xref: /linux-6.15/include/linux/blkdev.h (revision 06ed5c2b)
1 #ifndef _LINUX_BLKDEV_H
2 #define _LINUX_BLKDEV_H
3 
4 #include <linux/sched.h>
5 
6 #ifdef CONFIG_BLOCK
7 
8 #include <linux/major.h>
9 #include <linux/genhd.h>
10 #include <linux/list.h>
11 #include <linux/llist.h>
12 #include <linux/timer.h>
13 #include <linux/workqueue.h>
14 #include <linux/pagemap.h>
15 #include <linux/backing-dev.h>
16 #include <linux/wait.h>
17 #include <linux/mempool.h>
18 #include <linux/bio.h>
19 #include <linux/stringify.h>
20 #include <linux/gfp.h>
21 #include <linux/bsg.h>
22 #include <linux/smp.h>
23 #include <linux/rcupdate.h>
24 #include <linux/percpu-refcount.h>
25 
26 #include <asm/scatterlist.h>
27 
28 struct module;
29 struct scsi_ioctl_command;
30 
31 struct request_queue;
32 struct elevator_queue;
33 struct request_pm_state;
34 struct blk_trace;
35 struct request;
36 struct sg_io_hdr;
37 struct bsg_job;
38 struct blkcg_gq;
39 
40 #define BLKDEV_MIN_RQ	4
41 #define BLKDEV_MAX_RQ	128	/* Default maximum */
42 
43 /*
44  * Maximum number of blkcg policies allowed to be registered concurrently.
45  * Defined here to simplify include dependency.
46  */
47 #define BLKCG_MAX_POLS		2
48 
49 struct request;
50 typedef void (rq_end_io_fn)(struct request *, int);
51 
52 #define BLK_RL_SYNCFULL		(1U << 0)
53 #define BLK_RL_ASYNCFULL	(1U << 1)
54 
55 struct request_list {
56 	struct request_queue	*q;	/* the queue this rl belongs to */
57 #ifdef CONFIG_BLK_CGROUP
58 	struct blkcg_gq		*blkg;	/* blkg this request pool belongs to */
59 #endif
60 	/*
61 	 * count[], starved[], and wait[] are indexed by
62 	 * BLK_RW_SYNC/BLK_RW_ASYNC
63 	 */
64 	int			count[2];
65 	int			starved[2];
66 	mempool_t		*rq_pool;
67 	wait_queue_head_t	wait[2];
68 	unsigned int		flags;
69 };
70 
71 /*
72  * request command types
73  */
74 enum rq_cmd_type_bits {
75 	REQ_TYPE_FS		= 1,	/* fs request */
76 	REQ_TYPE_BLOCK_PC,		/* scsi command */
77 	REQ_TYPE_SENSE,			/* sense request */
78 	REQ_TYPE_PM_SUSPEND,		/* suspend request */
79 	REQ_TYPE_PM_RESUME,		/* resume request */
80 	REQ_TYPE_PM_SHUTDOWN,		/* shutdown request */
81 	REQ_TYPE_SPECIAL,		/* driver defined type */
82 	/*
83 	 * for ATA/ATAPI devices. this really doesn't belong here, ide should
84 	 * use REQ_TYPE_SPECIAL and use rq->cmd[0] with the range of driver
85 	 * private REQ_LB opcodes to differentiate what type of request this is
86 	 */
87 	REQ_TYPE_ATA_TASKFILE,
88 	REQ_TYPE_ATA_PC,
89 };
90 
91 #define BLK_MAX_CDB	16
92 
93 /*
94  * Try to put the fields that are referenced together in the same cacheline.
95  *
96  * If you modify this structure, make sure to update blk_rq_init() and
97  * especially blk_mq_rq_ctx_init() to take care of the added fields.
98  */
99 struct request {
100 	struct list_head queuelist;
101 	union {
102 		struct call_single_data csd;
103 		unsigned long fifo_time;
104 	};
105 
106 	struct request_queue *q;
107 	struct blk_mq_ctx *mq_ctx;
108 
109 	u64 cmd_flags;
110 	enum rq_cmd_type_bits cmd_type;
111 	unsigned long atomic_flags;
112 
113 	int cpu;
114 
115 	/* the following two fields are internal, NEVER access directly */
116 	unsigned int __data_len;	/* total data len */
117 	sector_t __sector;		/* sector cursor */
118 
119 	struct bio *bio;
120 	struct bio *biotail;
121 
122 	/*
123 	 * The hash is used inside the scheduler, and killed once the
124 	 * request reaches the dispatch list. The ipi_list is only used
125 	 * to queue the request for softirq completion, which is long
126 	 * after the request has been unhashed (and even removed from
127 	 * the dispatch list).
128 	 */
129 	union {
130 		struct hlist_node hash;	/* merge hash */
131 		struct list_head ipi_list;
132 	};
133 
134 	/*
135 	 * The rb_node is only used inside the io scheduler, requests
136 	 * are pruned when moved to the dispatch queue. So let the
137 	 * completion_data share space with the rb_node.
138 	 */
139 	union {
140 		struct rb_node rb_node;	/* sort/lookup */
141 		void *completion_data;
142 	};
143 
144 	/*
145 	 * Three pointers are available for the IO schedulers, if they need
146 	 * more they have to dynamically allocate it.  Flush requests are
147 	 * never put on the IO scheduler. So let the flush fields share
148 	 * space with the elevator data.
149 	 */
150 	union {
151 		struct {
152 			struct io_cq		*icq;
153 			void			*priv[2];
154 		} elv;
155 
156 		struct {
157 			unsigned int		seq;
158 			struct list_head	list;
159 			rq_end_io_fn		*saved_end_io;
160 		} flush;
161 	};
162 
163 	struct gendisk *rq_disk;
164 	struct hd_struct *part;
165 	unsigned long start_time;
166 #ifdef CONFIG_BLK_CGROUP
167 	struct request_list *rl;		/* rl this rq is alloced from */
168 	unsigned long long start_time_ns;
169 	unsigned long long io_start_time_ns;    /* when passed to hardware */
170 #endif
171 	/* Number of scatter-gather DMA addr+len pairs after
172 	 * physical address coalescing is performed.
173 	 */
174 	unsigned short nr_phys_segments;
175 #if defined(CONFIG_BLK_DEV_INTEGRITY)
176 	unsigned short nr_integrity_segments;
177 #endif
178 
179 	unsigned short ioprio;
180 
181 	void *special;		/* opaque pointer available for LLD use */
182 
183 	int tag;
184 	int errors;
185 
186 	/*
187 	 * when request is used as a packet command carrier
188 	 */
189 	unsigned char __cmd[BLK_MAX_CDB];
190 	unsigned char *cmd;
191 	unsigned short cmd_len;
192 
193 	unsigned int extra_len;	/* length of alignment and padding */
194 	unsigned int sense_len;
195 	unsigned int resid_len;	/* residual count */
196 	void *sense;
197 
198 	unsigned long deadline;
199 	struct list_head timeout_list;
200 	unsigned int timeout;
201 	int retries;
202 
203 	/*
204 	 * completion callback.
205 	 */
206 	rq_end_io_fn *end_io;
207 	void *end_io_data;
208 
209 	/* for bidi */
210 	struct request *next_rq;
211 };
212 
213 static inline unsigned short req_get_ioprio(struct request *req)
214 {
215 	return req->ioprio;
216 }
217 
218 /*
219  * State information carried for REQ_TYPE_PM_SUSPEND and REQ_TYPE_PM_RESUME
220  * requests. Some step values could eventually be made generic.
