Passthrough commands can always access the entire device, and thus
submitting them on partitions is an privelege escalation.

In hindsight we should have never allowed any passthrough commands on
partitions, but it's probably too late to change that decision now.

Fixes: e4fbcf32 ("nvme: identify-namespace without CAP_SYS_ADMIN")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

To prepare for passing down more information, replace the boolean
vec argument with a more extensible flags one.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

Open code __nvme_ioctl in the two callers to make future changes that
pass down additional paramters in the ioctl path easier.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

There are two issues in nvme_pci_enable():

 1) If pci_alloc_irq_vectors() fails, device is left enabled. Fix this by
    adding a goto disable statement.
 2) nvme_pci_configure_admin_queue could return -ENODEV, in this case,
    we will need to free IRQ properly.  Otherwise the following warning
    could be triggered:

[    5.286752] WARNING: CPU: 0 PID: 33 at kernel/irq/irqdomain.c:253 irq_domain_remove+0x12d/0x140
[    5.290547] Call Trace:
[    5.290626]  <TASK>
[    5.290695]  msi_remove_device_irq_domain+0xc9/0xf0
[    5.290843]  msi_device_data_release+0x15/0x80
[    5.290978]  release_nodes+0x58/0x90
[    5.293788] WARNING: CPU: 0 PID: 33 at kernel/irq/msi.c:276 msi_device_data_release+0x76/0x80
[    5.297573] Call Trace:
[    5.297651]  <TASK>
[    5.297719]  release_nodes+0x58/0x90
[    5.297831]  devres_release_all+0xef/0x140
[    5.298339]  device_unbind_cleanup+0x11/0xc0
[    5.298479]  really_probe+0x296/0x320

Fixes: a6ee7f19 ("nvme-pci: call nvme_pci_configure_admin_queue from nvme_pci_enable")
Co-developed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Tong Zhang <ztong0001@gmail.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

This mirrors the quirk added to Apple Silicon controllers in apple.c.
These controllers do not support the Active NS ID List command and
behave identically to the SoC version judging by existing user
reports/syslogs, so will need the same fix. This quirk reverts
back to NVMe 1.0 behavior and disables the broken commands.

Fixes: 811f4de0 ("nvme: avoid fallback to sequential scan due to transient issues")
Signed-off-by: Hector Martin <marcan@marcan.st>
Tested-by: Orlando Chamberlain <orlandoch.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

From the get-go, this driver and the ANS syslog have been complaining
about namespace identification. In 6.2-rc1, commit 811f4de0 ("nvme:
avoid fallback to sequential scan due to transient issues") regressed
the driver by no longer allowing fallback to sequential namespace scans,
leaving us with no namespaces.

It turns out that the real problem is that this controller claiming
NVMe 1.1 compat is treating the CNS field as a binary field, as in NVMe
1.0. This already has a quirk, NVME_QUIRK_IDENTIFY_CNS, so set it for
the controller to fix all this nonsense (including other errors
triggered by other CNS commands).

Fixes: 811f4de0 ("nvme: avoid fallback to sequential scan due to transient issues")
Fixes: 5bd2927a ("nvme-apple: Add initial Apple SoC NVMe driver")
Signed-off-by: Hector Martin <marcan@marcan.st>
Reviewed-by: Sven Peter <sven@svenpeter.dev>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Dan reports the following smatch detected the following:

  block/blk-cgroup.c:1863 blkcg_schedule_throttle() warn: sleeping in atomic context

caused by blkcg_schedule_throttle() calling blk_put_queue() in an
non-sleepable context.

blk_put_queue() acquired might_sleep() in 63f93fd6 ("block: mark
blk_put_queue as potentially blocking") which transferred the might_sleep()
from blk_free_queue().

blk_free_queue() acquired might_sleep() in e8c7d14a ("block: revert back
to synchronous request_queue removal") while turning request_queue removal
synchronous. However, this isn't necessary as nothing in the free path
actually requires sleeping.

It's pretty unusual to require a sleeping context in a put operation and
it's not needed in the first place. Let's drop it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Dan Carpenter <error27@gmail.com>
Link: https://lkml.kernel.org/r/Y7g3L6fntnTtOm63@kili
Cc: Christoph Hellwig <hch@lst.de>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Fixes: e8c7d14a ("block: revert back to synchronous request_queue removal") # v5.9+
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/Y7iFwjN+XzWvLv3y@slm.duckdns.orgSigned-off-by: Jens Axboe <axboe@kernel.dk>

Now that the SRCU Kconfig option is unconditionally selected, there is
no longer any point in selecting it.  Therefore, remove the "select SRCU"
Kconfig statements.
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This reverts commit f40eb998.

