1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/blkdev.h>
6 #include <linux/sbitmap.h>
7 #include <linux/lockdep.h>
8 #include <linux/scatterlist.h>
9 #include <linux/prefetch.h>
10 #include <linux/srcu.h>
13 struct blk_flush_queue;
15 #define BLKDEV_MIN_RQ 4
16 #define BLKDEV_DEFAULT_RQ 128
23 typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
27 typedef __u32 __bitwise req_flags_t;
29 /* drive already may have started this one */
30 #define RQF_STARTED ((__force req_flags_t)(1 << 1))
31 /* request for flush sequence */
32 #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4))
33 /* merge of different types, fail separately */
34 #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5))
35 /* don't call prep for this one */
36 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
37 /* use hctx->sched_tags */
38 #define RQF_SCHED_TAGS ((__force req_flags_t)(1 << 8))
39 /* use an I/O scheduler for this request */
40 #define RQF_USE_SCHED ((__force req_flags_t)(1 << 9))
41 /* vaguely specified driver internal error. Ignored by the block layer */
42 #define RQF_FAILED ((__force req_flags_t)(1 << 10))
43 /* don't warn about errors */
44 #define RQF_QUIET ((__force req_flags_t)(1 << 11))
45 /* account into disk and partition IO statistics */
46 #define RQF_IO_STAT ((__force req_flags_t)(1 << 13))
47 /* runtime pm request */
48 #define RQF_PM ((__force req_flags_t)(1 << 15))
49 /* on IO scheduler merge hash */
50 #define RQF_HASHED ((__force req_flags_t)(1 << 16))
51 /* track IO completion time */
52 #define RQF_STATS ((__force req_flags_t)(1 << 17))
53 /* Look at ->special_vec for the actual data payload instead of the
55 #define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18))
56 /* The per-zone write lock is held for this request */
57 #define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19))
58 /* ->timeout has been called, don't expire again */
59 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
60 #define RQF_RESV ((__force req_flags_t)(1 << 23))
62 /* flags that prevent us from merging requests: */
63 #define RQF_NOMERGE_FLAGS \
64 (RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
73 * Try to put the fields that are referenced together in the same cacheline.
75 * If you modify this structure, make sure to update blk_rq_init() and
76 * especially blk_mq_rq_ctx_init() to take care of the added fields.
79 struct request_queue *q;
80 struct blk_mq_ctx *mq_ctx;
81 struct blk_mq_hw_ctx *mq_hctx;
83 blk_opf_t cmd_flags; /* op and common flags */
91 /* the following two fields are internal, NEVER access directly */
92 unsigned int __data_len; /* total data len */
93 sector_t __sector; /* sector cursor */
99 struct list_head queuelist;
100 struct request *rq_next;
103 struct block_device *part;
104 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
105 /* Time that the first bio started allocating this request. */
108 /* Time that this request was allocated for this IO. */
110 /* Time that I/O was submitted to the device. */
111 u64 io_start_time_ns;
113 #ifdef CONFIG_BLK_WBT
114 unsigned short wbt_flags;
117 * rq sectors used for blk stats. It has the same value
118 * with blk_rq_sectors(rq), except that it never be zeroed
121 unsigned short stats_sectors;
124 * Number of scatter-gather DMA addr+len pairs after
125 * physical address coalescing is performed.
127 unsigned short nr_phys_segments;
129 #ifdef CONFIG_BLK_DEV_INTEGRITY
130 unsigned short nr_integrity_segments;
133 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
134 struct bio_crypt_ctx *crypt_ctx;
135 struct blk_crypto_keyslot *crypt_keyslot;
138 unsigned short ioprio;
140 enum mq_rq_state state;
143 unsigned long deadline;
146 * The hash is used inside the scheduler, and killed once the
147 * request reaches the dispatch list. The ipi_list is only used
148 * to queue the request for softirq completion, which is long
149 * after the request has been unhashed (and even removed from
150 * the dispatch list).
153 struct hlist_node hash; /* merge hash */
154 struct llist_node ipi_list;
158 * The rb_node is only used inside the io scheduler, requests
159 * are pruned when moved to the dispatch queue. special_vec must
160 * only be used if RQF_SPECIAL_PAYLOAD is set, and those cannot be
161 * insert into an IO scheduler.
