK_F_UNPRIVILEGED_DEV:

1) for command UBLK_CMD_ADD_DEV, it is always allowed, and user needs
to provide owner's uid/gid in this command, so that udev can set correct
ownership for the created ublk device, since the device owner uid/gid
can be queried via command of UBLK_CMD_GET_DEV_INFO.

2) for other control commands, they can only be run successfully if the
current user is allowed to access the specified ublk char device, for
running the permission check, path of the ublk char device has to be
provided by these commands.

Also add one control of command UBLK_CMD_GET_DEV_INFO2 which always
include the char dev path in payload since userspace may not have
knowledge if this device is created in unprivileged mode.

For applying this mechanism, system administrator needs to take
the following policies:

1) chmod 0666 /dev/ublk-control

2) change ownership of ublkcN & ublkbN
- chown owner_uid:owner_gid /dev/ublkcN
- chown owner_uid:owner_gid /dev/ublkbN

Both can be done via one simple udev rule.

Userspace:

https://github.com/ming1/ubdsrv/tree/unprivileged-ublk

'ublk add -t $TYPE --un_privileged=1' is for creating one un-privileged
ublk device if the user is un-privileged.

Link: https://lore.kernel.org/linux-block/YoOr6jBfgVm8GvWg@stefanha-x1.localdomain/
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-7-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: add module parameter of ublks_max for limiting max allowed ublk dev
Ming Lei [Fri, 6 Jan 2023 04:17:10 +0000 (12:17 +0800)]
ublk_drv: add module parameter of ublks_max for limiting max allowed ublk dev

Prepare for supporting unprivileged ublk device by limiting max number
ublk devices added. Otherwise too many ublk devices could be added by
un-trusted user, which can be thought as one DoS.

Reviewed-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-6-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: add device parameter UBLK_PARAM_TYPE_DEVT
Ming Lei [Fri, 6 Jan 2023 04:17:09 +0000 (12:17 +0800)]
ublk_drv: add device parameter UBLK_PARAM_TYPE_DEVT

Userspace side only knows device ID, but the associated path of ublkc* and
ublkb* could be changed by udev, and that depends on userspace's policy, so
add parameter of UBLK_PARAM_TYPE_DEVT for retrieving major/minor of the
ublkc* and ublkb*, then user may figure out major/minor of the ublk disks
he/she owns. With major/minor, it is easy to find the device node path.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-5-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: move ublk_get_device_from_id into ublk_ctrl_uring_cmd
Ming Lei [Fri, 6 Jan 2023 04:17:08 +0000 (12:17 +0800)]
ublk_drv: move ublk_get_device_from_id into ublk_ctrl_uring_cmd

It is annoying for each control command handler to get/put ublk
device and deal with failure.

Control command handler is simplified a lot by moving
ublk_get_device_from_id into ublk_ctrl_uring_cmd().

Reviewed-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-4-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: don't probe partitions if the ubq daemon isn't trusted
Ming Lei [Fri, 6 Jan 2023 04:17:07 +0000 (12:17 +0800)]
ublk_drv: don't probe partitions if the ubq daemon isn't trusted

If any ubq daemon is unprivileged, the ublk char device is allowed
for unprivileged user actually, and we can't trust the current user,
so not probe partitions.

Fixes: 71f28f3136af ("ublk_drv: add io_uring based userspace block driver")
Reviewed-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-3-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoublk_drv: remove nr_aborted_queues from ublk_device
Ming Lei [Fri, 6 Jan 2023 04:17:06 +0000 (12:17 +0800)]
ublk_drv: remove nr_aborted_queues from ublk_device

No one uses 'nr_aborted_queues' any more, so remove it.

