summaryrefslogtreecommitdiff
path: root/block/blk-sysfs.c
AgeCommit message (Collapse)Author
2020-10-09blk-mq: move cancel of hctx->run_work to the front of blk_exit_queueYang Yang
blk_exit_queue will free elevator_data, while blk_mq_run_work_fn will access it. Move cancel of hctx->run_work to the front of blk_exit_queue to avoid use-after-free. Fixes: 1b97871b501f ("blk-mq: move cancel of hctx->run_work into blk_mq_hw_sysfs_release") Signed-off-by: Yang Yang <yang.yang@vivo.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-10-09block: get rid of unnecessary local variableYufen Yu
Since whole elevator register is protectd by sysfs_lock, we don't need extras 'has_elevator'. Just use q->elevator directly. Signed-off-by: Yufen Yu <yuyufen@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-10-09block: invoke blk_mq_exit_sched no matter whether have .exit_schedYufen Yu
We will register debugfs for scheduler no matter whether it have defined callback funciton .exit_sched. So, blk_mq_exit_sched() is always needed to unregister debugfs. Also, q->elevator should be set as NULL after exiting scheduler. For now, since all register scheduler have defined .exit_sched, it will not cause any actual problem. But It will be more reasonable to do this change. Signed-off-by: Yufen Yu <yuyufen@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-09-24bdi: replace BDI_CAP_STABLE_WRITES with a queue and a sb flagChristoph Hellwig
The BDI_CAP_STABLE_WRITES is one of the few bits of information in the backing_dev_info shared between the block drivers and the writeback code. To help untangling the dependency replace it with a queue flag and a superblock flag derived from it. This also helps with the case of e.g. a file system requiring stable writes due to its own checksumming, but not forcing it on other users of the block device like the swap code. One downside is that we an't support the stable_pages_required bdi attribute in sysfs anymore. It is replaced with a queue attribute which also is writable for easier testing. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-09-24block: lift setting the readahead size into the block layerChristoph Hellwig
Drivers shouldn't really mess with the readahead size, as that is a VM concept. Instead set it based on the optimal I/O size by lifting the algorithm from the md driver when registering the disk. Also set bdi->io_pages there as well by applying the same scheme based on max_sectors. To ensure the limits work well for stacking drivers a new helper is added to update the readahead limits from the block limits, which is also called from disk_stack_limits. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Jan Kara <jack@suse.cz> Reviewed-by: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Acked-by: Coly Li <colyli@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-09-08block: make QUEUE_SYSFS_BIT_FNS more usefulChristoph Hellwig
Switch to the naming used by the other entries so that we can use the QUEUE_RW_ENTRY helper. Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-09-08block: add helper macros for queue sysfs entriesChristoph Hellwig
Add two helpers macros to avoid boilerplate code for the queue sysfs entries. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-07-15block: add max_active_zones to blk-sysfsNiklas Cassel
Add a new max_active zones definition in the sysfs documentation. This definition will be common for all devices utilizing the zoned block device support in the kernel. Export max_active_zones according to this new definition for NVMe Zoned Namespace devices, ZAC ATA devices (which are treated as SCSI devices by the kernel), and ZBC SCSI devices. Add the new max_active_zones member to struct request_queue, rather than as a queue limit, since this property cannot be split across stacking drivers. For SCSI devices, even though max active zones is not part of the ZBC/ZAC spec, export max_active_zones as 0, signifying "no limit". Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com> Reviewed-by: Javier González <javier@javigon.com> Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com> 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>
2020-07-15block: add max_open_zones to blk-sysfsNiklas Cassel
Add a new max_open_zones definition in the sysfs documentation. This definition will be common for all devices utilizing the zoned block device support in the kernel. Export max open zones according to this new definition for NVMe Zoned Namespace devices, ZAC ATA devices (which are treated as SCSI devices by the kernel), and ZBC SCSI devices. Add the new max_open_zones member to struct request_queue, rather than as a queue limit, since this property cannot be split across stacking drivers. Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com> Reviewed-by: Javier González <javier@javigon.com> Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com> 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>
2020-06-24block: create the request_queue debugfs_dir on registrationLuis Chamberlain