221  */
222 struct request_pm_state
223 {
224 	/* PM state machine step value, currently driver specific */
225 	int	pm_step;
226 	/* requested PM state value (S1, S2, S3, S4, ...) */
227 	u32	pm_state;
228 	void*	data;		/* for driver use */
229 };
230 
231 #include <linux/elevator.h>
232 
233 struct blk_queue_ctx;
234 
235 typedef void (request_fn_proc) (struct request_queue *q);
236 typedef void (make_request_fn) (struct request_queue *q, struct bio *bio);
237 typedef int (prep_rq_fn) (struct request_queue *, struct request *);
238 typedef void (unprep_rq_fn) (struct request_queue *, struct request *);
239 
240 struct bio_vec;
241 struct bvec_merge_data {
242 	struct block_device *bi_bdev;
243 	sector_t bi_sector;
244 	unsigned bi_size;
245 	unsigned long bi_rw;
246 };
247 typedef int (merge_bvec_fn) (struct request_queue *, struct bvec_merge_data *,
248 			     struct bio_vec *);
249 typedef void (softirq_done_fn)(struct request *);
250 typedef int (dma_drain_needed_fn)(struct request *);
251 typedef int (lld_busy_fn) (struct request_queue *q);
252 typedef int (bsg_job_fn) (struct bsg_job *);
253 
254 enum blk_eh_timer_return {
255 	BLK_EH_NOT_HANDLED,
256 	BLK_EH_HANDLED,
257 	BLK_EH_RESET_TIMER,
258 };
259 
260 typedef enum blk_eh_timer_return (rq_timed_out_fn)(struct request *);
261 
262 enum blk_queue_state {
263 	Queue_down,
264 	Queue_up,
265 };
266 
267 struct blk_queue_tag {
268 	struct request **tag_index;	/* map of busy tags */
269 	unsigned long *tag_map;		/* bit map of free/busy tags */
270 	int busy;			/* current depth */
271 	int max_depth;			/* what we will send to device */
272 	int real_max_depth;		/* what the array can hold */
273 	atomic_t refcnt;		/* map can be shared */
274 };
275 
276 #define BLK_SCSI_MAX_CMDS	(256)
277 #define BLK_SCSI_CMD_PER_LONG	(BLK_SCSI_MAX_CMDS / (sizeof(long) * 8))
278 
279 struct queue_limits {
280 	unsigned long		bounce_pfn;
281 	unsigned long		seg_boundary_mask;
282 
283 	unsigned int		max_hw_sectors;
284 	unsigned int		chunk_sectors;
285 	unsigned int		max_sectors;
286 	unsigned int		max_segment_size;
287 	unsigned int		physical_block_size;
288 	unsigned int		alignment_offset;
289 	unsigned int		io_min;
290 	unsigned int		io_opt;
291 	unsigned int		max_discard_sectors;
292 	unsigned int		max_write_same_sectors;
293 	unsigned int		discard_granularity;
294 	unsigned int		discard_alignment;
295 
296 	unsigned short		logical_block_size;
297 	unsigned short		max_segments;
298 	unsigned short		max_integrity_segments;
299 
300 	unsigned char		misaligned;
301 	unsigned char		discard_misaligned;
302 	unsigned char		cluster;
303 	unsigned char		discard_zeroes_data;
304 	unsigned char		raid_partial_stripes_expensive;
305 };
306 
307 struct request_queue {
308 	/*
309 	 * Together with queue_head for cacheline sharing
310 	 */
311 	struct list_head	queue_head;
312 	struct request		*last_merge;
313 	struct elevator_queue	*elevator;
314 	int			nr_rqs[2];	/* # allocated [a]sync rqs */
315 	int			nr_rqs_elvpriv;	/* # allocated rqs w/ elvpriv */
316 
317 	/*
318 	 * If blkcg is not used, @q->root_rl serves all requests.  If blkcg
319 	 * is used, root blkg allocates from @q->root_rl and all other
320 	 * blkgs from their own blkg->rl.  Which one to use should be
321 	 * determined using bio_request_list().
322 	 */
323 	struct request_list	root_rl;
324 
325 	request_fn_proc		*request_fn;
326 	make_request_fn		*make_request_fn;
327 	prep_rq_fn		*prep_rq_fn;
328 	unprep_rq_fn		*unprep_rq_fn;
329 	merge_bvec_fn		*merge_bvec_fn;
330 	softirq_done_fn		*softirq_done_fn;
331 	rq_timed_out_fn		*rq_timed_out_fn;
332 	dma_drain_needed_fn	*dma_drain_needed;
333 	lld_busy_fn		*lld_busy_fn;
334 
335 	struct blk_mq_ops	*mq_ops;
336 
337 	unsigned int		*mq_map;
338 
339 	/* sw queues */
340 	struct blk_mq_ctx __percpu	*queue_ctx;
341 	unsigned int		nr_queues;
342 
343 	/* hw dispatch queues */
344 	struct blk_mq_hw_ctx	**queue_hw_ctx;
345 	unsigned int		nr_hw_queues;
346 
347 	/*
348 	 * Dispatch queue sorting
349 	 */
350 	sector_t		end_sector;
351 	struct request		*boundary_rq;
352 
353 	/*
354 	 * Delayed queue handling
355 	 */
356 	struct delayed_work	delay_work;
357 
358 	struct backing_dev_info	backing_dev_info;
359 
360 	/*
361 	 * The queue owner gets to use this for whatever they like.
362 	 * ll_rw_blk doesn't touch it.
363 	 */
364 	void			*queuedata;
365 
366 	/*
367 	 * various queue flags, see QUEUE_* below
368 	 */
369 	unsigned long		queue_flags;
370 
371 	/*
372 	 * ida allocated id for this queue.  Used to index queues from
373 	 * ioctx.
374 	 */
375 	int			id;
376 
377 	/*
378 	 * queue needs bounce pages for pages above this limit
379 	 */
380 	gfp_t			bounce_gfp;
381 
382 	/*
383 	 * protects queue structures from reentrancy. ->__queue_lock should
384 	 * _never_ be used directly, it is queue private. always use
385 	 * ->queue_lock.
386 	 */
387 	spinlock_t		__queue_lock;
388 	spinlock_t		*queue_lock;
389 
390 	/*
391 	 * queue kobject
392 	 */
393 	struct kobject kobj;
394 
395 	/*
396 	 * mq queue kobject
397 	 */
398 	struct kobject mq_kobj;
399 
400 #ifdef CONFIG_PM_RUNTIME
401 	struct device		*dev;
402 	int			rpm_status;
403 	unsigned int		nr_pending;
404 #endif
405 
406 	/*
407 	 * queue settings
408 	 */
409 	unsigned long		nr_requests;	/* Max # of requests */
410 	unsigned int		nr_congestion_on;
411 	unsigned int		nr_congestion_off;
412 	unsigned int		nr_batching;
413 
414 	unsigned int		dma_drain_size;
415 	void			*dma_drain_buffer;
416 	unsigned int		dma_pad_mask;
417 	unsigned int		dma_alignment;
418 
419 	struct blk_queue_tag	*queue_tags;
420 	struct list_head	tag_busy_list;
421 
422 	unsigned int		nr_sorted;
423 	unsigned int		in_flight[2];
424 	/*
425 	 * Number of active block driver functions for which blk_drain_queue()
426 	 * must wait. Must be incremented around functions that unlock the
427 	 * queue_lock internally, e.g. scsi_request_fn().
