Group some variables based on their sizes to reduce holes.
On x86_64, this shrinks the size of 'struct nvmefc_fcp_req' from
112 to 104 bytes.
This structure is embedded in some other structures (nvme_fc_fcp_op
which itself is embedded in nvme_fcp_op_w_sgl), so it helps reducing the
size of these structures too.
nvmet: reorder fields in 'struct nvme_dhchap_queue_context'
Group some variables based on their sizes to reduce holes.
On x86_64, this shrinks the size of 'struct nvme_dhchap_queue_context' from
416 to 400 bytes.
This structure is kvcalloc()'ed in nvme_auth_init_ctrl(), so it is likely
that the allocation can be relatively big. Saving 16 bytes per structure
may might a slight difference.
nvmet: reorder fields in 'struct nvmf_ctrl_options'
Group some variables based on their sizes to reduce holes.
On x86_64, this shrinks the size of 'struct nvmf_ctrl_options' from 136 to
128 bytes.
When such a structure is allocated in nvmf_create_ctrl(), because of the
way memory allocation works, when 136 bytes were requested, 192 bytes were
allocated.
So this saves 64 bytes per allocation, 1 cache line to hold the whole
structure and a few cycles when zeroing the memory in nvmf_create_ctrl().
Group some variables based on their sizes to reduce holes.
On x86_64, this shrinks the size of 'struct nvme_ctrl' from 5368 to 5344
bytes when all CONFIG_* are defined.
This structure is embedded into some other structures, so it helps reducing
their size as well.
Group some variables based on their sizes to reduce holes.
On x86_64, this shrinks the size of 'struct nvmet_sq' from 472 to 464
bytes when CONFIG_NVME_TARGET_AUTH is defined.
This structure is embedded into some other structures, so it helps reducing
their sizes as well.
Yu Kuai [Sat, 10 Jun 2023 02:30:43 +0000 (10:30 +0800)]
blk-mq: fix potential io hang by wrong 'wake_batch'
In __blk_mq_tag_busy/idle(), updating 'active_queues' and calculating
'wake_batch' is not atomic:
t1: t2:
_blk_mq_tag_busy blk_mq_tag_busy
inc active_queues
// assume 1->2
inc active_queues
// 2 -> 3
blk_mq_update_wake_batch
// calculate based on 3
blk_mq_update_wake_batch
/* calculate based on 2, while active_queues is actually 3. */
Fix this problem by protecting them wih 'tags->lock', this is not a hot
path, so performance should not be concerned. And now that all writers
are inside the lock, switch 'actives_queues' from atomic to unsigned
int.
Store the file struct used as the holder in file->private_data as an
indicator that this file descriptor was opened exclusively to remove
the last use of FMODE_EXCL.
block: replace fmode_t with a block-specific type for block open flags
The only overlap between the block open flags mapped into the fmode_t and
other uses of fmode_t are FMODE_READ and FMODE_WRITE. Define a new
blk_mode_t instead for use in blkdev_get_by_{dev,path}, ->open and
->ioctl and stop abusing fmode_t.
block: use the holder as indication for exclusive opens
The current interface for exclusive opens is rather confusing as it
requires both the FMODE_EXCL flag and a holder. Remove the need to pass
FMODE_EXCL and just key off the exclusive open off a non-NULL holder.
For blkdev_put this requires adding the holder argument, which provides
better debug checking that only the holder actually releases the hold,
but at the same time allows removing the now superfluous mode argument.
btrfs: don't pass a holder for non-exclusive blkdev_get_by_path
Passing a holder to blkdev_get_by_path when FMODE_EXCL isn't set doesn't
make sense, so pass NULL instead and remove the holder argument from the
call chains the only end up in non-FMODE_EXCL blkdev_get_by_path calls.
Exclusive mode for device scanning is not used since commit 50d281fc434c
("btrfs: scan device in non-exclusive mode")".
bcache: don't pass a stack address to blkdev_get_by_path
sb is just an on-stack pointer that can easily be reused by other calls.
