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 /* may not be passed by ioscheduler */
32 #define RQF_SOFTBARRIER ((__force req_flags_t)(1 << 3))
33 /* request for flush sequence */
34 #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4))
35 /* merge of different types, fail separately */
36 #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5))
37 /* track inflight for MQ */
38 #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6))
39 /* don't call prep for this one */
40 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
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 /* queue has elevator attached */
61 #define RQF_ELV ((__force req_flags_t)(1 << 22))
62 #define RQF_RESV ((__force req_flags_t)(1 << 23))
64 /* flags that prevent us from merging requests: */
65 #define RQF_NOMERGE_FLAGS \
66 (RQF_STARTED | RQF_SOFTBARRIER | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
75 * Try to put the fields that are referenced together in the same cacheline.
77 * If you modify this structure, make sure to update blk_rq_init() and
78 * especially blk_mq_rq_ctx_init() to take care of the added fields.
81 struct request_queue *q;
82 struct blk_mq_ctx *mq_ctx;
83 struct blk_mq_hw_ctx *mq_hctx;
85 blk_opf_t cmd_flags; /* op and common flags */
93 /* the following two fields are internal, NEVER access directly */
94 unsigned int __data_len; /* total data len */
95 sector_t __sector; /* sector cursor */
101 struct list_head queuelist;
102 struct request *rq_next;
105 struct block_device *part;
106 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
107 /* Time that the first bio started allocating this request. */
110 /* Time that this request was allocated for this IO. */
112 /* Time that I/O was submitted to the device. */
113 u64 io_start_time_ns;
115 #ifdef CONFIG_BLK_WBT
116 unsigned short wbt_flags;
119 * rq sectors used for blk stats. It has the same value
120 * with blk_rq_sectors(rq), except that it never be zeroed
123 unsigned short stats_sectors;
126 * Number of scatter-gather DMA addr+len pairs after
127 * physical address coalescing is performed.
129 unsigned short nr_phys_segments;
131 #ifdef CONFIG_BLK_DEV_INTEGRITY
132 unsigned short nr_integrity_segments;
135 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
136 struct bio_crypt_ctx *crypt_ctx;
137 struct blk_crypto_keyslot *crypt_keyslot;
140 unsigned short ioprio;
142 enum mq_rq_state state;
145 unsigned long deadline;
148 * The hash is used inside the scheduler, and killed once the
149 * request reaches the dispatch list. The ipi_list is only used
150 * to queue the request for softirq completion, which is long
151 * after the request has been unhashed (and even removed from
152 * the dispatch list).
155 struct hlist_node hash; /* merge hash */
156 struct llist_node ipi_list;
160 * The rb_node is only used inside the io scheduler, requests
161 * are pruned when moved to the dispatch queue. So let the
162 * completion_data share space with the rb_node.
165 struct rb_node rb_node; /* sort/lookup */
166 struct bio_vec special_vec;
167 void *completion_data;
172 * Three pointers are available for the IO schedulers, if they need
173 * more they have to dynamically allocate it. Flush requests are
174 * never put on the IO scheduler. So let the flush fields share
175 * space with the elevator data.
185 struct list_head list;
186 rq_end_io_fn *saved_end_io;
191 struct __call_single_data csd;
196 * completion callback.