428 	 */
429 	unsigned int		request_fn_active;
430 
431 	unsigned int		rq_timeout;
432 	struct timer_list	timeout;
433 	struct list_head	timeout_list;
434 
435 	struct list_head	icq_list;
436 #ifdef CONFIG_BLK_CGROUP
437 	DECLARE_BITMAP		(blkcg_pols, BLKCG_MAX_POLS);
438 	struct blkcg_gq		*root_blkg;
439 	struct list_head	blkg_list;
440 #endif
441 
442 	struct queue_limits	limits;
443 
444 	/*
445 	 * sg stuff
446 	 */
447 	unsigned int		sg_timeout;
448 	unsigned int		sg_reserved_size;
449 	int			node;
450 #ifdef CONFIG_BLK_DEV_IO_TRACE
451 	struct blk_trace	*blk_trace;
452 #endif
453 	/*
454 	 * for flush operations
455 	 */
456 	unsigned int		flush_flags;
457 	unsigned int		flush_not_queueable:1;
458 	unsigned int		flush_queue_delayed:1;
459 	unsigned int		flush_pending_idx:1;
460 	unsigned int		flush_running_idx:1;
461 	unsigned long		flush_pending_since;
462 	struct list_head	flush_queue[2];
463 	struct list_head	flush_data_in_flight;
464 	struct request		*flush_rq;
465 	spinlock_t		mq_flush_lock;
466 
467 	struct list_head	requeue_list;
468 	spinlock_t		requeue_lock;
469 	struct work_struct	requeue_work;
470 
471 	struct mutex		sysfs_lock;
472 
473 	int			bypass_depth;
474 	int			mq_freeze_depth;
475 
476 #if defined(CONFIG_BLK_DEV_BSG)
477 	bsg_job_fn		*bsg_job_fn;
478 	int			bsg_job_size;
479 	struct bsg_class_device bsg_dev;
480 #endif
481 
482 #ifdef CONFIG_BLK_DEV_THROTTLING
483 	/* Throttle data */
484 	struct throtl_data *td;
485 #endif
486 	struct rcu_head		rcu_head;
487 	wait_queue_head_t	mq_freeze_wq;
488 	struct percpu_ref	mq_usage_counter;
489 	struct list_head	all_q_node;
490 
491 	struct blk_mq_tag_set	*tag_set;
492 	struct list_head	tag_set_list;
493 };
494 
495 #define QUEUE_FLAG_QUEUED	1	/* uses generic tag queueing */
496 #define QUEUE_FLAG_STOPPED	2	/* queue is stopped */
497 #define	QUEUE_FLAG_SYNCFULL	3	/* read queue has been filled */
498 #define QUEUE_FLAG_ASYNCFULL	4	/* write queue has been filled */
499 #define QUEUE_FLAG_DYING	5	/* queue being torn down */
500 #define QUEUE_FLAG_BYPASS	6	/* act as dumb FIFO queue */
501 #define QUEUE_FLAG_BIDI		7	/* queue supports bidi requests */
502 #define QUEUE_FLAG_NOMERGES     8	/* disable merge attempts */
503 #define QUEUE_FLAG_SAME_COMP	9	/* complete on same CPU-group */
504 #define QUEUE_FLAG_FAIL_IO     10	/* fake timeout */
505 #define QUEUE_FLAG_STACKABLE   11	/* supports request stacking */
506 #define QUEUE_FLAG_NONROT      12	/* non-rotational device (SSD) */
507 #define QUEUE_FLAG_VIRT        QUEUE_FLAG_NONROT /* paravirt device */
508 #define QUEUE_FLAG_IO_STAT     13	/* do IO stats */
509 #define QUEUE_FLAG_DISCARD     14	/* supports DISCARD */
510 #define QUEUE_FLAG_NOXMERGES   15	/* No extended merges */
511 #define QUEUE_FLAG_ADD_RANDOM  16	/* Contributes to random pool */
512 #define QUEUE_FLAG_SECDISCARD  17	/* supports SECDISCARD */
513 #define QUEUE_FLAG_SAME_FORCE  18	/* force complete on same CPU */
514 #define QUEUE_FLAG_DEAD        19	/* queue tear-down finished */
515 #define QUEUE_FLAG_INIT_DONE   20	/* queue is initialized */
516 #define QUEUE_FLAG_NO_SG_MERGE 21	/* don't attempt to merge SG segments*/
517 #define QUEUE_FLAG_SG_GAPS     22	/* queue doesn't support SG gaps */
518 
519 #define QUEUE_FLAG_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
520 				 (1 << QUEUE_FLAG_STACKABLE)	|	\
521 				 (1 << QUEUE_FLAG_SAME_COMP)	|	\
522 				 (1 << QUEUE_FLAG_ADD_RANDOM))
523 
524 #define QUEUE_FLAG_MQ_DEFAULT	((1 << QUEUE_FLAG_IO_STAT) |		\
525 				 (1 << QUEUE_FLAG_SAME_COMP))
526 
527 static inline void queue_lockdep_assert_held(struct request_queue *q)
528 {
529 	if (q->queue_lock)
530 		lockdep_assert_held(q->queue_lock);
531 }
532 
533 static inline void queue_flag_set_unlocked(unsigned int flag,
534 					   struct request_queue *q)
535 {
536 	__set_bit(flag, &q->queue_flags);
537 }
538 
539 static inline int queue_flag_test_and_clear(unsigned int flag,
540 					    struct request_queue *q)
541 {
542 	queue_lockdep_assert_held(q);
543 
544 	if (test_bit(flag, &q->queue_flags)) {
545 		__clear_bit(flag, &q->queue_flags);
546 		return 1;
547 	}
548 
549 	return 0;
550 }
551 
552 static inline int queue_flag_test_and_set(unsigned int flag,
553 					  struct request_queue *q)
554 {
555 	queue_lockdep_assert_held(q);
556 
557 	if (!test_bit(flag, &q->queue_flags)) {
558 		__set_bit(flag, &q->queue_flags);
559 		return 0;
560 	}
561 
562 	return 1;
563 }
564 
565 static inline void queue_flag_set(unsigned int flag, struct request_queue *q)
566 {
567 	queue_lockdep_assert_held(q);
568 	__set_bit(flag, &q->queue_flags);
569 }
570 
571 static inline void queue_flag_clear_unlocked(unsigned int flag,
572 					     struct request_queue *q)
573 {
574 	__clear_bit(flag, &q->queue_flags);
575 }
576 
577 static inline int queue_in_flight(struct request_queue *q)
578 {
579 	return q->in_flight[0] + q->in_flight[1];
580 }
581 
582 static inline void queue_flag_clear(unsigned int flag, struct request_queue *q)
583 {
584 	queue_lockdep_assert_held(q);
585 	__clear_bit(flag, &q->queue_flags);
586 }
587 
588 #define blk_queue_tagged(q)	test_bit(QUEUE_FLAG_QUEUED, &(q)->queue_flags)
589 #define blk_queue_stopped(q)	test_bit(QUEUE_FLAG_STOPPED, &(q)->queue_flags)
590 #define blk_queue_dying(q)	test_bit(QUEUE_FLAG_DYING, &(q)->queue_flags)
591 #define blk_queue_dead(q)	test_bit(QUEUE_FLAG_DEAD, &(q)->queue_flags)
592 #define blk_queue_bypass(q)	test_bit(QUEUE_FLAG_BYPASS, &(q)->queue_flags)
593 #define blk_queue_init_done(q)	test_bit(QUEUE_FLAG_INIT_DONE, &(q)->queue_flags)
594 #define blk_queue_nomerges(q)	test_bit(QUEUE_FLAG_NOMERGES, &(q)->queue_flags)
595 #define blk_queue_noxmerges(q)	\
596 	test_bit(QUEUE_FLAG_NOXMERGES, &(q)->queue_flags)
597 #define blk_queue_nonrot(q)	test_bit(QUEUE_FLAG_NONROT, &(q)->queue_flags)
598 #define blk_queue_io_stat(q)	test_bit(QUEUE_FLAG_IO_STAT, &(q)->queue_flags)
599 #define blk_queue_add_random(q)	test_bit(QUEUE_FLAG_ADD_RANDOM, &(q)->queue_flags)
600 #define blk_queue_stackable(q)	\
601 	test_bit(QUEUE_FLAG_STACKABLE, &(q)->queue_flags)
602 #define blk_queue_discard(q)	test_bit(QUEUE_FLAG_DISCARD, &(q)->queue_flags)
603 #define blk_queue_secdiscard(q)	(blk_queue_discard(q) && \
604 	test_bit(QUEUE_FLAG_SECDISCARD, &(q)->queue_flags))
605 
606 #define blk_noretry_request(rq) \
607 	((rq)->cmd_flags & (REQ_FAILFAST_DEV|REQ_FAILFAST_TRANSPORT| \
608 			     REQ_FAILFAST_DRIVER))
609 
610 #define blk_account_rq(rq) \
611 	(((rq)->cmd_flags & REQ_STARTED) && \
612 	 ((rq)->cmd_type == REQ_TYPE_FS))
613 
614 #define blk_pm_request(rq)	\
615 	((rq)->cmd_type == REQ_TYPE_PM_SUSPEND || \
616 	 (rq)->cmd_type == REQ_TYPE_PM_RESUME)
617 
618 #define blk_rq_cpu_valid(rq)	((rq)->cpu != -1)
619 #define blk_bidi_rq(rq)		((rq)->next_rq != NULL)
620 /* rq->queuelist of dequeued request must be list_empty() */
621 #define blk_queued_rq(rq)	(!list_empty(&(rq)->queuelist))
622 
623 #define list_entry_rq(ptr)	list_entry((ptr), struct request, queuelist)
624 
625 #define rq_data_dir(rq)		(((rq)->cmd_flags & 1) != 0)
626 
627 /*
628  * Driver can handle struct request, if it either has an old style
629  * request_fn defined, or is blk-mq based.
