This is for the next tuning pr code more readble patch, make
linux/nvme.h's changes separately.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The nvme keep-alive operation, which executes at a periodic interval,
could potentially sneak in while shutting down a fabric controller.
This may lead to a race between the fabric controller admin queue
destroy code path (invoked while shutting down controller) and hw/hctx
queue dispatcher called from the nvme keep-alive async request queuing
operation. This race could lead to the kernel crash shown below:
Call Trace:
autoremove_wake_function+0x0/0xbc (unreliable)
__blk_mq_sched_dispatch_requests+0x114/0x24c
blk_mq_sched_dispatch_requests+0x44/0x84
blk_mq_run_hw_queue+0x140/0x220
nvme_keep_alive_work+0xc8/0x19c [nvme_core]
process_one_work+0x200/0x4e0
worker_thread+0x340/0x504
kthread+0x138/0x140
start_kernel_thread+0x14/0x18
While shutting down fabric controller, if nvme keep-alive request sneaks
in then it would be flushed off. The nvme_keep_alive_end_io function is
then invoked to handle the end of the keep-alive operation which
decrements the admin->q_usage_counter and assuming this is the last/only
request in the admin queue then the admin->q_usage_counter becomes zero.
If that happens then blk-mq destroy queue operation (blk_mq_destroy_
queue()) which could be potentially running simultaneously on another
cpu (as this is the controller shutdown code path) would forward
progress and deletes the admin queue. So, now from this point onward
we are not supposed to access the admin queue resources. However the
issue here's that the nvme keep-alive thread running hw/hctx queue
dispatch operation hasn't yet finished its work and so it could still
potentially access the admin queue resource while the admin queue had
been already deleted and that causes the above crash.
The above kernel crash is regression caused due to changes implemented
in commit a54a93d0e3 ("nvme: move stopping keep-alive into
nvme_uninit_ctrl()"). Ideally we should stop keep-alive before destroyin
g the admin queue and freeing the admin tagset so that it wouldn't sneak
in during the shutdown operation. However we removed the keep alive stop
operation from the beginning of the controller shutdown code path in commit
a54a93d0e3 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()")
and added it under nvme_uninit_ctrl() which executes very late in the
shutdown code path after the admin queue is destroyed and its tagset is
removed. So this change created the possibility of keep-alive sneaking in
and interfering with the shutdown operation and causing observed kernel
crash.
To fix the observed crash, we decided to move nvme_stop_keep_alive() from
nvme_uninit_ctrl() to nvme_remove_admin_tag_set(). This change would ensure
that we don't forward progress and delete the admin queue until the keep-
alive operation is finished (if it's in-flight) or cancelled and that would
help contain the race condition explained above and hence avoid the crash.
Moving nvme_stop_keep_alive() to nvme_remove_admin_tag_set() instead of
adding nvme_stop_keep_alive() to the beginning of the controller shutdown
code path in nvme_stop_ctrl(), as was the case earlier before commit
a54a93d0e3 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()"),
would help save one callsite of nvme_stop_keep_alive().
Fixes: a54a93d0e3 ("nvme: move stopping keep-alive into nvme_uninit_ctrl()")
Link: https://lore.kernel.org/all/1a21f37b-0f2a-4745-8c56-4dc8628d3983@linux.ibm.com/
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This reverts commit d06923670b.
It was realized that the fix implemented to contain the race condition
among the keep alive task and the fabric shutdown code path in the commit
d06923670b5ia ("nvme: make keep-alive synchronous operation") is not
optimal. The reason being keep-alive runs under the workqueue and making
it synchronous would waste a workqueue context.
Furthermore, we later found that the above race condition is a regression
caused due to the changes implemented in commit a54a93d0e3 ("nvme: move
stopping keep-alive into nvme_uninit_ctrl()"). So we decided to revert the
commit d06923670b ("nvme: make keep-alive synchronous operation") and
then fix the regression.
Link: https://lore.kernel.org/all/196f4013-3bbf-43ff-98b4-9cb2a96c20c2@grimberg.me/
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Nilay Shroff <nilay@linux.ibm.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
If the device supports SGLs, use these for all user requests. This
format encodes the expected transfer length so it can catch short buffer
errors in a user command, whether it occurred accidently or maliciously.
