Modify xfs_mount_zones() to replace the call to blkdev_report_zones()
with blkdev_report_zones_cached() to speed-up mount operations.
Since this causes xfs_zone_validate_seq() to see zones with the
BLK_ZONE_COND_ACTIVE condition, this function is also modified to acept
this condition as valid.
With this change, mounting a freshly formatted large capacity (30 TB)
SMR HDD completes under 2s compared to over 4.7s before.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Modify btrfs_get_dev_zones() and btrfs_sb_log_location_bdev() to replace
the call to blkdev_report_zones() with blkdev_report_zones_cached() to
speed-up mount operations. btrfs_get_dev_zone_info() is also modified to
take into account the BLK_ZONE_COND_ACTIVE condition, which is
equivalent to either BLK_ZONE_COND_IMP_OPEN, BLK_ZONE_COND_EXP_OPEN or
BLK_ZONE_COND_CLOSED.
With this change, mounting a freshly formatted large capacity (30 TB)
SMR HDD completes under 100ms compared to over 1.8s before.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Acked-by: David Sterba <dsterba@suse.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Modify queue_zone_wplug_show() to include the condition of a zone write
plug to the zone_wplugs debugfs attribute of a zoned block device.
To improve readability and ease of use, rather than the zone condition
raw value, the zone condition name is given using blk_zone_cond_str().
Suggested-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make the output of the zone_wplugs debugfs attribute file more easily
readable by adding the name of the zone write plugs fields in the
output.
No functional changes.
Suggested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Introduce the new BLKREPORTZONESV2 ioctl command to allow user
applications access to the fast zone report implemented by
blkdev_report_zones_cached(). This new ioctl is defined as number 142
and is documented in include/uapi/linux/fs.h.
Unlike the existing BLKREPORTZONES ioctl, this new ioctl uses the flags
field of struct blk_zone_report also as an input. If the user sets the
BLK_ZONE_REP_CACHED flag as an input, then blkdev_report_zones_cached()
is used to generate the zone report using cached zone information. If
this flag is not set, then BLKREPORTZONESV2 behaves in the same manner
as BLKREPORTZONES and the zone report is generated by accessing the
zoned device.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Introduce the function blkdev_report_zones_cached() to provide a fast
report zone built using the blkdev_get_zone_info() function, which gets
zone information from a disk zones_cond array or zone write plugs.
For a large capacity SMR drive, such fast report zone can be completed
in a few milliseconds compared to several seconds completion times
when the report zone is obtained from the device.
The zone report is built in the same manner as with the regular
blkdev_report_zones() function, that is, the first zone reported is the
one containing the specified start sector and the report is limited to
the specified number of zones (nr_zones argument). The information for
each zone in the report is obtained using blkdev_get_zone_info().
For zoned devices that do not use zone write plug resources,
using blkdev_get_zone_info() is inefficient as the zone report would
be very slow, generated one zone at a time. To avoid this,
blkdev_report_zones_cached() falls back to calling
blkdev_do_report_zones() to execute a regular zone report. In this case,
the .report_active field of struct blk_report_zones_args is set to true
to report zone conditions using the BLK_ZONE_COND_ACTIVE condition in
place of the implicit open, explicit open and closed conditions.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Introduce the function blkdev_get_zone_info() to obtain a single zone
information from cached zone data, that is, either from the zone write
plug for the target zone if it exists and from the disk zones_cond
array otherwise.
Since sequential zones that do not have a zone write plug are either
full, empty or in a bad state (read-only or offline), the zone write
pointer can be inferred from the zone condition cached in the disk
zones_cond array. For sequential zones that have a zone write plug, the
zone condition and zone write pointer are obtained from the condition
and write pointer offset managed with the zone write plug. This allows
obtaining the information for a zone much more quickly than having to
execute a report zones command on the device.
blkdev_get_zone_info() falls back to using a regular zone report if the
target zone is flagged as needing an update with the
BLK_ZONE_WPLUG_NEED_WP_UPDATE flag, or if the target device does not
use zone write plugs (i.e. a device mapper device). In this case, the
new function blkdev_report_zone_fallback() is used and the zone
condition is reported consistantly with the cahced report, that is, the
BLK_ZONE_COND_ACTIVE condition is used in place of the implicit open,
explicit open and closed conditions. This is achieved by adding the
.report_active field to struct blk_report_zones_args and by having
disk_report_zone() sets the correct zone condition if .report_active is
true.