630  */
631 static inline bool queue_is_rq_based(struct request_queue *q)
632 {
633 	return q->request_fn || q->mq_ops;
634 }
635 
636 static inline unsigned int blk_queue_cluster(struct request_queue *q)
637 {
638 	return q->limits.cluster;
639 }
640 
641 /*
642  * We regard a request as sync, if either a read or a sync write
643  */
644 static inline bool rw_is_sync(unsigned int rw_flags)
645 {
646 	return !(rw_flags & REQ_WRITE) || (rw_flags & REQ_SYNC);
647 }
648 
649 static inline bool rq_is_sync(struct request *rq)
650 {
651 	return rw_is_sync(rq->cmd_flags);
652 }
653 
654 static inline bool blk_rl_full(struct request_list *rl, bool sync)
655 {
656 	unsigned int flag = sync ? BLK_RL_SYNCFULL : BLK_RL_ASYNCFULL;
657 
658 	return rl->flags & flag;
659 }
660 
661 static inline void blk_set_rl_full(struct request_list *rl, bool sync)
662 {
663 	unsigned int flag = sync ? BLK_RL_SYNCFULL : BLK_RL_ASYNCFULL;
664 
665 	rl->flags |= flag;
666 }
667 
668 static inline void blk_clear_rl_full(struct request_list *rl, bool sync)
669 {
670 	unsigned int flag = sync ? BLK_RL_SYNCFULL : BLK_RL_ASYNCFULL;
671 
672 	rl->flags &= ~flag;
673 }
674 
675 static inline bool rq_mergeable(struct request *rq)
676 {
677 	if (rq->cmd_type != REQ_TYPE_FS)
678 		return false;
679 
680 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
681 		return false;
682 
683 	return true;
684 }
685 
686 static inline bool blk_check_merge_flags(unsigned int flags1,
687 					 unsigned int flags2)
688 {
689 	if ((flags1 & REQ_DISCARD) != (flags2 & REQ_DISCARD))
690 		return false;
691 
692 	if ((flags1 & REQ_SECURE) != (flags2 & REQ_SECURE))
693 		return false;
694 
695 	if ((flags1 & REQ_WRITE_SAME) != (flags2 & REQ_WRITE_SAME))
696 		return false;
697 
698 	return true;
699 }
700 
701 static inline bool blk_write_same_mergeable(struct bio *a, struct bio *b)
702 {
703 	if (bio_data(a) == bio_data(b))
704 		return true;
705 
706 	return false;
707 }
708 
709 /*
710  * q->prep_rq_fn return values
711  */
712 #define BLKPREP_OK		0	/* serve it */
713 #define BLKPREP_KILL		1	/* fatal error, kill */
714 #define BLKPREP_DEFER		2	/* leave on queue */
715 
716 extern unsigned long blk_max_low_pfn, blk_max_pfn;
717 
718 /*
719  * standard bounce addresses:
720  *
721  * BLK_BOUNCE_HIGH	: bounce all highmem pages
722  * BLK_BOUNCE_ANY	: don't bounce anything
723  * BLK_BOUNCE_ISA	: bounce pages above ISA DMA boundary
724  */
725 
726 #if BITS_PER_LONG == 32
727 #define BLK_BOUNCE_HIGH		((u64)blk_max_low_pfn << PAGE_SHIFT)
728 #else
729 #define BLK_BOUNCE_HIGH		-1ULL
730 #endif
731 #define BLK_BOUNCE_ANY		(-1ULL)
732 #define BLK_BOUNCE_ISA		(DMA_BIT_MASK(24))
733 
734 /*
735  * default timeout for SG_IO if none specified
736  */
737 #define BLK_DEFAULT_SG_TIMEOUT	(60 * HZ)
738 #define BLK_MIN_SG_TIMEOUT	(7 * HZ)
739 
740 #ifdef CONFIG_BOUNCE
741 extern int init_emergency_isa_pool(void);
742 extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
743 #else
744 static inline int init_emergency_isa_pool(void)
745 {
746 	return 0;
747 }
748 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
749 {
750 }
751 #endif /* CONFIG_MMU */
752 
753 struct rq_map_data {
754 	struct page **pages;
755 	int page_order;
756 	int nr_entries;
757 	unsigned long offset;
758 	int null_mapped;
759 	int from_user;
760 };
761 
762 struct req_iterator {
763 	struct bvec_iter iter;
764 	struct bio *bio;
765 };
766 
767 /* This should not be used directly - use rq_for_each_segment */
768 #define for_each_bio(_bio)		\
769 	for (; _bio; _bio = _bio->bi_next)
770 #define __rq_for_each_bio(_bio, rq)	\
771 	if ((rq->bio))			\
772 		for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
773 
774 #define rq_for_each_segment(bvl, _rq, _iter)			\
775 	__rq_for_each_bio(_iter.bio, _rq)			\
776 		bio_for_each_segment(bvl, _iter.bio, _iter.iter)
777 
778 #define rq_iter_last(bvec, _iter)				\
779 		(_iter.bio->bi_next == NULL &&			\
780 		 bio_iter_last(bvec, _iter.iter))
781 
782 #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
783 # error	"You should define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE for your platform"
784 #endif
785 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
786 extern void rq_flush_dcache_pages(struct request *rq);
787 #else
788 static inline void rq_flush_dcache_pages(struct request *rq)
789 {
790 }
791 #endif
792 
793 extern int blk_register_queue(struct gendisk *disk);
794 extern void blk_unregister_queue(struct gendisk *disk);
795 extern void generic_make_request(struct bio *bio);
796 extern void blk_rq_init(struct request_queue *q, struct request *rq);
797 extern void blk_put_request(struct request *);
798 extern void __blk_put_request(struct request_queue *, struct request *);
799 extern struct request *blk_get_request(struct request_queue *, int, gfp_t);
800 extern struct request *blk_make_request(struct request_queue *, struct bio *,
801 					gfp_t);
802 extern void blk_rq_set_block_pc(struct request *);
803 extern void blk_requeue_request(struct request_queue *, struct request *);
804 extern void blk_add_request_payload(struct request *rq, struct page *page,
805 		unsigned int len);
806 extern int blk_rq_check_limits(struct request_queue *q, struct request *rq);
807 extern int blk_lld_busy(struct request_queue *q);
808 extern int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
809 			     struct bio_set *bs, gfp_t gfp_mask,
810 			     int (*bio_ctr)(struct bio *, struct bio *, void *),
811 			     void *data);
812 extern void blk_rq_unprep_clone(struct request *rq);
813 extern int blk_insert_cloned_request(struct request_queue *q,
814 				     struct request *rq);
815 extern void blk_delay_queue(struct request_queue *, unsigned long);
816 extern void blk_recount_segments(struct request_queue *, struct bio *);
817 extern int scsi_verify_blk_ioctl(struct block_device *, unsigned int);
818 extern int scsi_cmd_blk_ioctl(struct block_device *, fmode_t,
819 			      unsigned int, void __user *);
820 extern int scsi_cmd_ioctl(struct request_queue *, struct gendisk *, fmode_t,
821 			  unsigned int, void __user *);
822 extern int sg_scsi_ioctl(struct request_queue *, struct gendisk *, fmode_t,
823 			 struct scsi_ioctl_command __user *);
824 
825 extern void blk_queue_bio(struct request_queue *q, struct bio *bio);
826 
827 /*
828  * A queue has just exitted congestion.  Note this in the global counter of
829  * congested queues, and wake up anyone who was waiting for requests to be
830  * put back.
831  */
832 static inline void blk_clear_queue_congested(struct request_queue *q, int sync)
833 {
834 	clear_bdi_congested(&q->backing_dev_info, sync);
835 }
836 
837 /*
838  * A queue has just entered congestion.  Flag that in the queue's VM-visible
839  * state flags and increment the global gounter of congested queues.