198 rq_end_io_fn *end_io;
202 static inline enum req_op req_op(const struct request *req)
204 return req->cmd_flags & REQ_OP_MASK;
207 static inline bool blk_rq_is_passthrough(struct request *rq)
209 return blk_op_is_passthrough(req_op(rq));
212 static inline unsigned short req_get_ioprio(struct request *req)
217 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
219 #define rq_dma_dir(rq) \
220 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
222 #define rq_list_add(listptr, rq) do { \
223 (rq)->rq_next = *(listptr); \
227 #define rq_list_add_tail(lastpptr, rq) do { \
228 (rq)->rq_next = NULL; \
230 *(lastpptr) = &rq->rq_next; \
233 #define rq_list_pop(listptr) \
235 struct request *__req = NULL; \
236 if ((listptr) && *(listptr)) { \
237 __req = *(listptr); \
238 *(listptr) = __req->rq_next; \
243 #define rq_list_peek(listptr) \
245 struct request *__req = NULL; \
246 if ((listptr) && *(listptr)) \
247 __req = *(listptr); \
251 #define rq_list_for_each(listptr, pos) \
252 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
254 #define rq_list_for_each_safe(listptr, pos, nxt) \
255 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
256 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
258 #define rq_list_next(rq) (rq)->rq_next
259 #define rq_list_empty(list) ((list) == (struct request *) NULL)
262 * rq_list_move() - move a struct request from one list to another
263 * @src: The source list @rq is currently in
264 * @dst: The destination list that @rq will be appended to
265 * @rq: The request to move
266 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
268 static inline void rq_list_move(struct request **src, struct request **dst,
269 struct request *rq, struct request *prev)
272 prev->rq_next = rq->rq_next;
275 rq_list_add(dst, rq);
279 * enum blk_eh_timer_return - How the timeout handler should proceed
280 * @BLK_EH_DONE: The block driver completed the command or will complete it at
282 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
283 * request to complete.
285 enum blk_eh_timer_return {
290 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
291 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
294 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
297 struct blk_mq_hw_ctx {
299 /** @lock: Protects the dispatch list. */
302 * @dispatch: Used for requests that are ready to be
303 * dispatched to the hardware but for some reason (e.g. lack of
304 * resources) could not be sent to the hardware. As soon as the
305 * driver can send new requests, requests at this list will
306 * be sent first for a fairer dispatch.
308 struct list_head dispatch;
310 * @state: BLK_MQ_S_* flags. Defines the state of the hw
311 * queue (active, scheduled to restart, stopped).
314 } ____cacheline_aligned_in_smp;
317 * @run_work: Used for scheduling a hardware queue run at a later time.
319 struct delayed_work run_work;
320 /** @cpumask: Map of available CPUs where this hctx can run. */
321 cpumask_var_t cpumask;
323 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
324 * selection from @cpumask.
328 * @next_cpu_batch: Counter of how many works left in the batch before
329 * changing to the next CPU.
333 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
337 * @sched_data: Pointer owned by the IO scheduler attached to a request
338 * queue. It's up to the IO scheduler how to use this pointer.
342 * @queue: Pointer to the request queue that owns this hardware context.
344 struct request_queue *queue;
345 /** @fq: Queue of requests that need to perform a flush operation. */
346 struct blk_flush_queue *fq;
349 * @driver_data: Pointer to data owned by the block driver that created
355 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
356 * pending request in that software queue.
358 struct sbitmap ctx_map;
361 * @dispatch_from: Software queue to be used when no scheduler was
364 struct blk_mq_ctx *dispatch_from;
366 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
367 * decide if the hw_queue is busy using Exponential Weighted Moving
370 unsigned int dispatch_busy;
372 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
374 /** @nr_ctx: Number of software queues. */
375 unsigned short nr_ctx;
376 /** @ctxs: Array of software queues. */
377 struct blk_mq_ctx **ctxs;
379 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
380 spinlock_t dispatch_wait_lock;
382 * @dispatch_wait: Waitqueue to put requests when there is no tag
383 * available at the moment, to wait for another try in the future.
385 wait_queue_entry_t dispatch_wait;
388 * @wait_index: Index of next available dispatch_wait queue to insert
394 * @tags: Tags owned by the block driver. A tag at this set is only
395 * assigned when a request is dispatched from a hardware queue.
397 struct blk_mq_tags *tags;
399 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
400 * scheduler associated with a request queue, a tag is assigned when
401 * that request is allocated. Else, this member is not used.
403 struct blk_mq_tags *sched_tags;
405 /** @queued: Number of queued requests. */
406 unsigned long queued;
407 /** @run: Number of dispatched requests. */
410 /** @numa_node: NUMA node the storage adapter has been connected to. */
411 unsigned int numa_node;
412 /** @queue_num: Index of this hardware queue. */
413 unsigned int queue_num;
416 * @nr_active: Number of active requests. Only used when a tag set is
417 * shared across request queues.
