There is currently no way to see that the needs_check flag has been set
in the metadata. Display 'needs_check' in the thin-pool status if it is
set in the thinp metadata.
Also, update thinp documentation.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
This fixes an issue where running out of data space would cause the
thin-pool's metadata to become read-only. There was no reason to make
metadata read-only -- calling set_pool_mode() with PM_READ_ONLY was a
misguided way to error all queued and future write IOs. We can
accomplish the same by degrading from PM_OUT_OF_DATA_SPACE to
PM_OUT_OF_DATA_SPACE with error_if_no_space enabled.
Otherwise, the use of PM_READ_ONLY could cause a race where commit() was
started before the PM_READ_ONLY transition but dm_pool_commit_metadata()
would go on to fail because the block manager had transitioned to
read-only. The return of -EPERM from dm_pool_commit_metadata(), due to
attempting to commit while in read-only mode, caused the thin-pool to
set 'needs_check' because a metadata_operation_failed(). This needless
cascade of failures makes life for users more difficult than needed.
Reported-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Given the pool's cell_sort_array holds 8192 pointers it triggers an
order 5 allocation via kmalloc. This order 5 allocation is prone to
failure as system memory gets more fragmented over time.
Fix this by allocating the cell_sort_array using vmalloc.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Use EOPNOTSUPP, rather than EINVAL, error code when user attempts to
send the pool a message. Otherwise usespace is led to believe the
message failed due to invalid argument.
Reported-by: Zdenek Kabelac <zkabelac@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Previously REQ_DISCARD bios have been split into block sized chunks
before submission to the thin target. There are a couple of issues with
this:
- If the block size is small, a large discard request can
get broken up into a great many bios which is both slow and causes
a lot of memory pressure.
- The thin pool block size and the discard granularity for the
underlying data device need to be compatible if we want to passdown
the discard.
This patch relaxes the block size granularity for thin devices. It
makes use of the recent range locking added to the bio_prison to
quiesce a whole range of thin blocks before unmapping them. Once a
thin range has been unmapped the discard can then be passed down to
the data device for those sub ranges where the data blocks are no
longer used (ie. they weren't shared in the first place).
This patch also doesn't make any apologies about open-coding portions
of block core as a means to supporting async discard completions in the
near-term -- if/when late bio splitting lands it'll all get cleaned up.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
The overwrite has only ever about optimizing away the need to zero a
block if the entire block was being overwritten. As such it is only
relevant when zeroing is enabled.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Commit c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for
non-chains") regressed all existing callers that followed this pattern:
1) saving a bio's original bi_end_io
2) wiring up an intermediate bi_end_io
3) restoring the original bi_end_io from intermediate bi_end_io
4) calling bio_endio() to execute the restored original bi_end_io
The regression was due to BIO_CHAIN only ever getting set if
bio_inc_remaining() is called. For the above pattern it isn't set until
step 3 above (step 2 would've needed to establish BIO_CHAIN). As such
the first bio_endio(), in step 2 above, never decremented __bi_remaining
before calling the intermediate bi_end_io -- leaving __bi_remaining with
the value 1 instead of 0. When bio_inc_remaining() occurred during step
3 it brought it to a value of 2. When the second bio_endio() was
called, in step 4 above, it should've called the original bi_end_io but
it didn't because there was an extra reference that wasn't dropped (due
to atomic operations being optimized away since BIO_CHAIN wasn't set
upfront).
Fix this issue by removing the __bi_remaining management complexity for
all callers that use the above pattern -- bio_chain() is the only
interface that _needs_ to be concerned with __bi_remaining. For the
above pattern callers just expect the bi_end_io they set to get called!
Remove bio_endio_nodec() and also remove all bio_inc_remaining() calls
that aren't associated with the bio_chain() interface.
Also, the bio_inc_remaining() interface has been moved local to bio.c.
Fixes: c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for non-chains")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Struct bio has an atomic ref count for chained bio's, and we use this
to know when to end IO on the bio. However, most bio's are not chained,
so we don't need to always introduce this atomic operation as part of
ending IO.
Add a helper to elevate the bi_remaining count, and flag the bio as
now actually needing the decrement at end_io time. Rename the field
to __bi_remaining to catch any current users of this doing the
incrementing manually.
For high IOPS workloads, this reduces the overhead of bio_endio()
substantially.