In preparation for using blkdev_get_zone_info() in upcoming file systems
changes, also export this function as a GPL symbol.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for implementing cached report zone, split the main part
of the code of blkdev_report_zones() into the helper function
blkdev_do_report_zones(), with this new helper taking as argument a
struct blk_report_zones_args pointer instead of a report callback
function and its private argument.
No functional changes.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The function blk_revalidate_zone_cond() already caches the condition of
all zones of a zoned block device in the zones_cond array of a gendisk.
However, the zone conditions are updated only when the device is scanned
or revalidated.
Implement tracking of the runtime changes to zone conditions using
the new cond field in struct blk_zone_wplug. The size of this structure
remains 112 Bytes as the new field replaces the 4 Bytes padding at the
end of the structure.
Beause zones that do not have a zone write plug can be in the empty,
implicit open, explicit open or full condition, the zones_cond array of
a disk is used to track the conditions, of zones that do not have a zone
write plug. The condition of such zone is updated in the disk zones_cond
array when a zone reset, reset all or finish operation is executed, and
also when a zone write plug is removed from the disk hash table when the
zone becomes full.
Since a device may automatically close an implicitly open zone when
writing to an empty or closed zone, if the total number of open zones
has reached the device limit, the BLK_ZONE_COND_IMP_OPEN and
BLK_ZONE_COND_CLOSED zone conditions cannot be precisely tracked. To
overcome this, the zone condition BLK_ZONE_COND_ACTIVE is introduced to
represent a zone that has the condition BLK_ZONE_COND_IMP_OPEN,
BLK_ZONE_COND_EXP_OPEN or BLK_ZONE_COND_CLOSED. This follows the
definition of an active zone as defined in the NVMe Zoned Namespace
specifications. As such, for a zoned device that has a limit on the
maximum number of open zones, we will never have more zones in the
BLK_ZONE_COND_ACTIVE condition than the device limit. This is compatible
with the SCSI ZBC and ATA ZAC specifications for SMR HDDs as these
devices do not have a limit on the number of active zones.
The function disk_zone_wplug_set_wp_offset() is modified to use the new
helper disk_zone_wplug_update_cond() to update a zone write plug
condition whenever a zone write plug write offset is updated on
submission or merging of write BIOs to a zone.
The functions blk_zone_reset_bio_endio(), blk_zone_reset_all_bio_endio()
and blk_zone_finish_bio_endio() are modified to update the condition of
the zones targeted by reset, reset_all and finish operations, either
using though disk_zone_wplug_set_wp_offset() for zones that have a
zone write plug, or using the disk_zone_set_cond() helper to update the
zones_cond array of the disk for zones that do not have a zone write
plug.
When a zone write plug is removed from the disk hash table (when the
zone becomes empty or full), the condition of struct blk_zone_wplug is
used to update the disk zones_cond array. Conversely, when a zone write
plug is added to the disk hash table, the zones_cond array is used to
initialize the zone write plug condition.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The conv_zones_bitmap field of struct gendisk is used to define a bitmap
to identify the conventional zones of a zoned block device. The bit for
a zone is set in this bitmap if the zone is a conventional one, that is,
if the zone type is BLK_ZONE_TYPE_CONVENTIONAL. For such zone, this
always corresponds to the zone condition BLK_ZONE_COND_NOT_WP.
In other words, conv_zones_bitmap tracks a single condition of the
zones of a zoned block device.