We were only creating the request_queue debugfs_dir only for make_request block drivers (multiqueue), but never for request-based block drivers. We did this as we were only creating non-blktrace additional debugfs files on that directory for make_request drivers. However, since blktrace *always* creates that directory anyway, we special-case the use of that directory on blktrace. Other than this being an eye-sore, this exposes request-based block drivers to the same debugfs fragile race that used to exist with make_request block drivers where if we start adding files onto that directory we can later run a race with a double removal of dentries on the directory if we don't deal with this carefully on blktrace. Instead, just simplify things by always creating the request_queue debugfs_dir on request_queue registration. Rename the mutex also to reflect the fact that this is used outside of the blktrace context. Signed-off-by: Luis Chamberlain <mcgrof@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-06-24block: revert back to synchronous request_queue removalLuis Chamberlain
Commit dc9edc44de6c ("block: Fix a blk_exit_rl() regression") merged on v4.12 moved the work behind blk_release_queue() into a workqueue after a splat floated around which indicated some work on blk_release_queue() could sleep in blk_exit_rl(). This splat would be possible when a driver called blk_put_queue() or blk_cleanup_queue() (which calls blk_put_queue() as its final call) from an atomic context. blk_put_queue() decrements the refcount for the request_queue kobject, and upon reaching 0 blk_release_queue() is called. Although blk_exit_rl() is now removed through commit db6d99523560 ("block: remove request_list code") on v5.0, we reserve the right to be able to sleep within blk_release_queue() context. The last reference for the request_queue must not be called from atomic context. *When* the last reference to the request_queue reaches 0 varies, and so let's take the opportunity to document when that is expected to happen and also document the context of the related calls as best as possible so we can avoid future issues, and with the hopes that the synchronous request_queue removal sticks. We revert back to synchronous request_queue removal because asynchronous removal creates a regression with expected userspace interaction with several drivers. An example is when removing the loopback driver, one uses ioctls from userspace to do so, but upon return and if successful, one expects the device to be removed. Likewise if one races to add another device the new one may not be added as it is still being removed. This was expected behavior before and it now fails as the device is still present and busy still. Moving to asynchronous request_queue removal could have broken many scripts which relied on the removal to have been completed if there was no error. Document this expectation as well so that this doesn't regress userspace again. Using asynchronous request_queue removal however has helped us find other bugs. In the future we can test what could break with this arrangement by enabling CONFIG_DEBUG_KOBJECT_RELEASE. While at it, update the docs with the context expectations for the request_queue / gendisk refcount decrement, and make these expectations explicit by using might_sleep(). Fixes: dc9edc44de6c ("block: Fix a blk_exit_rl() regression") Suggested-by: Nicolai Stange <nstange@suse.de> Signed-off-by: Luis Chamberlain <mcgrof@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Cc: Bart Van Assche <bvanassche@acm.org> Cc: Omar Sandoval <osandov@fb.com> Cc: Hannes Reinecke <hare@suse.com> Cc: Nicolai Stange <nstange@suse.de> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: yu kuai <yukuai3@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-05-12block: Introduce REQ_OP_ZONE_APPENDKeith Busch
Define REQ_OP_ZONE_APPEND to append-write sectors to a zone of a zoned block device. This is a no-merge write operation. A zone append write BIO must: * Target a zoned block device * Have a sector position indicating the start sector of the target zone * The target zone must be a sequential write zone * The BIO must not cross a zone boundary * The BIO size must not be split to ensure that a single range of LBAs is written with a single command. Implement these checks in generic_make_request_checks() using the helper function blk_check_zone_append(). To avoid write append BIO splitting, introduce the new max_zone_append_sectors queue limit attribute and ensure that a BIO size is always lower than this limit. Export this new limit through sysfs and check these limits in bio_full(). Also when a LLDD can't dispatch a request to a specific zone, it will return BLK_STS_ZONE_RESOURCE indicating this request needs to be delayed, e.g. because the zone it will be dispatched to is still write-locked. If this happens set the request aside in a local list to continue trying dispatching requests such as READ requests or a WRITE/ZONE_APPEND requests targetting other zones. This way we can still keep a high queue depth without starving other requests even if one request can't be served due to zone write-locking. Finally, make sure that the bio sector position indicates the actual write position as indicated by the device on completion. Signed-off-by: Keith Busch <kbusch@kernel.org> [ jth: added zone-append specific add_page and merge_page helpers ] Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-10-07block: Remove "dying" checks from sysfs callbacksBart Van Assche