840  */
841 static inline void blk_set_queue_congested(struct request_queue *q, int sync)
842 {
843 	set_bdi_congested(&q->backing_dev_info, sync);
844 }
845 
846 extern void blk_start_queue(struct request_queue *q);
847 extern void blk_stop_queue(struct request_queue *q);
848 extern void blk_sync_queue(struct request_queue *q);
849 extern void __blk_stop_queue(struct request_queue *q);
850 extern void __blk_run_queue(struct request_queue *q);
851 extern void blk_run_queue(struct request_queue *);
852 extern void blk_run_queue_async(struct request_queue *q);
853 extern int blk_rq_map_user(struct request_queue *, struct request *,
854 			   struct rq_map_data *, void __user *, unsigned long,
855 			   gfp_t);
856 extern int blk_rq_unmap_user(struct bio *);
857 extern int blk_rq_map_kern(struct request_queue *, struct request *, void *, unsigned int, gfp_t);
858 extern int blk_rq_map_user_iov(struct request_queue *, struct request *,
859 			       struct rq_map_data *, const struct sg_iovec *,
860 			       int, unsigned int, gfp_t);
861 extern int blk_execute_rq(struct request_queue *, struct gendisk *,
862 			  struct request *, int);
863 extern void blk_execute_rq_nowait(struct request_queue *, struct gendisk *,
864 				  struct request *, int, rq_end_io_fn *);
865 
866 static inline struct request_queue *bdev_get_queue(struct block_device *bdev)
867 {
868 	return bdev->bd_disk->queue;
869 }
870 
871 /*
872  * blk_rq_pos()			: the current sector
873  * blk_rq_bytes()		: bytes left in the entire request
874  * blk_rq_cur_bytes()		: bytes left in the current segment
875  * blk_rq_err_bytes()		: bytes left till the next error boundary
876  * blk_rq_sectors()		: sectors left in the entire request
877  * blk_rq_cur_sectors()		: sectors left in the current segment
878  */
879 static inline sector_t blk_rq_pos(const struct request *rq)
880 {
881 	return rq->__sector;
882 }
883 
884 static inline unsigned int blk_rq_bytes(const struct request *rq)
885 {
886 	return rq->__data_len;
887 }
888 
889 static inline int blk_rq_cur_bytes(const struct request *rq)
890 {
891 	return rq->bio ? bio_cur_bytes(rq->bio) : 0;
892 }
893 
894 extern unsigned int blk_rq_err_bytes(const struct request *rq);
895 
896 static inline unsigned int blk_rq_sectors(const struct request *rq)
897 {
898 	return blk_rq_bytes(rq) >> 9;
899 }
900 
901 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
902 {
903 	return blk_rq_cur_bytes(rq) >> 9;
904 }
905 
906 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
907 						     unsigned int cmd_flags)
908 {
909 	if (unlikely(cmd_flags & REQ_DISCARD))
910 		return min(q->limits.max_discard_sectors, UINT_MAX >> 9);
911 
912 	if (unlikely(cmd_flags & REQ_WRITE_SAME))
913 		return q->limits.max_write_same_sectors;
914 
915 	return q->limits.max_sectors;
916 }
917 
918 /*
919  * Return maximum size of a request at given offset. Only valid for
920  * file system requests.
921  */
922 static inline unsigned int blk_max_size_offset(struct request_queue *q,
923 					       sector_t offset)
924 {
925 	if (!q->limits.chunk_sectors)
926 		return q->limits.max_sectors;
927 
928 	return q->limits.chunk_sectors -
929 			(offset & (q->limits.chunk_sectors - 1));
930 }
931 
932 static inline unsigned int blk_rq_get_max_sectors(struct request *rq)
933 {
934 	struct request_queue *q = rq->q;
935 
936 	if (unlikely(rq->cmd_type == REQ_TYPE_BLOCK_PC))
937 		return q->limits.max_hw_sectors;
938 
939 	if (!q->limits.chunk_sectors)
940 		return blk_queue_get_max_sectors(q, rq->cmd_flags);
941 
942 	return min(blk_max_size_offset(q, blk_rq_pos(rq)),
943 			blk_queue_get_max_sectors(q, rq->cmd_flags));
944 }
945 
946 static inline unsigned int blk_rq_count_bios(struct request *rq)
947 {
948 	unsigned int nr_bios = 0;
949 	struct bio *bio;
950 
951 	__rq_for_each_bio(bio, rq)
952 		nr_bios++;
953 
954 	return nr_bios;
955 }
956 
957 /*
958  * Request issue related functions.
959  */
960 extern struct request *blk_peek_request(struct request_queue *q);
961 extern void blk_start_request(struct request *rq);
962 extern struct request *blk_fetch_request(struct request_queue *q);
963 
964 /*
965  * Request completion related functions.
966  *
967  * blk_update_request() completes given number of bytes and updates
968  * the request without completing it.
969  *
970  * blk_end_request() and friends.  __blk_end_request() must be called
971  * with the request queue spinlock acquired.
972  *
973  * Several drivers define their own end_request and call
974  * blk_end_request() for parts of the original function.
975  * This prevents code duplication in drivers.
976  */
977 extern bool blk_update_request(struct request *rq, int error,
978 			       unsigned int nr_bytes);
979 extern void blk_finish_request(struct request *rq, int error);
980 extern bool blk_end_request(struct request *rq, int error,
981 			    unsigned int nr_bytes);
982 extern void blk_end_request_all(struct request *rq, int error);
983 extern bool blk_end_request_cur(struct request *rq, int error);
984 extern bool blk_end_request_err(struct request *rq, int error);
985 extern bool __blk_end_request(struct request *rq, int error,
986 			      unsigned int nr_bytes);
987 extern void __blk_end_request_all(struct request *rq, int error);
988 extern bool __blk_end_request_cur(struct request *rq, int error);
989 extern bool __blk_end_request_err(struct request *rq, int error);
990 
991 extern void blk_complete_request(struct request *);
992 extern void __blk_complete_request(struct request *);
993 extern void blk_abort_request(struct request *);
994 extern void blk_unprep_request(struct request *);
995 
996 /*
997  * Access functions for manipulating queue properties
998  */
999 extern struct request_queue *blk_init_queue_node(request_fn_proc *rfn,
1000 					spinlock_t *lock, int node_id);
1001 extern struct request_queue *blk_init_queue(request_fn_proc *, spinlock_t *);
1002 extern struct request_queue *blk_init_allocated_queue(struct request_queue *,
1003 						      request_fn_proc *, spinlock_t *);
1004 extern void blk_cleanup_queue(struct request_queue *);
1005 extern void blk_queue_make_request(struct request_queue *, make_request_fn *);
1006 extern void blk_queue_bounce_limit(struct request_queue *, u64);
1007 extern void blk_limits_max_hw_sectors(struct queue_limits *, unsigned int);
1008 extern void blk_queue_max_hw_sectors(struct request_queue *, unsigned int);
1009 extern void blk_queue_chunk_sectors(struct request_queue *, unsigned int);
1010 extern void blk_queue_max_segments(struct request_queue *, unsigned short);
1011 extern void blk_queue_max_segment_size(struct request_queue *, unsigned int);
1012 extern void blk_queue_max_discard_sectors(struct request_queue *q,
1013 		unsigned int max_discard_sectors);
1014 extern void blk_queue_max_write_same_sectors(struct request_queue *q,
1015 		unsigned int max_write_same_sectors);
1016 extern void blk_queue_logical_block_size(struct request_queue *, unsigned short);
1017 extern void blk_queue_physical_block_size(struct request_queue *, unsigned int);
1018 extern void blk_queue_alignment_offset(struct request_queue *q,
1019 				       unsigned int alignment);
1020 extern void blk_limits_io_min(struct queue_limits *limits, unsigned int min);
1021 extern void blk_queue_io_min(struct request_queue *q, unsigned int min);