There are apparently still users out there of this driver. While we'd
love to remove it to ease the maintenance burden, let's reinstate it
for now until better (userspace) solutions can be developed.

Link: https://lore.kernel.org/lkml/20230104190115.ceglfefco475ev6c@pali/Reported-by: Pali Rohár <pali@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This reverts commit 85d6ce58.

We're reinstating the pktcdvd driver, which needs this API.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This reverts commit db1c7d77.

We're reinstating the pktcdvd driver, which needs this API.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Most of control command handlers may sleep, so return -EAGAIN in case
of IO_URING_F_NONBLOCK to defer the handling into io wq context.

Fixes: 71f28f31 ("ublk_drv: add io_uring based userspace block driver")
Reported-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20230104133235.836536-1-ming.lei@redhat.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

If we split a bio marked with REQ_NOWAIT, then we can trigger spurious
EAGAIN if constituent parts of that split bio end up failing request
allocations. Parts will complete just fine, but just a single failure
in one of the chained bios will yield an EAGAIN final result for the
parent bio.

Return EAGAIN early if we end up needing to split such a bio, which
allows for saner recovery handling.

Cc: stable@vger.kernel.org # 5.15+
Link: https://github.com/axboe/liburing/issues/766Reported-by: Michael Kelley <mikelley@microsoft.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This can't happen right now, but in preparation for allowing
bio_split_to_limits() returning NULL if it ended the bio, check for it
in all the callers.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Pull NVMe fixes from Christoph:

"nvme fixes for Linux 6.2

 - fix various problems in handling the Command Supported and Effects log
   (Christoph Hellwig)
 - don't allow unprivileged passthrough of commands that don't transfer
   data but modify logical block content (Christoph Hellwig)
 - add a features and quirks policy document (Christoph Hellwig)
 - fix some really nasty code that was correct but made smatch complain
   (Sagi Grimberg)"

* tag 'nvme-6.2-2022-12-29' of git://git.infradead.org/nvme:
  nvme-auth: fix smatch warning complaints
  nvme: consult the CSE log page for unprivileged passthrough
  nvme: also return I/O command effects from nvme_command_effects
  nvmet: don't defer passthrough commands with trivial effects to the workqueue
  nvmet: set the LBCC bit for commands that modify data
  nvmet: use NVME_CMD_EFFECTS_CSUPP instead of open coding it
  nvme: fix the NVME_CMD_EFFECTS_CSE_MASK definition
  docs, nvme: add a feature and quirk policy document

When initializing auth context, there may be no secrets passed
by the user. Make return code explicit when returning successfully.

smatch warnings:
drivers/nvme/host/auth.c:950 nvme_auth_init_ctrl() warn: missing error code? 'ret'
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Commands like Write Zeros can change the contents of a namespaces without
actually transferring data.  To protect against this, check the Commands
Supported and Effects log is supported by the controller for any
unprivileg command passthrough and refuse unprivileged passthrough if the
command has any effects that can change data or metadata.

Note: While the Commands Support and Effects log page has only been
mandatory since NVMe 2.0, it is widely supported because Windows requires
it for any command passthrough from userspace.

Fixes: e4fbcf32 ("nvme: identify-namespace without CAP_SYS_ADMIN")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

To be able to use the Commands Supported and Effects Log for allowing
unprivileged passtrough, it needs to be corretly reported for I/O
commands as well.  Return the I/O command effects from
nvme_command_effects, and also add a default list of effects for the
NVM command set.  For other command sets, the Commands Supported and
Effects log is required to be present already.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Mask out the "Command Supported" and "Logical Block Content Change" bits
and only defer execution of commands that have non-trivial effects to
the workqueue for synchronous execution.  This allows to execute admin
commands asynchronously on controllers that provide a Command Supported
and Effects log page, and will keep allowing to execute Write commands
asynchronously once command effects on I/O commands are taken into
account.

Fixes: c1fef73f ("nvmet: add passthru code to process commands")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

Write, Write Zeroes, Zone append and a Zone Reset through
Zone Management Send modify the logical block content of a namespace,
so make sure the LBCC bit is reported for them.