Reviewed-by: ZiyangZhang <ZiyangZhang@linux.alibaba.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230106041711.914434-2-ming.lei@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: don't allow multiple bios for IOCB_NOWAIT issue
Jens Axboe [Mon, 16 Jan 2023 15:55:53 +0000 (08:55 -0700)]
block: don't allow multiple bios for IOCB_NOWAIT issue

If we're doing a large IO request which needs to be split into multiple
bios for issue, then we can run into the same situation as the below
marked commit fixes - parts will complete just fine, one or more parts
will fail to allocate a request. This will result in a partially
completed read or write request, where the caller gets EAGAIN even though
parts of the IO completed just fine.

Do the same for large bios as we do for splits - fail a NOWAIT request
with EAGAIN. This isn't technically fixing an issue in the below marked
patch, but for stable purposes, we should have either none of them or
both.

This depends on: 613b14884b85 ("block: handle bio_split_to_limits() NULL return")

Cc: stable@vger.kernel.org # 5.15+
Fixes: 9cea62b2cbab ("block: don't allow splitting of a REQ_NOWAIT bio")
Link: https://github.com/axboe/liburing/issues/766
Reported-and-tested-by: Michael Kelley <mikelley@microsoft.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: drbd_insert_interval(): Clarify comment
Andreas Gruenbacher [Fri, 13 Jan 2023 12:35:38 +0000 (13:35 +0100)]
drbd: drbd_insert_interval(): Clarify comment

Signed-off-by: Andreas Gruenbacher <agruen@linbit.com>
Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-9-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: interval tree: make removing an "empty" interval a no-op
Lars Ellenberg [Fri, 13 Jan 2023 12:35:37 +0000 (13:35 +0100)]
drbd: interval tree: make removing an "empty" interval a no-op

Trying to remove an "empty" (just initialized, or "cleared") interval
from the tree, this results in an endless loop.

As we typically protect the tree with a spinlock_irq,
the result is a hung system.

Be nice to error cleanup code paths, ignore removal of empty intervals.

Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-8-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoMAINTAINERS: add drbd headers
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:36 +0000 (13:35 +0100)]
MAINTAINERS: add drbd headers

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-7-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: remove macros using require_context
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:35 +0000 (13:35 +0100)]
drbd: remove macros using require_context

This require_context attribute originated in a proposed sparse patch by
Philipp Reisner back in 2008. Johannes Berg had a different solution to
a similar problem, and that patch "won" in the end; so the require_context
thing never got merged. The whole history can be read at [0].

DRBD kept using these annotations anyway for a while. Nowadays, on a
modern unmodified sparse, they obviously do nothing, and they are hardly
used anymore anyway.

So, just remove the definitions of these macros.

[0] https://www.spinics.net/lists/linux-sparse/msg01150.html

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-6-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: remove unnecessary assignment in vli_encode_bits
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:34 +0000 (13:35 +0100)]
drbd: remove unnecessary assignment in vli_encode_bits

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-5-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: make limits unsigned
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:33 +0000 (13:35 +0100)]
drbd: make limits unsigned

These are almost always used as unsigned integers, so mark them as such.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-4-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: fix DRBD_VOLUME_MAX 65535 -> 65534
Robert Altnoeder [Fri, 13 Jan 2023 12:35:32 +0000 (13:35 +0100)]
drbd: fix DRBD_VOLUME_MAX 65535 -> 65534

The protocol uses -1 as a reserved value for
'no specific volume', and since the protocol field
is a 16 bit unsigned value, -1 is converted to
65535. Therefore, limit the range of valid volume
numbers to [0, 65534].