164 struct rb_node rb_node; /* sort/lookup */
165 struct bio_vec special_vec;
169 * Three pointers are available for the IO schedulers, if they need
170 * more they have to dynamically allocate it.
179 rq_end_io_fn *saved_end_io;
185 * completion callback.
187 rq_end_io_fn *end_io;
191 static inline enum req_op req_op(const struct request *req)
193 return req->cmd_flags & REQ_OP_MASK;
196 static inline bool blk_rq_is_passthrough(struct request *rq)
198 return blk_op_is_passthrough(rq->cmd_flags);
201 static inline unsigned short req_get_ioprio(struct request *req)
206 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
208 #define rq_dma_dir(rq) \
209 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
211 #define rq_list_add(listptr, rq) do { \
212 (rq)->rq_next = *(listptr); \
216 #define rq_list_add_tail(lastpptr, rq) do { \
217 (rq)->rq_next = NULL; \
219 *(lastpptr) = &rq->rq_next; \
222 #define rq_list_pop(listptr) \
224 struct request *__req = NULL; \
225 if ((listptr) && *(listptr)) { \
226 __req = *(listptr); \
227 *(listptr) = __req->rq_next; \
232 #define rq_list_peek(listptr) \
234 struct request *__req = NULL; \
235 if ((listptr) && *(listptr)) \
236 __req = *(listptr); \
240 #define rq_list_for_each(listptr, pos) \
241 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
243 #define rq_list_for_each_safe(listptr, pos, nxt) \
244 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
245 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
247 #define rq_list_next(rq) (rq)->rq_next
248 #define rq_list_empty(list) ((list) == (struct request *) NULL)
251 * rq_list_move() - move a struct request from one list to another
252 * @src: The source list @rq is currently in
253 * @dst: The destination list that @rq will be appended to
254 * @rq: The request to move
255 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
257 static inline void rq_list_move(struct request **src, struct request **dst,
258 struct request *rq, struct request *prev)
261 prev->rq_next = rq->rq_next;
264 rq_list_add(dst, rq);
268 * enum blk_eh_timer_return - How the timeout handler should proceed
269 * @BLK_EH_DONE: The block driver completed the command or will complete it at
271 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
272 * request to complete.
274 enum blk_eh_timer_return {
279 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
280 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
283 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
286 struct blk_mq_hw_ctx {
288 /** @lock: Protects the dispatch list. */
291 * @dispatch: Used for requests that are ready to be
292 * dispatched to the hardware but for some reason (e.g. lack of
293 * resources) could not be sent to the hardware. As soon as the
294 * driver can send new requests, requests at this list will
295 * be sent first for a fairer dispatch.
297 struct list_head dispatch;
299 * @state: BLK_MQ_S_* flags. Defines the state of the hw
300 * queue (active, scheduled to restart, stopped).
303 } ____cacheline_aligned_in_smp;
306 * @run_work: Used for scheduling a hardware queue run at a later time.
308 struct delayed_work run_work;
309 /** @cpumask: Map of available CPUs where this hctx can run. */
310 cpumask_var_t cpumask;
312 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
313 * selection from @cpumask.
317 * @next_cpu_batch: Counter of how many works left in the batch before
318 * changing to the next CPU.
322 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
326 * @sched_data: Pointer owned by the IO scheduler attached to a request
327 * queue. It's up to the IO scheduler how to use this pointer.
331 * @queue: Pointer to the request queue that owns this hardware context.
333 struct request_queue *queue;
334 /** @fq: Queue of requests that need to perform a flush operation. */
335 struct blk_flush_queue *fq;
338 * @driver_data: Pointer to data owned by the block driver that created
344 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
345 * pending request in that software queue.
347 struct sbitmap ctx_map;
350 * @dispatch_from: Software queue to be used when no scheduler was
353 struct blk_mq_ctx *dispatch_from;
355 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
356 * decide if the hw_queue is busy using Exponential Weighted Moving
359 unsigned int dispatch_busy;
361 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
363 /** @nr_ctx: Number of software queues. */
364 unsigned short nr_ctx;
365 /** @ctxs: Array of software queues. */
366 struct blk_mq_ctx **ctxs;
368 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
369 spinlock_t dispatch_wait_lock;
371 * @dispatch_wait: Waitqueue to put requests when there is no tag
372 * available at the moment, to wait for another try in the future.