Block drivers must call del_gendisk() before blk_cleanup_queue(). del_gendisk() calls kobject_del() and kobject_del() waits until any ongoing sysfs callback functions have finished. In other words, the sysfs callback functions won't be called for a queue in the dying state. Hence remove the "dying" checks from the sysfs callback functions. Cc: Christoph Hellwig <hch@infradead.org> Cc: Ming Lei <ming.lei@redhat.com> Cc: Hannes Reinecke <hare@suse.com> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-09-27rq-qos: get rid of redundant wbt_update_limits()Yufen Yu
We have updated limits after calling wbt_set_min_lat(). No need to update again. Reviewed-by: Bob Liu <bob.liu@oracle.com> Signed-off-by: Yufen Yu <yuyufen@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-09-26block: don't release queue's sysfs lock during switching elevatorMing Lei
cecf5d87ff20 ("block: split .sysfs_lock into two locks") starts to release & acquire sysfs_lock before registering/un-registering elevator queue during switching elevator for avoiding potential deadlock from showing & storing 'queue/iosched' attributes and removing elevator's kobject. Turns out there isn't such deadlock because 'q->sysfs_lock' isn't required in .show & .store of queue/iosched's attributes, and just elevator's sysfs lock is acquired in elv_iosched_store() and elv_iosched_show(). So it is safe to hold queue's sysfs lock when registering/un-registering elevator queue. The biggest issue is that commit cecf5d87ff20 assumes that concurrent write on 'queue/scheduler' can't happen. However, this assumption isn't true, because kernfs_fop_write() only guarantees that concurrent write aren't called on the same open file, but the write could be from different open on the file. So we can't release & re-acquire queue's sysfs lock during switching elevator, otherwise use-after-free on elevator could be triggered. Fixes the issue by not releasing queue's sysfs lock during switching elevator. Fixes: cecf5d87ff20 ("block: split .sysfs_lock into two locks") Cc: Christoph Hellwig <hch@infradead.org> Cc: Hannes Reinecke <hare@suse.com> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-09-17Merge tag 'for-5.4/block-2019-09-16' of git://git.kernel.dk/linux-blockLinus Torvalds
Pull block updates from Jens Axboe: - Two NVMe pull requests: - ana log parse fix from Anton - nvme quirks support for Apple devices from Ben - fix missing bio completion tracing for multipath stack devices from Hannes and Mikhail - IP TOS settings for nvme rdma and tcp transports from Israel - rq_dma_dir cleanups from Israel - tracing for Get LBA Status command from Minwoo - Some nvme-tcp cleanups from Minwoo, Potnuri and Myself - Some consolidation between the fabrics transports for handling the CAP register - reset race with ns scanning fix for fabrics (move fabrics commands to a dedicated request queue with a different lifetime from the admin request queue)." - controller reset and namespace scan races fixes - nvme discovery log change uevent support - naming improvements from Keith - multiple discovery controllers reject fix from James - some regular cleanups from various people - Series fixing (and re-fixing) null_blk debug printing and nr_devices checks (André) - A few pull requests from Song, with fixes from Andy, Guoqing, Guilherme, Neil, Nigel, and Yufen. - REQ_OP_ZONE_RESET_ALL support (Chaitanya) - Bio merge handling unification (Christoph) - Pick default elevator correctly for devices with special needs (Damien) - Block stats fixes (Hou) - Timeout and support devices nbd fixes (Mike) - Series fixing races around elevator switching and device add/remove (Ming) - sed-opal cleanups (Revanth) - Per device weight support for BFQ (Fam) - Support for blk-iocost, a new model that can properly account cost of IO workloads. (Tejun) - blk-cgroup writeback fixes (Tejun) - paride queue init fixes (zhengbin) - blk_set_runtime_active() cleanup (Stanley) - Block segment mapping optimizations (Bart) - lightnvm fixes (Hans/Minwoo/YueHaibing) - Various little fixes and cleanups * tag 'for-5.4/block-2019-09-16' of git://git.kernel.dk/linux-block: (186 commits) null_blk: format pr_* logs with pr_fmt null_blk: match the type of parameter nr_devices null_blk: do not fail the module load with zero devices block: also check RQF_STATS in blk_mq_need_time_stamp() block: make rq sector size accessible for block stats bfq: Fix bfq linkage error raid5: use bio_end_sector in r5_next_bio raid5: remove STRIPE_OPS_REQ_PENDING md: add feature flag MD_FEATURE_RAID0_LAYOUT md/raid0: avoid RAID0 data corruption due to layout confusion. raid5: don't set STRIPE_HANDLE to stripe which is in batch list raid5: don't increment read_errors on EILSEQ return nvmet: fix a wrong error status returned in error log page nvme: send discovery log page change events to userspace nvme: add uevent variables for controller devices nvme: enable aen regardless of the presence of I/O queues nvme-fabrics: allow discovery subsystems accept a kato nvmet: Use PTR_ERR_OR_ZERO() in nvmet_init_discovery() nvme: Remove redundant assignment of cq vector nvme: Assign subsys instance from first ctrl ...
