| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Current code allows the module to be unloaded even if there are
pending data structures, such as localports and controllers on
the localports, that have yet to hit their reference counting
to remove them.
Fix by having exit entrypoint explicitly delete every controller,
which in turn will remove references on the remoteports and localports
causing them to be deleted as well. The exit entrypoint, after
initiating the deletes, will wait for the last localport to be deleted
before continuing.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
According to commit a10674bf2406 ("tcp: detecting the misuse of
.sendpage for Slab objects") and previous discussion, tcp_sendpage
should not be used for pages that is managed by SLAB, as SLAB is not
taking page reference counters into consideration.
Signed-off-by: Mikhail Skorzhinskii <mskorzhinskiy@solarflare.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
There was a few false alarms sighted on target side about wrong data
digest while performing high throughput load to XFS filesystem shared
through NVMoF TCP.
This flag tells the rest of the kernel to ensure that the data buffer
does not change while the write is in flight. It incurs a performance
penalty, so only enable it when it is actually needed, i.e. when we are
calculating data digests.
Although even with this change in place, ext2 users can steel experience
false positives, as ext2 is not respecting this flag. This may be apply
to vfat as well.
Signed-off-by: Mikhail Skorzhinskii <mskorzhinskiy@solarflare.com>
Signed-off-by: Mike Playle <mplayle@solarflare.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Adding this hint for the sake of convenience.
It was spotted that a few times people spent some time before
understanding what is exactly wrong in configuration process. This
should save a few time in such situations, especially for people who
is not very confident with NVMe requirements.
Signed-off-by: Mikhail Skorzhinskii <mskorzhinskiy@solarflare.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
nvme_ns_remove() will first set the NVME_NS_REMOVING flag before removing
it from the list at the very last step.
So to avoid selecting a namespace in nvme_find_path() which is about to be
removed check the NVME_NS_REMOVING flag, too, when selecting a new path.
Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
When we have a singular list in nvme_round_robin_path() we still
need to check its validity.
Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Factor our a common helper to check if a path has been disabled
by something other than the per-namespace ANA state.
Signed-off-by: Hannes Reinecke <hare@suse.com>
[hch: split from a bigger patch]
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
>From the NVMe 1.4 spec:
NSFEAT bit 4 if set to 1: indicates that the fields NPWG, NPWA, NPDG, NPDA,
and NOWS are defined for this namespace and should be used by the host for
I/O optimization;
[ ... ]
Namespace Preferred Write Granularity (NPWG): This field indicates the
smallest recommended write granularity in logical blocks for this namespace.
This is a 0's based value. The size indicated should be less than or equal
to Maximum Data Transfer Size (MDTS) that is specified in units of minimum
memory page size. The value of this field may change if the namespace is
reformatted. The size should be a multiple of Namespace Preferred Write
Alignment (NPWA). Refer to section 8.25 for how this field is utilized to
improve performance and endurance.
[ ... ]
Each Write, Write Uncorrectable, or Write Zeroes commands should address a
multiple of Namespace Preferred Write Granularity (NPWG) (refer to Figure
245) and Stream Write Size (SWS) (refer to Figure 515) logical blocks (as
expressed in the NLB field), and the SLBA field of the command should be
aligned to Namespace Preferred Write Alignment (NPWA) (refer to Figure 245)
for best performance.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Several new fields have been introduced in version 1.4 of the NVMe spec
at offsets that were defined as reserved in version 1.3d of the NVMe
spec. Update the definition of the nvme_id_ns data structure such that
it is in sync with version 1.4 of the NVMe spec. This change preserves
backwards compatibility.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Make the NVMe NAWUN, NAWUPF, NACWU, NPWG, NPWA, NPDG and NOWS attributes
available to initator systems for the block backend.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
The trace log for 'delete I/O submission queue' and 'delete I/O
completion queue' command will look like as below:
kworker/u49:1-3438 [003] .... 6693.070865: nvme_setup_cmd: nvme0: qid=0, cmdid=11, nsid=0, flags=0x0, meta=0x0, cmd=(nvme_admin_delete_sq sqid=1)
kworker/u49:1-3438 [003] .... 6693.071171: nvme_setup_cmd: nvme0: qid=0, cmdid=8, nsid=0, flags=0x0, meta=0x0, cmd=(nvme_admin_delete_cq cqid=24)
Signed-off-by: Tom Wu <tomwu@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Israel Rukshin <israelr@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
There are two spelling mistakes in trace_seq_printf messages, fix these.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
When running a NVMe device that is attached to a addressing
challenged PCIe root port that requires bounce buffering, our
request sizes can easily overflow the swiotlb bounce buffer
size. Limit the maximum I/O size to the limit exposed by
the DMA mapping subsystem.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Atish Patra <Atish.Patra@wdc.com>
Tested-by: Atish Patra <Atish.Patra@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
|
|
Modify nvme_alloc_sq_cmds() to call pci_free_p2pmem() to free the memory
it allocated using pci_alloc_p2pmem() in case pci_p2pmem_virt_to_bus()
returns null.