421 /** @cpuhp_online: List to store request if CPU is going to die */
422 struct hlist_node cpuhp_online;
423 /** @cpuhp_dead: List to store request if some CPU die. */
424 struct hlist_node cpuhp_dead;
425 /** @kobj: Kernel object for sysfs. */
428 #ifdef CONFIG_BLK_DEBUG_FS
430 * @debugfs_dir: debugfs directory for this hardware queue. Named
431 * as cpu<cpu_number>.
433 struct dentry *debugfs_dir;
434 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
435 struct dentry *sched_debugfs_dir;
439 * @hctx_list: if this hctx is not in use, this is an entry in
440 * q->unused_hctx_list.
442 struct list_head hctx_list;
446 * struct blk_mq_queue_map - Map software queues to hardware queues
447 * @mq_map: CPU ID to hardware queue index map. This is an array
448 * with nr_cpu_ids elements. Each element has a value in the range
449 * [@queue_offset, @queue_offset + @nr_queues).
450 * @nr_queues: Number of hardware queues to map CPU IDs onto.
451 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
452 * driver to map each hardware queue type (enum hctx_type) onto a distinct
453 * set of hardware queues.
455 struct blk_mq_queue_map {
456 unsigned int *mq_map;
457 unsigned int nr_queues;
458 unsigned int queue_offset;
462 * enum hctx_type - Type of hardware queue
463 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
464 * @HCTX_TYPE_READ: Just for READ I/O.
465 * @HCTX_TYPE_POLL: Polled I/O of any kind.
466 * @HCTX_MAX_TYPES: Number of types of hctx.
477 * struct blk_mq_tag_set - tag set that can be shared between request queues
478 * @ops: Pointers to functions that implement block driver behavior.
479 * @map: One or more ctx -> hctx mappings. One map exists for each
480 * hardware queue type (enum hctx_type) that the driver wishes
481 * to support. There are no restrictions on maps being of the
482 * same size, and it's perfectly legal to share maps between
484 * @nr_maps: Number of elements in the @map array. A number in the range
485 * [1, HCTX_MAX_TYPES].
486 * @nr_hw_queues: Number of hardware queues supported by the block driver that
487 * owns this data structure.
488 * @queue_depth: Number of tags per hardware queue, reserved tags included.
489 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
491 * @cmd_size: Number of additional bytes to allocate per request. The block
492 * driver owns these additional bytes.
493 * @numa_node: NUMA node the storage adapter has been connected to.
494 * @timeout: Request processing timeout in jiffies.
495 * @flags: Zero or more BLK_MQ_F_* flags.
496 * @driver_data: Pointer to data owned by the block driver that created this
498 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
501 * Shared set of tags. Has @nr_hw_queues elements. If set,
502 * shared by all @tags.
503 * @tag_list_lock: Serializes tag_list accesses.
504 * @tag_list: List of the request queues that use this tag set. See also
505 * request_queue.tag_set_list.
506 * @srcu: Use as lock when type of the request queue is blocking
507 * (BLK_MQ_F_BLOCKING).
509 struct blk_mq_tag_set {
510 const struct blk_mq_ops *ops;
511 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
512 unsigned int nr_maps;
513 unsigned int nr_hw_queues;
514 unsigned int queue_depth;
515 unsigned int reserved_tags;
516 unsigned int cmd_size;
518 unsigned int timeout;
522 struct blk_mq_tags **tags;
524 struct blk_mq_tags *shared_tags;
526 struct mutex tag_list_lock;
527 struct list_head tag_list;
528 struct srcu_struct *srcu;
532 * struct blk_mq_queue_data - Data about a request inserted in a queue
534 * @rq: Request pointer.
535 * @last: If it is the last request in the queue.
537 struct blk_mq_queue_data {
542 typedef bool (busy_tag_iter_fn)(struct request *, void *);
545 * struct blk_mq_ops - Callback functions that implements block driver
550 * @queue_rq: Queue a new request from block IO.