For controllers that support SGL data mode, this is a viable mitigation
to CVE-2023-6238. For controllers that don't support SGLs, log a warning
in the passthrough path since not having the capability can corrupt
data if the interface is not used correctly.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This provides a little more context when reading the code than hardcoded
magic numbers.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Supporting this mode allows creating and merging multi-segment metadata
requests that wouldn't be possible otherwise. It also allows directly
using user space requests that straddle physically discontiguous pages.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The code currently uses list_for_each_entry_rcu() while holding an SRCU
lock, triggering false positive warnings with CONFIG_PROVE_RCU=y
enabled:
drivers/nvme/host/multipath.c:168 RCU-list traversed in non-reader section!!
drivers/nvme/host/multipath.c:227 RCU-list traversed in non-reader section!!
drivers/nvme/host/multipath.c:260 RCU-list traversed in non-reader section!!
While the list is properly protected by SRCU lock, the code uses the
wrong list traversal primitive. Replace list_for_each_entry_rcu() with
list_for_each_entry_srcu() to correctly indicate SRCU-based protection
and eliminate the false warning.
Signed-off-by: Breno Leitao <leitao@debian.org>
Fixes: be647e2c76 ("nvme: use srcu for iterating namespace list")
Signed-off-by: Keith Busch <kbusch@kernel.org>
Pull MD fixes from Song:
"This set contains a fix for a W=1 warning, by John Garry, and a
MAINTAINERS update."
* tag 'md-6.13-20241115' of https://git.kernel.org/pub/scm/linux/kernel/git/mdraid/linux:
MAINTAINERS: Update git tree for mdraid subsystem
md/raid5: Increase r5conf.cache_name size
A previous commit changed how requests are linked in the plug structure,
but unlike the previous method, it uses a new type for it rather than
struct request. The latter is available even for !CONFIG_BLOCK, while
struct rq_list is now. Move it outside CONFIG_BLOCK.
Reported-by: Nathan Chancellor <nathan@kernel.org>
Fixes: a3396b9999 ("block: add a rq_list type")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
seq_printf is costly. For each block device, 19 decimal values are
yielded in /proc/diskstats via seq_printf. On a system with 16 logical
block devices, profiling for open/read/close sequences shows seq_printf
took ~75% samples of diskstats_show:
diskstats_show(92.626% 2269372/2450040)
seq_printf(76.026% 1725313/2269372)
vsnprintf(99.163% 1710866/1725313)
format_decode(26.597% 455040/1710866)
number(19.554% 334542/1710866)
memcpy_orig(4.183% 71570/1710866)
...
srso_return_thunk(0.009% 148/1725313)
part_stat_read_all(8.030% 182236/2269372)
One million rounds of open/read/close /proc/diskstats takes:
real 0m37.687s
user 0m0.264s
sys 0m32.911s
On average, each sequence tooks ~0.032ms
With this patch, most decimal values are yield via seq_put_decimal_ull,
performance is significantly improved:
real 0m20.792s
user 0m0.316s
sys 0m20.463s
On average, each sequence tooks ~0.020ms, a ~37.5% improvement.
Signed-off-by: David Wang <00107082@163.com>
Link: https://lore.kernel.org/r/20241108054500.4251-1-00107082@163.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
LIFO ordering for batched completions is a bit unexpected and also
defeats some merging optimizations in e.g. the XFS buffered write
code. Now that we can easily add the request to the tail of the list
do that.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241113152050.157179-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add requests to the tail of the list instead of the front so that they
are queued up in submission order.
Remove the re-reordering in blk_mq_dispatch_plug_list, virtio_queue_rqs
and nvme_queue_rqs now that the list is ordered as expected.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241113152050.157179-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Replace the semi-open coded request list helpers with a proper rq_list
type that mirrors the bio_list and has head and tail pointers. Besides
better type safety this actually allows to insert at the tail of the
list, which will be useful soon.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241113152050.157179-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_flush_plug_list submits requests in the reverse order that they
were submitted, which leads to a rather suboptimal I/O pattern
especially in rotational devices. Fix this by rewriting virtio_queue_rqs
so that it always pops the requests from the passed in request list, and
then adds them to the head of a local submit list. This actually
simplifies the code a bit as it removes the complicated list splicing,
at the cost of extra updates of the rq_next pointer. As that should be
cache hot anyway it should be an easy price to pay.