In preparation for tracking more zone conditions, change
conv_zones_bitmap into an array of zone conditions, using 1 byte per
zone. This increases the memory usage from 1 bit per zone to 1 byte per
zone, that is, from 16 KiB to about 100 KiB for a 30 TB SMR HDD with 256
MiB zones. This is a trade-off to allow fast cached report zones later
on top of this change.
Rename the conv_zones_bitmap field of struct gendisk to zones_cond. Add
a blk_revalidate_zone_cond() function to initialize the zones_cond array
of a disk during device scan and to update it on device revalidation.
Move the allocation of the zones_cond array to
disk_revalidate_zone_resources(), making sure that this array is always
allocated, even for devices that do not need zone write plugs (zone
resources), to ensure that bdev_zone_is_seq() can be re-implemented to
use the zone condition array in place of the conv zones bitmap.
Finally, the function bdev_zone_is_seq() is rewritten to use a test on
the condition of the target zone.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Reorganize the fields of struct blk_zone_wplug to remove a hole after
the wp_offset field and avoid having the bio_work structure split
between 2 cache lines.
No functional changes.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit b76b840fd9 ("dm: Fix dm-zoned-reclaim zone write pointer
alignment") introduced an indirect call for the callback function of a
report zones executed with blkdev_report_zones(). This is necessary so
that the function disk_zone_wplug_sync_wp_offset() can be called to
refresh a zone write plug zone write pointer offset after a write error.
However, this solution makes following the path of a zone information
harder to understand.
Clean this up by introducing the new blk_report_zones_args structure to
define a zone report callback and its private data and introduce the
helper function disk_report_zone() which calls both
disk_zone_wplug_sync_wp_offset() and the zone report user callback
function for all zones of a zone report. This helper function must be
called by all block device drivers that implement the report zones
block operation in order to correctly report a zone information.
All block device drivers supporting the report_zones block operation are
updated to use this new scheme.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Modify disk_update_zone_resources() to freeze the device queue before
updating the number of zones, zone capacity and other zone related
resources. The locking order resulting from the call to
queue_limits_commit_update_frozen() is preserved, that is, the queue
limits lock is first taken by calling queue_limits_start_update() before
freezing the queue, and the queue is unfrozen after executing
queue_limits_commit_update(), which replaces the call to
queue_limits_commit_update_frozen().
This change ensures that there are no in-flights I/Os when the zone
resources are updated due to a zone revalidation. In case of error when
the limits are applied, directly call disk_free_zone_resources() from
disk_update_zone_resources() while the disk queue is still frozen to
avoid needing to freeze & unfreeze the queue again in
blk_revalidate_disk_zones(), thus simplifying that function code a
little.
Fixes: 0b83c86b44 ("block: Prevent potential deadlock in blk_revalidate_disk_zones()")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The functions blk_zone_wplug_handle_reset_or_finish() and
blk_zone_wplug_handle_reset_all() both modify the zone write pointer
offset of zone write plugs that are the target of a reset, reset all or
finish zone management operation. However, these functions do this
modification before the BIO is executed. So if the zone operation fails,
the modified zone write pointer offsets become invalid.
Avoid this by modifying the zone write pointer offset of a zone write
plug that is the target of a zone management operation when the
operation completes. To do so, modify blk_zone_bio_endio() to call the
new function blk_zone_mgmt_bio_endio() which in turn calls the functions
blk_zone_reset_all_bio_endio(), blk_zone_reset_bio_endio() or
blk_zone_finish_bio_endio() depending on the operation of the completed
BIO, to modify a zone write plug write pointer offset accordingly.
These functions are called only if the BIO execution was successful.
Fixes: dd291d77cc ("block: Introduce zone write plugging")
Cc: stable@vger.kernel.org
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently the automatic block layer PI generation allocates the integrity
buffer using kmalloc, and thus could deadlock, or fail I/O request due
to memory pressure.
Fix this by adding a mempool, and capping the maximum I/O size on PI
capable devices to not exceed the allocation size of the mempool.