374 wait_queue_entry_t dispatch_wait;
377 * @wait_index: Index of next available dispatch_wait queue to insert
383 * @tags: Tags owned by the block driver. A tag at this set is only
384 * assigned when a request is dispatched from a hardware queue.
386 struct blk_mq_tags *tags;
388 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
389 * scheduler associated with a request queue, a tag is assigned when
390 * that request is allocated. Else, this member is not used.
392 struct blk_mq_tags *sched_tags;
394 /** @run: Number of dispatched requests. */
397 /** @numa_node: NUMA node the storage adapter has been connected to. */
398 unsigned int numa_node;
399 /** @queue_num: Index of this hardware queue. */
400 unsigned int queue_num;
403 * @nr_active: Number of active requests. Only used when a tag set is
404 * shared across request queues.
408 /** @cpuhp_online: List to store request if CPU is going to die */
409 struct hlist_node cpuhp_online;
410 /** @cpuhp_dead: List to store request if some CPU die. */
411 struct hlist_node cpuhp_dead;
412 /** @kobj: Kernel object for sysfs. */
415 #ifdef CONFIG_BLK_DEBUG_FS
417 * @debugfs_dir: debugfs directory for this hardware queue. Named
418 * as cpu<cpu_number>.
420 struct dentry *debugfs_dir;
421 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
422 struct dentry *sched_debugfs_dir;
426 * @hctx_list: if this hctx is not in use, this is an entry in
427 * q->unused_hctx_list.
429 struct list_head hctx_list;
433 * struct blk_mq_queue_map - Map software queues to hardware queues
434 * @mq_map: CPU ID to hardware queue index map. This is an array
435 * with nr_cpu_ids elements. Each element has a value in the range
436 * [@queue_offset, @queue_offset + @nr_queues).
437 * @nr_queues: Number of hardware queues to map CPU IDs onto.
438 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
439 * driver to map each hardware queue type (enum hctx_type) onto a distinct
440 * set of hardware queues.
442 struct blk_mq_queue_map {
443 unsigned int *mq_map;
444 unsigned int nr_queues;
445 unsigned int queue_offset;
449 * enum hctx_type - Type of hardware queue
450 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
451 * @HCTX_TYPE_READ: Just for READ I/O.
452 * @HCTX_TYPE_POLL: Polled I/O of any kind.
453 * @HCTX_MAX_TYPES: Number of types of hctx.
464 * struct blk_mq_tag_set - tag set that can be shared between request queues
465 * @ops: Pointers to functions that implement block driver behavior.
466 * @map: One or more ctx -> hctx mappings. One map exists for each
467 * hardware queue type (enum hctx_type) that the driver wishes
468 * to support. There are no restrictions on maps being of the
469 * same size, and it's perfectly legal to share maps between
471 * @nr_maps: Number of elements in the @map array. A number in the range
472 * [1, HCTX_MAX_TYPES].
473 * @nr_hw_queues: Number of hardware queues supported by the block driver that
474 * owns this data structure.
475 * @queue_depth: Number of tags per hardware queue, reserved tags included.
476 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
478 * @cmd_size: Number of additional bytes to allocate per request. The block
479 * driver owns these additional bytes.
480 * @numa_node: NUMA node the storage adapter has been connected to.
481 * @timeout: Request processing timeout in jiffies.
482 * @flags: Zero or more BLK_MQ_F_* flags.
483 * @driver_data: Pointer to data owned by the block driver that created this
485 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
488 * Shared set of tags. Has @nr_hw_queues elements. If set,
489 * shared by all @tags.
490 * @tag_list_lock: Serializes tag_list accesses.
491 * @tag_list: List of the request queues that use this tag set. See also
492 * request_queue.tag_set_list.
493 * @srcu: Use as lock when type of the request queue is blocking
494 * (BLK_MQ_F_BLOCKING).
496 struct blk_mq_tag_set {
497 const struct blk_mq_ops *ops;
498 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
499 unsigned int nr_maps;
500 unsigned int nr_hw_queues;
501 unsigned int queue_depth;
502 unsigned int reserved_tags;
503 unsigned int cmd_size;
505 unsigned int timeout;
509 struct blk_mq_tags **tags;
511 struct blk_mq_tags *shared_tags;
513 struct mutex tag_list_lock;
514 struct list_head tag_list;
515 struct srcu_struct *srcu;
519 * struct blk_mq_queue_data - Data about a request inserted in a queue
521 * @rq: Request pointer.