Switch to use the bcache-wide bcache_kobj instead as there is no need to
claim per-bcache device anyway.
bdev_check_media_change should only ever be called for the whole device.
Pass a gendisk to make that explicit and rename the function to
disk_check_media_change.
cdrom: track if a cdrom_device_info was opened for data
Set a flag when a cdrom_device_info is opened for writing, instead of
trying to figure out this at release time. This will allow to eventually
remove the mode argument to the ->release block_device_operation as
nothing but the CDROM drivers uses that argument.
block: also call ->open for incremental partition opens
For whole devices ->open is called for each open, but for partitions it
is only called on the first open of a partition, e.g.:
open("/dev/vdb", ...)
open("/dev/vdb", ...)
- 2 call to ->open
open("/dev/vdb1", ...)
open("/dev/vdb", ...)
- 2 call to ->open
open("/dev/vdb", ...)
open("/dev/vdb", ...)
- just open call to ->open
This is problematic as various block drivers look at open flags and
might not do all the required setup if the earlier open was with an
odd flag like O_NDELAY or the magic 3 ioctl-only open mode.
Guoqing Jiang [Wed, 24 May 2023 07:00:21 +0000 (15:00 +0800)]
block/rnbd: introduce rnbd_access_modes
Add one new array (marked with __maybe_unused to prevent gcc warning about
"defined but not used" with W=1), then we can remove rnbd_access_mode_str
and rnbd-common.c accordingly.
Andy Shevchenko [Fri, 10 Mar 2023 16:45:44 +0000 (18:45 +0200)]
pktcdvd: Get rid of pkt_seq_show() forward declaration
The code can be neater without forward declarations.
Get rid of pkt_seq_show() forward declaration. This
will also allow futher cleanups to be cleaner.
Andy Shevchenko [Fri, 10 Mar 2023 16:45:43 +0000 (18:45 +0200)]
pktcdvd: use sysfs_emit() to instead of scnprintf()
Follow the advice of the Documentation/filesystems/sysfs.rst and show()
should only use sysfs_emit() or sysfs_emit_at() when formatting the
value to be returned to user space.
Waiman Long [Tue, 6 Jun 2023 18:07:24 +0000 (14:07 -0400)]
blk-cgroup: Reinit blkg_iostat_set after clearing in blkcg_reset_stats()
When blkg_alloc() is called to allocate a blkcg_gq structure
with the associated blkg_iostat_set's, there are 2 fields within
blkg_iostat_set that requires proper initialization - blkg & sync.
The former field was introduced by commit 3b8cc6298724 ("blk-cgroup:
Optimize blkcg_rstat_flush()") while the later one was introduced by
commit f73316482977 ("blk-cgroup: reimplement basic IO stats using
cgroup rstat").
Unfortunately those fields in the blkg_iostat_set's are not properly
re-initialized when they are cleared in v1's blkcg_reset_stats(). This
can lead to a kernel panic due to NULL pointer access of the blkg
pointer. The missing initialization of sync is less problematic and
can be a problem in a debug kernel due to missing lockdep initialization.
Fix these problems by re-initializing them after memory clearing.
Yu Kuai [Tue, 6 Jun 2023 01:14:38 +0000 (09:14 +0800)]
blk-ioc: fix recursive spin_lock/unlock_irq() in ioc_clear_queue()
Recursive spin_lock/unlock_irq() is not safe, because spin_unlock_irq()
will enable irq unconditionally:
spin_lock_irq queue_lock -> disable irq
spin_lock_irq ioc->lock
spin_unlock_irq ioc->lock -> enable irq
/*
* AA dead lock will be triggered if current context is preempted by irq,
* and irq try to hold queue_lock again.
*/
spin_unlock_irq queue_lock
Fix this problem by using spin_lock/unlock() directly for 'ioc->lock'.
Zhong Jinghua [Mon, 5 Jun 2023 12:21:59 +0000 (20:21 +0800)]
nbd: Add the maximum limit of allocated index in nbd_dev_add
If the index allocated by idr_alloc greater than MINORMASK >> part_shift,
the device number will overflow, resulting in failure to create a block
device.