Link: https://lore.kernel.org/linux-block/20251103101653.2083310-1-hch@lst.de/
Signed-off-by: Jens Axboe <axboe@kernel.dk>
* autopi-deadlock:
block: make bio auto-integrity deadlock safe
block: blocking mempool_alloc doesn't fail
The current block layer automatic integrity protection allocates the
actual integrity buffer, which has three problems:
- because it happens at the bottom of the I/O stack and doesn't use a
mempool it can deadlock under load
- because the data size in a bio is almost unbounded when using lage
folios it can relatively easily exceed the maximum kmalloc size
- even when it does not exceed the maximum kmalloc size, it could
exceed the maximum segment size of the device
Fix this by limiting the I/O size so that we can allocate at least a
2MiB integrity buffer, i.e. 128MiB for 8 byte PI and 512 byte integrity
intervals, and create a mempool as a last resort for this maximum size,
mirroring the scheme used for bvecs. As a nice upside none of this
can fail now, so we remove the error handling and open code the
trivial addition of the bip vec.
The new allocation helpers sit outside of bio-integrity-auto.c because
I plan to reuse them for file system based PI in the near future.
Fixes: 7ba1ba12ee ("block: Block layer data integrity support")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Anuj Gupta <anuj20.g@samsung.com>
Reviewed-by: Kanchan Joshi <joshi.k@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Refactor ublk_thread to be a thread-local variable instead of storing
it in ublk_dev:
- Remove pthread_t thread field from struct ublk_thread and move it to
struct ublk_thread_info
- Remove struct ublk_thread array from struct ublk_dev, reducing memory
footprint
- Define struct ublk_thread as local variable in __ublk_io_handler_fn()
instead of accessing it from dev->threads[]
- Extract main IO handling logic into __ublk_io_handler_fn() which is
marked as noinline
- Move CPU affinity setup to ublk_io_handler_fn() before calling
__ublk_io_handler_fn()
- Update ublk_thread_set_sched_affinity() to take struct ublk_thread_info *
instead of struct ublk_thread *, and use pthread_setaffinity_np()
instead of sched_setaffinity()
- Reorder struct ublk_thread fields to group related state together
This change makes each thread's ublk_thread structure truly local to
the thread, improving cache locality and reducing memory usage.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move ublk_thread_set_sched_affinity() call before ublk_thread_init()
to ensure memory allocations during thread initialization occur on
the correct NUMA node. This leverages Linux's first-touch memory
policy for better NUMA locality.
Also convert ublk_thread_set_sched_affinity() to use
pthread_setaffinity_np() instead of sched_setaffinity(), as the
pthread API is the proper interface for setting thread affinity in
multithreaded programs.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Convert ublk_queue to use struct_size() for allocation.
Changes in this commit:
1. Update ublk_init_queue() to use struct_size(ubq, ios, depth)
instead of manual size calculation (sizeof(struct ublk_queue) +
depth * sizeof(struct ublk_io)).
This provides better type safety and makes the code more maintainable
by using standard kernel macro for flexible array handling.
Meantime annotate ublk_queue.ios by __counted_by().
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Implement NUMA-friendly memory allocation for ublk driver to improve
performance on multi-socket systems.
This commit includes the following changes:
1. Rename __queues to queues, dropping the __ prefix since the field is
now accessed directly throughout the codebase rather than only through
the ublk_get_queue() helper.
2. Remove the queue_size field from struct ublk_device as it is no longer
needed.
3. Move queue allocation and deallocation into ublk_init_queue() and
ublk_deinit_queue() respectively, improving encapsulation. This
simplifies ublk_init_queues() and ublk_deinit_queues() to just
iterate and call the per-queue functions.
4. Add ublk_get_queue_numa_node() helper function to determine the
appropriate NUMA node for a queue by finding the first CPU mapped
to that queue via tag_set.map[HCTX_TYPE_DEFAULT].mq_map[] and
converting it to a NUMA node using cpu_to_node(). This function is
called internally by ublk_init_queue() to determine the allocation
node.