2019-09-12block: fix race between switching elevator and removing queuesMing Lei
cecf5d87ff20 ("block: split .sysfs_lock into two locks") starts to release & actuire sysfs_lock again during switching elevator. So it isn't enough to prevent switching elevator from happening by simply clearing QUEUE_FLAG_REGISTERED with holding sysfs_lock, because in-progress switch still can move on after re-acquiring the lock, meantime the flag of QUEUE_FLAG_REGISTERED won't get checked. Fixes this issue by checking 'q->elevator' directly & locklessly after q->kobj is removed in blk_unregister_queue(), this way is safe because q->elevator can't be changed at that time. Fixes: cecf5d87ff20 ("block: split .sysfs_lock into two locks") Cc: Christoph Hellwig <hch@infradead.org> Cc: Hannes Reinecke <hare@suse.com> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Mike Snitzer <snitzer@redhat.com> Cc: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-08-27block: split .sysfs_lock into two locksMing Lei
The kernfs built-in lock of 'kn->count' is held in sysfs .show/.store path. Meantime, inside block's .show/.store callback, q->sysfs_lock is required. However, when mq & iosched kobjects are removed via blk_mq_unregister_dev() & elv_unregister_queue(), q->sysfs_lock is held too. This way causes AB-BA lock because the kernfs built-in lock of 'kn-count' is required inside kobject_del() too, see the lockdep warning[1]. On the other hand, it isn't necessary to acquire q->sysfs_lock for both blk_mq_unregister_dev() & elv_unregister_queue() because clearing REGISTERED flag prevents storing to 'queue/scheduler' from being happened. Also sysfs write(store) is exclusive, so no necessary to hold the lock for elv_unregister_queue() when it is called in switching elevator path. So split .sysfs_lock into two: one is still named as .sysfs_lock for covering sync .store, the other one is named as .sysfs_dir_lock for covering kobjects and related status change. sysfs itself can handle the race between add/remove kobjects and showing/storing attributes under kobjects. For switching scheduler via storing to 'queue/scheduler', we use the queue flag of QUEUE_FLAG_REGISTERED with .sysfs_lock for avoiding the race, then we can avoid to hold .sysfs_lock during removing/adding kobjects. [1] lockdep warning ====================================================== WARNING: possible circular locking dependency detected 5.3.0-rc3-00044-g73277fc75ea0 #1380 Not tainted ------------------------------------------------------ rmmod/777 is trying to acquire lock: 00000000ac50e981 (kn->count#202){++++}, at: kernfs_remove_by_name_ns+0x59/0x72 but task is already holding lock: 00000000fb16ae21 (&q->sysfs_lock){+.+.}, at: blk_unregister_queue+0x78/0x10b which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #1 (&q->sysfs_lock){+.+.}: __lock_acquire+0x95f/0xa2f lock_acquire+0x1b4/0x1e8 __mutex_lock+0x14a/0xa9b blk_mq_hw_sysfs_show+0x63/0xb6 sysfs_kf_seq_show+0x11f/0x196 seq_read+0x2cd/0x5f2 vfs_read+0xc7/0x18c ksys_read+0xc4/0x13e do_syscall_64+0xa7/0x295 entry_SYSCALL_64_after_hwframe+0x49/0xbe -> #0 (kn->count#202){++++}: check_prev_add+0x5d2/0xc45 validate_chain+0xed3/0xf94 __lock_acquire+0x95f/0xa2f lock_acquire+0x1b4/0x1e8 __kernfs_remove+0x237/0x40b kernfs_remove_by_name_ns+0x59/0x72 remove_files+0x61/0x96 sysfs_remove_group+0x81/0xa4 sysfs_remove_groups+0x3b/0x44 kobject_del+0x44/0x94 blk_mq_unregister_dev+0x83/0xdd blk_unregister_queue+0xa0/0x10b del_gendisk+0x259/0x3fa null_del_dev+0x8b/0x1c3 [null_blk] null_exit+0x5c/0x95 [null_blk] __se_sys_delete_module+0x204/0x337 do_syscall_64+0xa7/0x295 entry_SYSCALL_64_after_hwframe+0x49/0xbe other info that might help us debug this: Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&q->sysfs_lock); lock(kn->count#202); lock(&q->sysfs_lock); lock(kn->count#202); *** DEADLOCK *** 2 locks held by rmmod/777: #0: 00000000e69bd9de (&lock){+.+.}, at: null_exit+0x2e/0x95 [null_blk] #1: 00000000fb16ae21 (&q->sysfs_lock){+.+.}, at: blk_unregister_queue+0x78/0x10b stack backtrace: CPU: 0 PID: 777 Comm: rmmod Not tainted 5.3.0-rc3-00044-g73277fc75ea0 #1380 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS ?-20180724_192412-buildhw-07.phx4 Call Trace: dump_stack+0x9a/0xe6 check_noncircular+0x207/0x251 ? print_circular_bug+0x32a/0x32a ? find_usage_backwards+0x84/0xb0 check_prev_add+0x5d2/0xc45 validate_chain+0xed3/0xf94 ? check_prev_add+0xc45/0xc45 ? mark_lock+0x11b/0x804 ? check_usage_forwards+0x1ca/0x1ca __lock_acquire+0x95f/0xa2f lock_acquire+0x1b4/0x1e8 ? kernfs_remove_by_name_ns+0x59/0x72 __kernfs_remove+0x237/0x40b ? kernfs_remove_by_name_ns+0x59/0x72 ? kernfs_next_descendant_post+0x7d/0x7d ? strlen+0x10/0x23 ? strcmp+0x22/0x44 kernfs_remove_by_name_ns+0x59/0x72 remove_files+0x61/0x96 sysfs_remove_group+0x81/0xa4 sysfs_remove_groups+0x3b/0x44 kobject_del+0x44/0x94 blk_mq_unregister_dev+0x83/0xdd blk_unregister_queue+0xa0/0x10b del_gendisk+0x259/0x3fa ? disk_events_poll_msecs_store+0x12b/0x12b ? check_flags+0x1ea/0x204 ? mark_held_locks+0x1f/0x7a null_del_dev+0x8b/0x1c3 [null_blk] null_exit+0x5c/0x95 [null_blk] __se_sys_delete_module+0x204/0x337 ? free_module+0x39f/0x39f ? blkcg_maybe_throttle_current+0x8a/0x718 ? rwlock_bug+0x62/0x62 ? __blkcg_punt_bio_submit+0xd0/0xd0 ? trace_hardirqs_on_thunk+0x1a/0x20 ? mark_held_locks+0x1f/0x7a ? do_syscall_64+0x4c/0x295 do_syscall_64+0xa7/0x295 entry_SYSCALL_64_after_hwframe+0x49/0xbe RIP: 0033:0x7fb696cdbe6b Code: 73 01 c3 48 8b 0d 1d 20 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 008 RSP: 002b:00007ffec9588788 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 RAX: ffffffffffffffda RBX: 0000559e589137c0 RCX: 00007fb696cdbe6b RDX: 000000000000000a RSI: 0000000000000800 RDI: 0000559e58913828 RBP: 0000000000000000 R08: 00007ffec9587701 R09: 0000000000000000 R10: 00007fb696d4eae0 R11: 0000000000000206 R12: 00007ffec95889b0 R13: 00007ffec95896b3 R14: 0000559e58913260 R15: 0000559e589137c0 Cc: Christoph Hellwig <hch@infradead.org> Cc: Hannes Reinecke <hare@suse.com> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-08-27block: add helper for checking if queue is registeredMing Lei