552 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
553 const struct blk_mq_queue_data *);
556 * @commit_rqs: If a driver uses bd->last to judge when to submit
557 * requests to hardware, it must define this function. In case of errors
558 * that make us stop issuing further requests, this hook serves the
559 * purpose of kicking the hardware (which the last request otherwise
562 void (*commit_rqs)(struct blk_mq_hw_ctx *);
565 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
566 * that each request belongs to the same queue. If the driver doesn't
567 * empty the @rqlist completely, then the rest will be queued
568 * individually by the block layer upon return.
570 void (*queue_rqs)(struct request **rqlist);
573 * @get_budget: Reserve budget before queue request, once .queue_rq is
574 * run, it is driver's responsibility to release the
575 * reserved budget. Also we have to handle failure case
576 * of .get_budget for avoiding I/O deadlock.
578 int (*get_budget)(struct request_queue *);
581 * @put_budget: Release the reserved budget.
583 void (*put_budget)(struct request_queue *, int);
586 * @set_rq_budget_token: store rq's budget token
588 void (*set_rq_budget_token)(struct request *, int);
590 * @get_rq_budget_token: retrieve rq's budget token
592 int (*get_rq_budget_token)(struct request *);
595 * @timeout: Called on request timeout.
597 enum blk_eh_timer_return (*timeout)(struct request *);
600 * @poll: Called to poll for completion of a specific tag.
602 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
605 * @complete: Mark the request as complete.
607 void (*complete)(struct request *);
610 * @init_hctx: Called when the block layer side of a hardware queue has
611 * been set up, allowing the driver to allocate/init matching
614 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
616 * @exit_hctx: Ditto for exit/teardown.
618 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
621 * @init_request: Called for every command allocated by the block layer
622 * to allow the driver to set up driver specific data.
624 * Tag greater than or equal to queue_depth is for setting up
627 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
628 unsigned int, unsigned int);
630 * @exit_request: Ditto for exit/teardown.
632 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
636 * @cleanup_rq: Called before freeing one request which isn't completed
637 * yet, and usually for freeing the driver private data.
639 void (*cleanup_rq)(struct request *);
642 * @busy: If set, returns whether or not this queue currently is busy.
644 bool (*busy)(struct request_queue *);
647 * @map_queues: This allows drivers specify their own queue mapping by
648 * overriding the setup-time function that builds the mq_map.
650 void (*map_queues)(struct blk_mq_tag_set *set);
652 #ifdef CONFIG_BLK_DEBUG_FS
654 * @show_rq: Used by the debugfs implementation to show driver-specific
655 * information about a request.
657 void (*show_rq)(struct seq_file *m, struct request *rq);
662 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
663 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
665 * Set when this device requires underlying blk-mq device for
668 BLK_MQ_F_STACKING = 1 << 2,
669 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
670 BLK_MQ_F_BLOCKING = 1 << 5,
671 /* Do not allow an I/O scheduler to be configured. */
672 BLK_MQ_F_NO_SCHED = 1 << 6,
674 * Select 'none' during queue registration in case of a single hwq
675 * or shared hwqs instead of 'mq-deadline'.
677 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
678 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
679 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
681 BLK_MQ_S_STOPPED = 0,
682 BLK_MQ_S_TAG_ACTIVE = 1,
683 BLK_MQ_S_SCHED_RESTART = 2,
685 /* hw queue is inactive after all its CPUs become offline */
686 BLK_MQ_S_INACTIVE = 3,
688 BLK_MQ_MAX_DEPTH = 10240,
690 BLK_MQ_CPU_WORK_BATCH = 8,
692 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
693 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
694 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
695 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
696 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
697 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
699 #define BLK_MQ_NO_HCTX_IDX (-1U)
701 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
702 struct lock_class_key *lkclass);
703 #define blk_mq_alloc_disk(set, queuedata) \
705 static struct lock_class_key __key; \
707 __blk_mq_alloc_disk(set, queuedata, &__key); \
709 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
710 struct lock_class_key *lkclass);
711 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
712 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
713 struct request_queue *q);
714 void blk_mq_destroy_queue(struct request_queue *);
716 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
717 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
718 const struct blk_mq_ops *ops, unsigned int queue_depth,
719 unsigned int set_flags);
720 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
722 void blk_mq_free_request(struct request *rq);
724 bool blk_mq_queue_inflight(struct request_queue *q);
727 /* return when out of requests */
728 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
729 /* allocate from reserved pool */
730 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
732 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
735 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
736 blk_mq_req_flags_t flags);
737 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
738 blk_opf_t opf, blk_mq_req_flags_t flags,
739 unsigned int hctx_idx);
742 * Tag address space map.
