Pre-incrementing ->cq_head can't be done in memory because OOB value
can be observed by another context.
This devalues space savings compared to original code :-\
$ ./scripts/bloat-o-meter ../vmlinux-000 ../obj/vmlinux
add/remove: 0/0 grow/shrink: 0/4 up/down: 0/-32 (-32)
Function old new delta
nvme_poll_irqdisable 464 456 -8
nvme_poll 455 447 -8
nvme_irq 388 380 -8
nvme_dev_disable 955 947 -8
But the code is minimal now: one read for head, one read for q_depth,
one increment, one comparison, single instruction phase bit update and
one write for new head.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Reported-by: John Garry <john.garry@huawei.com>
Tested-by: John Garry <john.garry@huawei.com>
Fixes: e2a366a4b0 ("nvme-pci: slimmer CQ head update")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When jumping to the out_put_disk label, we will call put_disk(), which will
trigger a call to disk_release(), which calls blk_put_queue().
Later in the cleanup code, we do blk_cleanup_queue(), which will also call
blk_put_queue().
Putting the queue twice is incorrect, and will generate a KASAN splat.
Set the disk->queue pointer to NULL, before calling put_disk(), so that the
first call to blk_put_queue() will not free the queue.
The second call to blk_put_queue() uses another pointer to the same queue,
so this call will still free the queue.
Fixes: 85136c0102 ("lightnvm: simplify geometry enumeration")
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
-----BEGIN PGP SIGNATURE-----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=3woC
-----END PGP SIGNATURE-----
Merge tag 'block-5.7-2020-04-10' of git://git.kernel.dk/linux-block
Pull block fixes from Jens Axboe:
"Here's a set of fixes that should go into this merge window. This
contains:
- NVMe pull request from Christoph with various fixes
- Better discard support for loop (Evan)
- Only call ->commit_rqs() if we have queued IO (Keith)
- blkcg offlining fixes (Tejun)
- fix (and fix the fix) for busy partitions"
* tag 'block-5.7-2020-04-10' of git://git.kernel.dk/linux-block:
block: fix busy device checking in blk_drop_partitions again
block: fix busy device checking in blk_drop_partitions
nvmet-rdma: fix double free of rdma queue
blk-mq: don't commit_rqs() if none were queued
nvme-fc: Revert "add module to ops template to allow module references"
nvme: fix deadlock caused by ANA update wrong locking
nvmet-rdma: fix bonding failover possible NULL deref
loop: Better discard support for block devices
loop: Report EOPNOTSUPP properly
nvmet: fix NULL dereference when removing a referral
nvme: inherit stable pages constraint in the mpath stack device
blkcg: don't offline parent blkcg first
blkcg: rename blkcg->cgwb_refcnt to ->online_pin and always use it
nvme-tcp: fix possible crash in recv error flow
nvme-tcp: don't poll a non-live queue
nvme-tcp: fix possible crash in write_zeroes processing
nvmet-fc: fix typo in comment
nvme-rdma: Replace comma with a semicolon
nvme-fcloop: fix deallocation of working context
nvme: fix compat address handling in several ioctls
In case rdma accept fails at nvmet_rdma_queue_connect(), release work is
scheduled. Later on, a new RDMA CM event may arrive since we didn't
destroy the cm-id and call nvmet_rdma_queue_connect_fail(), which
schedule another release work. This will cause calling
nvmet_rdma_free_queue twice. To fix this we implicitly destroy the cm_id
with non-zero ret code, which guarantees that new rdma_cm events will
not arrive afterwards. Also add a qp pointer to nvmet_rdma_queue
structure, so we can use it when the cm_id pointer is NULL or was
destroyed.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Suggested-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The original patch was to resolve the lldd being able to be unloaded
while being used to talk to the boot device of the system. However, the
end result of the original patch is that any driver unload while a nvme
controller is live via the lldd is now being prohibited. Given the module
reference, the module teardown routine can't be called, thus there's no
way, other than manual actions to terminate the controllers.