1022 extern void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt);
1023 extern void blk_queue_io_opt(struct request_queue *q, unsigned int opt);
1024 extern void blk_set_default_limits(struct queue_limits *lim);
1025 extern void blk_set_stacking_limits(struct queue_limits *lim);
1026 extern int blk_stack_limits(struct queue_limits *t, struct queue_limits *b,
1027 			    sector_t offset);
1028 extern int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev,
1029 			    sector_t offset);
1030 extern void disk_stack_limits(struct gendisk *disk, struct block_device *bdev,
1031 			      sector_t offset);
1032 extern void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b);
1033 extern void blk_queue_dma_pad(struct request_queue *, unsigned int);
1034 extern void blk_queue_update_dma_pad(struct request_queue *, unsigned int);
1035 extern int blk_queue_dma_drain(struct request_queue *q,
1036 			       dma_drain_needed_fn *dma_drain_needed,
1037 			       void *buf, unsigned int size);
1038 extern void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn);
1039 extern void blk_queue_segment_boundary(struct request_queue *, unsigned long);
1040 extern void blk_queue_prep_rq(struct request_queue *, prep_rq_fn *pfn);
1041 extern void blk_queue_unprep_rq(struct request_queue *, unprep_rq_fn *ufn);
1042 extern void blk_queue_merge_bvec(struct request_queue *, merge_bvec_fn *);
1043 extern void blk_queue_dma_alignment(struct request_queue *, int);
1044 extern void blk_queue_update_dma_alignment(struct request_queue *, int);
1045 extern void blk_queue_softirq_done(struct request_queue *, softirq_done_fn *);
1046 extern void blk_queue_rq_timed_out(struct request_queue *, rq_timed_out_fn *);
1047 extern void blk_queue_rq_timeout(struct request_queue *, unsigned int);
1048 extern void blk_queue_flush(struct request_queue *q, unsigned int flush);
1049 extern void blk_queue_flush_queueable(struct request_queue *q, bool queueable);
1050 extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev);
1051 
1052 extern int blk_rq_map_sg(struct request_queue *, struct request *, struct scatterlist *);
1053 extern int blk_bio_map_sg(struct request_queue *q, struct bio *bio,
1054 			  struct scatterlist *sglist);
1055 extern void blk_dump_rq_flags(struct request *, char *);
1056 extern long nr_blockdev_pages(void);
1057 
1058 bool __must_check blk_get_queue(struct request_queue *);
1059 struct request_queue *blk_alloc_queue(gfp_t);
1060 struct request_queue *blk_alloc_queue_node(gfp_t, int);
1061 extern void blk_put_queue(struct request_queue *);
1062 
1063 /*
1064  * block layer runtime pm functions
1065  */
1066 #ifdef CONFIG_PM_RUNTIME
1067 extern void blk_pm_runtime_init(struct request_queue *q, struct device *dev);
1068 extern int blk_pre_runtime_suspend(struct request_queue *q);
1069 extern void blk_post_runtime_suspend(struct request_queue *q, int err);
1070 extern void blk_pre_runtime_resume(struct request_queue *q);
1071 extern void blk_post_runtime_resume(struct request_queue *q, int err);
1072 #else
1073 static inline void blk_pm_runtime_init(struct request_queue *q,
1074 	struct device *dev) {}
1075 static inline int blk_pre_runtime_suspend(struct request_queue *q)
1076 {
1077 	return -ENOSYS;
1078 }
1079 static inline void blk_post_runtime_suspend(struct request_queue *q, int err) {}
1080 static inline void blk_pre_runtime_resume(struct request_queue *q) {}
1081 static inline void blk_post_runtime_resume(struct request_queue *q, int err) {}
1082 #endif
1083 
1084 /*
1085  * blk_plug permits building a queue of related requests by holding the I/O
1086  * fragments for a short period. This allows merging of sequential requests
1087  * into single larger request. As the requests are moved from a per-task list to
1088  * the device's request_queue in a batch, this results in improved scalability
1089  * as the lock contention for request_queue lock is reduced.
1090  *
1091  * It is ok not to disable preemption when adding the request to the plug list
1092  * or when attempting a merge, because blk_schedule_flush_list() will only flush
1093  * the plug list when the task sleeps by itself. For details, please see
1094  * schedule() where blk_schedule_flush_plug() is called.
1095  */
1096 struct blk_plug {
1097 	struct list_head list; /* requests */
1098 	struct list_head mq_list; /* blk-mq requests */
1099 	struct list_head cb_list; /* md requires an unplug callback */
1100 };
1101 #define BLK_MAX_REQUEST_COUNT 16
1102 
1103 struct blk_plug_cb;
1104 typedef void (*blk_plug_cb_fn)(struct blk_plug_cb *, bool);
1105 struct blk_plug_cb {
1106 	struct list_head list;
1107 	blk_plug_cb_fn callback;
1108 	void *data;
1109 };
1110 extern struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug,
1111 					     void *data, int size);
1112 extern void blk_start_plug(struct blk_plug *);
1113 extern void blk_finish_plug(struct blk_plug *);
1114 extern void blk_flush_plug_list(struct blk_plug *, bool);
1115 
1116 static inline void blk_flush_plug(struct task_struct *tsk)
1117 {
1118 	struct blk_plug *plug = tsk->plug;
1119 
1120 	if (plug)
1121 		blk_flush_plug_list(plug, false);
1122 }
1123 
1124 static inline void blk_schedule_flush_plug(struct task_struct *tsk)
1125 {
1126 	struct blk_plug *plug = tsk->plug;
1127 
1128 	if (plug)
1129 		blk_flush_plug_list(plug, true);
1130 }
1131 
1132 static inline bool blk_needs_flush_plug(struct task_struct *tsk)
1133 {
1134 	struct blk_plug *plug = tsk->plug;
1135 
1136 	return plug &&
1137 		(!list_empty(&plug->list) ||
1138 		 !list_empty(&plug->mq_list) ||
1139 		 !list_empty(&plug->cb_list));
1140 }
1141 
1142 /*
1143  * tag stuff
1144  */
1145 #define blk_rq_tagged(rq) \
1146 	((rq)->mq_ctx || ((rq)->cmd_flags & REQ_QUEUED))
1147 extern int blk_queue_start_tag(struct request_queue *, struct request *);
1148 extern struct request *blk_queue_find_tag(struct request_queue *, int);
1149 extern void blk_queue_end_tag(struct request_queue *, struct request *);
1150 extern int blk_queue_init_tags(struct request_queue *, int, struct blk_queue_tag *);
1151 extern void blk_queue_free_tags(struct request_queue *);
1152 extern int blk_queue_resize_tags(struct request_queue *, int);
1153 extern void blk_queue_invalidate_tags(struct request_queue *);
1154 extern struct blk_queue_tag *blk_init_tags(int);
1155 extern void blk_free_tags(struct blk_queue_tag *);
1156 
1157 static inline struct request *blk_map_queue_find_tag(struct blk_queue_tag *bqt,
1158 						int tag)
1159 {
1160 	if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
1161 		return NULL;
1162 	return bqt->tag_index[tag];
1163 }
1164 
1165 #define BLKDEV_DISCARD_SECURE  0x01    /* secure discard */
1166 
1167 extern int blkdev_issue_flush(struct block_device *, gfp_t, sector_t *);
1168 extern int blkdev_issue_discard(struct block_device *bdev, sector_t sector,
1169 		sector_t nr_sects, gfp_t gfp_mask, unsigned long flags);
1170 extern int blkdev_issue_write_same(struct block_device *bdev, sector_t sector,
1171 		sector_t nr_sects, gfp_t gfp_mask, struct page *page);
1172 extern int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
1173 			sector_t nr_sects, gfp_t gfp_mask);
1174 static inline int sb_issue_discard(struct super_block *sb, sector_t block,
1175 		sector_t nr_blocks, gfp_t gfp_mask, unsigned long flags)
1176 {