Signed-off-by: Robert Altnoeder <robert.altnoeder@linbit.com>
Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-3-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: adjust drbd_limits license header
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:31 +0000 (13:35 +0100)]
drbd: adjust drbd_limits license header

See also commit 93c68cc46a07 ("drbd: use consistent license"). We only
want to license drbd under GPL-2.0, so use the corresponding SPDX header
consistently.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123538.144276-2-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: split off drbd_config into separate file
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:06 +0000 (13:35 +0100)]
drbd: split off drbd_config into separate file

To be more similar to what we do in the out-of-tree module and ease the
upstreaming process.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123506.144082-4-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: drop API_VERSION define
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:05 +0000 (13:35 +0100)]
drbd: drop API_VERSION define

Use the genetlink api version as defined in drbd_genl_api.h.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123506.144082-3-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agodrbd: split off drbd_buildtag into separate file
Christoph Böhmwalder [Fri, 13 Jan 2023 12:35:04 +0000 (13:35 +0100)]
drbd: split off drbd_buildtag into separate file

To be more similar to what we do in the out-of-tree module and ease the
upstreaming process.

Signed-off-by: Christoph Böhmwalder <christoph.boehmwalder@linbit.com>
Reviewed-by: Joel Colledge <joel.colledge@linbit.com>
Link: https://lore.kernel.org/r/20230113123506.144082-2-christoph.boehmwalder@linbit.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: add a BUILD_BUG_ON() for adding more bio flags than we have space
Jens Axboe [Mon, 9 Jan 2023 17:45:56 +0000 (10:45 -0700)]
block: add a BUILD_BUG_ON() for adding more bio flags than we have space

We have BIO_FLAG_LAST in the enum for bio specific flags, but it's
not used to check that we're not exceeding the size of them. Add
such a check.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: save user max_sectors limit
Keith Busch [Thu, 5 Jan 2023 20:51:46 +0000 (12:51 -0800)]
block: save user max_sectors limit

The user can set the max_sectors limit to any valid value via sysfs
/sys/block/<dev>/queue/max_sectors_kb attribute. If the device limits
are ever rescanned, though, the limit reverts back to the potentially
artificially low BLK_DEF_MAX_SECTORS value.

Preserve the user's setting as the max_sectors limit as long as it's
valid. The user can reset back to defaults by writing 0 to the sysfs
file.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230105205146.3610282-3-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock: make BLK_DEF_MAX_SECTORS unsigned
Keith Busch [Thu, 5 Jan 2023 20:51:45 +0000 (12:51 -0800)]
block: make BLK_DEF_MAX_SECTORS unsigned

This is used as an unsigned value, so define it that way to avoid
having to cast it.

Suggested-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Link: https://lore.kernel.org/r/20230105205146.3610282-2-kbusch@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: balance I/O injection among underutilized actuators
Davide Zini [Tue, 3 Jan 2023 14:55:03 +0000 (15:55 +0100)]
block, bfq: balance I/O injection among underutilized actuators

Upon the invocation of its dispatch function, BFQ returns the next I/O
request of the in-service bfq_queue, unless some exception holds. One
such exception is that there is some underutilized actuator, different
from the actuator for which the in-service queue contains I/O, and
that some other bfq_queue happens to contain I/O for such an
actuator. In this case, the next I/O request of the latter bfq_queue,
and not of the in-service bfq_queue, is returned (I/O is injected from
that bfq_queue). To find such an actuator, a linear scan, in
increasing index order, is performed among actuators.

Performing a linear scan entails a prioritization among actuators: an
underutilized actuator may be considered for injection only if all
actuators with a lower index are currently fully utilized, or if there
is no pending I/O for any lower-index actuator that happens to be
underutilized.

This commits breaks this prioritization and tends to distribute
injection uniformly across actuators. This is obtained by adding the
following condition to the linear scan: even if an actuator A is
underutilized, A is however skipped if its load is higher than that of
the next actuator.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Davide Zini <davidezini2@gmail.com>
Link: https://lore.kernel.org/r/20230103145503.71712-9-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: inject I/O to underutilized actuators
Davide Zini [Tue, 3 Jan 2023 14:55:02 +0000 (15:55 +0100)]
block, bfq: inject I/O to underutilized actuators

The main service scheme of BFQ for sync I/O is serving one sync
bfq_queue at a time, for a while. In particular, BFQ enforces this
scheme when it deems the latter necessary to boost throughput or
to preserve service guarantees. Unfortunately, when BFQ enforces
this policy, only one actuator at a time gets served for a while,
because each bfq_queue contains I/O only for one actuator. The
other actuators may remain underutilized.