Fixes: 863fbae929 ("nvme_fc: add module to ops template to allow module references")
Cc: <stable@vger.kernel.org> # v5.4+
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The deadlock combines 4 flows in parallel:
- ns scanning (triggered from reconnect)
- request timeout
- ANA update (triggered from reconnect)
- I/O coming into the mpath device
(1) ns scanning triggers disk revalidation -> update disk info ->
freeze queue -> but blocked, due to (2)
(2) timeout handler reference the g_usage_counter - > but blocks in
the transport .timeout() handler, due to (3)
(3) the transport timeout handler (indirectly) calls nvme_stop_queue() ->
which takes the (down_read) namespaces_rwsem - > but blocks, due to (4)
(4) ANA update takes the (down_write) namespaces_rwsem -> calls
nvme_mpath_set_live() -> which synchronize the ns_head srcu
(see commit 504db087aa) -> but blocks, due to (5)
(5) I/O came into nvme_mpath_make_request -> took srcu_read_lock ->
direct_make_request > blk_queue_enter -> but blocked, due to (1)
==> the request queue is under freeze -> deadlock.
The fix is making ANA update take a read lock as the namespaces list
is not manipulated, it is just the ns and ns->head that are being
updated (which is protected with the ns->head lock).
Fixes: 0d0b660f21 ("nvme: add ANA support")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
RDMA_CM_EVENT_ADDR_CHANGE event occur in the case of bonding failover
on normal as well as on listening cm_ids. Hence this event will
immediately trigger a NULL dereference trying to disconnect a queue
for a cm_id that actually belongs to the port.
To fix this we provide a different handler for the listener cm_ids
that will defer a work to disable+(re)enable the port which essentially
destroys and setups another listener cm_id
Reported-by: Alex Lyakas <alex@zadara.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Tested-by: Alex Lyakas <alex@zadara.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
update changing all our txt files to rst ones. Excluding that, we
have the usual driver updates (qla2xxx, ufs, lpfc, zfcp, ibmvfc,
pm80xx, aacraid), a treewide update for scnprintf and some other minor
updates. The major core update is Hannes moving functions out of the
aacraid driver and into the core.
Signed-off-by: James E.J. Bottomley <jejb@linux.ibm.com>
-----BEGIN PGP SIGNATURE-----
iJwEABMIAEQWIQTnYEDbdso9F2cI+arnQslM7pishQUCXoYKiyYcamFtZXMuYm90
dG9tbGV5QGhhbnNlbnBhcnRuZXJzaGlwLmNvbQAKCRDnQslM7pishSasAP4iGwSB
Y8tFaZgWadu76+wj5MdqTBoXdhnIuFF0rZG3pQEAiIKdsfQlbSFdm75+gUtx5hG/
GOilX/pJczTRJDCGNis=
=g7Sk
-----END PGP SIGNATURE-----
Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull SCSI updates from James Bottomley:
"This series has a huge amount of churn because it pulls in Mauro's doc
update changing all our txt files to rst ones.
Excluding that, we have the usual driver updates (qla2xxx, ufs, lpfc,
zfcp, ibmvfc, pm80xx, aacraid), a treewide update for scnprintf and
some other minor updates.
The major core change is Hannes moving functions out of the aacraid
driver and into the core"
* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (223 commits)
scsi: aic7xxx: aic97xx: Remove FreeBSD-specific code
scsi: ufs: Do not rely on prefetched data
scsi: dc395x: remove dc395x_bios_param
scsi: libiscsi: Fix error count for active session
scsi: hpsa: correct race condition in offload enabled
scsi: message: fusion: Replace zero-length array with flexible-array member
scsi: qedi: Add PCI shutdown handler support
scsi: qedi: Add MFW error recovery process
scsi: ufs: Enable block layer runtime PM for well-known logical units
scsi: ufs-qcom: Override devfreq parameters
scsi: ufshcd: Let vendor override devfreq parameters
scsi: ufshcd: Update the set frequency to devfreq
scsi: ufs: Resume ufs host before accessing ufs device
scsi: ufs-mediatek: customize the delay for enabling host
scsi: ufs: make HCE polling more compact to improve initialization latency
scsi: ufs: allow custom delay prior to host enabling
scsi: ufs-mediatek: use common delay function
scsi: ufs: introduce common and flexible delay function
scsi: ufs: use an enum for host capabilities
scsi: ufs: fix uninitialized tx_lanes in ufshcd_disable_tx_lcc()
...
When item release is called, the parent is already null. We need the
parent to pass to nvmet_referral_disable so hook it up to
->disconnect_notify.