Hou Tao [Fri, 28 Apr 2023 07:44:04 +0000 (15:44 +0800)]
blk-ioprio: Introduce promote-to-rt policy
Since commit a78418e6a04c ("block: Always initialize bio IO priority on
submit"), bio->bi_ioprio will never be IOPRIO_CLASS_NONE when calling
blkcg_set_ioprio(), so there will be no way to promote the io-priority
of one cgroup to IOPRIO_CLASS_RT, because bi_ioprio will always be
greater than or equals to IOPRIO_CLASS_RT.
It seems possible to call blkcg_set_ioprio() first then try to
initialize bi_ioprio later in bio_set_ioprio(), but this doesn't work
for bio in which bi_ioprio is already initialized (e.g., direct-io), so
introduce a new promote-to-rt policy to promote the iopriority of bio to
IOPRIO_CLASS_RT if the ioprio is not already RT.
For none-to-rt policy, although it doesn't work now, but considering
that its purpose was also to override the io-priority to RT and allowing
for a smoother transition, just keep it and treat it as an alias of
the promote-to-rt policy.
Li Nan [Sat, 27 May 2023 09:19:04 +0000 (17:19 +0800)]
blk-iocost: use spin_lock_irqsave in adjust_inuse_and_calc_cost
adjust_inuse_and_calc_cost() use spin_lock_irq() and IRQ will be enabled
when unlock. DEADLOCK might happen if we have held other locks and disabled
IRQ before invoking it.
Fix it by using spin_lock_irqsave() instead, which can keep IRQ state
consistent with before when unlock.
================================
WARNING: inconsistent lock state 5.10.0-02758-g8e5f91fd772f #26 Not tainted
--------------------------------
inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
kworker/2:3/388 [HC0[0]:SC0[0]:HE0:SE1] takes: ffff888118c00c28 (&bfqd->lock){?.-.}-{2:2}, at: spin_lock_irq ffff888118c00c28 (&bfqd->lock){?.-.}-{2:2}, at: bfq_bio_merge+0x141/0x390
{IN-HARDIRQ-W} state was registered at:
__lock_acquire+0x3d7/0x1070
lock_acquire+0x197/0x4a0
__raw_spin_lock_irqsave
_raw_spin_lock_irqsave+0x3b/0x60
bfq_idle_slice_timer_body
bfq_idle_slice_timer+0x53/0x1d0
__run_hrtimer+0x477/0xa70
__hrtimer_run_queues+0x1c6/0x2d0
hrtimer_interrupt+0x302/0x9e0
local_apic_timer_interrupt
__sysvec_apic_timer_interrupt+0xfd/0x420
run_sysvec_on_irqstack_cond
sysvec_apic_timer_interrupt+0x46/0xa0
asm_sysvec_apic_timer_interrupt+0x12/0x20
irq event stamp: 837522
hardirqs last enabled at (837521): [<ffffffff84b9419d>] __raw_spin_unlock_irqrestore
hardirqs last enabled at (837521): [<ffffffff84b9419d>] _raw_spin_unlock_irqrestore+0x3d/0x40
hardirqs last disabled at (837522): [<ffffffff84b93fa3>] __raw_spin_lock_irq
hardirqs last disabled at (837522): [<ffffffff84b93fa3>] _raw_spin_lock_irq+0x43/0x50
softirqs last enabled at (835852): [<ffffffff84e00558>] __do_softirq+0x558/0x8ec
softirqs last disabled at (835845): [<ffffffff84c010ff>] asm_call_irq_on_stack+0xf/0x20
other info that might help us debug this:
Possible unsafe locking scenario:
mtd: block2mtd: don't call early_lookup_bdev after the system is running
early_lookup_bdev is supposed to only be called from the early boot
code, but mdtblock_early_get_bdev is called as a general fallback when
lookup_bdev fails, which is problematic because early_lookup_bdev
bypasses all normal path based permission checking, and might cause
problems with certain container environments renaming devices.