Actually, BFQ may serve (inject) extra I/O, taken from other
bfq_queues, in parallel with that of the in-service queue. This
injection mechanism may provide the ground for dealing also with
the above actuator-underutilization problem. Yet BFQ does not take
the actuator load into account when choosing which queue to pick
extra I/O from. In addition, BFQ may happen to inject extra I/O
only when the in-service queue is temporarily empty.

In view of these facts, this commit extends the
injection mechanism in such a way that the latter:
(1) takes into account also the actuator load;
(2) checks such a load on each dispatch, and injects I/O for an
    underutilized actuator, if there is one and there is I/O for it.

To perform the check in (2), this commit introduces a load
threshold, currently set to 4.  A linear scan of each actuator is
performed, until an actuator is found for which the following two
conditions hold: the load of the actuator is below the threshold,
and there is at least one non-in-service queue that contains I/O
for that actuator. If such a pair (actuator, queue) is found, then
the head request of that queue is returned for dispatch, instead
of the head request of the in-service queue.

We have set the threshold, empirically, to the minimum possible
value for which an actuator is fully utilized, or close to be
fully utilized. By doing so, injected I/O 'steals' as few
drive-queue slots as possibile to the in-service queue. This
reduces as much as possible the probability that the service of
I/O from the in-service bfq_queue gets delayed because of slot
exhaustion, i.e., because all the slots of the drive queue are
filled with I/O injected from other queues (NCQ provides for 32
slots).

This new mechanism also counters actuator underutilization in the
case of asymmetric configurations of bfq_queues. Namely if there
are few bfq_queues containing I/O for some actuators and many
bfq_queues containing I/O for other actuators. Or if the
bfq_queues containing I/O for some actuators have lower weights
than the other bfq_queues.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Davide Zini <davidezini2@gmail.com>
Link: https://lore.kernel.org/r/20230103145503.71712-8-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: retrieve independent access ranges from request queue
Federico Gavioli [Tue, 3 Jan 2023 14:55:01 +0000 (15:55 +0100)]
block, bfq: retrieve independent access ranges from request queue

This patch implements the code to gather the content of the
independent_access_ranges structure from the request_queue and copy
it into the queue's bfq_data. This copy is done at queue initialization.

We copy the access ranges into the bfq_data to avoid taking the queue
lock each time we access the ranges.

This implementation, however, puts a limit to the maximum independent
ranges supported by the scheduler. Such a limit is equal to the constant
BFQ_MAX_ACTUATORS. This limit was placed to avoid the allocation of
dynamic memory.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Co-developed-by: Rory Chen <rory.c.chen@seagate.com>
Signed-off-by: Rory Chen <rory.c.chen@seagate.com>
Signed-off-by: Federico Gavioli <f.gavioli97@gmail.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Link: https://lore.kernel.org/r/20230103145503.71712-7-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: split also async bfq_queues on a per-actuator basis
Davide Zini [Tue, 3 Jan 2023 14:55:00 +0000 (15:55 +0100)]
block, bfq: split also async bfq_queues on a per-actuator basis

Similarly to sync bfq_queues, also async bfq_queues need to be split
on a per-actuator basis.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Davide Zini <davidezini2@gmail.com>
Link: https://lore.kernel.org/r/20230103145503.71712-6-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: turn bfqq_data into an array in bfq_io_cq
Paolo Valente [Tue, 3 Jan 2023 14:54:59 +0000 (15:54 +0100)]
block, bfq: turn bfqq_data into an array in bfq_io_cq

When a bfq_queue Q is merged with another queue, several pieces of
information are saved about Q. These pieces are stored in the
bfqq_data field in the bfq_io_cq data structure of the process
associated with Q.