Reported-by: Tony Asleson <tasleson@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
If the backing device require stable pages, we need to set it on the
stack mpath device as well. This applies to rdma/fc transports when
doing data integrity and tcp transport calculating digests.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
If the target misbehaves and sends us unexpected payload we
need to make sure to fail the controller and stop processing
the input stream. We clear the rd_enabled flag and stop
the io_work, but we may still requeue it if we still have pending
sends and then in the next invocation we will process the input
stream as the check is only in the .data_ready upcall.
To fix this we need to make sure not to self-requeue io_work
upon a recv flow error.
This fixes the crash:
nvme nvme2: receive failed: -22
BUG: unable to handle page fault for address: ffffbeb5816c3b48
nvme_ns_head_make_request: 29 callbacks suppressed
block nvme0n5: no usable path - requeuing I/O
block nvme0n5: no usable path - requeuing I/O
block nvme0n7: no usable path - requeuing I/O
block nvme0n7: no usable path - requeuing I/O
block nvme0n3: no usable path - requeuing I/O
block nvme0n3: no usable path - requeuing I/O
block nvme0n3: no usable path - requeuing I/O
block nvme0n7: no usable path - requeuing I/O
block nvme0n3: no usable path - requeuing I/O
block nvme0n3: no usable path - requeuing I/O
#PF: supervisor read access inkernel mode
#PF: error_code(0x0000) - not-present page
PGD 1039157067 P4D 1039157067 PUD 103915a067 PMD 102719f067 PTE 0
Oops: 0000 [#1] SMP PTI
CPU: 8 PID: 411 Comm: kworker/8:1H Not tainted 5.3.0-40-generic #32~18.04.1-Ubuntu
Hardware name: Supermicro Super Server/X10SRi-F, BIOS 2.0 12/17/2015
Workqueue: nvme_tcp_wq nvme_tcp_io_work [nvme_tcp]
RIP: 0010:nvme_tcp_recv_skb+0x2ae/0xb50 [nvme_tcp]
RSP: 0018:ffffbeb5806cfd10 EFLAGS: 00010246
RAX: ffffbeb5816c3b48 RBX: 00000000000003d0 RCX: 0000000000000008
RDX: 00000000000003d0 RSI: 0000000000000001 RDI: ffff9a3040684b40
RBP: ffffbeb5806cfd90 R08: 0000000000000000 R09: ffffffff946e6900
R10: ffffbeb5806cfce0 R11: 0000000000000001 R12: 0000000000000000
R13: ffff9a2ff86501c0 R14: 00000000000003d0 R15: ffff9a30b85f2798
FS: 0000000000000000(0000) GS:ffff9a30bf800000(0000) knlGS:0000000000000000
CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: ffffbeb5816c3b48 CR3: 000000088400a006 CR4: 00000000003626e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
tcp_read_sock+0x8c/0x290
? __release_sock+0x9d/0xe0
? nvme_tcp_write_space+0xb0/0xb0 [nvme_tcp]
nvme_tcp_io_work+0x4b4/0x830 [nvme_tcp]
? finish_task_switch+0x163/0x270
process_one_work+0x1fd/0x3f0
worker_thread+0x34/0x410
kthread+0x121/0x140
? process_one_work+0x3f0/0x3f0
? kthread_park+0xb0/0xb0
ret_from_fork+0x35/0x40
Reported-by: Roy Shterman <roys@lightbitslabs.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
In error recovery we might be removing the queue so check we
can actually poll before we do.
Reported-by: Mark Wunderlich <mark.wunderlich@intel.com>
Tested-by: Mark Wunderlich <mark.wunderlich@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We cannot look at blk_rq_payload_bytes without first checking
that the request has a mappable physical segments first (e.g.
blk_rq_nr_phys_segments(rq) != 0) and only then to take the
request payload bytes. This caused us to send a wrong sgl to
the target or even dereference a non-existing buffer in case
we actually got to the data send sequence (if it was in-capsule).
Reported-by: Tony Asleson <tasleson@redhat.com>
Suggested-by: Chaitanya Kulkarni <Chaitanya.Kulkarni@wdc.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Fix typo in comment: about should be abort
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chiatanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Use a semicolon at the end of an assignment expression.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
There's been a longstanding bug of LS completions which freed ls ops,
particularly the disconnect LS, while executing on a work context that
is in the memory being free. Not a good thing to do.
Rework LS handling to make callbacks in the rport context rather than
the ls_request context.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
On a 32-bit kernel, the upper bits of userspace addresses passed via
various ioctls are silently ignored by the nvme driver.