Tested-by: Robert Elliott <elliott@hp.com>
Acked-by: Kent Overstreet <kent.overstreet@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <axboe@fb.com>
It was always intended that a read to an unprovisioned block will return
zeroes regardless of whether the pool is in read-only or read-write
mode. thin_bio_map() was inconsistent with its handling of such reads
when the pool is in read-only mode, it now properly zero-fills the bios
it returns in response to unprovisioned block reads.
Eliminate thin_bio_map()'s special read-only mode handling of -ENODATA
and just allow the IO to be deferred to the worker which will result in
pool->process_bio() handling the IO (which already properly zero-fills
reads to unprovisioned blocks).
Reported-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
stacking ontop of blk-mq devices. This blk-mq support changes the
model request-based DM uses for cloning a request to relying on
calling blk_get_request() directly from the underlying blk-mq device.
Early consumer of this code is Intel's emerging NVMe hardware; thanks
to Keith Busch for working on, and pushing for, these changes.
- A few other small fixes and cleanups across other DM targets.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQEcBAABAgAGBQJU3NRnAAoJEMUj8QotnQNavG0H/3yogMcHvKg9H+w0WmUQdwhN
w99Wj3nkquAw2sm9yahKlAMBNY53iu/LHmC6/PaTpJetgdH7y1foTrRa0qjyeB2D
DgNr8mOzxSxzX6CX9V8JMwqzky9XoG2IOt/7FeQQOpMqp4T1M2zgvbZtpl0lK/f3
lNaNBFpl+47NbGssD/WbtfI4Yy3hX0u406yGmQN5DxRyGTWD2AFqpA76g2mp8vrp
wmw259gPr4oLhj3pDc0GkuiVn59ZR2Zp+2gs0jD5uKlDL84VP/nE+WNB+ny1Mnmt
cOg8Q+W6/OosL66MKBHNsF0QS6DXNo5UvsN9fHGa5IUJw7Tsa11ZEPKHZGEbQw4=
=RiN2
-----END PGP SIGNATURE-----
Merge tag 'dm-3.20-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm
Pull device mapper changes from Mike Snitzer:
- The most significant change this cycle is request-based DM now
supports stacking ontop of blk-mq devices. This blk-mq support
changes the model request-based DM uses for cloning a request to
relying on calling blk_get_request() directly from the underlying
blk-mq device.
An early consumer of this code is Intel's emerging NVMe hardware;
thanks to Keith Busch for working on, and pushing for, these changes.
- A few other small fixes and cleanups across other DM targets.
* tag 'dm-3.20-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
dm: inherit QUEUE_FLAG_SG_GAPS flags from underlying queues
dm snapshot: remove unnecessary NULL checks before vfree() calls
dm mpath: simplify failure path of dm_multipath_init()
dm thin metadata: remove unused dm_pool_get_data_block_size()
dm ioctl: fix stale comment above dm_get_inactive_table()
dm crypt: update url in CONFIG_DM_CRYPT help text
dm bufio: fix time comparison to use time_after_eq()
dm: use time_in_range() and time_after()
dm raid: fix a couple integer overflows
dm table: train hybrid target type detection to select blk-mq if appropriate
dm: allocate requests in target when stacking on blk-mq devices
dm: prepare for allocating blk-mq clone requests in target
dm: submit stacked requests in irq enabled context
dm: split request structure out from dm_rq_target_io structure
dm: remove exports for request-based interfaces without external callers
To be future-proof and for better readability the time comparisons are modified
to use time_in_range() and time_after() instead of plain, error-prone math.
Signed-off-by: Manuel Schölling <manuel.schoelling@gmx.de>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
You can't modify the metadata in these modes. It's better to fail these
messages immediately than let the block-manager deny write locks on
metadata blocks. Otherwise these failed metadata changes will trigger
'needs_check' to get set in the metadata superblock -- requiring repair
using the thin_check utility.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Commit 80e96c5484 ("dm thin: do not allow thin device activation
while pool is suspended") delayed the initialization of a new thin
device's refcount and completion until after this new thin was added
to the pool's active_thins list and the pool lock is released. This
opens a race with a worker thread that walks the list and calls
thin_get/put, noticing that the refcount goes to 0 and calling
complete, freezing up the system and giving the oops below:
kernel: BUG: unable to handle kernel NULL pointer dereference at (null)
kernel: IP: [<ffffffff810d360b>] __wake_up_common+0x2b/0x90
kernel: Call Trace:
kernel: [<ffffffff810d3683>] __wake_up_locked+0x13/0x20
kernel: [<ffffffff810d3dc7>] complete+0x37/0x50
kernel: [<ffffffffa0595c50>] thin_put+0x20/0x30 [dm_thin_pool]
kernel: [<ffffffffa059aab7>] do_worker+0x667/0x870 [dm_thin_pool]
kernel: [<ffffffff816a8a4c>] ? __schedule+0x3ac/0x9a0
kernel: [<ffffffff810b1aef>] process_one_work+0x14f/0x400
kernel: [<ffffffff810b206b>] worker_thread+0x6b/0x490
kernel: [<ffffffff810b2000>] ? rescuer_thread+0x260/0x260
kernel: [<ffffffff810b6a7b>] kthread+0xdb/0x100
kernel: [<ffffffff810b69a0>] ? kthread_create_on_node+0x170/0x170
kernel: [<ffffffff816ad7ec>] ret_from_fork+0x7c/0xb0
kernel: [<ffffffff810b69a0>] ? kthread_create_on_node+0x170/0x170
Set the thin device's initial refcount and initialize the completion
before adding it to the pool's active_thins list in thin_ctr().