Yet, with a multi-actuator drive, a process may get associated with
multiple bfq_queues: one queue for each of the N actuators. Each of
these queues may undergo a merge. So, the bfq_io_cq data structure
must be able to accommodate the above information for N queues.

This commit solves this problem by turning the bfqq_data scalar field
into an array of N elements (and by changing code so as to handle
this array).

This solution is written under the assumption that bfq_queues
associated with different actuators cannot be cross-merged. This
assumption holds naturally with basic queue merging: the latter is
triggered by spatial locality, and sectors for different actuators are
not close to each other (apart from the corner case of the last
sectors served by a given actuator and the first sectors served by the
next actuator). As for stable cross-merging, the assumption here is
that it is disabled.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Gabriele Felici <felicigb@gmail.com>
Signed-off-by: Gianmarco Lusvardi <glusvardi@posteo.net>
Signed-off-by: Giulio Barabino <giuliobarabino99@gmail.com>
Signed-off-by: Emiliano Maccaferri <inbox@emilianomaccaferri.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Link: https://lore.kernel.org/r/20230103145503.71712-5-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: move io_cq-persistent bfqq data into a dedicated struct
Paolo Valente [Tue, 3 Jan 2023 14:54:58 +0000 (15:54 +0100)]
block, bfq: move io_cq-persistent bfqq data into a dedicated struct

With a multi-actuator drive, a process may get associated with multiple
bfq_queues: one queue for each of the N actuators. So, the bfq_io_cq
data structure must be able to accommodate its per-queue persistent
information for N queues. Currently it stores this information for
just one queue, in several scalar fields.

This is a preparatory commit for moving to accommodating persistent
information for N queues. In particular, this commit packs all the
above scalar fields into a single data structure. Then there is now
only one field, in bfq_io_cq, that stores all the above information. This
scalar field will then be turned into an array by a following commit.

Suggested-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Gianmarco Lusvardi <glusvardi@posteo.net>
Signed-off-by: Giulio Barabino <giuliobarabino99@gmail.com>
Signed-off-by: Emiliano Maccaferri <inbox@emilianomaccaferri.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Link: https://lore.kernel.org/r/20230103145503.71712-4-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: forbid stable merging of queues associated with different actuators
Paolo Valente [Tue, 3 Jan 2023 14:54:57 +0000 (15:54 +0100)]
block, bfq: forbid stable merging of queues associated with different actuators

If queues associated with different actuators are merged, then control
is lost on each actuator. Therefore some actuator may be
underutilized, and throughput may decrease. This problem cannot occur
with basic queue merging, because the latter is triggered by spatial
locality, and sectors for different actuators are not close to each
other. Yet it may happen with stable merging. To address this issue,
this commit prevents stable merging from occurring among queues
associated with different actuators.

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Link: https://lore.kernel.org/r/20230103145503.71712-3-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoblock, bfq: split sync bfq_queues on a per-actuator basis
Paolo Valente [Tue, 3 Jan 2023 14:54:56 +0000 (15:54 +0100)]
block, bfq: split sync bfq_queues on a per-actuator basis

Single-LUN multi-actuator SCSI drives, as well as all multi-actuator
SATA drives appear as a single device to the I/O subsystem [1].  Yet
they address commands to different actuators internally, as a function
of Logical Block Addressing (LBAs). A given sector is reachable by
only one of the actuators. For example, Seagate’s Serial Advanced
Technology Attachment (SATA) version contains two actuators and maps
the lower half of the SATA LBA space to the lower actuator and the
upper half to the upper actuator.

Evidently, to fully utilize actuators, no actuator must be left idle
or underutilized while there is pending I/O for it. The block layer
must somehow control the load of each actuator individually. This
commit lays the ground for allowing BFQ to provide such a per-actuator
control.