However on a 64-bit kernel running a compat task, these upper bits are
not ignored and are in fact required to be zero for the ioctls to work.
Unfortunately, this difference matters. 32-bit smartctl submits the
NVME_IOCTL_ADMIN_CMD ioctl with garbage in these upper bits because it
seems the pointer value it puts into the nvme_passthru_cmd structure is
sign extended. This works fine on 32-bit kernels but fails on a 64-bit
one because (at least on my setup) the addresses smartctl uses are
consistently above 2G. For example:
# smartctl -x /dev/nvme0n1
smartctl 7.1 2019-12-30 r5022 [x86_64-linux-5.5.11] (local build)
Copyright (C) 2002-19, Bruce Allen, Christian Franke, www.smartmontools.org
Read NVMe Identify Controller failed: NVME_IOCTL_ADMIN_CMD: Bad address
Since changing 32-bit kernels to actually check all of the submitted
address bits now would break existing userspace, this patch fixes the
compat problem by explicitly zeroing the upper bits in the compat case.
This enables 32-bit smartctl to work on a 64-bit kernel.
Signed-off-by: Nick Bowler <nbowler@draconx.ca>
Signed-off-by: Christoph Hellwig <hch@lst.de>
-----BEGIN PGP SIGNATURE-----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=6Glq
-----END PGP SIGNATURE-----
Merge tag 'for-5.7/block-2020-03-29' of git://git.kernel.dk/linux-block
Pull block updates from Jens Axboe:
- Online capacity resizing (Balbir)
- Number of hardware queue change fixes (Bart)
- null_blk fault injection addition (Bart)
- Cleanup of queue allocation, unifying the node/no-node API
(Christoph)
- Cleanup of genhd, moving code to where it makes sense (Christoph)
- Cleanup of the partition handling code (Christoph)
- disk stat fixes/improvements (Konstantin)
- BFQ improvements (Paolo)
- Various fixes and improvements
* tag 'for-5.7/block-2020-03-29' of git://git.kernel.dk/linux-block: (72 commits)
block: return NULL in blk_alloc_queue() on error
block: move bio_map_* to blk-map.c
Revert "blkdev: check for valid request queue before issuing flush"
block: simplify queue allocation
bcache: pass the make_request methods to blk_queue_make_request
null_blk: use blk_mq_init_queue_data
block: add a blk_mq_init_queue_data helper
block: move the ->devnode callback to struct block_device_operations
block: move the part_stat* helpers from genhd.h to a new header
block: move block layer internals out of include/linux/genhd.h
block: move guard_bio_eod to bio.c
block: unexport get_gendisk
block: unexport disk_map_sector_rcu
block: unexport disk_get_part
block: mark part_in_flight and part_in_flight_rw static
block: mark block_depr static
block: factor out requeue handling from dispatch code
block/diskstats: replace time_in_queue with sum of request times
block/diskstats: accumulate all per-cpu counters in one pass
block/diskstats: more accurate approximation of io_ticks for slow disks
...
Current make_request based drivers use either blk_alloc_queue_node or
blk_alloc_queue to allocate a queue, and then set up the make_request_fn
function pointer and a few parameters using the blk_queue_make_request
helper. Simplify this by passing the make_request pointer to
blk_alloc_queue, and while at it merge the _node variant into the main
helper by always passing a node_id, and remove the superfluous gfp_mask
parameter. A lower-level __blk_alloc_queue is kept for the blk-mq case.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Lift the common namespace identifier reporting between the shared
namespace and new nshead cases into common code. This also means
one less lock is held while doing I/O.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
There is no non __-prefixed version, so make the name a little more
readable.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Move the handling of an error into the function from the caller, and
only do it for an actual error on the admin command itself, not the
command parsing, as that should be enough to deal with devices claiming
a bogus version compliance.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The transition to LIVE state should not fail in case of a new controller.
Moving to DELETING state before nvme_tcp_create_ctrl() allocates all the
resources may leads to NULL dereference at teardown flow (e.g., IO tagset,
admin_q, connect_q).
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The transition to LIVE state should not fail in case of a new controller.
Moving to DELETING state before nvme_tcp_create_ctrl() allocates all the
resources may leads to NULL dereference at teardown flow (e.g., IO tagset,
admin_q, connect_q).