1177 	return blkdev_issue_discard(sb->s_bdev, block << (sb->s_blocksize_bits - 9),
1178 				    nr_blocks << (sb->s_blocksize_bits - 9),
1179 				    gfp_mask, flags);
1180 }
1181 static inline int sb_issue_zeroout(struct super_block *sb, sector_t block,
1182 		sector_t nr_blocks, gfp_t gfp_mask)
1183 {
1184 	return blkdev_issue_zeroout(sb->s_bdev,
1185 				    block << (sb->s_blocksize_bits - 9),
1186 				    nr_blocks << (sb->s_blocksize_bits - 9),
1187 				    gfp_mask);
1188 }
1189 
1190 extern int blk_verify_command(unsigned char *cmd, fmode_t has_write_perm);
1191 
1192 enum blk_default_limits {
1193 	BLK_MAX_SEGMENTS	= 128,
1194 	BLK_SAFE_MAX_SECTORS	= 255,
1195 	BLK_DEF_MAX_SECTORS	= 1024,
1196 	BLK_MAX_SEGMENT_SIZE	= 65536,
1197 	BLK_SEG_BOUNDARY_MASK	= 0xFFFFFFFFUL,
1198 };
1199 
1200 #define blkdev_entry_to_request(entry) list_entry((entry), struct request, queuelist)
1201 
1202 static inline unsigned long queue_bounce_pfn(struct request_queue *q)
1203 {
1204 	return q->limits.bounce_pfn;
1205 }
1206 
1207 static inline unsigned long queue_segment_boundary(struct request_queue *q)
1208 {
1209 	return q->limits.seg_boundary_mask;
1210 }
1211 
1212 static inline unsigned int queue_max_sectors(struct request_queue *q)
1213 {
1214 	return q->limits.max_sectors;
1215 }
1216 
1217 static inline unsigned int queue_max_hw_sectors(struct request_queue *q)
1218 {
1219 	return q->limits.max_hw_sectors;
1220 }
1221 
1222 static inline unsigned short queue_max_segments(struct request_queue *q)
1223 {
1224 	return q->limits.max_segments;
1225 }
1226 
1227 static inline unsigned int queue_max_segment_size(struct request_queue *q)
1228 {
1229 	return q->limits.max_segment_size;
1230 }
1231 
1232 static inline unsigned short queue_logical_block_size(struct request_queue *q)
1233 {
1234 	int retval = 512;
1235 
1236 	if (q && q->limits.logical_block_size)
1237 		retval = q->limits.logical_block_size;
1238 
1239 	return retval;
1240 }
1241 
1242 static inline unsigned short bdev_logical_block_size(struct block_device *bdev)
1243 {
1244 	return queue_logical_block_size(bdev_get_queue(bdev));
1245 }
1246 
1247 static inline unsigned int queue_physical_block_size(struct request_queue *q)
1248 {
1249 	return q->limits.physical_block_size;
1250 }
1251 
1252 static inline unsigned int bdev_physical_block_size(struct block_device *bdev)
1253 {
1254 	return queue_physical_block_size(bdev_get_queue(bdev));
1255 }
1256 
1257 static inline unsigned int queue_io_min(struct request_queue *q)
1258 {
1259 	return q->limits.io_min;
1260 }
1261 
1262 static inline int bdev_io_min(struct block_device *bdev)
1263 {
1264 	return queue_io_min(bdev_get_queue(bdev));
1265 }
1266 
1267 static inline unsigned int queue_io_opt(struct request_queue *q)
1268 {
1269 	return q->limits.io_opt;
1270 }
1271 
1272 static inline int bdev_io_opt(struct block_device *bdev)
1273 {
1274 	return queue_io_opt(bdev_get_queue(bdev));
1275 }
1276 
1277 static inline int queue_alignment_offset(struct request_queue *q)
1278 {
1279 	if (q->limits.misaligned)
1280 		return -1;
1281 
1282 	return q->limits.alignment_offset;
1283 }
1284 
1285 static inline int queue_limit_alignment_offset(struct queue_limits *lim, sector_t sector)
1286 {
1287 	unsigned int granularity = max(lim->physical_block_size, lim->io_min);
1288 	unsigned int alignment = (sector << 9) & (granularity - 1);
1289 
1290 	return (granularity + lim->alignment_offset - alignment)
1291 		& (granularity - 1);
1292 }
1293 
1294 static inline int bdev_alignment_offset(struct block_device *bdev)
1295 {
1296 	struct request_queue *q = bdev_get_queue(bdev);
1297 
1298 	if (q->limits.misaligned)
1299 		return -1;
1300 
1301 	if (bdev != bdev->bd_contains)
1302 		return bdev->bd_part->alignment_offset;
1303 
1304 	return q->limits.alignment_offset;
1305 }
1306 
1307 static inline int queue_discard_alignment(struct request_queue *q)
1308 {
1309 	if (q->limits.discard_misaligned)
1310 		return -1;
1311 
1312 	return q->limits.discard_alignment;
1313 }
1314 
1315 static inline int queue_limit_discard_alignment(struct queue_limits *lim, sector_t sector)
1316 {
1317 	unsigned int alignment, granularity, offset;
1318 
1319 	if (!lim->max_discard_sectors)
1320 		return 0;
1321 
1322 	/* Why are these in bytes, not sectors? */
1323 	alignment = lim->discard_alignment >> 9;
1324 	granularity = lim->discard_granularity >> 9;
1325 	if (!granularity)
1326 		return 0;
1327 
1328 	/* Offset of the partition start in 'granularity' sectors */
1329 	offset = sector_div(sector, granularity);
1330 
1331 	/* And why do we do this modulus *again* in blkdev_issue_discard()? */
1332 	offset = (granularity + alignment - offset) % granularity;
1333 
1334 	/* Turn it back into bytes, gaah */
1335 	return offset << 9;
1336 }
1337 
1338 static inline int bdev_discard_alignment(struct block_device *bdev)
1339 {
1340 	struct request_queue *q = bdev_get_queue(bdev);
1341 
1342 	if (bdev != bdev->bd_contains)
1343 		return bdev->bd_part->discard_alignment;
1344 
1345 	return q->limits.discard_alignment;
1346 }
1347 
1348 static inline unsigned int queue_discard_zeroes_data(struct request_queue *q)
1349 {
1350 	if (q->limits.max_discard_sectors && q->limits.discard_zeroes_data == 1)
1351 		return 1;
1352 
1353 	return 0;
1354 }
1355 
1356 static inline unsigned int bdev_discard_zeroes_data(struct block_device *bdev)
1357 {
1358 	return queue_discard_zeroes_data(bdev_get_queue(bdev));
1359 }
1360 
1361 static inline unsigned int bdev_write_same(struct block_device *bdev)
1362 {
1363 	struct request_queue *q = bdev_get_queue(bdev);
1364 
1365 	if (q)
1366 		return q->limits.max_write_same_sectors;
1367 
1368 	return 0;
1369 }
1370 
1371 static inline int queue_dma_alignment(struct request_queue *q)
1372 {
1373 	return q ? q->dma_alignment : 511;
1374 }
1375 
1376 static inline int blk_rq_aligned(struct request_queue *q, unsigned long addr,
1377 				 unsigned int len)
1378 {
1379 	unsigned int alignment = queue_dma_alignment(q) | q->dma_pad_mask;
1380 	return !(addr & alignment) && !(len & alignment);
1381 }
1382 
1383 /* assumes size > 256 */
1384 static inline unsigned int blksize_bits(unsigned int size)
1385 {
1386 	unsigned int bits = 8;
1387 	do {
1388 		bits++;
1389 		size >>= 1;
1390 	} while (size > 256);
1391 	return bits;
1392 }
1393 
1394 static inline unsigned int block_size(struct block_device *bdev)
1395 {
1396 	return bdev->bd_block_size;
1397 }
1398 
1399 static inline bool queue_flush_queueable(struct request_queue *q)
1400 {
1401 	return !q->flush_not_queueable;
1402 }
1403 
1404 typedef struct {struct page *v;} Sector;
1405 
1406 unsigned char *read_dev_sector(struct block_device *, sector_t, Sector *);
1407 
1408 static inline void put_dev_sector(Sector p)
1409 {
1410 	page_cache_release(p.v);
1411 }
1412 
1413 struct work_struct;
1414 int kblockd_schedule_work(struct work_struct *work);
1415 int kblockd_schedule_delayed_work(struct delayed_work *dwork, unsigned long delay);
1416 int kblockd_schedule_delayed_work_on(int cpu, struct delayed_work *dwork, unsigned long delay);
1417 
1418 #ifdef CONFIG_BLK_CGROUP
1419 /*
1420  * This should not be using sched_clock(). A real patch is in progress
1421  * to fix this up, until that is in place we need to disable preemption
1422  * around sched_clock() in this function and set_io_start_time_ns().