522 * @last: If it is the last request in the queue.
524 struct blk_mq_queue_data {
529 typedef bool (busy_tag_iter_fn)(struct request *, void *);
532 * struct blk_mq_ops - Callback functions that implements block driver
537 * @queue_rq: Queue a new request from block IO.
539 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
540 const struct blk_mq_queue_data *);
543 * @commit_rqs: If a driver uses bd->last to judge when to submit
544 * requests to hardware, it must define this function. In case of errors
545 * that make us stop issuing further requests, this hook serves the
546 * purpose of kicking the hardware (which the last request otherwise
549 void (*commit_rqs)(struct blk_mq_hw_ctx *);
552 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
553 * that each request belongs to the same queue. If the driver doesn't
554 * empty the @rqlist completely, then the rest will be queued
555 * individually by the block layer upon return.
557 void (*queue_rqs)(struct request **rqlist);
560 * @get_budget: Reserve budget before queue request, once .queue_rq is
561 * run, it is driver's responsibility to release the
562 * reserved budget. Also we have to handle failure case
563 * of .get_budget for avoiding I/O deadlock.
565 int (*get_budget)(struct request_queue *);
568 * @put_budget: Release the reserved budget.
570 void (*put_budget)(struct request_queue *, int);
573 * @set_rq_budget_token: store rq's budget token
575 void (*set_rq_budget_token)(struct request *, int);
577 * @get_rq_budget_token: retrieve rq's budget token
579 int (*get_rq_budget_token)(struct request *);
582 * @timeout: Called on request timeout.
584 enum blk_eh_timer_return (*timeout)(struct request *);
587 * @poll: Called to poll for completion of a specific tag.
589 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
592 * @complete: Mark the request as complete.
594 void (*complete)(struct request *);
597 * @init_hctx: Called when the block layer side of a hardware queue has
598 * been set up, allowing the driver to allocate/init matching
601 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
603 * @exit_hctx: Ditto for exit/teardown.
605 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
608 * @init_request: Called for every command allocated by the block layer
609 * to allow the driver to set up driver specific data.
611 * Tag greater than or equal to queue_depth is for setting up
614 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
615 unsigned int, unsigned int);
617 * @exit_request: Ditto for exit/teardown.
619 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
623 * @cleanup_rq: Called before freeing one request which isn't completed
624 * yet, and usually for freeing the driver private data.
626 void (*cleanup_rq)(struct request *);
629 * @busy: If set, returns whether or not this queue currently is busy.
631 bool (*busy)(struct request_queue *);
634 * @map_queues: This allows drivers specify their own queue mapping by
635 * overriding the setup-time function that builds the mq_map.
637 void (*map_queues)(struct blk_mq_tag_set *set);
639 #ifdef CONFIG_BLK_DEBUG_FS
641 * @show_rq: Used by the debugfs implementation to show driver-specific
642 * information about a request.
644 void (*show_rq)(struct seq_file *m, struct request *rq);
649 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
650 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
652 * Set when this device requires underlying blk-mq device for
655 BLK_MQ_F_STACKING = 1 << 2,
656 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
657 BLK_MQ_F_BLOCKING = 1 << 5,
658 /* Do not allow an I/O scheduler to be configured. */
659 BLK_MQ_F_NO_SCHED = 1 << 6,
661 * Select 'none' during queue registration in case of a single hwq
662 * or shared hwqs instead of 'mq-deadline'.