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Calling nvme_sysfs_delete() when the controller is in the middle of
creation may cause several bugs. If the controller is in NEW state we
remove delete_controller file and don't delete the controller. The user
will not be able to use nvme disconnect command on that controller again,
although the controller may be active. Other bugs may happen if the
controller is in the middle of create_ctrl callback and
nvme_do_delete_ctrl() starts. For example, freeing I/O tagset at
nvme_do_delete_ctrl() before it was allocated at create_ctrl callback.
To fix all those races don't allow the user to delete the controller
before it was fully created.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Put the ctrl reference count at nvme_uninit_ctrl as opposed to
nvme_init_ctrl which takes it. This decrease the reference count at the
core layer instead of decreasing it on each transport separately.
Also move the call of nvme_uninit_ctrl at PCI driver after calling to
nvme_release_prp_pools and nvme_dev_unmap, in order to put the reference
count after using the dev. This is safe because those functions use
nvme_dev which is freed only later at nvme_pci_free_ctrl.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
In case nvme_sysfs_delete() is called by the user before taking the ctrl
reference count, the ctrl may be freed during the creation and cause the
bug. Take the reference as soon as the controller is externally visible,
which is done by cdev_device_add() in nvme_init_ctrl(). Also take the
reference count at the core layer instead of taking it on each transport
separately.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Destroy the resources in the same order like in nvme_probe error flow to
improve code readability.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The return code of nvme_delete_ctrl_sync is never used, so change it to
void.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Improve code readability.
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
ida instances allocate some internal memory in addition to the base
'struct ida'. Use ida_destroy() to release that memory at module_exit().
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently 32 bit application gets ENOTTY when it calls
compat_ioctl with NVME_IOCTL_SUBMIT_IO in 64 bit kernel.
The cause is that the results of sizeof(struct nvme_user_io),
which is used to define NVME_IOCTL_SUBMIT_IO,
are not same between 32 bit compiler and 64 bit compiler.
* 32 bit: the result of sizeof nvme_user_io is 44.
* 64 bit: the result of sizeof nvme_user_io is 48.
64 bit compiler seems to add 32 bit padding for multiple of 8 bytes.
This patch adds a compat_ioctl handler.
The handler replaces NVME_IOCTL_SUBMIT_IO32 with NVME_IOCTL_SUBMIT_IO
in case 32 bit application calls compat_ioctl for submit in 64 bit kernel.
Then, it calls nvme_ioctl as usual.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Masahiro Yamada (KIOXIA) <masahiro31.yamada@kioxia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
If we have a 4-byte data digest to send to the wire, but we
have more data to send, set MSG_MORE to tell the stack
that more is coming.
Reviewed-by: Mark Wunderlich <mark.wunderlich@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Since snprintf() returns the would-be-output size instead of the
actual output size, the succeeding calls may go beyond the given
buffer limit. Fix it by replacing with scnprintf().
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The nvme multipath error handling defaults to controller reset if the
error is unknown. There are, however, no existing nvme status codes that
indicate a reset should be used, and resetting causes unnecessary
disruption to the rest of IO.
Change nvme's error handling to first check if failover should happen.
If not, let the normal error handling take over rather than reset the
controller.
Based-on-a-patch-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: John Meneghini <johnm@netapp.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Current nvmet-rdma code allocates MR pool budget based on queue size,
assuming both host and target use the same "max_pages_per_mr" count.
After limiting the mdts value for RDMA controllers, we know the factor
of maximum MR's per IO operation. Thus, make sure MR pool will be
sufficient for the required IO depth and IO size.
That is, say host's SQ size is 100, then the MR pool budget allocated
currently at target will also be 100 MRs. But 100 IO WRITE Requests
with 256 sg_count(IO size above 1MB) require 200 MRs when target's
"max_pages_per_mr" is 128.
Reported-by: Krishnamraju Eraparaju <krishna2@chelsio.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Set the maximal data transfer size to be 1MB (currently mdts is
unlimited). This will allow calculating the amount of MR's that
one ctrl should allocate to fulfill it's capabilities.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Some transports, such as RDMA, would like to set the Maximum Data
Transfer Size (MDTS) according to device/port/ctrl characteristics.