Makes sure not to call pci_free_p2pmem() if pci_alloc_p2pmem() returned
NULL, which can happen if CONFIG_PCI_P2PDMA is not configured.
The current implementation is not expected to leak since
pci_p2pmem_virt_to_bus() is expected to fail only if pci_alloc_p2pmem()
returns null. However, checking the return value of pci_alloc_p2pmem()
is more explicit.
Signed-off-by: Alan Mikhak <alan.mikhak@sifive.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Only request an IRQ mapping for read queues if at least one read queue
is being allocted, as nvme_pci_map_queues() will later on ignore the
unnecessary mapping request should nvme_dev_add() request such an IRQ
mapping even though no read queues are being allocated. However,
nvme_dev_add() can avoid making the request by checking the number of
read queues without assuming. This would bring it more in line with
nvme_setup_irqs() and nvme_calc_irq_sets().
Signed-off-by: Alan Mikhak <alan.mikhak@sifive.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
Since Linux 5.0 drivers can safely set the largest DMA mask supported
by the device, and don't need fallbacks to work around the dma mapping
implementations.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
|
|
Fix sparse warning:
drivers/nvme/host/pci.c:2926:25: warning:
symbol 'nvme_dev_pm_ops' was not declared. Should it be static?
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
With additional debugging enabled, seeing warnings for suspicious RCU
usage or Sleeping function called from invalid context.
These both map to allocation of a work structure which is currently
GFP_KERNEL, meaning it can sleep. For the RCU warning, the sequence was
sleeping while holding the RCU lock.
Convert the allocation to GFP_ATOMIC.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
With extra debug on, inconsistent lock state warnings are being called
out as the tfcp_req->reqlock is being taken out without irq, while some
calling sequences have the sequence in a softirq state.
Change the lock taking/release to raise/drop irq.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
|
|
The iolatency controller is based on rq_qos. It increments on
rq_qos_throttle() and decrements on either rq_qos_cleanup() or
rq_qos_done_bio(). a3fb01ba5af0 fixes the double accounting issue where
blk_mq_make_request() may call both rq_qos_cleanup() and
rq_qos_done_bio() on REQ_NO_WAIT. So checking STS_AGAIN prevents the
double decrement.
The above works upstream as the only way we can get STS_AGAIN is from
blk_mq_get_request() failing. The STS_AGAIN handling isn't a real
problem as bio_endio() skipping only happens on reserved tag allocation
failures which can only be caused by driver bugs and already triggers
WARN.
However, the fix creates a not so great dependency on how STS_AGAIN can
be propagated. Internally, we (Facebook) carry a patch that kills read
ahead if a cgroup is io congested or a fatal signal is pending. This
combined with chained bios progagate their bi_status to the parent is
not already set can can cause the parent bio to not clean up properly
even though it was successful. This consequently leaks the inflight
counter and can hang all IOs under that blkg.
To nip the adverse interaction early, this removes the rq_qos_cleanup()
callback in iolatency in favor of cleaning up always on the
rq_qos_done_bio() path.
Fixes: a3fb01ba5af0 ("blk-iolatency: only account submitted bios")
Debugged-by: Tejun Heo <tj@kernel.org>
Debugged-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Fix a regression introduced when removing bi_phys_segments for Write Zeroes
requests, which need to have a segment count of zero, as they don't have a
payload.
Fixes: 14ccb66b3f58 ("block: remove the bi_phys_segments field in struct bio")
Reported-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Move the blk_mq_bio_to_request() call in front of the if-statement.
Cc: Hannes Reinecke <hare@suse.com>
Cc: Omar Sandoval <osandov@fb.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
No code that occurs between blk_mq_get_ctx() and blk_mq_put_ctx() depends
on preemption being disabled for its correctness. Since removing the CPU
preemption calls does not measurably affect performance, simplify the
blk-mq code by removing the blk_mq_put_ctx() function and also by not
disabling preemption in blk_mq_get_ctx().