664 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
665 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
666 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
668 BLK_MQ_S_STOPPED = 0,
669 BLK_MQ_S_TAG_ACTIVE = 1,
670 BLK_MQ_S_SCHED_RESTART = 2,
672 /* hw queue is inactive after all its CPUs become offline */
673 BLK_MQ_S_INACTIVE = 3,
675 BLK_MQ_MAX_DEPTH = 10240,
677 BLK_MQ_CPU_WORK_BATCH = 8,
679 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
680 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
681 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
682 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
683 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
684 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
686 #define BLK_MQ_NO_HCTX_IDX (-1U)
688 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
689 struct lock_class_key *lkclass);
690 #define blk_mq_alloc_disk(set, queuedata) \
692 static struct lock_class_key __key; \
694 __blk_mq_alloc_disk(set, queuedata, &__key); \
696 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
697 struct lock_class_key *lkclass);
698 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
699 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
700 struct request_queue *q);
701 void blk_mq_destroy_queue(struct request_queue *);
703 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
704 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
705 const struct blk_mq_ops *ops, unsigned int queue_depth,
706 unsigned int set_flags);
707 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
709 void blk_mq_free_request(struct request *rq);
710 int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
711 unsigned int poll_flags);
713 bool blk_mq_queue_inflight(struct request_queue *q);
716 /* return when out of requests */
717 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
718 /* allocate from reserved pool */
719 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
721 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
724 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
725 blk_mq_req_flags_t flags);
726 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
727 blk_opf_t opf, blk_mq_req_flags_t flags,
728 unsigned int hctx_idx);
731 * Tag address space map.
734 unsigned int nr_tags;
735 unsigned int nr_reserved_tags;
736 unsigned int active_queues;
738 struct sbitmap_queue bitmap_tags;
739 struct sbitmap_queue breserved_tags;
741 struct request **rqs;
742 struct request **static_rqs;
743 struct list_head page_list;
746 * used to clear request reference in rqs[] before freeing one
752 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
755 if (tag < tags->nr_tags) {
756 prefetch(tags->rqs[tag]);
757 return tags->rqs[tag];
764 BLK_MQ_UNIQUE_TAG_BITS = 16,
765 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
768 u32 blk_mq_unique_tag(struct request *rq);
770 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
772 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
775 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
777 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
781 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
782 * @rq: target request.
784 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
786 return READ_ONCE(rq->state);
789 static inline int blk_mq_request_started(struct request *rq)
791 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
794 static inline int blk_mq_request_completed(struct request *rq)
796 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
801 * Set the state to complete when completing a request from inside ->queue_rq.
802 * This is used by drivers that want to ensure special complete actions that
803 * need access to the request are called on failure, e.g. by nvme for
806 static inline void blk_mq_set_request_complete(struct request *rq)
808 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
812 * Complete the request directly instead of deferring it to softirq or
813 * completing it another CPU. Useful in preemptible instead of an interrupt.
815 static inline void blk_mq_complete_request_direct(struct request *rq,
816 void (*complete)(struct request *rq))
818 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
822 void blk_mq_start_request(struct request *rq);
823 void blk_mq_end_request(struct request *rq, blk_status_t error);
824 void __blk_mq_end_request(struct request *rq, blk_status_t error);
825 void blk_mq_end_request_batch(struct io_comp_batch *ib);
828 * Only need start/end time stamping if we have iostat or
829 * blk stats enabled, or using an IO scheduler.
831 static inline bool blk_mq_need_time_stamp(struct request *rq)
834 * passthrough io doesn't use iostat accounting, cgroup stats
835 * and io scheduler functionalities.
837 if (blk_rq_is_passthrough(rq))
839 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
842 static inline bool blk_mq_is_reserved_rq(struct request *rq)
844 return rq->rq_flags & RQF_RESV;
848 * Batched completions only work when there is no I/O error and no special
851 static inline bool blk_mq_add_to_batch(struct request *req,
852 struct io_comp_batch *iob, int ioerror,
853 void (*complete)(struct io_comp_batch *))
856 * blk_mq_end_request_batch() can't end request allocated from
859 if (!iob || (req->rq_flags & RQF_SCHED_TAGS) || ioerror ||
860 (req->end_io && !blk_rq_is_passthrough(req)))
864 iob->complete = complete;
865 else if (iob->complete != complete)
867 iob->need_ts |= blk_mq_need_time_stamp(req);
868 rq_list_add(&iob->req_list, req);
872 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
873 void blk_mq_kick_requeue_list(struct request_queue *q);
874 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
875 void blk_mq_complete_request(struct request *rq);
876 bool blk_mq_complete_request_remote(struct request *rq);
877 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
878 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
879 void blk_mq_stop_hw_queues(struct request_queue *q);
880 void blk_mq_start_hw_queues(struct request_queue *q);
881 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
882 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
883 void blk_mq_quiesce_queue(struct request_queue *q);
884 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
885 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
886 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
887 void blk_mq_unquiesce_queue(struct request_queue *q);
888 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
889 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
890 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
891 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
892 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
893 busy_tag_iter_fn *fn, void *priv);
894 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
895 void blk_mq_freeze_queue(struct request_queue *q);
896 void blk_mq_unfreeze_queue(struct request_queue *q);
897 void blk_freeze_queue_start(struct request_queue *q);
898 void blk_mq_freeze_queue_wait(struct request_queue *q);
899 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
900 unsigned long timeout);
902 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
903 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
905 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
907 unsigned int blk_mq_rq_cpu(struct request *rq);
909 bool __blk_should_fake_timeout(struct request_queue *q);
910 static inline bool blk_should_fake_timeout(struct request_queue *q)
912 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
913 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
914 return __blk_should_fake_timeout(q);
919 * blk_mq_rq_from_pdu - cast a PDU to a request
920 * @pdu: the PDU (Protocol Data Unit) to be casted
924 * Driver command data is immediately after the request. So subtract request
925 * size to get back to the original request.