This will enable the transport to set the optimal MDTS according to
controller needs and device capabilities. Add a new nvmet transport
op that is called during ctrl identification. This will not effect
transports that don't implement this option. The return value of the new
op is according to the NVMe spec definition for MDTS.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
If we failed to receive data from the socket, don't try
to further process it, we will for sure be handling a queue
error at this point. While no issue was seen with the
current behavior thus far, its safer to cease socket processing
if we detected an error.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Consolidate the request failure handling code to where
it is being fetched (nvme_tcp_try_send).
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
MAXH2CDATA is not zero based. Also no reason to limit ourselves to
1M transfers as we can do more easily. Make this an arbitrary limit
of 16M.
Reported-by: Wenhua Liu <liuw@vmware.com>
Cc: stable@vger.kernel.org # v5.0+
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, queue io_cpu assignment is done sequentially for default,
read and poll queues based on queue id. This causes miss-alignment between
context of CPU initiating I/O and the I/O worker thread processing
queued requests or completions.
Change to modify queue io_cpu assignment to take into account queue
maps offset. Each queue io_cpu will start at zero for each queue map.
This essentially aligns read/poll queues to start over the same range as
default queues.
Testing performed by Mark with:
- ram device (nvmet)
- single CPU core (pinned)
- 100% 4k reads
- engine io_uring (not using sq_thread option)
- hipri flag set
Micro-benchmark results show a net gain of:
- increase of 18%-29% in IOPs
- reduction of 16%-22% in average latency
- reduction of 7%-23% in 99.99% latency
Baseline:
========
QDepth/Batch | IOPs [k] | Avg. Lat [us] | 99.99% Lat [us]
-----------------------------------------------------------------
1/1 | 32.4 | 30.11 | 50.94
32/8 | 179 | 168.20 | 371
CPU alignment:
=============
QDepth/Batch | IOPs [k] | Avg. Lat [us] | 99.99% Lat [us]
-----------------------------------------------------------------
1/1 | 38.5 | 25.18 | 39.16
32/8 | 231 | 130.75 | 343
Reported-by: Mark Wunderlich <mark.wunderlich@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The timeout handler can use the existing nvme_poll() if it needs to
check a polled queue, allowing nvme_poll_irqdisable() to handle only
irq driven queues for the remaining callers.
Signed-off-by: Keith Busch <kbusch@kernel.org>
Completion handling had been done in two steps: find all new completions
under a lock, then handle those completions outside the lock. This was
done to make the locked section as short as possible so that other
threads using the same lock wait less time.
The driver no longer shares locks during completion, and is in fact
lockless for interrupt driven queues, so the optimization no longer
serves its original purpose. Replace the two-pass completion queue
handler with a single pass that completes entries immediately.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The only user for tagged completion was for timeout handling. That user,
though, really only cares if the timed out command is completed, which
we can safely check within the timeout handler.
Remove the tag check to simplify completion handling.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
For set feature command when setting up NVME_FEAT_NUM_QUEUES, check
Number of I/O Completion Queues Requested (NCQR) and Number of I/O
Submission Queues Requested (NSQR) before we proceed, for invalid values
(i.e. 65535) return an appropriate NVMe invalid field status.
Signed-off-by: Amit Engel <Amit.Engel@dell.com>
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
After initialization, nvme_wait_ready checks for readiness every 100ms,
even though the drive may be ready far sooner than that. This delays
system boot by hundreds of milliseconds. Reduce the delay, checking for
readiness every millisecond instead.
Boot-time tests on an AWS c5.12xlarge:
Before:
[ 0.546936] initcall nvme_init+0x0/0x5b returned 0 after 37 usecs
...
[ 0.764178] nvme nvme0: 2/0/0 default/read/poll queues
[ 0.768424] nvme0n1: p1
[ 0.774132] EXT4-fs (nvme0n1p1): mounted filesystem with ordered data mode. Opts: (null)
[ 0.774146] VFS: Mounted root (ext4 filesystem) on device 259:1.
...
[ 0.788141] Run /sbin/init as init process
After:
[ 0.537088] initcall nvme_init+0x0/0x5b returned 0 after 37 usecs
...
[ 0.543457] nvme nvme0: 2/0/0 default/read/poll queues
[ 0.548473] nvme0n1: p1
[ 0.554339] EXT4-fs (nvme0n1p1): mounted filesystem with ordered data mode. Opts: (null)
[ 0.554344] VFS: Mounted root (ext4 filesystem) on device 259:1.
...
[ 0.567931] Run /sbin/init as init process
Signed-off-by: Josh Triplett <josh@joshtriplett.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