There are 4 users which check if queue is registered, so add one helper to check it. Cc: Christoph Hellwig <hch@infradead.org> Cc: Hannes Reinecke <hare@suse.com> Cc: Greg KH <gregkh@linuxfoundation.org> Cc: Mike Snitzer <snitzer@redhat.com> Cc: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-08-12blk-mq: move cancel of requeue_work to the front of blk_exit_queuezhengbin
blk_exit_queue will free elevator_data, while blk_mq_requeue_work will access it. Move cancel of requeue_work to the front of blk_exit_queue to avoid use-after-free. blk_exit_queue blk_mq_requeue_work __elevator_exit blk_mq_run_hw_queues blk_mq_exit_sched blk_mq_run_hw_queue dd_exit_queue blk_mq_hctx_has_pending kfree(elevator_data) blk_mq_sched_has_work dd_has_work Fixes: fbc2a15e3433 ("blk-mq: move cancel of requeue_work into blk_mq_release") Cc: stable@vger.kernel.org Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: zhengbin <zhengbin13@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-06-06block: free sched's request pool in blk_cleanup_queueMing Lei
In theory, IO scheduler belongs to request queue, and the request pool of sched tags belongs to the request queue too. However, the current tags allocation interfaces are re-used for both driver tags and sched tags, and driver tags is definitely host wide, and doesn't belong to any request queue, same with its request pool. So we need tagset instance for freeing request of sched tags. Meantime, blk_mq_free_tag_set() often follows blk_cleanup_queue() in case of non-BLK_MQ_F_TAG_SHARED, this way requires that request pool of sched tags to be freed before calling blk_mq_free_tag_set(). Commit 47cdee29ef9d94e ("block: move blk_exit_queue into __blk_release_queue") moves blk_exit_queue into __blk_release_queue for simplying the fast path in generic_make_request(), then causes oops during freeing requests of sched tags in __blk_release_queue(). Fix the above issue by move freeing request pool of sched tags into blk_cleanup_queue(), this way is safe becasue queue has been frozen and no any in-queue requests at that time. Freeing sched tags has to be kept in queue's release handler becasue there might be un-completed dispatch activity which might refer to sched tags. Cc: Bart Van Assche <bvanassche@acm.org> Cc: Christoph Hellwig <hch@lst.de> Fixes: 47cdee29ef9d94e485eb08f962c74943023a5271 ("block: move blk_exit_queue into __blk_release_queue") Tested-by: Yi Zhang <yi.zhang@redhat.com> Reported-by: kernel test robot <rong.a.chen@intel.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-05-29block: move blk_exit_queue into __blk_release_queueMing Lei
Commit 498f6650aec8 ("block: Fix a race between the cgroup code and request queue initialization") moves what blk_exit_queue does into blk_cleanup_queue() for fixing issue caused by changing back queue lock. However, after legacy request IO path is killed, driver queue lock won't be used at all, and there isn't story for changing back queue lock. Then the issue addressed by Commit 498f6650aec8 doesn't exist any more. So move move blk_exit_queue into __blk_release_queue. This patch basically reverts the following two commits: 498f6650aec8 block: Fix a race between the cgroup code and request queue initialization 24ecc3585348 block: Ensure that a request queue is dissociated from the cgroup controller Cc: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-04-22block: don't show io_timeout if driver has no timeout handlerWeiping Zhang
If the low level driver has no timeout handler, the /sys/block/<disk>/queue/io_timeout will not be displayed. Reviewed-by: Bart Van Assche <bvanassche@acm.org> Signed-off-by: Weiping Zhang <zhangweiping@didiglobal.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-03-20block: add BLK_MQ_POLL_CLASSIC for hybrid poll and return EINVAL for ↵Yufen Yu
unexpected value For q->poll_nsec == -1, means doing classic poll, not hybrid poll. We introduce a new flag BLK_MQ_POLL_CLASSIC to replace -1, which may make code much easier to read. Additionally, since val is an int obtained with kstrtoint(), val can be a negative value other than -1, so return -EINVAL for that case. Thanks to Damien Le Moal for some good suggestion. Reviewed-by: Damien Le Moal <damien.lemoal@wdc.com> Signed-off-by: Yufen Yu <yuyufen@huawei.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-02-11block: avoid setting wbt_lat_usec to current valueAleksei Zakharov