Signed-off-by: Marc Dionne <marc.dionne@your-file-system.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Discard bios and thin device deletion have the potential to release data
blocks. If the thin-pool is in out-of-data-space mode, and blocks were
released, transition the thin-pool back to full write mode.
The correct time to do this is just after the thin-pool metadata commit.
It cannot be done before the commit because the space maps will not
allow immediate reuse of the data blocks in case there's a rollback
following power failure.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
When the pool was in PM_OUT_OF_SPACE mode its process_prepared_discard
function pointer was incorrectly being set to
process_prepared_discard_passdown rather than process_prepared_discard.
This incorrect function pointer meant the discard was being passed down,
but not effecting the mapping. As such any discard that was issued, in
an attempt to reclaim blocks, would not successfully free data space.
Reported-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Simplify the pool_io_hints code that works to establish a max_sectors
value that is a power-of-2 factor of the thin-pool's blocksize. The
biggest associated improvement is that the DM thin-pool is no longer
concerning itself with the data device's max_hw_sectors when adjusting
max_sectors.
This fixes the relative fragility of the original "dm thin: adjust
max_sectors_kb based on thinp blocksize" commit that only became
apparent when testing was performed using a DM thin-pool ontop of a
virtio_blk device. One proposed upstream patch detailed the problems
inherent in virtio_blk: https://lkml.org/lkml/2014/11/20/611
So even though virtio_blk incorrectly set its max_hw_sectors it actually
helped make it clear that we need DM thinp to be tolerant of any future
Linux driver that incorrectly sets max_hw_sectors.
We only need to be concerned with modifying the thin-pool device's
max_sectors limit if it is smaller than the thin-pool's blocksize. In
this case the value of max_sectors does become a limiting factor when
upper layers (e.g. filesystems) construct their bios. But if the
hardware can support IOs larger than the thin-pool's blocksize the user
is encouraged to adjust the thin-pool's data device's max_sectors
accordingly -- doing so will enable the thin-pool to inherit the
established user-defined max_sectors.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Before this change it was expected that userspace would first suspend
all active thin devices, reload/resize the thin-pool target, then resume
all active thin devices. Now the thin-pool suspend/resume will trigger
the suspend/resume of all active thins via appropriate calls to
dm_internal_suspend and dm_internal_resume.
Store the mapped_device for each thin device in struct thin_c to make
these calls possible.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Otherwise IO could be issued to the pool while it is suspended.
Care was taken to properly interlock between the thin and thin-pool
targets when accessing the pool's 'suspended' flag. The thin_ctr will
not add a new thin device to the pool's active_thins list if the pool is
susepended.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-by: Joe Thornber <ejt@redhat.com>
As long as struct thin_c is in the list, anyone can grab a reference of
it. Consequently, we must wait for the reference count to drop to zero
*after* we remove the structure from the list, not before.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Ranges will be placed in the same cell if they overlap.
Range locking is a prerequisite for more efficient multi-block discard
support in both the cache and thin-provisioning targets.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Sort the cells in logical block order before processing each cell in
process_thin_deferred_cells(). This significantly improves the ondisk
layout on rotational storage, whereby improving read performance.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
This use of direct submission in process_shared_bio() reduces latency
for submitting bios in the shared cell by avoiding adding those bios to
the deferred list and waiting for the next iteration of the worker.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
This use of direct submission in process_prepared_mapping() reduces
latency for submitting bios in a cell by avoiding adding those bios to
the deferred list and waiting for the next iteration of the worker.