BFQ associates an I/O-request sync bfq_queue with each process doing
synchronous I/O, or with a group of processes, in case of queue
merging. Then BFQ serves one bfq_queue at a time. While in service, a
bfq_queue is emptied in request-position order. Yet the same process,
or group of processes, may generate I/O for different actuators. In
this case, different streams of I/O (each for a different actuator)
get all inserted into the same sync bfq_queue. So there is basically
no individual control on when each stream is served, i.e., on when the
I/O requests of the stream are picked from the bfq_queue and
dispatched to the drive.

This commit enables BFQ to control the service of each actuator
individually for synchronous I/O, by simply splitting each sync
bfq_queue into N queues, one for each actuator. In other words, a sync
bfq_queue is now associated to a pair (process, actuator). As a
consequence of this split, the per-queue proportional-share policy
implemented by BFQ will guarantee that the sync I/O generated for each
actuator, by each process, receives its fair share of service.

This is just a preparatory patch. If the I/O of the same process
happens to be sent to different queues, then each of these queues may
undergo queue merging. To handle this event, the bfq_io_cq data
structure must be properly extended. In addition, stable merging must
be disabled to avoid loss of control on individual actuators. Finally,
also async queues must be split. These issues are described in detail
and addressed in next commits. As for this commit, although multiple
per-process bfq_queues are provided, the I/O of each process or group
of processes is still sent to only one queue, regardless of the
actuator the I/O is for. The forwarding to distinct bfq_queues will be
enabled after addressing the above issues.

[1] https://www.linaro.org/blog/budget-fair-queueing-bfq-linux-io-scheduler-optimizations-for-multi-actuator-sata-hard-drives/

Reviewed-by: Damien Le Moal <damien.lemoal@opensource.wdc.com>
Signed-off-by: Gabriele Felici <felicigb@gmail.com>
Signed-off-by: Carmine Zaccagnino <carmine@carminezacc.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Link: https://lore.kernel.org/r/20230103145503.71712-2-paolo.valente@linaro.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
21 months agoLinux 6.2-rc6
Linus Torvalds [Sun, 29 Jan 2023 21:59:43 +0000 (13:59 -0800)]
Linux 6.2-rc6

21 months agoMerge tag 'irq_urgent_for_v6.2_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 29 Jan 2023 19:26:49 +0000 (11:26 -0800)]
Merge tag 'irq_urgent_for_v6.2_rc6' of git://git./linux/kernel/git/tip/tip

Pull irq fix from Borislav Petkov:

 - Cleanup the firmware node for the new IRQ MSI domain properly, to
   avoid leaking memory

* tag 'irq_urgent_for_v6.2_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  genirq/msi: Free the fwnode created by msi_create_device_irq_domain()

21 months agoMerge tag 'x86_urgent_for_v6.2_rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Sun, 29 Jan 2023 19:17:34 +0000 (11:17 -0800)]
Merge tag 'x86_urgent_for_v6.2_rc6' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:

 - Start checking for -mindirect-branch-cs-prefix clang support too now
   that LLVM 16 will support it

 - Fix a NULL ptr deref when suspending with Xen PV

 - Have a SEV-SNP guest check explicitly for features enabled by the
   hypervisor and fail gracefully if some are unsupported by the guest
   instead of failing in a non-obvious and hard-to-debug way

 - Fix a MSI descriptor leakage under Xen

 - Mark Xen's MSI domain as supporting MSI-X

 - Prevent legacy PIC interrupts from being resent in software by
   marking them level triggered, as they should be, which lead to a NULL
   ptr deref

* tag 'x86_urgent_for_v6.2_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/build: Move '-mindirect-branch-cs-prefix' out of GCC-only block
  acpi: Fix suspend with Xen PV
  x86/sev: Add SEV-SNP guest feature negotiation support
  x86/pci/xen: Fixup fallout from the PCI/MSI overhaul
  x86/pci/xen: Set MSI_FLAG_PCI_MSIX support in Xen MSI domain
  x86/i8259: Mark legacy PIC interrupts with IRQ_LEVEL