There's no reason to set wbt min lat and freeze request queue if current value is the same. Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Aleksei Zakharov <zakharov.a.g@yandex.ru> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2019-02-10blk-sysfs: Rework documention of __blk_release_queueMarcos Paulo de Souza
The Notes section of the comment was removed, because now blk_release_queue can only be executed from blk_cleanup_queue (being called when the q->kobj reaches zero), and also blk_init_queue was removed in a1ce35fa4985. Signed-off-by: Marcos Paulo de Souza <marcos.souza.org@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-12-28Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsiLinus Torvalds
Pull SCSI updates from James Bottomley: "This is mostly update of the usual drivers: smarpqi, lpfc, qedi, megaraid_sas, libsas, zfcp, mpt3sas, hisi_sas. Additionally, we have a pile of annotation, unused variable and minor updates. The big API change is the updates for Christoph's DMA rework which include removing the DISABLE_CLUSTERING flag. And finally there are a couple of target tree updates" * tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (259 commits) scsi: isci: request: mark expected switch fall-through scsi: isci: remote_node_context: mark expected switch fall-throughs scsi: isci: remote_device: Mark expected switch fall-throughs scsi: isci: phy: Mark expected switch fall-through scsi: iscsi: Capture iscsi debug messages using tracepoints scsi: myrb: Mark expected switch fall-throughs scsi: megaraid: fix out-of-bound array accesses scsi: mpt3sas: mpt3sas_scsih: Mark expected switch fall-through scsi: fcoe: remove set but not used variable 'port' scsi: smartpqi: call pqi_free_interrupts() in pqi_shutdown() scsi: smartpqi: fix build warnings scsi: smartpqi: update driver version scsi: smartpqi: add ofa support scsi: smartpqi: increase fw status register read timeout scsi: smartpqi: bump driver version scsi: smartpqi: add smp_utils support scsi: smartpqi: correct lun reset issues scsi: smartpqi: correct volume status scsi: smartpqi: do not offline disks for transient did no connect conditions scsi: smartpqi: allow for larger raid maps ...
2018-12-18scsi: block: remove the cluster flagChristoph Hellwig
Now that the the SCSI layer replaced the use of the cluster flag with segment size limits and the DMA boundary we can remove the cluster flag from the block layer. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2018-12-17blk-mq: enable IO poll if .nr_queues of type poll > 0Ming Lei
The queue mapping of type poll only exists when set->map[HCTX_TYPE_POLL].nr_queues is bigger than zero, so enhance the constraint by checking .nr_queues of type poll before enabling IO poll. Otherwise IO race & timeout can be observed when running block/007. Cc: Jeff Moyer <jmoyer@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-12-04block: only allow polling if a poll queue_map existsChristoph Hellwig
This avoids having to have differnet mq_ops for different setups with or without poll queues. Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-28block: add io timeout to sysfsWeiping Zhang
Give a interface to adjust io timeout(ms) by device. Signed-off-by: Weiping Zhang <zhangweiping@didiglobal.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-16block: add queue_is_mq() helperJens Axboe
Various spots check for q->mq_ops being non-NULL, but provide a helper to do this instead. Where the ->mq_ops != NULL check is redundant, remove it. Since mq == rq-based now that legacy is gone, get rid of the queue_is_rq_based() and just use queue_is_mq() everywhere. Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-15block: remove the queue_lock indirectionChristoph Hellwig
With the legacy request path gone there is no good reason to keep queue_lock as a pointer, we can always use the embedded lock now. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Fixed floppy and blk-cgroup missing conversions and half done edits. Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-15block: use atomic bitops for ->queue_flagsChristoph Hellwig
->queue_flags is generally not set or cleared in the fast path, and also generally set or cleared one flag at a time. Make use of the normal atomic bitops for it so that we don't need to take the queue_lock, which is otherwise mostly unused in the core block layer now. Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-07block: remove dead elevator codeJens Axboe
This removes a bunch of core and elevator related code. On the core front, we remove anything related to queue running, draining, initialization, plugging, and congestions. We also kill anything related to request allocation, merging, retrieval, and completion. Remove any checking for single queue IO schedulers, as they no longer exist. This means we can also delete a bunch of code related to request issue, adding, completion, etc - and all the SQ related ops and helpers. Also kill the load_default_modules(), as all that did was provide for a way to load the default single queue elevator. Tested-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-11-07block: remove legacy rq taggingJens Axboe
It's now unused, kill it. Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Omar Sandoval <osandov@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-10-31block: call rq_qos_exit() after queue is frozenMing Lei