Cc: Hannes Reinecke <hare@suse.com>
Cc: Omar Sandoval <osandov@fb.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
cmpxchg() with an immediate value could be replaced with less expensive
xchg(). The same true if new value don't _depend_ on the old one.
In the second block, atomic_cmpxchg() return value isn't checked, so
after atomic_cmpxchg() -> atomic_xchg() conversion it could be replaced
with atomic_set(). Comparison with atomic_read() in the second chunk was
left as an optimisation (if that was the initial intention).
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
'bio->bi_iter.bi_size' is 'unsigned int', which at most hold 4G - 1
bytes.
Before 07173c3ec276 ("block: enable multipage bvecs"), one bio can
include very limited pages, and usually at most 256, so the fs bio
size won't be bigger than 1M bytes most of times.
Since we support multi-page bvec, in theory one fs bio really can
be added > 1M pages, especially in case of hugepage, or big writeback
with too many dirty pages. Then there is chance in which .bi_size
is overflowed.
Fixes this issue by using bio_full() to check if the added segment may
overflow .bi_size.
Cc: Liu Yiding <liuyd.fnst@cn.fujitsu.com>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: linux-xfs@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Cc: stable@vger.kernel.org
Fixes: 07173c3ec276 ("block: enable multipage bvecs")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Check whether the shadow mbr does fit in the provided space on the
target. Also a proper firmware should handle this case and return an
error we may prevent problems or even damage with crappy firmwares.
Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Signed-off-by: David Kozub <zub@linux.fjfi.cvut.cz>
Reviewed-by: Scott Bauer <sbauer@plzdonthack.me>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Allow modification of the shadow mbr. If the shadow mbr is not marked as
done, this data will be presented read only as the device content. Only
after marking the shadow mbr as done and unlocking a locking range the
actual content is accessible.
Co-authored-by: David Kozub <zub@linux.fjfi.cvut.cz>
Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Signed-off-by: David Kozub <zub@linux.fjfi.cvut.cz>
Reviewed-by: Scott Bauer <sbauer@plzdonthack.me>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Enable users to mark the shadow mbr as done without completely
deactivating the shadow mbr feature. This may be useful on reboots,
when the power to the disk is not disconnected in between and the shadow
mbr stores the required boot files. Of course, this saves also the
(few) commands required to enable the feature if it is already enabled
and one only wants to mark the shadow mbr as done.
Co-authored-by: David Kozub <zub@linux.fjfi.cvut.cz>
Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Signed-off-by: David Kozub <zub@linux.fjfi.cvut.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed by: Scott Bauer <sbauer@plzdonthack.me>
Reviewed-by: Jon Derrick <jonathan.derrick@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
If we pass pages through an iov_iter we always already have a reference
in the caller. Thus remove the ITER_BVEC_FLAG_NO_REF and don't take
reference to pages by default for bvec backed iov_iters.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of duplicating it.
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of duplicating it.
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of duplicating it.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of duplicating it.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of open coding it.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Use bio_release_pages instead of open coding it.
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
A lot of callers of bio_release_pages also want to mark the released
pages as dirty. Add a mark_dirty parameter to avoid a second
relatively expensive bio_for_each_segment_all loop.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Move the BIO_NO_PAGE_REF check into bio_release_pages instead of
duplicating it in both callers.
Also make the function available outside of bio.c so that we can
reuse it in other direct I/O implementations.
Reviewed-by: Minwoo Im <minwoo.im.dev@gmail.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
In commit af7ddd8a627c
("Merge tag 'dma-mapping-4.21' of git://git.infradead.org/users/hch/dma-mapping"),
dma_alloc_coherent has already zeroed the memory.
So memset is not needed.
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
In commit af7ddd8a627c
("Merge tag 'dma-mapping-4.21' of git://git.infradead.org/users/hch/dma-mapping"),
dma_alloc_coherent has already zeroed the memory.
So memset is not needed.
Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
'who' an unsigned variable in stucture opal_session_info
can never be lesser than zero. Hence, the condition
"who < OPAL_ADMIN1" can never be true.
Signed-off-by: Revanth Rajashekar <revanth.rajashekar@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
PSID is a 32 character password printed on the drive label,
to prove its physical access. This PSID reverttper function
is very useful to regain the control over the drive when it
is locked and the user can no longer access it because of some
failures. However, *all the data on the drive is completely
erased*. This method is advisable only when the user is exhausted
of all other recovery methods.