1423  */
1424 static inline void set_start_time_ns(struct request *req)
1425 {
1426 	preempt_disable();
1427 	req->start_time_ns = sched_clock();
1428 	preempt_enable();
1429 }
1430 
1431 static inline void set_io_start_time_ns(struct request *req)
1432 {
1433 	preempt_disable();
1434 	req->io_start_time_ns = sched_clock();
1435 	preempt_enable();
1436 }
1437 
1438 static inline uint64_t rq_start_time_ns(struct request *req)
1439 {
1440         return req->start_time_ns;
1441 }
1442 
1443 static inline uint64_t rq_io_start_time_ns(struct request *req)
1444 {
1445         return req->io_start_time_ns;
1446 }
1447 #else
1448 static inline void set_start_time_ns(struct request *req) {}
1449 static inline void set_io_start_time_ns(struct request *req) {}
1450 static inline uint64_t rq_start_time_ns(struct request *req)
1451 {
1452 	return 0;
1453 }
1454 static inline uint64_t rq_io_start_time_ns(struct request *req)
1455 {
1456 	return 0;
1457 }
1458 #endif
1459 
1460 #define MODULE_ALIAS_BLOCKDEV(major,minor) \
1461 	MODULE_ALIAS("block-major-" __stringify(major) "-" __stringify(minor))
1462 #define MODULE_ALIAS_BLOCKDEV_MAJOR(major) \
1463 	MODULE_ALIAS("block-major-" __stringify(major) "-*")
1464 
1465 #if defined(CONFIG_BLK_DEV_INTEGRITY)
1466 
1467 #define INTEGRITY_FLAG_READ	2	/* verify data integrity on read */
1468 #define INTEGRITY_FLAG_WRITE	4	/* generate data integrity on write */
1469 
1470 struct blk_integrity_exchg {
1471 	void			*prot_buf;
1472 	void			*data_buf;
1473 	sector_t		sector;
1474 	unsigned int		data_size;
1475 	unsigned short		sector_size;
1476 	const char		*disk_name;
1477 };
1478 
1479 typedef void (integrity_gen_fn) (struct blk_integrity_exchg *);
1480 typedef int (integrity_vrfy_fn) (struct blk_integrity_exchg *);
1481 typedef void (integrity_set_tag_fn) (void *, void *, unsigned int);
1482 typedef void (integrity_get_tag_fn) (void *, void *, unsigned int);
1483 
1484 struct blk_integrity {
1485 	integrity_gen_fn	*generate_fn;
1486 	integrity_vrfy_fn	*verify_fn;
1487 	integrity_set_tag_fn	*set_tag_fn;
1488 	integrity_get_tag_fn	*get_tag_fn;
1489 
1490 	unsigned short		flags;
1491 	unsigned short		tuple_size;
1492 	unsigned short		sector_size;
1493 	unsigned short		tag_size;
1494 
1495 	const char		*name;
1496 
1497 	struct kobject		kobj;
1498 };
1499 
1500 extern bool blk_integrity_is_initialized(struct gendisk *);
1501 extern int blk_integrity_register(struct gendisk *, struct blk_integrity *);
1502 extern void blk_integrity_unregister(struct gendisk *);
1503 extern int blk_integrity_compare(struct gendisk *, struct gendisk *);
1504 extern int blk_rq_map_integrity_sg(struct request_queue *, struct bio *,
1505 				   struct scatterlist *);
1506 extern int blk_rq_count_integrity_sg(struct request_queue *, struct bio *);
1507 extern int blk_integrity_merge_rq(struct request_queue *, struct request *,
1508 				  struct request *);
1509 extern int blk_integrity_merge_bio(struct request_queue *, struct request *,
1510 				   struct bio *);
1511 
1512 static inline
1513 struct blk_integrity *bdev_get_integrity(struct block_device *bdev)
1514 {
1515 	return bdev->bd_disk->integrity;
1516 }
1517 
1518 static inline struct blk_integrity *blk_get_integrity(struct gendisk *disk)
1519 {
1520 	return disk->integrity;
1521 }
1522 
1523 static inline int blk_integrity_rq(struct request *rq)
1524 {
1525 	if (rq->bio == NULL)
1526 		return 0;
1527 
1528 	return bio_integrity(rq->bio);
1529 }
1530 
1531 static inline void blk_queue_max_integrity_segments(struct request_queue *q,
1532 						    unsigned int segs)
1533 {
1534 	q->limits.max_integrity_segments = segs;
1535 }
1536 
1537 static inline unsigned short
1538 queue_max_integrity_segments(struct request_queue *q)
1539 {
1540 	return q->limits.max_integrity_segments;
1541 }
1542 
1543 #else /* CONFIG_BLK_DEV_INTEGRITY */
1544 
1545 struct bio;
1546 struct block_device;
1547 struct gendisk;
1548 struct blk_integrity;
1549 
1550 static inline int blk_integrity_rq(struct request *rq)
1551 {
1552 	return 0;
1553 }
1554 static inline int blk_rq_count_integrity_sg(struct request_queue *q,
1555 					    struct bio *b)
1556 {
1557 	return 0;
1558 }
1559 static inline int blk_rq_map_integrity_sg(struct request_queue *q,
1560 					  struct bio *b,
1561 					  struct scatterlist *s)
1562 {
1563 	return 0;
1564 }
1565 static inline struct blk_integrity *bdev_get_integrity(struct block_device *b)
1566 {
1567 	return 0;
1568 }
1569 static inline struct blk_integrity *blk_get_integrity(struct gendisk *disk)
1570 {
1571 	return NULL;
1572 }
1573 static inline int blk_integrity_compare(struct gendisk *a, struct gendisk *b)
1574 {
1575 	return 0;
1576 }
1577 static inline int blk_integrity_register(struct gendisk *d,
1578 					 struct blk_integrity *b)
1579 {
1580 	return 0;
1581 }
1582 static inline void blk_integrity_unregister(struct gendisk *d)
1583 {
1584 }
1585 static inline void blk_queue_max_integrity_segments(struct request_queue *q,
1586 						    unsigned int segs)
1587 {
1588 }
1589 static inline unsigned short queue_max_integrity_segments(struct request_queue *q)
1590 {
1591 	return 0;
1592 }
1593 static inline int blk_integrity_merge_rq(struct request_queue *rq,
1594 					 struct request *r1,
1595 					 struct request *r2)
1596 {
1597 	return 0;
1598 }
1599 static inline int blk_integrity_merge_bio(struct request_queue *rq,
1600 					  struct request *r,
1601 					  struct bio *b)
1602 {
1603 	return 0;
1604 }
1605 static inline bool blk_integrity_is_initialized(struct gendisk *g)
1606 {
1607 	return 0;
1608 }
1609 
1610 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1611 
1612 struct block_device_operations {
1613 	int (*open) (struct block_device *, fmode_t);
1614 	void (*release) (struct gendisk *, fmode_t);
1615 	int (*rw_page)(struct block_device *, sector_t, struct page *, int rw);
1616 	int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
1617 	int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
1618 	int (*direct_access) (struct block_device *, sector_t,
1619 						void **, unsigned long *);
1620 	unsigned int (*check_events) (struct gendisk *disk,
1621 				      unsigned int clearing);
1622 	/* ->media_changed() is DEPRECATED, use ->check_events() instead */
1623 	int (*media_changed) (struct gendisk *);
1624 	void (*unlock_native_capacity) (struct gendisk *);
1625 	int (*revalidate_disk) (struct gendisk *);
1626 	int (*getgeo)(struct block_device *, struct hd_geometry *);
1627 	/* this callback is with swap_lock and sometimes page table lock held */
1628 	void (*swap_slot_free_notify) (struct block_device *, unsigned long);
1629 	struct module *owner;
1630 };
1631 
1632 extern int __blkdev_driver_ioctl(struct block_device *, fmode_t, unsigned int,
1633 				 unsigned long);
1634 extern int bdev_read_page(struct block_device *, sector_t, struct page *);
1635 extern int bdev_write_page(struct block_device *, sector_t, struct page *,
1636 						struct writeback_control *);
1637 #else /* CONFIG_BLOCK */
1638 
1639 struct block_device;
1640 
1641 /*
1642  * stubs for when the block layer is configured out
1643  */
1644 #define buffer_heads_over_limit 0
1645 
1646 static inline long nr_blockdev_pages(void)
1647 {
1648 	return 0;
1649 }
1650 
1651 struct blk_plug {
1652 };
1653 
1654 static inline void blk_start_plug(struct blk_plug *plug)
1655 {
1656 }
1657 
1658 static inline void blk_finish_plug(struct blk_plug *plug)
1659 {
1660 }
1661 
1662 static inline void blk_flush_plug(struct task_struct *task)
1663 {
1664 }
1665 
1666 static inline void blk_schedule_flush_plug(struct task_struct *task)
1667 {
1668 }
1669 
1670 
1671 static inline bool blk_needs_flush_plug(struct task_struct *tsk)
1672 {
1673 	return false;
1674 }
1675 
1676 static inline int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask,
1677 				     sector_t *error_sector)
1678 {
1679 	return 0;
1680 }
1681 
1682 #endif /* CONFIG_BLOCK */
1683 
1684 #endif
1685