rq_qos_exit() removes the current q->rq_qos, this action has to be done after queue is frozen, otherwise the IO queue path may never be waken up, then IO hang is caused. So fixes this issue by moving rq_qos_exit() after queue is frozen. Cc: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-10-25block: Introduce blk_revalidate_disk_zones()Damien Le Moal
Drivers exposing zoned block devices have to initialize and maintain correctness (i.e. revalidate) of the device zone bitmaps attached to the device request queue (seq_zones_bitmap and seq_zones_wlock). To simplify coding this, introduce a generic helper function blk_revalidate_disk_zones() suitable for most (and likely all) cases. This new function always update the seq_zones_bitmap and seq_zones_wlock bitmaps as well as the queue nr_zones field when called for a disk using a request based queue. For a disk using a BIO based queue, only the number of zones is updated since these queues do not have schedulers and so do not need the zone bitmaps. With this change, the zone bitmap initialization code in sd_zbc.c can be replaced with a call to this function in sd_zbc_read_zones(), which is called from the disk revalidate block operation method. A call to blk_revalidate_disk_zones() is also added to the null_blk driver for devices created with the zoned mode enabled. Finally, to ensure that zoned devices created with dm-linear or dm-flakey expose the correct number of zones through sysfs, a call to blk_revalidate_disk_zones() is added to dm_table_set_restrictions(). The zone bitmaps allocated and initialized with blk_revalidate_disk_zones() are freed automatically from __blk_release_queue() using the block internal function blk_queue_free_zone_bitmaps(). Reviewed-by: Hannes Reinecke <hare@suse.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-10-25block: Expose queue nr_zones in sysfsDamien Le Moal
Expose through sysfs the nr_zones field of struct request_queue. Exposing this value helps in debugging disk issues as well as facilitating scripts based use of the disk (e.g. blktests). For zoned block devices, the nr_zones field indicates the total number of zones of the device calculated using the known disk capacity and zone size. This number of zones is always 0 for regular block devices. Since nr_zones is defined conditionally with CONFIG_BLK_DEV_ZONED, introduce the blk_queue_nr_zones() function to return the correct value for any device, regardless if CONFIG_BLK_DEV_ZONED is set. Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.com> Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-08-23blk-wbt: don't maintain inflight counts if disabledJens Axboe
A previous commit removed the ability to have per-rq flags. We used those flags to maintain inflight counts. Since we don't have those anymore, we have to always maintain inflight counts, even if wbt is disabled. This is clearly suboptimal. Add a queue quiesce around changing the wbt latency settings from sysfs to work around this. With that, we can reliably put the enabled check in our bio_to_wbt_flags(), since we know the WBT_TRACKED flag will be consistent for the lifetime of the request. Fixes: c1c80384c8f ("block: remove external dependency on wbt_flags") Reviewed-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-08-11blkcg: Make blkg_root_lookup() work for queues in bypass modeBart Van Assche
For legacy queues the only call of blkg_root_lookup() happens after bypass mode has been enabled. Since blkg_lookup() returns NULL for queues in bypass mode, modify the blkg_root_lookup() such that it no longer depends on bypass mode. Rename the function into blk_queue_root_blkg() as suggested by Tejun. Suggested-by: Tejun Heo <tj@kernel.org> Fixes: 6bad9b210a22 ("blkcg: Introduce blkg_root_lookup()") Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Cc: Tejun Heo <tj@kernel.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-08-09block: Ensure that a request queue is dissociated from the cgroup controllerBart Van Assche
Several block drivers call alloc_disk() followed by put_disk() if something fails before device_add_disk() is called without calling blk_cleanup_queue(). Make sure that also for this scenario a request queue is dissociated from the cgroup controller. This patch avoids that loading the parport_pc, paride and pf drivers triggers the following kernel crash: BUG: KASAN: null-ptr-deref in pi_init+0x42e/0x580 [paride] Read of size 4 at addr 0000000000000008 by task modprobe/744 Call Trace: dump_stack+0x9a/0xeb kasan_report+0x139/0x350 pi_init+0x42e/0x580 [paride] pf_init+0x2bb/0x1000 [pf] do_one_initcall+0x8e/0x405 do_init_module+0xd9/0x2f2 load_module+0x3ab4/0x4700 SYSC_finit_module+0x176/0x1a0 do_syscall_64+0xee/0x2b0 entry_SYSCALL_64_after_hwframe+0x42/0xb7 Reported-by: Alexandru Moise <00moses.alexander00@gmail.com> Fixes: a063057d7c73 ("block: Fix a race between request queue removal and the block cgroup controller") # v4.17 Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Tested-by: Alexandru Moise <00moses.alexander00@gmail.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Cc: Tejun Heo <tj@kernel.org> Cc: Christoph Hellwig <hch@lst.de> Cc: Ming Lei <ming.lei@redhat.com> Cc: Alexandru Moise <00moses.alexander00@gmail.com> Cc: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: <stable@vger.kernel.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-07-09blk-rq-qos: refactor out common elements of blk-wbtJosef Bacik