PSID capabilities are described in:
https://trustedcomputinggroup.org/wp-content/uploads/TCG_Storage-Opal_Feature_Set_PSID_v1.00_r1.00.pdf
Signed-off-by: Revanth Rajashekar <revanth.rajashekar@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Several block layer users who are not kernel developers do not know that
the word 'segment' refers to an element in a DMA scatter/gather list. Make
the block layer documentation easier to understand by stating explicitly
what the word 'segment' stands for.
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Commit f9824952ee1c ("block: update sysfs documentation") # v5.0 broke the
alphabetical order of the sysfs attribute names. List queue sysfs attribute
names alphabetically.
Cc: Damien Le Moal <damien.lemoal@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Fix the spelling of the wbt_lat_usec sysfs attribute.
Fixes: 87760e5eef35 ("block: hook up writeback throttling") # v4.10.
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
In null_handle_cmd() when device is configured as zoned, variable op is
decalred as an int, where it is used to hold values of type
REQ_OP_XXX which is of type enum req_opf. Change the type from
int to enum req_opf.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
In reboot tests on several devices we were seeing a "use after free"
when slub_debug or KASAN was enabled. The kernel complained about:
Unable to handle kernel paging request at virtual address 6b6b6c2b
...which is a classic sign of use after free under slub_debug. The
stack crawl in kgdb looked like:
0 test_bit (addr=<optimized out>, nr=<optimized out>)
1 bfq_bfqq_busy (bfqq=<optimized out>)
2 bfq_select_queue (bfqd=<optimized out>)
3 __bfq_dispatch_request (hctx=<optimized out>)
4 bfq_dispatch_request (hctx=<optimized out>)
5 0xc056ef00 in blk_mq_do_dispatch_sched (hctx=0xed249440)
6 0xc056f728 in blk_mq_sched_dispatch_requests (hctx=0xed249440)
7 0xc0568d24 in __blk_mq_run_hw_queue (hctx=0xed249440)
8 0xc0568d94 in blk_mq_run_work_fn (work=<optimized out>)
9 0xc024c5c4 in process_one_work (worker=0xec6d4640, work=0xed249480)
10 0xc024cff4 in worker_thread (__worker=0xec6d4640)
Digging in kgdb, it could be found that, though bfqq looked fine,
bfqq->bic had been freed.
Through further digging, I postulated that perhaps it is illegal to
access a "bic" (AKA an "icq") after bfq_exit_icq() had been called
because the "bic" can be freed at some point in time after this call
is made. I confirmed that there certainly were cases where the exact
crashing code path would access the "bic" after bfq_exit_icq() had
been called. Sspecifically I set the "bfqq->bic" to (void *)0x7 and
saw that the bic was 0x7 at the time of the crash.
To understand a bit more about why this crash was fairly uncommon (I
saw it only once in a few hundred reboots), you can see that much of
the time bfq_exit_icq_fbqq() fully frees the bfqq and thus it can't
access the ->bic anymore. The only case it doesn't is if
bfq_put_queue() sees a reference still held.
However, even in the case when bfqq isn't freed, the crash is still
rare. Why? I tracked what happened to the "bic" after the exit
routine. It doesn't get freed right away. Rather,
put_io_context_active() eventually called put_io_context() which
queued up freeing on a workqueue. The freeing then actually happened
later than that through call_rcu(). Despite all these delays, some
extra debugging showed that all the hoops could be jumped through in
time and the memory could be freed causing the original crash. Phew!
To make a long story short, assuming it truly is illegal to access an
icq after the "exit_icq" callback is finished, this patch is needed.
Cc: stable@vger.kernel.org
Reviewed-by: Paolo Valente <paolo.valente@unimore.it>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Now we have counters for how many times jouranl is reclaimed, how many
times cached dirty btree nodes are flushed, but we don't know how many
jouranl buckets are really reclaimed.
This patch adds reclaimed_journal_buckets into struct cache_set, this
is an increasing only counter, to tell how many journal buckets are
reclaimed since cache set runs. From all these three counters (reclaim,
reclaimed_journal_buckets, flush_write), we can have idea how well
current journal space reclaim code works.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
This patch improves performance for btree_flush_write() in following
ways,
- Use another spinlock journal.flush_write_lock to replace the very
hot journal.lock. We don't have to use journal.lock here, selecting
candidate btree nodes takes a lot of time, hold journal.lock here will
block other jouranling threads and drop the overall I/O performance.