But this direct submission exposes the potential for a race between
releasing a cell and incrementing deferred set. Fix this by introducing
dm_cell_visit_release() and refactoring inc_remap_and_issue_cell()
accordingly.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
This avoids dropping the cell, so increases the probability that other
bios will collect within the cell, rather than being passed individually
to the worker.
Also add required process_cell and process_discard_cell error handling
wrappers and set associated pool-mode function pointers accordingly.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
When processing a discard bio, if the block is already quiesced do the
discard immediately rather than adding the mapping to a list for the
next iteration of the worker thread.
Discarding a fully provisioned 100G thin volume with 64k block size goes
from 860s to 95s with this change.
Clearly there's something wrong with the worker architecture, more
investigation needed.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Introduce thin_merge so that any additional constraints from the data
volume may be taken into account when determing the maximum number of
sectors that can be issued relative to the specified logical offset.
This is particularly important if/when the data volume is layered ontop
of a more sophisticated device (e.g. dm-raid or some other DM target).
Reviewed-by: Heinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Allows for filesystems to submit bios that are a factor of the thinp
blocksize, improving dm-thinp efficiency (particularly when the data
volume is RAID).
Also set io_min to max_sectors_kb if it is a factor of the thinp
blocksize.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Throttle IO based on the time it's taking the worker to do one loop.
There were reports of hung task timeouts occuring and it was observed
that the excessively long avgqu-sz (as reported by iostat) was
contributing to these hung tasks.
Throttling definitely helps dm-thinp perform better under heavy IO load
(without being detremental by being overzealous). It reduces avgqu-sz
drastically, e.g.: from 60K to ~6K, and even as low as 150 once metadata
is cached by bufio, when dirty_ratio=5, dirty_background_ratio=2. And
avgqu-sz stays at or below 30K even with dirty_ratio=20,
dirty_background_ratio=10.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Prefetch metadata at the start of the worker thread and then again every
128th bio processed from the deferred list.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Previously it was using a fixed sized hash table. There are times
when very many concurrent cells are held (such as when processing a very
large discard). When this happens the hash table performance becomes
very poor.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Avoids normal IO racing with discard.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Before, if the block layer's limit stacking didn't establish an
optimal_io_size that was compatible with the thin-pool's data block size
we'd set optimal_io_size to the data block size and minimum_io_size to 0
(which the block layer adjusts to be physical_block_size).
Update pool_io_hints() to set both minimum_io_size and optimal_io_size
to the thin-pool's data block size. This fixes an issue reported where
mkfs.xfs would create more XFS Allocation Groups on thinp volumes than
on a normal linear LV of comparable size, see:
https://bugzilla.redhat.com/show_bug.cgi?id=1003227
Reported-by: Chris Murphy <lists@colorremedies.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Track the size of any external origin. Previously the external origin's
size had to be a multiple of the thin-pool's block size, that is no
longer a requirement. In addition, snapshots that are larger than the
external origin are now supported.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Previously we used separate boolean values to track quiescing and
copying actions. By switching to an atomic_t we can support blocks that
need a partial copy and partial zero.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
DM thinp already checks whether the discard_granularity of the data
device is a factor of the thin-pool block size. But when using the
dm-thin-pool's discard passdown support, DM thinp was not selecting the
max of the underlying data device's discard_granularity and the
thin-pool's block size.
Update set_discard_limits() to set discard_granularity to the max of
these values. This enables blkdev_issue_discard() to properly align the
discards that are sent to the DM thin device on a full block boundary.
As such each discard will now cover an entire DM thin-pool block and the
block will be reclaimed.
Reported-by: Zdenek Kabelac <zkabelac@redhat.com>
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org
Update the DM thin provisioning target's allocation failure error to be
consistent with commit a9d6ceb8 ("[SCSI] return ENOSPC on thin
provisioning failure").
The DM thin target now returns -ENOSPC rather than -EIO when
block allocation fails due to the pool being out of data space (and
the 'error_if_no_space' thin-pool feature is enabled).
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-By: Joe Thornber <ejt@redhat.com>
Factor out a pool_work interface that noflush_work makes use of to wait
for and complete work items (in terms of a proper completion struct).
Allows discontinuing the use of a custom completion in terms of atomic_t
and wait_event.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Commit 85ad643b ("dm thin: add timeout to stop out-of-data-space mode
holding IO forever") introduced a fixed 60 second timeout. Users may
want to either disable or modify this timeout.