blkcg-qos is going to do essentially what wbt does, only on a cgroup basis. Break out the common code that will be shared between blkcg-qos and wbt into blk-rq-qos.* so they can both utilize the same infrastructure. Signed-off-by: Josef Bacik <jbacik@fb.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-05-30block: convert bounce, q->bio_split to bioset_init()/mempool_init()Kent Overstreet
Convert the core block functionality to embedded bio sets. Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-05-24block drivers/block: Use octal not symbolic permissionsJoe Perches
Convert the S_<FOO> symbolic permissions to their octal equivalents as using octal and not symbolic permissions is preferred by many as more readable. see: https://lkml.org/lkml/2016/8/2/1945 Done with automated conversion via: $ ./scripts/checkpatch.pl -f --types=SYMBOLIC_PERMS --fix-inplace <files...> Miscellanea: o Wrapped modified multi-line calls to a single line where appropriate o Realign modified multi-line calls to open parenthesis Signed-off-by: Joe Perches <joe@perches.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-05-14block: Add sysfs entry for fua supportKent Overstreet
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-03-08block: Introduce blk_queue_flag_{set,clear,test_and_{set,clear}}()Bart Van Assche
Introduce functions that modify the queue flags and that protect these modifications with the request queue lock. Except for moving one wake_up_all() call from inside to outside a critical section, this patch does not change any functionality. Cc: Christoph Hellwig <hch@lst.de> Cc: Hannes Reinecke <hare@suse.de> Cc: Ming Lei <ming.lei@redhat.com> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-02-28block: Fix a race between request queue removal and the block cgroup controllerBart Van Assche
Avoid that the following race can occur: blk_cleanup_queue() blkcg_print_blkgs() spin_lock_irq(lock) (1) spin_lock_irq(blkg->q->queue_lock) (2,5) q->queue_lock = &q->__queue_lock (3) spin_unlock_irq(lock) (4) spin_unlock_irq(blkg->q->queue_lock) (6) (1) take driver lock; (2) busy loop for driver lock; (3) override driver lock with internal lock; (4) unlock driver lock; (5) can take driver lock now; (6) but unlock internal lock. This change is safe because only the SCSI core and the NVME core keep a reference on a request queue after having called blk_cleanup_queue(). Neither driver accesses any of the removed data structures between its blk_cleanup_queue() and blk_put_queue() calls. Reported-by: Joseph Qi <joseph.qi@linux.alibaba.com> Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com> Cc: Jan Kara <jack@suse.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-18block: Protect less code with sysfs_lock in blk_{un,}register_queue()Bart Van Assche
The __blk_mq_register_dev(), blk_mq_unregister_dev(), elv_register_queue() and elv_unregister_queue() calls need to be protected with sysfs_lock but other code in these functions not. Hence protect only this code with sysfs_lock. This patch fixes a locking inversion issue in blk_unregister_queue() and also in an error path of blk_register_queue(): it is not allowed to hold sysfs_lock around the kobject_del(&q->kobj) call. Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-15block: allow gendisk's request_queue registration to be deferredMike Snitzer
Since I can remember DM has forced the block layer to allow the allocation and initialization of the request_queue to be distinct operations. Reason for this is block/genhd.c:add_disk() has requires that the request_queue (and associated bdi) be tied to the gendisk before add_disk() is called -- because add_disk() also deals with exposing the request_queue via blk_register_queue(). DM's dynamic creation of arbitrary device types (and associated request_queue types) requires the DM device's gendisk be available so that DM table loads can establish a master/slave relationship with subordinate devices that are referenced by loaded DM tables -- using bd_link_disk_holder(). But until these DM tables, and their associated subordinate devices, are known DM cannot know what type of request_queue it needs -- nor what its queue_limits should be. This chicken and egg scenario has created all manner of problems for DM and, at times, the block layer. Summary of changes: - Add device_add_disk_no_queue_reg() and add_disk_no_queue_reg() variant that drivers may use to add a disk without also calling blk_register_queue(). Driver must call blk_register_queue() once its request_queue is fully initialized. - Return early from blk_unregister_queue() if QUEUE_FLAG_REGISTERED is not set. It won't be set if driver used add_disk_no_queue_reg() but driver encounters an error and must del_gendisk() before calling blk_register_queue(). - Export blk_register_queue(). These changes allow DM to use add_disk_no_queue_reg() to anchor its gendisk as the "master" for master/slave relationships DM must establish with subordinate devices referenced in DM tables that get loaded. Once all "slave" devices for a DM device are known its request_queue can be properly initialized and then advertised via sysfs -- important improvement being that no request_queue resource initialization performed by blk_register_queue() is missed for DM devices anymore. Signed-off-by: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>