927 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
929 return pdu - sizeof(struct request);
933 * blk_mq_rq_to_pdu - cast a request to a PDU
934 * @rq: the request to be casted
936 * Return: pointer to the PDU
938 * Driver command data is immediately after the request. So add request to get
941 static inline void *blk_mq_rq_to_pdu(struct request *rq)
946 #define queue_for_each_hw_ctx(q, hctx, i) \
947 xa_for_each(&(q)->hctx_table, (i), (hctx))
949 #define hctx_for_each_ctx(hctx, ctx, i) \
950 for ((i) = 0; (i) < (hctx)->nr_ctx && \
951 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
953 static inline void blk_mq_cleanup_rq(struct request *rq)
955 if (rq->q->mq_ops->cleanup_rq)
956 rq->q->mq_ops->cleanup_rq(rq);
959 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
960 unsigned int nr_segs)
962 rq->nr_phys_segments = nr_segs;
963 rq->__data_len = bio->bi_iter.bi_size;
964 rq->bio = rq->biotail = bio;
965 rq->ioprio = bio_prio(bio);
968 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
969 struct lock_class_key *key);
971 static inline bool rq_is_sync(struct request *rq)
973 return op_is_sync(rq->cmd_flags);
976 void blk_rq_init(struct request_queue *q, struct request *rq);
977 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
978 struct bio_set *bs, gfp_t gfp_mask,
979 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
980 void blk_rq_unprep_clone(struct request *rq);
981 blk_status_t blk_insert_cloned_request(struct request *rq);
985 unsigned long offset;
986 unsigned short page_order;
987 unsigned short nr_entries;
992 int blk_rq_map_user(struct request_queue *, struct request *,
993 struct rq_map_data *, void __user *, unsigned long, gfp_t);
994 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
995 void __user *, unsigned long, gfp_t, bool, int, bool, int);
996 int blk_rq_map_user_iov(struct request_queue *, struct request *,
997 struct rq_map_data *, const struct iov_iter *, gfp_t);
998 int blk_rq_unmap_user(struct bio *);
999 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
1000 unsigned int, gfp_t);
1001 int blk_rq_append_bio(struct request *rq, struct bio *bio);
1002 void blk_execute_rq_nowait(struct request *rq, bool at_head);
1003 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1004 bool blk_rq_is_poll(struct request *rq);
1006 struct req_iterator {
1007 struct bvec_iter iter;
1011 #define __rq_for_each_bio(_bio, rq) \
1013 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1015 #define rq_for_each_segment(bvl, _rq, _iter) \
1016 __rq_for_each_bio(_iter.bio, _rq) \
1017 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1019 #define rq_for_each_bvec(bvl, _rq, _iter) \
1020 __rq_for_each_bio(_iter.bio, _rq) \
1021 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1023 #define rq_iter_last(bvec, _iter) \
1024 (_iter.bio->bi_next == NULL && \
1025 bio_iter_last(bvec, _iter.iter))
1028 * blk_rq_pos() : the current sector
1029 * blk_rq_bytes() : bytes left in the entire request
1030 * blk_rq_cur_bytes() : bytes left in the current segment
1031 * blk_rq_sectors() : sectors left in the entire request
1032 * blk_rq_cur_sectors() : sectors left in the current segment
1033 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1035 static inline sector_t blk_rq_pos(const struct request *rq)
1037 return rq->__sector;
1040 static inline unsigned int blk_rq_bytes(const struct request *rq)
1042 return rq->__data_len;
1045 static inline int blk_rq_cur_bytes(const struct request *rq)
1049 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1050 return rq->bio->bi_iter.bi_size;
1051 return bio_iovec(rq->bio).bv_len;
1054 static inline unsigned int blk_rq_sectors(const struct request *rq)
1056 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1059 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1061 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1064 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1066 return rq->stats_sectors;
1070 * Some commands like WRITE SAME have a payload or data transfer size which
1071 * is different from the size of the request. Any driver that supports such
1072 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1073 * calculate the data transfer size.