21 months agoMerge tag 'input-for-v6.2-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor...
Linus Torvalds [Sun, 29 Jan 2023 19:06:47 +0000 (11:06 -0800)]
Merge tag 'input-for-v6.2-rc5' of git://git./linux/kernel/git/dtor/input

Pull input fixes from Dmitry Torokhov:

 - touchpads on HP 15-* laptops switched back to PS/2 emulation mode

 - a quirk for Clevo PCX0DX/TUXEDO XP1511 to make sure keyboard is
   responding after resume

* tag 'input-for-v6.2-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: i8042 - add Clevo PCX0DX to i8042 quirk table
  Revert "Input: synaptics - switch touchpad on HP Laptop 15-da3001TU to RMI mode"

21 months agoMerge tag 'cxl-fixes-for-6.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Sun, 29 Jan 2023 18:47:22 +0000 (10:47 -0800)]
Merge tag 'cxl-fixes-for-6.2-rc6' of git://git./linux/kernel/git/cxl/cxl

Pull cxl fixes from Dan Williams:
 "A couple of fixes for bugs introduced during the merge window. One is
  a regression, the other was a bug in the CXL AER handler:

   - Fix a crash regression due to module load order of cxl_pmem.ko

   - Fix wrong register offset read in CXL AER handling path"

* tag 'cxl-fixes-for-6.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
  cxl/pmem: Fix nvdimm unregistration when cxl_pmem driver is absent
  cxl: fix cxl_report_and_clear() RAS UE addr mis-assignment

21 months agoRevert "mm/compaction: fix set skip in fast_find_migrateblock"
Vlastimil Babka [Fri, 13 Jan 2023 17:33:45 +0000 (18:33 +0100)]
Revert "mm/compaction: fix set skip in fast_find_migrateblock"

This reverts commit 7efc3b7261030da79001c00d92bc3392fd6c664c.

We have got openSUSE reports (Link 1) for 6.1 kernel with khugepaged
stalling CPU for long periods of time.  Investigation of tracepoint data
shows that compaction is stuck in repeating fast_find_migrateblock()
based migrate page isolation, and then fails to migrate all isolated
pages.

Commit 7efc3b726103 ("mm/compaction: fix set skip in fast_find_migrateblock")
was suspected as it was merged in 6.1 and in theory can indeed remove a
termination condition for fast_find_migrateblock() under certain
conditions, as it removes a place that always marks a scanned pageblock
from being re-scanned.  There are other such places, but those can be
skipped under certain conditions, which seems to match the tracepoint
data.

Testing of revert also appears to have resolved the issue, thus revert
the commit until a more robust solution for the original problem is
developed.

It's also likely this will fix qemu stalls with 6.1 kernel reported in
Link 2, but that is not yet confirmed.

Link: https://bugzilla.suse.com/show_bug.cgi?id=1206848
Link: https://lore.kernel.org/kvm/b8017e09-f336-3035-8344-c549086c2340@kernel.org/
Link: https://lore.kernel.org/lkml/20230125134434.18017-1-mgorman@techsingularity.net/
Fixes: 7efc3b726103 ("mm/compaction: fix set skip in fast_find_migrateblock")
Cc: <stable@vger.kernel.org>
Tested-by: Pedro Falcato <pedro.falcato@gmail.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
21 months agoFix up more non-executable files marked executable
Linus Torvalds [Sat, 28 Jan 2023 19:17:57 +0000 (11:17 -0800)]
Fix up more non-executable files marked executable

Joe found another DT file that shouldn't be executable, and that
frustrated me enough that I went hunting with this script:

    git ls-files -s |
        grep '^100755' |
        cut -f2 |
        xargs grep -L '^#!'

and that found another file that shouldn't have been marked executable
either, despite being in the scripts directory.

Maybe these two are the last ones at least for now.  But I'm sure we'll
be back in a few years, fixing things up again.