Fixes: 0e9911fa76 ("virtio-blk: support mq_ops->queue_rqs()")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241113152050.157179-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blk_mq_flush_plug_list submits requests in the reverse order that they
were submitted, which leads to a rather suboptimal I/O pattern especially
in rotational devices. Fix this by rewriting nvme_queue_rqs so that it
always pops the requests from the passed in request list, and then adds
them to the head of a local submit list. This actually simplifies the
code a bit as it removes the complicated list splicing, at the cost of
extra updates of the rq_next pointer. As that should be cache hot
anyway it should be an easy price to pay.
Fixes: d62cbcf62f ("nvme: add support for mq_ops->queue_rqs()")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241113152050.157179-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Call blk_validate_limits on the queue limits used for zone append
splitting so that calculated values get filled in and any stacking
conflicts get cought.
Without this there isn't a max_zone_append_sectors limits as of commit
559218d43e ("block: pre-calculate max_zone_append_sectors").
Fixes: 559218d43e ("block: pre-calculate max_zone_append_sectors")
Reported-by: Yi Zhang <yi.zhang@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20241113084541.34315-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
While block drivers do the validation as part of committing them to the
queue, users that use the limit outside of a block device context have
to validate the limits and fill in the calculated values as well.
So far btrfs is the only user of queue limits without a block device,
and it has gotten away with that more or less by accident. But with
commit 559218d43e ("block: pre-calculate max_zone_append_sectors")
this became fatal for setups that have small max zone append size,
as it won't be limited now.
Export blk_validate_limits so that it can be called directly from btrfs.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20241113084541.34315-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull NVMe updates from Keith:
"nvme updates for Linux 6.13
- Use uring_cmd helper (Pavel)
- Host Memory Buffer allocation enhancements (Christoph)
- Target persistent reservation support (Guixin)
- Persistent reservation tracing (Guixen)
- NVMe 2.1 specification support (Keith)
- Rotational Meta Support (Matias, Wang, Keith)
- Volatile cache detection enhancment (Guixen)"
* tag 'nvme-6.13-2024-11-13' of git://git.infradead.org/nvme: (22 commits)
nvmet: add tracing of reservation commands
nvme: parse reservation commands's action and rtype to string
nvmet: report ns's vwc not present
nvme: check ns's volatile write cache not present
nvme: add rotational support
nvme: use command set independent id ns if available
nvmet: support for csi identify ns
nvmet: implement rotational media information log
nvmet: implement endurance groups
nvmet: declare 2.1 version compliance
nvmet: implement crto property
nvmet: implement supported features log
nvmet: implement supported log pages
nvmet: implement active command set ns list
nvmet: implement id ns for nvm command set
nvmet: support reservation feature
nvme: add reservation command's defines
nvme-core: remove repeated wq flags
nvmet: make nvmet_wq visible in sysfs
nvme-pci: use dma_alloc_noncontigous if possible
...
Add tracing of reservation commands, including register, acquire,
release and report, and also parse the action and rtype to string
to make the trace log more human-readable.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Parse reservation commands's action(including rrega, racqa and rrela)
and rtype to string to make the trace log more human-readable.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, we report that controller has vwc even though the ns may
not have vwc. Report ns's vwc not present when not buffered_io or
backdev doesn't have vwc.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
For compiling with W=1, the following warning can be seen:
drivers/md/raid5.c: In function ‘setup_conf’:
drivers/md/raid5.c:2423:12: error: ‘%s’ directive output may be truncated writing up to 31 bytes into a region of size between 16 and 26 [-Werror=format-truncation=]
"raid%d-%s", conf->level, mdname(conf->mddev));
^~
drivers/md/raid5.c:2422:3: note: ‘snprintf’ output between 7 and 48 bytes into a destination of size 32
snprintf(conf->cache_name[0], namelen,
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
"raid%d-%s", conf->level, mdname(conf->mddev));
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
Increase the array size to avoid this warning.