- Only select flushing btree node from c->btree_cache list. When the
machine has a large system memory, mca cache may have a huge number of
cached btree nodes. Iterating all the cached nodes will take a lot
of CPU time, and most of the nodes on c->btree_cache_freeable and
c->btree_cache_freed lists are cleared and have need to flush. So only
travel mca list c->btree_cache to select flushing btree node should be
enough for most of the cases.
- Don't iterate whole c->btree_cache list, only reversely select first
BTREE_FLUSH_NR btree nodes to flush. Iterate all btree nodes from
c->btree_cache and select the oldest journal pin btree nodes consumes
huge number of CPU cycles if the list is huge (push and pop a node
into/out of a heap is expensive). The last several dirty btree nodes
on the tail of c->btree_cache list are earlest allocated and cached
btree nodes, they are relative to the oldest journal pin btree nodes.
Therefore only flushing BTREE_FLUSH_NR btree nodes from tail of
c->btree_cache probably includes the oldest journal pin btree nodes.
In my testing, the above change decreases 50%+ CPU consumption when
journal space is full. Some times IOPS drops to 0 for 5-8 seconds,
comparing blocking I/O for 120+ seconds in previous code, this is much
better. Maybe there is room to improve in future, but at this momment
the fix looks fine and performs well in my testing.
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
There is a race between mca_reap(), btree_node_free() and journal code
btree_flush_write(), which results very rare and strange deadlock or
panic and are very hard to reproduce.
Let me explain how the race happens. In btree_flush_write() one btree
node with oldest journal pin is selected, then it is flushed to cache
device, the select-and-flush is a two steps operation. Between these two
steps, there are something may happen inside the race window,
- The selected btree node was reaped by mca_reap() and allocated to
other requesters for other btree node.
- The slected btree node was selected, flushed and released by mca
shrink callback bch_mca_scan().
When btree_flush_write() tries to flush the selected btree node, firstly
b->write_lock is held by mutex_lock(). If the race happens and the
memory of selected btree node is allocated to other btree node, if that
btree node's write_lock is held already, a deadlock very probably
happens here. A worse case is the memory of the selected btree node is
released, then all references to this btree node (e.g. b->write_lock)
will trigger NULL pointer deference panic.
This race was introduced in commit cafe56359144 ("bcache: A block layer
cache"), and enlarged by commit c4dc2497d50d ("bcache: fix high CPU
occupancy during journal"), which selected 128 btree nodes and flushed
them one-by-one in a quite long time period.
Such race is not easy to reproduce before. On a Lenovo SR650 server with
48 Xeon cores, and configure 1 NVMe SSD as cache device, a MD raid0
device assembled by 3 NVMe SSDs as backing device, this race can be
observed around every 10,000 times btree_flush_write() gets called. Both
deadlock and kernel panic all happened as aftermath of the race.
The idea of the fix is to add a btree flag BTREE_NODE_journal_flush. It
is set when selecting btree nodes, and cleared after btree nodes
flushed. Then when mca_reap() selects a btree node with this bit set,
this btree node will be skipped. Since mca_reap() only reaps btree node
without BTREE_NODE_journal_flush flag, such race is avoided.
Once corner case should be noticed, that is btree_node_free(). It might
be called in some error handling code path. For example the following
code piece from btree_split(),
2149 err_free2:
2150 bkey_put(b->c, &n2->key);
2151 btree_node_free(n2);
2152 rw_unlock(true, n2);
2153 err_free1:
2154 bkey_put(b->c, &n1->key);
2155 btree_node_free(n1);
2156 rw_unlock(true, n1);
At line 2151 and 2155, the btree node n2 and n1 are released without
mac_reap(), so BTREE_NODE_journal_flush also needs to be checked here.
If btree_node_free() is called directly in such error handling path,
and the selected btree node has BTREE_NODE_journal_flush bit set, just
delay for 1 us and retry again. In this case this btree node won't
be skipped, just retry until the BTREE_NODE_journal_flush bit cleared,
and free the btree node memory.
Fixes: cafe56359144 ("bcache: A block layer cache")
Signed-off-by: Coly Li <colyli@suse.de>
Reported-and-tested-by: kbuild test robot <lkp@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
James Smart
Mikhail Skorzhinskii
Hannes Reinecke
Bart Van Assche
Tom Wu
Colin Ian King
Christoph Hellwig
Alan Mikhak
YueHaibing
Dennis Zhou
Pavel Begunkov
Ming Lei
Jonas Rabenstein
Fuqian Huang
Revanth Rajashekar
Chaitanya Kulkarni
Douglas Anderson
Coly Li