745 unsigned int nr_tags;
746 unsigned int nr_reserved_tags;
748 atomic_t active_queues;
750 struct sbitmap_queue bitmap_tags;
751 struct sbitmap_queue breserved_tags;
753 struct request **rqs;
754 struct request **static_rqs;
755 struct list_head page_list;
758 * used to clear request reference in rqs[] before freeing one
764 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
767 if (tag < tags->nr_tags) {
768 prefetch(tags->rqs[tag]);
769 return tags->rqs[tag];
776 BLK_MQ_UNIQUE_TAG_BITS = 16,
777 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
780 u32 blk_mq_unique_tag(struct request *rq);
782 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
784 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
787 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
789 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
793 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
794 * @rq: target request.
796 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
798 return READ_ONCE(rq->state);
801 static inline int blk_mq_request_started(struct request *rq)
803 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
806 static inline int blk_mq_request_completed(struct request *rq)
808 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
813 * Set the state to complete when completing a request from inside ->queue_rq.
814 * This is used by drivers that want to ensure special complete actions that
815 * need access to the request are called on failure, e.g. by nvme for
818 static inline void blk_mq_set_request_complete(struct request *rq)
820 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
824 * Complete the request directly instead of deferring it to softirq or
825 * completing it another CPU. Useful in preemptible instead of an interrupt.
827 static inline void blk_mq_complete_request_direct(struct request *rq,
828 void (*complete)(struct request *rq))
830 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
834 void blk_mq_start_request(struct request *rq);
835 void blk_mq_end_request(struct request *rq, blk_status_t error);
836 void __blk_mq_end_request(struct request *rq, blk_status_t error);
837 void blk_mq_end_request_batch(struct io_comp_batch *ib);
840 * Only need start/end time stamping if we have iostat or
841 * blk stats enabled, or using an IO scheduler.
843 static inline bool blk_mq_need_time_stamp(struct request *rq)
845 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_ELV));
848 static inline bool blk_mq_is_reserved_rq(struct request *rq)
850 return rq->rq_flags & RQF_RESV;
854 * Batched completions only work when there is no I/O error and no special
857 static inline bool blk_mq_add_to_batch(struct request *req,
858 struct io_comp_batch *iob, int ioerror,
859 void (*complete)(struct io_comp_batch *))
861 if (!iob || (req->rq_flags & RQF_ELV) || ioerror ||
862 (req->end_io && !blk_rq_is_passthrough(req)))
866 iob->complete = complete;
867 else if (iob->complete != complete)
869 iob->need_ts |= blk_mq_need_time_stamp(req);
870 rq_list_add(&iob->req_list, req);
874 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
875 void blk_mq_kick_requeue_list(struct request_queue *q);
876 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
877 void blk_mq_complete_request(struct request *rq);
878 bool blk_mq_complete_request_remote(struct request *rq);
879 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
880 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
881 void blk_mq_stop_hw_queues(struct request_queue *q);
882 void blk_mq_start_hw_queues(struct request_queue *q);
883 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
884 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
885 void blk_mq_quiesce_queue(struct request_queue *q);
886 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
887 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
888 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
889 void blk_mq_unquiesce_queue(struct request_queue *q);
890 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
891 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
892 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
893 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
894 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
895 busy_tag_iter_fn *fn, void *priv);
896 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
897 void blk_mq_freeze_queue(struct request_queue *q);
898 void blk_mq_unfreeze_queue(struct request_queue *q);
899 void blk_freeze_queue_start(struct request_queue *q);
900 void blk_mq_freeze_queue_wait(struct request_queue *q);
901 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
902 unsigned long timeout);
904 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
905 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
907 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
909 unsigned int blk_mq_rq_cpu(struct request *rq);
911 bool __blk_should_fake_timeout(struct request_queue *q);
912 static inline bool blk_should_fake_timeout(struct request_queue *q)
914 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
915 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
916 return __blk_should_fake_timeout(q);
921 * blk_mq_rq_from_pdu - cast a PDU to a request
922 * @pdu: the PDU (Protocol Data Unit) to be casted
926 * Driver command data is immediately after the request. So subtract request
927 * size to get back to the original request.