5. Allocate each queue structure on its local NUMA node using
kvzalloc_node() in ublk_init_queue().
6. Allocate the I/O command buffer on the same NUMA node using
alloc_pages_node().
This reduces memory access latency on multi-socket NUMA systems by
ensuring each queue's data structures are local to the CPUs that
access them.
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Move ublk_add_tag_set() before ublk_init_queues() in the device
initialization path. This allows us to use the blk-mq CPU-to-queue
mapping established by the tag_set to determine the appropriate
NUMA node for each queue allocation.
The error handling paths are also reordered accordingly.
Reviewed-by: Caleb Sander Mateos <csander@purestorage.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use kmap_local_page() instead of kmap() to avoid
CPU contention.
kmap() uses a global set of mapping slots that can cause contention
between multiple CPUs, while kmap_local_page() uses per-CPU slots
eliminating this contention. It also ensures non-sleeping operation
and provides better cache locality.
Convert kmap() to kmap_local_page() as it aligns with ongoing
kernel efforts to modernize kmap() usage for better multi-core
scalability.
Signed-off-by: Shi Hao <i.shihao.999@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The ftrace blktrace path allocates buffers and writes trace events but
was using the wrong recording function. After
commit 4d8bc7bd4f ("blktrace: move ftrace blk_io_tracer to blk_io_trace2"),
the ftrace interface was moved to use blk_io_trace2 format, but
__blk_add_trace() still called record_blktrace_event() which writes in
blk_io_trace (v1) format.
This causes critical data corruption:
- blk_io_trace (v1) has 32-bit 'action' field at offset 28
- blk_io_trace2 (v2) has 32-bit 'pid' at offset 28 and 64-bit 'action'
at offset 32
- When record_blktrace_event() writes to a v2 buffer:
* Writing pid (offset 32 in v1) corrupts the v2 action field
* Writing action (offset 28 in v1) corrupts the v2 pid field
* The 64-bit action is truncated to 32-bit via lower_32_bits()
Fix by:
1. Adding version switch to select correct format (v1 vs v2)
2. Calling appropriate recording function based on version
3. Defaulting to v2 for ftrace (as intended by commit 4d8bc7bd4f)
4. Adding WARN_ONCE for unexpected version values
Without this patch :-
linux-block (for-next) # sh reproduce_blktrace_bug.sh
dd-14242 [033] d..1. 3903.022308: Unknown action 36a2
dd-14242 [033] d..1. 3903.022333: Unknown action 36a2
dd-14242 [033] d..1. 3903.022365: Unknown action 36a2
dd-14242 [033] d..1. 3903.022366: Unknown action 36a2
dd-14242 [033] d..1. 3903.022369: Unknown action 36a2
The action field is corrupted because:
- ftrace allocated blk_io_trace2 buffer (64 bytes)
- But called record_blktrace_event() (writes v1, 48 bytes)
- Field offsets don't match, causing corruption
The hex value shown 0x30e3 is actually a PID, not an action code!
linux-block (for-next) #
linux-block (for-next) #
linux-block (for-next) # sh reproduce_blktrace_bug.sh
Trace output looks correct:
dd-2420 [019] d..1. 59.641742: 251,0 Q RS 0 + 8 [dd]
dd-2420 [019] d..1. 59.641775: 251,0 G RS 0 + 8 [dd]
dd-2420 [019] d..1. 59.641784: 251,0 P N [dd]
dd-2420 [019] d..1. 59.641785: 251,0 U N [dd] 1
dd-2420 [019] d..1. 59.641788: 251,0 D RS 0 + 8 [dd]
Fixes: 4d8bc7bd4f ("blktrace: move ftrace blk_io_tracer to blk_io_trace2")
Signed-off-by: Chaitanya Kulkarni <ckulkarnilinux@gmail.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The WARN_ON_ONCE introduced in
commit f9ee38bbf7 ("blktrace: add block trace commands for zone operations")
triggers kernel warnings when zone operations are traced with blktrace
version 1. This can spam the kernel log during normal operation with
zoned block devices when userspace is using the legacy blktrace
protocol.