Switch to only call early_lookup_bdev when block2mtd is built-in and the
system state in not running yet.
Note that this strictly speaking changes the kernel ABI as the PARTUUID=
and PARTLABEL= style syntax is now not available during a running
systems. They never were intended for that, but this breaks things
we'll have to figure out a way to make them available again. But if
avoidable in any way I'd rather avoid that.
PM: hibernate: don't use early_lookup_bdev in resume_store
resume_store is a sysfs attribute written during normal kernel runtime,
and it should not use the early_lookup_bdev API that bypasses all normal
path based permission checking, and might cause problems with certain
container environments renaming devices.
Switch to lookup_bdev, which does a normal path lookup instead, and fall
back to trying to parse a numeric dev_t just like early_lookup_bdev did.
Note that this strictly speaking changes the kernel ABI as the PARTUUID=
and PARTLABEL= style syntax is now not available during a running
systems. They never were intended for that, but this breaks things
we'll have to figure out a way to make them available again. But if
avoidable in any way I'd rather avoid that.
dm: only call early_lookup_bdev from early boot context
early_lookup_bdev is supposed to only be called from the early boot
code, but dm_get_device calls it as a general fallback when lookup_bdev
fails, which is problematic because early_lookup_bdev bypasses all normal
path based permission checking, and might cause problems with certain
container environments renaming devices.
Switch to only call early_lookup_bdev when dm is built-in and the system
state in not running yet. This means it is still available when tables
are constructed by dm-init.c from the kernel command line, but not
otherwise.
Note that this strictly speaking changes the kernel ABI as the PARTUUID=
and PARTLABEL= style syntax is now not available during a running
systems. They never were intended for that, but this breaks things
we'll have to figure out a way to make them available again. But if
avoidable in any way I'd rather avoid that.
init: clear root_wait on all invalid root= strings
Instead of only clearing root_wait in devt_from_partuuid when the UUID
format was invalid, do that in parse_root_device for all strings that
failed to parse.
name_to_dev_t has a very misleading name, that doesn't make clear
it should only be used by the early init code, and also has a bad
calling convention that doesn't allow returning different kinds of
errors. Rename it to early_lookup_bdev to make the use case clear,
and return an errno, where -EINVAL means the string could not be
parsed, and -ENODEV means it the string was valid, but there was
no device found for it.
Also stub out the whole call for !CONFIG_BLOCK as all the non-block
root cases are always covered in the caller.
init: handle ubi/mtd root mounting like all other root types
Assign a Root_Generic magic value for UBI/MTD root and handle the root
mounting in mount_root like all other root types. Besides making the
code more clear this also means that UBI/MTD root can be used together
with an initrd (not that anyone should care).
Also factor parsing of the root name into a helper now that it can
be easily done and will get more complicated with subsequent patches.
PM: hibernate: move finding the resume device out of software_resume
software_resume can be called either from an init call in the boot code,
or from sysfs once the system has finished booting, and the two
invocation methods this can't race with each other.
For the latter case we did just parse the suspend device manually, while
the former might not have one. Split software_resume so that the search
only happens for the boot case, which also means the special lockdep
nesting annotation can go away as the system transition mutex can be
taken a little later and doesn't have the sysfs locking nest inside it.
Wire up the shutdown method to shut down the file system when the
underlying block device is marked dead. Add a new message to
clearly distinguish this shutdown reason from other shutdowns.
Add a new ->shutdown super operation that can be used to tell the file
system to shut down, and call it from newly created holder ops when the
block device under a file system shuts down.
This only covers the main block device for "simple" file systems using
get_tree_bdev / mount_bdev. File systems their own get_tree method
or opening additional devices will need to set up their own
blk_holder_ops.
Add a mark_dead method to blk_holder_ops that is called from blk_mark_disk_dead
to notify the holder that the block device it is using has been marked dead.
Add a new blk_holder_ops structure, which is passed to blkdev_get_by_* and
installed in the block_device for exclusive claims. It will be used to
allow the block layer to call back into the user of the block device for
thing like notification of a removed device or a device resize.
projects / linux.git / log