Fixes: b5d0b38c0475 ("nvmet: add Command Set Identifier support")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

Use NVME_CMD_EFFECTS_CSUPP instead of open coding it and assign a
single value to multiple array entries instead of repeated assignments.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

3 << 16 does not generate the correct mask for bits 16, 17 and 18.
Use the GENMASK macro to generate the correct mask instead.

Fixes: 84fef62d ("nvme: check admin passthru command effects")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>

This adds a document about what specification features are supported by
the Linux NVMe driver, and what qualifies for a quirk if an implementation
has problems following the specification.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Acked-by: Jonathan Corbet <corbet@lwn.net>

Update the core sqsize field in addition to the PCIe-specific
q_depth field as the core tagset allocation helpers rely on it.

Fixes: 0da7feaa ("nvme-pci: use the tagset alloc/free helpers")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Hugh Dickins <hughd@google.com>
Link: https://lore.kernel.org/r/20221225103234.226794-3-hch@lst.deSigned-off-by: Jens Axboe <axboe@kernel.dk>

While the CAP.MQES field in NVMe is a 0s based filed with a natural one
off, we also need to account for the queue wrap condition and fix undo
the one off again in nvme_alloc_io_tag_set.  This was never properly
done by the fabrics drivers, but they don't seem to care because there
is no actual physical queue that can wrap around, but it became a
problem when converting over the PCIe driver.  Also add back the
BLK_MQ_MAX_DEPTH check that was lost in the same commit.

Fixes: 0da7feaa ("nvme-pci: use the tagset alloc/free helpers")
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Hugh Dickins <hughd@google.com>
Link: https://lore.kernel.org/r/20221225103234.226794-2-hch@lst.deSigned-off-by: Jens Axboe <axboe@kernel.dk>

Commit 64dc8c73 ("block, bfq: fix possible uaf for 'bfqq->bic'")
will access 'bic->bfqq' in bic_set_bfqq(), however, bfq_exit_icq_bfqq()
can free bfqq first, and then call bic_set_bfqq(), which will cause uaf.

Fix the problem by moving bfq_exit_bfqq() behind bic_set_bfqq().

Fixes: 64dc8c73 ("block, bfq: fix possible uaf for 'bfqq->bic'")
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20221226030605.1437081-1-yukuai1@huaweicloud.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Pull NVMe fixes from Christoph:

"nvme fixes for Linux 6.2

 - fix doorbell buffer value endianness (Klaus Jensen)
 - fix Linux vs NVMe page size mismatch (Keith Busch)
 - fix a potential use memory access beyong the allocation limit
   (Keith Busch)
 - fix a multipath vs blktrace NULL pointer dereference
   (Yanjun Zhang)"

* tag 'nvme-6.2-2022-12-22' of git://git.infradead.org/nvme:
  nvme: fix multipath crash caused by flush request when blktrace is enabled
  nvme-pci: fix page size checks
  nvme-pci: fix mempool alloc size
  nvme-pci: fix doorbell buffer value endianness

The flush request initialized by blk_kick_flush has NULL bio,
and it may be dealt with nvme_end_req during io completion.
When blktrace is enabled, nvme_trace_bio_complete with multipath
activated trying to access NULL pointer bio from flush request
results in the following crash:

[ 2517.831677] BUG: kernel NULL pointer dereference, address: 000000000000001a
[ 2517.835213] #PF: supervisor read access in kernel mode
[ 2517.838724] #PF: error_code(0x0000) - not-present page
[ 2517.842222] PGD 7b2d51067 P4D 0
[ 2517.845684] Oops: 0000 [#1] SMP NOPTI
[ 2517.849125] CPU: 2 PID: 732 Comm: kworker/2:1H Kdump: loaded Tainted: G S                5.15.67-0.cl9.x86_64 #1
[ 2517.852723] Hardware name: XFUSION 2288H V6/BC13MBSBC, BIOS 1.13 07/27/2022
[ 2517.856358] Workqueue: nvme_tcp_wq nvme_tcp_io_work [nvme_tcp]
[ 2517.859993] RIP: 0010:blk_add_trace_bio_complete+0x6/0x30
[ 2517.863628] Code: 1f 44 00 00 48 8b 46 08 31 c9 ba 04 00 10 00 48 8b 80 50 03 00 00 48 8b 78 50 e9 e5 fe ff ff 0f 1f 44 00 00 41 54 49 89 f4 55 <0f> b6 7a 1a 48 89 d5 e8 3e 1c 2b 00 48 89 ee 4c 89 e7 5d 89 c1 ba
[ 2517.871269] RSP: 0018:ff7f6a008d9dbcd0 EFLAGS: 00010286
[ 2517.875081] RAX: ff3d5b4be00b1d50 RBX: 0000000002040002 RCX: ff3d5b0a270f2000
[ 2517.878966] RDX: 0000000000000000 RSI: ff3d5b0b021fb9f8 RDI: 0000000000000000
[ 2517.882849] RBP: ff3d5b0b96a6fa00 R08: 0000000000000001 R09: 0000000000000000
[ 2517.886718] R10: 000000000000000c R11: 000000000000000c R12: ff3d5b0b021fb9f8
[ 2517.890575] R13: 0000000002000000 R14: ff3d5b0b021fb1b0 R15: 0000000000000018
[ 2517.894434] FS:  0000000000000000(0000) GS:ff3d5b42bfc80000(0000) knlGS:0000000000000000
[ 2517.898299] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2517.902157] CR2: 000000000000001a CR3: 00000004f023e005 CR4: 0000000000771ee0
[ 2517.906053] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 2517.909930] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 2517.913761] PKRU: 55555554
[ 2517.917558] Call Trace:
[ 2517.921294]  <TASK>
[ 2517.924982]  nvme_complete_rq+0x1c3/0x1e0 [nvme_core]
[ 2517.928715]  nvme_tcp_recv_pdu+0x4d7/0x540 [nvme_tcp]
[ 2517.932442]  nvme_tcp_recv_skb+0x4f/0x240 [nvme_tcp]
[ 2517.936137]  ? nvme_tcp_recv_pdu+0x540/0x540 [nvme_tcp]
[ 2517.939830]  tcp_read_sock+0x9c/0x260
[ 2517.943486]  nvme_tcp_try_recv+0x65/0xa0 [nvme_tcp]
[ 2517.947173]  nvme_tcp_io_work+0x64/0x90 [nvme_tcp]
[ 2517.950834]  process_one_work+0x1e8/0x390
[ 2517.954473]  worker_thread+0x53/0x3c0
[ 2517.958069]  ? process_one_work+0x390/0x390
[ 2517.961655]  kthread+0x10c/0x130
[ 2517.965211]  ? set_kthread_struct+0x40/0x40
[ 2517.968760]  ret_from_fork+0x1f/0x30
[ 2517.972285]  </TASK>

To avoid this situation, add a NULL check for req->bio before
calling trace_block_bio_complete.
Signed-off-by: Yanjun Zhang <zhangyanjun@cestc.cn>
Signed-off-by: Christoph Hellwig <hch@lst.de>

The size allocated out of the dma pool is at most NVME_CTRL_PAGE_SIZE,
which may be smaller than the PAGE_SIZE.

Fixes: c61b82c7 ("nvme-pci: fix PRP pool size")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Convert the max size to bytes to match the units of the divisor that
calculates the worst-case number of PRP entries.

The result is used to determine how many PRP Lists are required. The
code was previously rounding this to 1 list, but we can require 2 in the
worst case. In that scenario, the driver would corrupt memory beyond the
size provided by the mempool.

While unlikely to occur (you'd need a 4MB in exactly 127 phys segments
on a queue that doesn't support SGLs), this memory corruption has been
observed by kfence.

Cc: Jens Axboe <axboe@kernel.dk>
Fixes: 943e942e ("nvme-pci: limit max IO size and segments to avoid high order allocations")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

When using shadow doorbells, the event index and the doorbell values are
written to host memory. Prior to this patch, the values written would
erroneously be written in host endianness. This causes trouble on
big-endian platforms. Fix this by adding missing endian conversions.

This issue was noticed by Guenter while testing various big-endian
platforms under QEMU[1]. A similar fix required for hw/nvme in QEMU is
up for review as well[2].

  [1]: https://lore.kernel.org/qemu-devel/20221209110022.GA3396194@roeck-us.net/
  [2]: https://lore.kernel.org/qemu-devel/20221212114409.34972-4-its@irrelevant.dk/

Fixes: f9f38e33 ("nvme: improve performance for virtual NVMe devices")
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>

Since commit:

b99182c5 ("bio: add pcpu caching for non-polling bio_put")

we support bio caching for IRQ based IO as well, hence there's no need
to manually clear REQ_ALLOC_CACHE if we disable polling on a request.
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

For blk-mq, queue release handler is usually called after
blk_mq_freeze_queue_wait() returns. However, the
q_usage_counter->release() handler may not be run yet at that time, so
this can cause a use-after-free.