1075 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1077 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1078 return rq->special_vec.bv_len;
1079 return blk_rq_bytes(rq);
1083 * Return the first full biovec in the request. The caller needs to check that
1084 * there are any bvecs before calling this helper.
1086 static inline struct bio_vec req_bvec(struct request *rq)
1088 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1089 return rq->special_vec;
1090 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1093 static inline unsigned int blk_rq_count_bios(struct request *rq)
1095 unsigned int nr_bios = 0;
1098 __rq_for_each_bio(bio, rq)
1104 void blk_steal_bios(struct bio_list *list, struct request *rq);
1107 * Request completion related functions.
1109 * blk_update_request() completes given number of bytes and updates
1110 * the request without completing it.
1112 bool blk_update_request(struct request *rq, blk_status_t error,
1113 unsigned int nr_bytes);
1114 void blk_abort_request(struct request *);
1117 * Number of physical segments as sent to the device.
1119 * Normally this is the number of discontiguous data segments sent by the
1120 * submitter. But for data-less command like discard we might have no
1121 * actual data segments submitted, but the driver might have to add it's
1122 * own special payload. In that case we still return 1 here so that this
1123 * special payload will be mapped.
1125 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1127 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1129 return rq->nr_phys_segments;
1133 * Number of discard segments (or ranges) the driver needs to fill in.
1134 * Each discard bio merged into a request is counted as one segment.
1136 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1138 return max_t(unsigned short, rq->nr_phys_segments, 1);
1141 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1142 struct scatterlist *sglist, struct scatterlist **last_sg);
1143 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1144 struct scatterlist *sglist)
1146 struct scatterlist *last_sg = NULL;
1148 return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1150 void blk_dump_rq_flags(struct request *, char *);
1152 #ifdef CONFIG_BLK_DEV_ZONED
1153 static inline unsigned int blk_rq_zone_no(struct request *rq)
1155 return disk_zone_no(rq->q->disk, blk_rq_pos(rq));
1158 static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1160 return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq));
1164 * blk_rq_is_seq_zoned_write() - Check if @rq requires write serialization.
1165 * @rq: Request to examine.
1167 * Note: REQ_OP_ZONE_APPEND requests do not require serialization.
1169 static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1171 return op_needs_zoned_write_locking(req_op(rq)) &&
1172 blk_rq_zone_is_seq(rq);
1175 bool blk_req_needs_zone_write_lock(struct request *rq);
1176 bool blk_req_zone_write_trylock(struct request *rq);
1177 void __blk_req_zone_write_lock(struct request *rq);
1178 void __blk_req_zone_write_unlock(struct request *rq);
1180 static inline void blk_req_zone_write_lock(struct request *rq)
1182 if (blk_req_needs_zone_write_lock(rq))
1183 __blk_req_zone_write_lock(rq);
1186 static inline void blk_req_zone_write_unlock(struct request *rq)
1188 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1189 __blk_req_zone_write_unlock(rq);
1192 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1194 return rq->q->disk->seq_zones_wlock &&
1195 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock);
1198 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1200 if (!blk_req_needs_zone_write_lock(rq))
1202 return !blk_req_zone_is_write_locked(rq);
1204 #else /* CONFIG_BLK_DEV_ZONED */
1205 static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1210 static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1215 static inline void blk_req_zone_write_lock(struct request *rq)
1219 static inline void blk_req_zone_write_unlock(struct request *rq)
1222 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1227 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1231 #endif /* CONFIG_BLK_DEV_ZONED */
1233 #endif /* BLK_MQ_H */