Currently blktrace implementation drops newly added REQ_OP_ZONE_XXX
when blktrace userspce version is set to 1.
Remove the WARN_ON_ONCE and quietly filter these events. Add a
rate-limited debug message to help diagnose potential issues without
flooding the kernel log. The debug message can be enabled via dynamic
debug when needed for troubleshooting.
This approach is more appropriate as encountering zone operations with
blktrace v1 is an expected condition that should be handled gracefully
rather than warned about, since users may be running older blktrace
userspace tools that only support version 1 of the protocol.
With this patch :-
linux-block (for-next) # git log -1
commit c8966006a0971d2b4bf94c0426eb7e4407c6853f (HEAD -> for-next)
Author: Chaitanya Kulkarni <ckulkarnilinux@gmail.com>
Date: Mon Oct 27 19:26:53 2025 -0700
blktrace: use debug print to report dropped events
linux-block (for-next) # cdblktests
blktests (master) # ./check blktrace
blktrace/001 (blktrace zone management command tracing) [passed]
runtime 3.805s ... 3.889s
blktests (master) # dmesg -c
blktests (master) # echo "file kernel/trace/blktrace.c +p" > /sys/kernel/debug/dynamic_debug/control
blktests (master) # ./check blktrace
blktrace/001 (blktrace zone management command tracing) [passed]
runtime 3.889s ... 3.881s
blktests (master) # dmesg -c
[ 77.826237] blktrace: blktrace v1 cannot trace zone operation 0x1000190001
[ 77.826260] blktrace: blktrace v1 cannot trace zone operation 0x1000190004
[ 77.826282] blktrace: blktrace v1 cannot trace zone operation 0x1001490007
[ 77.826288] blktrace: blktrace v1 cannot trace zone operation 0x1001890008
[ 77.826343] blktrace: blktrace v1 cannot trace zone operation 0x1000190001
[ 77.826347] blktrace: blktrace v1 cannot trace zone operation 0x1000190004
[ 77.826350] blktrace: blktrace v1 cannot trace zone operation 0x1001490007
[ 77.826354] blktrace: blktrace v1 cannot trace zone operation 0x1001890008
[ 77.826373] blktrace: blktrace v1 cannot trace zone operation 0x1000190001
[ 77.826377] blktrace: blktrace v1 cannot trace zone operation 0x1000190004
blktests (master) # echo "file kernel/trace/blktrace.c -p" > /sys/kernel/debug/dynamic_debug/control
blktests (master) # ./check blktrace
blktrace/001 (blktrace zone management command tracing) [passed]
runtime 3.881s ... 3.824s
blktests (master) # dmesg -c
blktests (master) #
Reported-by: syzbot+153e64c0aa875d7e4c37@syzkaller.appspotmail.com
Fixes: f9ee38bbf7 ("blktrace: add block trace commands for zone operations")
Signed-off-by: Chaitanya Kulkarni <ckulkarnilinux@gmail.com>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Trace zone write plugging operations on block devices.
As tracing of zoned block commands needs the upper 32bit of the widened
64bit action, only add traces to blktrace if user-space has requested
version 2 of the blktrace protocol.
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Expose ZONE APPEND completions as a block trace completion action to
blktrace.
As tracing of zoned block commands needs the upper 32bit of the widened
64bit action, only add traces to blktrace if user-space has requested
version 2 of the blktrace protocol.
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add block trace commands for zone operations. These commands can only be
handled with version 2 of the blktrace protocol. For version 1, warn if a
command that does not fit into the 16 bits reserved for the command in
this version is passed in.