Signed-off-by: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241112161019.4154616-2-john.g.garry@oracle.com
Signed-off-by: Song Liu <song@kernel.org>
When the VWC of a namespace does not exist, the BLK_FEAT_WRITE_CACHE
flag should not be set when registering the block device, regardless
of whether the controller supports VWC.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Rotational devices, such as hard-drives, can be detected using
the rotational bit in the namespace independent identify namespace
data structure. Make the bit visible to the block layer through the
rotational queue setting.
Signed-off-by: Wang Yugui <wangyugui@e16-tech.com>
Reviewed-by: Matias Bjørling <matias.bjorling@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The NVMe 2.0 specification adds an independent identify namespace
data structure that contains generic attributes that apply to all
namespace types. Some attributes carry over from the NVM command set
identify namespace data structure, and others are new.
Currently, the data structure only considered when CRIMS is enabled or
when the namespace type is key-value.
However, the independent namespace data structure is mandatory for
devices that implement features from the 2.0+ specification. Therefore,
we can check this data structure first. If unavailable, retrieve the
generic attributes from the NVM command set identify namespace data
structure.
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Implements reporting the I/O Command Set Independent Identify Namespace
command.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Most of the information is stubbed. Supporting these commands is a
requirement for supporting rotational media.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Most of the returned information is just stubbed data. The target must
support these in order to report rotational media. Since this driver
doesn't know any better, each namespace is its own endurance group with
the engid value matching the nsid.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The target driver implements all the mandatory logs, identifications,
features, and properties up to nvme sepcification 2.1.
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This property is required for nvme 2.1. The target only supports ready
with media, so this is just the same value as CAP.TO.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This is required for nvme 2.1 for targets that support multiple command
sets. We support NVM and ZNS, so are required to support this
identification.
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
We don't report anything here, but it's a mandatory identification for
nvme 2.1.
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Matias Bjørling <matias.bjorling@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This patch implements the reservation feature, including:
1. reservation register(register, unregister and replace).
2. reservation acquire(acquire, preempt, preempt and abort).
3. reservation release(release and clear).
4. reservation report.
5. set feature and get feature of reservation notify mask.
6. get log page of reservation event.
Not supported:
1. persistent reservation through power loss.
Test cases:
Use nvme-cli and fio to test all implemented sub features:
1. use nvme resv-register to register host a registrant or
unregister or replace a new key.
2. use nvme resv-acquire to set host to the holder, and use fio
to send read and write io in all reservation type. And also
test preempt and "preempt and abort".
3. use nvme resv-report to show all registrants and reservation
status.
4. use nvme resv-release to release all registrants.
5. use nvme get-log to get events generated by the preceding
operations.
In addition, make reservation configurable, one can set ns to
support reservation before enable ns. The default of resv_enable
is false.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Dmitry Bogdanov <d.bogdanov@yadro.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
max_zone_append_sectors differs from all other queue limits in that the
final value used is not stored in the queue_limits but needs to be
obtained using queue_limits_max_zone_append_sectors helper. This not
only adds (tiny) extra overhead to the I/O path, but also can be easily
forgotten in file system code.
Add a new max_hw_zone_append_sectors value to queue_limits which is
set by the driver, and calculate max_zone_append_sectors from that and
the other inputs in blk_validate_zoned_limits, similar to how
max_sectors is calculated to fix this.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20241104073955.112324-3-hch@lst.de
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20241108154657.845768-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For zoned devices, write zeroes must be split at the zone boundary
which is represented as chunk_sectors. For other uses like the
internally RAIDed NVMe devices it is probably at least useful.
Enhance get_max_io_size to know about write zeroes and use it in
bio_split_write_zeroes. Also add a comment about the seemingly
nonsensical zero max_write_zeroes limit.
Fixes: 885fa13f65 ("block: implement splitting of REQ_OP_WRITE_ZEROES bios")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Link: https://lore.kernel.org/r/20241104062647.91160-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add proper bio_split() error handling. For any error, call
raid_end_bio_io() and return.
For the case of an in the write path, we need to undo the increment in
the rdev pending count and NULLify the r1_bio->bios[] pointers.
For read path failure, we need to undo rdev pending count increment from
the earlier read_balance() call.
Reviewed-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: John Garry <john.g.garry@oracle.com>
Link: https://lore.kernel.org/r/20241111112150.3756529-6-john.g.garry@oracle.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