929 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
931 return pdu - sizeof(struct request);
935 * blk_mq_rq_to_pdu - cast a request to a PDU
936 * @rq: the request to be casted
938 * Return: pointer to the PDU
940 * Driver command data is immediately after the request. So add request to get
943 static inline void *blk_mq_rq_to_pdu(struct request *rq)
948 #define queue_for_each_hw_ctx(q, hctx, i) \
949 xa_for_each(&(q)->hctx_table, (i), (hctx))
951 #define hctx_for_each_ctx(hctx, ctx, i) \
952 for ((i) = 0; (i) < (hctx)->nr_ctx && \
953 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
955 static inline void blk_mq_cleanup_rq(struct request *rq)
957 if (rq->q->mq_ops->cleanup_rq)
958 rq->q->mq_ops->cleanup_rq(rq);
961 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
962 unsigned int nr_segs)
964 rq->nr_phys_segments = nr_segs;
965 rq->__data_len = bio->bi_iter.bi_size;
966 rq->bio = rq->biotail = bio;
967 rq->ioprio = bio_prio(bio);
970 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
971 struct lock_class_key *key);
973 static inline bool rq_is_sync(struct request *rq)
975 return op_is_sync(rq->cmd_flags);
978 void blk_rq_init(struct request_queue *q, struct request *rq);
979 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
980 struct bio_set *bs, gfp_t gfp_mask,
981 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
982 void blk_rq_unprep_clone(struct request *rq);
983 blk_status_t blk_insert_cloned_request(struct request *rq);
987 unsigned long offset;
988 unsigned short page_order;
989 unsigned short nr_entries;
994 int blk_rq_map_user(struct request_queue *, struct request *,
995 struct rq_map_data *, void __user *, unsigned long, gfp_t);
996 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
997 void __user *, unsigned long, gfp_t, bool, int, bool, int);
998 int blk_rq_map_user_iov(struct request_queue *, struct request *,
999 struct rq_map_data *, const struct iov_iter *, gfp_t);
1000 int blk_rq_unmap_user(struct bio *);
1001 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
1002 unsigned int, gfp_t);
1003 int blk_rq_append_bio(struct request *rq, struct bio *bio);
1004 void blk_execute_rq_nowait(struct request *rq, bool at_head);
1005 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1006 bool blk_rq_is_poll(struct request *rq);
1008 struct req_iterator {
1009 struct bvec_iter iter;
1013 #define __rq_for_each_bio(_bio, rq) \
1015 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1017 #define rq_for_each_segment(bvl, _rq, _iter) \
1018 __rq_for_each_bio(_iter.bio, _rq) \
1019 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1021 #define rq_for_each_bvec(bvl, _rq, _iter) \
1022 __rq_for_each_bio(_iter.bio, _rq) \
1023 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1025 #define rq_iter_last(bvec, _iter) \
1026 (_iter.bio->bi_next == NULL && \
1027 bio_iter_last(bvec, _iter.iter))
1030 * blk_rq_pos() : the current sector
1031 * blk_rq_bytes() : bytes left in the entire request
1032 * blk_rq_cur_bytes() : bytes left in the current segment
1033 * blk_rq_sectors() : sectors left in the entire request
1034 * blk_rq_cur_sectors() : sectors left in the current segment
1035 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1037 static inline sector_t blk_rq_pos(const struct request *rq)
1039 return rq->__sector;
1042 static inline unsigned int blk_rq_bytes(const struct request *rq)
1044 return rq->__data_len;
1047 static inline int blk_rq_cur_bytes(const struct request *rq)
1051 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1052 return rq->bio->bi_iter.bi_size;
1053 return bio_iovec(rq->bio).bv_len;
1056 static inline unsigned int blk_rq_sectors(const struct request *rq)
1058 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1061 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1063 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1066 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1068 return rq->stats_sectors;
1072 * Some commands like WRITE SAME have a payload or data transfer size which
1073 * is different from the size of the request. Any driver that supports such
1074 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1075 * calculate the data transfer size.