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Differentiate between blk_io_trace and blk_io_trace2 when relaying to
user-space depending on which version has been requested by the blktrace
utility.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add definitions for the extended version of the blktrace protocol using a
wider action type to be able to record new actions in the kernel.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pass struct blk_user_trace_setup2 to blktrace_setup_finalize(). This
prepares for the incoming extension of the blktrace protocol with a 64bit
act_mask.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add definitions for a version 2 of the blk_user_trace_setup ioctl. This
new ioctl will enable a different struct layout of the binary data passed
to user-space when using a new version of the blktrace utility requesting
the new struct layout.
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split do_blk_trace_setup into two functions, this is done to prepare for
an incoming new BLKTRACESETUP2 ioctl(2) which can receive extended
parameters from user-space.
Also move the size verification logic to the callers in preparation for
using a new internal structure later.
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Change the internal use of the action in blktrace to 64bit. Although for
now only the lower 32bits will be used.
With the upcoming version 2 of the blktrace user-space protocol the upper
32bit will also be utilized.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Untangle the if/else sequence setting the trace action in
__blk_add_trace() and turn it into a switch statement for better
extensibility.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split out the code relaying a blktrace event to user-space using relayfs.
This enables adding a second version supporting a new version of the
protocol.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Factor out the recording of a blktrace event into its own function,
deduplicating the code.
This also enables recording different versions of the blktrace protocol
later on.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
De-duplicate the calculation of the trace length instead of doing the
calculation twice, once for calling trace_buffer_lock_reserve() and once
for calling relay_reserve().
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Despite its name, the block layer is fine with segments smaller that the
"min_segment_size" limit. The value is an optimization limit indicating
the largest segment that can be used without considering boundary
limits. Smaller segments can take a fast path, so give it a name that
reflects that: max_fast_segment_size.
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Change struct size calculation to use struct_size()
to align with new recommended practices[1] which quotes:
"Another common case to avoid is calculating the size of a structure with
a trailing array of others structures, as in:
header = kzalloc(sizeof(*header) + count * sizeof(*header->item),
GFP_KERNEL);
Instead, use the helper:
header = kzalloc(struct_size(header, item, count), GFP_KERNEL);"
Signed-off-by: Mehdi Ben Hadj Khelifa <mehdi.benhadjkhelifa@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit c807ab520f ("block/mq-deadline: Add I/O priority support")
modified the behavior of request flag BLK_MQ_INSERT_AT_HEAD from
dispatching a request before other requests into dispatching a request
before other requests with the same I/O priority. This is not correct since
BLK_MQ_INSERT_AT_HEAD is used when requeuing requests and also when a flush
request is inserted. Both types of requests should be dispatched as soon
as possible. Hence, make the mq-deadline I/O scheduler again ignore the I/O
priority for BLK_MQ_INSERT_AT_HEAD requests.
Cc: Damien Le Moal <dlemoal@kernel.org>
Cc: Yu Kuai <yukuai@kernel.org>
Reported-by: chengkaitao <chengkaitao@kylinos.cn>
Closes: https://lore.kernel.org/linux-block/20251009155253.14611-1-pilgrimtao@gmail.com/
Fixes: c807ab520f ("block/mq-deadline: Add I/O priority support")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Damien Le Moalv <dlemoal@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull scheduler fixes from Borislav Petkov:
- Make sure the check for lost pelt idle time is done unconditionally
to have correct lost idle time accounting
- Stop the deadline server task before a CPU goes offline
* tag 'sched_urgent_for_v6.18_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/fair: Fix pelt lost idle time detection
sched/deadline: Stop dl_server before CPU goes offline
Pull perf fixes from Borislav Petkov:
- Make sure perf reporting works correctly in setups using
overlayfs or FUSE
- Move the uprobe optimization to a better location logically
* tag 'perf_urgent_for_v6.18_rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/core: Fix MMAP2 event device with backing files
perf/core: Fix MMAP event path names with backing files
perf/core: Fix address filter match with backing files
uprobe: Move arch_uprobe_optimize right after handlers execution