Fix the issue by moving percpu_ref_exit() into blk_free_queue_rcu().
Since ->release() is called with rcu read lock held, it is agreed that
the race should be covered in caller per discussion from the two links.
Reported-by: Zhang Wensheng <zhangwensheng@huaweicloud.com>
Reported-by: Zhong Jinghua <zhongjinghua@huawei.com>
Link: https://lore.kernel.org/linux-block/Y5prfOjyyjQKUrtH@T590/T/#u
Link: https://lore.kernel.org/lkml/Y4%2FmzMd4evRg9yDi@fedora/
Cc: Hillf Danton <hdanton@sina.com>
Cc: Yu Kuai <yukuai3@huawei.com>
Cc: Dennis Zhou <dennis@kernel.org>
Fixes: 2b0d3d3e ("percpu_ref: reduce memory footprint of percpu_ref in fast path")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20221215021629.74870-1-ming.lei@redhat.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

The 'bfqd->num_groups_with_pending_reqs' is used when
CONFIG_BFQ_GROUP_IOSCHED is enabled, so let the variables and processes
take effect when CONFIG_BFQ_GROUP_IOSCHED is enabled.

Cc: Yu Kuai <yukuai3@huawei.com>
Signed-off-by: Yuwei Guan <Yuwei.Guan@zeekrlife.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Link: https://lore.kernel.org/r/20221110112622.389332-1-Yuwei.Guan@zeekrlife.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

When a gendisk is successfully initialized but add_disk() fails such as when
a loop device has invalid number of minor device numbers specified,
blkcg_init_disk() is called during init and then blkcg_exit_disk() during
error handling. Unfortunately, iolatency gets initialized in the former but
doesn't get cleaned up in the latter.

This is because, in non-error cases, the cleanup is performed by
del_gendisk() calling rq_qos_exit(), the assumption being that rq_qos
policies, iolatency being one of them, can only be activated once the disk
is fully registered and visible. That assumption is true for wbt and iocost,
but not so for iolatency as it gets initialized before add_disk() is called.

It is desirable to lazy-init rq_qos policies because they are optional
features and add to hot path overhead once initialized - each IO has to walk
all the registered rq_qos policies. So, we want to switch iolatency to lazy
init too. However, that's a bigger change. As a fix for the immediate
problem, let's just add an extra call to rq_qos_exit() in blkcg_exit_disk().
This is safe because duplicate calls to rq_qos_exit() become noop's.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: darklight2357@icloud.com
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Fixes: d7067512 ("block: introduce blk-iolatency io controller")
Cc: stable@vger.kernel.org # v4.19+
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/Y5TQ5gm3O4HXrXR3@slm.duckdns.orgSigned-off-by: Jens Axboe <axboe@kernel.dk>

Currently, the max_loop commandline argument can be used to specify how
many loop block devices are created at init time. If it is not
specified on the commandline, CONFIG_BLK_DEV_LOOP_MIN_COUNT loop block
devices will be created.

The max_loop commandline argument can be used to override the value of
CONFIG_BLK_DEV_LOOP_MIN_COUNT. However, when max_loop is set to 0
through the commandline, the current logic treats it as if it had not
been set, and creates CONFIG_BLK_DEV_LOOP_MIN_COUNT devices anyway.

Fix this by starting max_loop off as set to CONFIG_BLK_DEV_LOOP_MIN_COUNT.
This preserves the intended behavior of creating
CONFIG_BLK_DEV_LOOP_MIN_COUNT loop block devices if the max_loop
commandline parameter is not specified, and allowing max_loop to
be respected for all values, including 0.

This allows environments that can create all of their required loop
block devices on demand to not have to unnecessarily preallocate loop
block devices.

Fixes: 73285082 ("remove artificial software max_loop limit")
Cc: stable@vger.kernel.org
Cc: Ken Chen <kenchen@google.com>
Signed-off-by: Isaac J. Manjarres <isaacmanjarres@google.com>
Link: https://lore.kernel.org/r/20221208212902.765781-1-isaacmanjarres@google.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Since gcc13, each member of an enum has the same type as the enum [1]. And
that is inherited from its members. Provided:
  VTIME_PER_SEC_SHIFT     = 37,
  VTIME_PER_SEC           = 1LLU << VTIME_PER_SEC_SHIFT,
  ...
  AUTOP_CYCLE_NSEC        = 10LLU * NSEC_PER_SEC,
the named type is unsigned long.