Fixes: 8c6789f4e2d4 ("ASoC: dt-bindings: Add Everest ES8326 audio CODEC")
Fixes: 4d8e5cd233db ("locking/atomics: Fix scripts/atomic/ script permissions")
Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
21 months agoMerge tag '6.2-rc5-ksmbd-server-fixes' of git://git.samba.org/ksmbd
Linus Torvalds [Sat, 28 Jan 2023 18:52:51 +0000 (10:52 -0800)]
Merge tag '6.2-rc5-ksmbd-server-fixes' of git://git.samba.org/ksmbd

Pull ksmbd server fixes from Steve French:
 "Four smb3 server fixes, all also for stable:

   - fix for signing bug

   - fix to more strictly check packet length

   - add a max connections parm to limit simultaneous connections

   - fix error message flood that can occur with newer Samba xattr
     format"

* tag '6.2-rc5-ksmbd-server-fixes' of git://git.samba.org/ksmbd:
  ksmbd: downgrade ndr version error message to debug
  ksmbd: limit pdu length size according to connection status
  ksmbd: do not sign response to session request for guest login
  ksmbd: add max connections parameter

21 months agoMerge tag '6.2-rc5-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Sat, 28 Jan 2023 01:41:47 +0000 (17:41 -0800)]
Merge tag '6.2-rc5-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs fix from Steve French:
 "Fix for reconnect oops in smbdirect (RDMA), also is marked for stable"

* tag '6.2-rc5-smb3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: Fix oops due to uncleared server->smbd_conn in reconnect

21 months agoMerge tag 'block-6.2-2023-01-27' of git://git.kernel.dk/linux
Linus Torvalds [Sat, 28 Jan 2023 00:16:57 +0000 (16:16 -0800)]
Merge tag 'block-6.2-2023-01-27' of git://git.kernel.dk/linux

Pull block fixes from Jens Axboe:
 "Minor tweaks for this release:

   - NVMe pull request via Christoph:
        - Flush initial scan_work for async probe (Keith Busch)
        - Fix passthrough csi check (Keith Busch)
        - Fix nvme-fc initialization order (Ross Lagerwall)

   - Fix for tearing down non-started device in ublk (Ming)"

* tag 'block-6.2-2023-01-27' of git://git.kernel.dk/linux:
  block: ublk: move ublk_chr_class destroying after devices are removed
  nvme: fix passthrough csi check
  nvme-pci: flush initial scan_work for async probe
  nvme-fc: fix initialization order

21 months agoMerge tag 'io_uring-6.2-2023-01-27' of git://git.kernel.dk/linux
Linus Torvalds [Sat, 28 Jan 2023 00:15:06 +0000 (16:15 -0800)]
Merge tag 'io_uring-6.2-2023-01-27' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:
 "Two small fixes for this release:

   - Sanitize how async prep is done for drain requests, so we ensure
     that it always gets done (Dylan)

   - A ring provided buffer recycling fix for multishot receive (me)"

* tag 'io_uring-6.2-2023-01-27' of git://git.kernel.dk/linux:
  io_uring: always prep_async for drain requests
  io_uring/net: cache provided buffer group value for multishot receives

21 months agoMerge tag 'hardening-v6.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/kees...
Linus Torvalds [Sat, 28 Jan 2023 00:09:12 +0000 (16:09 -0800)]
Merge tag 'hardening-v6.2-rc6' of git://git./linux/kernel/git/kees/linux

Pull hardening fixes from Kees Cook:

 - Split slow memcpy tests into MEMCPY_SLOW_KUNIT_TEST

 - Reorganize gcc-plugin includes for GCC 13

 - Silence bcache memcpy run-time false positive warnings

* tag 'hardening-v6.2-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
  bcache: Silence memcpy() run-time false positive warnings
  gcc-plugins: Reorganize gimple includes for GCC 13
  kunit: memcpy: Split slow memcpy tests into MEMCPY_SLOW_KUNIT_TEST