Allow the out-of-data-space timeout to be configured using the
'no_space_timeout' dm-thin-pool module param. Setting it to 0 will
disable the timeout, resulting in IO being queued until more data space
is added to the thin-pool.
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 3.14+
If the pool runs out of data space, dm-thin can be configured to
either error IOs that would trigger provisioning, or hold those IOs
until the pool is resized. Unfortunately, holding IOs until the pool is
resized can result in a cascade of tasks hitting the hung_task_timeout,
which may render the system unavailable.
Add a fixed timeout so IOs can only be held for a maximum of 60 seconds.
If LVM is going to resize a thin-pool that is out of data space it needs
to be prompt about it.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 3.14+
Commit 3e1a0699 ("dm thin: fix out of data space handling") introduced
a regression in the metadata commit() method by returning an error if
the pool is in PM_OUT_OF_DATA_SPACE mode. This oversight caused a thin
device to return errors even if the default queue_if_no_space ENOSPC
handling mode is used.
Fix commit() to only fail if pool is in PM_READ_ONLY or PM_FAIL mode.
Reported-by: qindehua@163.com
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 3.14+
Use INIT_WORK_ONSTACK to silence "ODEBUG: object is on stack, but not
annotated".
Reported-by: Zdeněk Kabeláč <zkabelac@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Commit c140e1c4e2 ("dm thin: use per thin device deferred bio lists")
introduced the use of an rculist for all active thin devices. The use
of rcu_read_lock() in process_deferred_bios() can result in a BUG if a
dm_bio_prison_cell must be allocated as a side-effect of bio_detain():
BUG: sleeping function called from invalid context at mm/mempool.c:203
in_atomic(): 1, irqs_disabled(): 0, pid: 6, name: kworker/u8:0
3 locks held by kworker/u8:0/6:
#0: ("dm-" "thin"){.+.+..}, at: [<ffffffff8106be42>] process_one_work+0x192/0x550
#1: ((&pool->worker)){+.+...}, at: [<ffffffff8106be42>] process_one_work+0x192/0x550
#2: (rcu_read_lock){.+.+..}, at: [<ffffffff816360b5>] do_worker+0x5/0x4d0
We can't process deferred bios with the rcu lock held, since
dm_bio_prison_cell allocation may block if the bio-prison's cell mempool
is exhausted.
To fix:
- Introduce a refcount and completion field to each thin_c
- Add thin_get/put methods for adjusting the refcount. If the refcount
hits zero then the completion is triggered.
- Initialise refcount to 1 when creating thin_c
- When iterating the active_thins list we thin_get() whilst the rcu
lock is held.
- After the rcu lock is dropped we process the deferred bios for that
thin.
- When destroying a thin_c we thin_put() and then wait for the
completion -- to avoid a race between the worker thread iterating
from that thin_c and destroying the thin_c.
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Commit c140e1c4e2 ("dm thin: use per thin device deferred bio lists")
incorrectly stopped disabling irqs when taking the pool's spinlock.
Irqs must be disabled when taking the pool's spinlock otherwise a thread
could spin_lock(), then get interrupted to service thin_endio() in
interrupt context, which would then deadlock in spin_lock_irqsave().
Signed-off-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
A thin-pool will allocate blocks using FIFO order for all thin devices
which share the thin-pool. Because of this simplistic allocation the
thin-pool's space can become fragmented quite easily; especially when
multiple threads are requesting blocks in parallel.
Sort each thin device's deferred_bio_list based on logical sector to
help reduce fragmentation of the thin-pool's ondisk layout.
The following tables illustrate the realized gains/potential offered by
sorting each thin device's deferred_bio_list. An "io size"-sized random
read of the device would result in "seeks/io" fragments being read, with
an average "distance/seek" between each fragment.
Data was written to a single thin device using multiple threads via
iozone (8 threads, 64K for both the block_size and io_size).
unsorted:
io size seeks/io distance/seek
--------------------------------------
4k 0.000 0b
16k 0.013 11m
64k 0.065 11m
256k 0.274 10m
1m 1.109 10m
4m 4.411 10m
16m 17.097 11m
64m 60.055 13m
256m 148.798 25m
1g 809.929 21m
sorted:
io size seeks/io distance/seek
--------------------------------------
4k 0.000 0b
16k 0.000 1g
64k 0.001 1g
256k 0.003 1g
1m 0.011 1g
4m 0.045 1g
16m 0.181 1g
64m 0.747 1011m
256m 3.299 1g
1g 14.373 1g
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Acked-by: Joe Thornber <ejt@redhat.com>
Mike Snitzer