This generates warnings with gcc-13:
  block/blk-iocost.c: In function 'ioc_weight_prfill':
  block/blk-iocost.c:3037:37: error: format '%u' expects argument of type 'unsigned int', but argument 4 has type 'long unsigned int'

  block/blk-iocost.c: In function 'ioc_weight_show':
  block/blk-iocost.c:3047:34: error: format '%u' expects argument of type 'unsigned int', but argument 3 has type 'long unsigned int'

So split the anonymous enum with large values to a separate enum, so
that they don't affect other members.

[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=36113

Cc: Martin Liska <mliska@suse.cz>
Cc: Tejun Heo <tj@kernel.org>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: cgroups@vger.kernel.org
Cc: linux-block@vger.kernel.org
Signed-off-by: Jiri Slaby (SUSE) <jirislaby@kernel.org>
Link: https://lore.kernel.org/r/20221213120826.17446-1-jirislaby@kernel.orgSigned-off-by: Jens Axboe <axboe@kernel.dk>

Just to make the code a litter cleaner, there are no functional changes.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20221214033155.3455754-3-yukuai1@huaweicloud.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

The return value is not used, hence remove it.
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20221214033155.3455754-2-yukuai1@huaweicloud.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Our test report a uaf for 'bfqq->bic' in 5.10:

==================================================================
BUG: KASAN: use-after-free in bfq_select_queue+0x378/0xa30

CPU: 6 PID: 2318352 Comm: fsstress Kdump: loaded Not tainted 5.10.0-60.18.0.50.h602.kasan.eulerosv2r11.x86_64 #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58-20220320_160524-szxrtosci10000 04/01/2014
Call Trace:
 bfq_select_queue+0x378/0xa30
 bfq_dispatch_request+0xe8/0x130
 blk_mq_do_dispatch_sched+0x62/0xb0
 __blk_mq_sched_dispatch_requests+0x215/0x2a0
 blk_mq_sched_dispatch_requests+0x8f/0xd0
 __blk_mq_run_hw_queue+0x98/0x180
 __blk_mq_delay_run_hw_queue+0x22b/0x240
 blk_mq_run_hw_queue+0xe3/0x190
 blk_mq_sched_insert_requests+0x107/0x200
 blk_mq_flush_plug_list+0x26e/0x3c0
 blk_finish_plug+0x63/0x90
 __iomap_dio_rw+0x7b5/0x910
 iomap_dio_rw+0x36/0x80
 ext4_dio_read_iter+0x146/0x190 [ext4]
 ext4_file_read_iter+0x1e2/0x230 [ext4]
 new_sync_read+0x29f/0x400
 vfs_read+0x24e/0x2d0
 ksys_read+0xd5/0x1b0
 do_syscall_64+0x33/0x40
 entry_SYSCALL_64_after_hwframe+0x61/0xc6

Commit 3bc5e683 ("bfq: Split shared queues on move between cgroups")
changes that move process to a new cgroup will allocate a new bfqq to
use, however, the old bfqq and new bfqq can point to the same bic:

1) Initial state, two process with io in the same cgroup.

Process 1       Process 2
 (BIC1)          (BIC2)
  |  Λ            |  Λ
  |  |            |  |
  V  |            V  |
  bfqq1           bfqq2

2) bfqq1 is merged to bfqq2.

Process 1       Process 2
 (BIC1)          (BIC2)
  |               |
   \-------------\|
                  V
  bfqq1           bfqq2(coop)

3) Process 1 exit, then issue new io(denoce IOA) from Process 2.

 (BIC2)
  |  Λ
  |  |
  V  |
  bfqq2(coop)

4) Before IOA is completed, move Process 2 to another cgroup and issue io.

Process 2
 (BIC2)
   Λ
   |\--------------\
   |                V
  bfqq2           bfqq3

Now that BIC2 points to bfqq3, while bfqq2 and bfqq3 both point to BIC2.
If all the requests are completed, and Process 2 exit, BIC2 will be
freed while there is no guarantee that bfqq2 will be freed before BIC2.

Fix the problem by clearing bfqq->bic while bfqq is detached from bic.

Fixes: 3bc5e683 ("bfq: Split shared queues on move between cgroups")
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://lore.kernel.org/r/20221214030430.3304151-1-yukuai1@huaweicloud.comSigned-off-by: Jens Axboe <axboe@kernel.dk>