1077 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1079 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1080 return rq->special_vec.bv_len;
1081 return blk_rq_bytes(rq);
1085 * Return the first full biovec in the request. The caller needs to check that
1086 * there are any bvecs before calling this helper.
1088 static inline struct bio_vec req_bvec(struct request *rq)
1090 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1091 return rq->special_vec;
1092 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1095 static inline unsigned int blk_rq_count_bios(struct request *rq)
1097 unsigned int nr_bios = 0;
1100 __rq_for_each_bio(bio, rq)
1106 void blk_steal_bios(struct bio_list *list, struct request *rq);
1109 * Request completion related functions.
1111 * blk_update_request() completes given number of bytes and updates
1112 * the request without completing it.
1114 bool blk_update_request(struct request *rq, blk_status_t error,
1115 unsigned int nr_bytes);
1116 void blk_abort_request(struct request *);
1119 * Number of physical segments as sent to the device.
1121 * Normally this is the number of discontiguous data segments sent by the
1122 * submitter. But for data-less command like discard we might have no
1123 * actual data segments submitted, but the driver might have to add it's
1124 * own special payload. In that case we still return 1 here so that this
1125 * special payload will be mapped.
1127 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1129 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1131 return rq->nr_phys_segments;
1135 * Number of discard segments (or ranges) the driver needs to fill in.
1136 * Each discard bio merged into a request is counted as one segment.
1138 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1140 return max_t(unsigned short, rq->nr_phys_segments, 1);
1143 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1144 struct scatterlist *sglist, struct scatterlist **last_sg);
1145 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1146 struct scatterlist *sglist)
1148 struct scatterlist *last_sg = NULL;
1150 return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1152 void blk_dump_rq_flags(struct request *, char *);
1154 #ifdef CONFIG_BLK_DEV_ZONED
1155 static inline unsigned int blk_rq_zone_no(struct request *rq)
1157 return disk_zone_no(rq->q->disk, blk_rq_pos(rq));
1160 static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1162 return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq));
1165 bool blk_req_needs_zone_write_lock(struct request *rq);
1166 bool blk_req_zone_write_trylock(struct request *rq);
1167 void __blk_req_zone_write_lock(struct request *rq);
1168 void __blk_req_zone_write_unlock(struct request *rq);
1170 static inline void blk_req_zone_write_lock(struct request *rq)
1172 if (blk_req_needs_zone_write_lock(rq))
1173 __blk_req_zone_write_lock(rq);
1176 static inline void blk_req_zone_write_unlock(struct request *rq)
1178 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1179 __blk_req_zone_write_unlock(rq);
1182 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1184 return rq->q->disk->seq_zones_wlock &&
1185 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock);
1188 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1190 if (!blk_req_needs_zone_write_lock(rq))
1192 return !blk_req_zone_is_write_locked(rq);
1194 #else /* CONFIG_BLK_DEV_ZONED */
1195 static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1200 static inline void blk_req_zone_write_lock(struct request *rq)
1204 static inline void blk_req_zone_write_unlock(struct request *rq)
1207 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1212 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1216 #endif /* CONFIG_BLK_DEV_ZONED */
1218 #endif /* BLK_MQ_H */