xref: /linux-6.15/include/linux/blkdev.h (revision 930cc144)
1 #ifndef _LINUX_BLKDEV_H
2 #define _LINUX_BLKDEV_H
3 
4 #ifdef CONFIG_BLOCK
5 
6 #include <linux/sched.h>
7 #include <linux/major.h>
8 #include <linux/genhd.h>
9 #include <linux/list.h>
10 #include <linux/timer.h>
11 #include <linux/workqueue.h>
12 #include <linux/pagemap.h>
13 #include <linux/backing-dev.h>
14 #include <linux/wait.h>
15 #include <linux/mempool.h>
16 #include <linux/bio.h>
17 #include <linux/module.h>
18 #include <linux/stringify.h>
19 #include <linux/gfp.h>
20 #include <linux/bsg.h>
21 #include <linux/smp.h>
22 
23 #include <asm/scatterlist.h>
24 
25 struct scsi_ioctl_command;
26 
27 struct request_queue;
28 struct elevator_queue;
29 typedef struct elevator_queue elevator_t;
30 struct request_pm_state;
31 struct blk_trace;
32 struct request;
33 struct sg_io_hdr;
34 
35 #define BLKDEV_MIN_RQ	4
36 #define BLKDEV_MAX_RQ	128	/* Default maximum */
37 
38 struct request;
39 typedef void (rq_end_io_fn)(struct request *, int);
40 
41 struct request_list {
42 	int count[2];
43 	int starved[2];
44 	int elvpriv;
45 	mempool_t *rq_pool;
46 	wait_queue_head_t wait[2];
47 };
48 
49 /*
50  * request command types
51  */
52 enum rq_cmd_type_bits {
53 	REQ_TYPE_FS		= 1,	/* fs request */
54 	REQ_TYPE_BLOCK_PC,		/* scsi command */
55 	REQ_TYPE_SENSE,			/* sense request */
56 	REQ_TYPE_PM_SUSPEND,		/* suspend request */
57 	REQ_TYPE_PM_RESUME,		/* resume request */
58 	REQ_TYPE_PM_SHUTDOWN,		/* shutdown request */
59 	REQ_TYPE_SPECIAL,		/* driver defined type */
60 	REQ_TYPE_LINUX_BLOCK,		/* generic block layer message */
61 	/*
62 	 * for ATA/ATAPI devices. this really doesn't belong here, ide should
63 	 * use REQ_TYPE_SPECIAL and use rq->cmd[0] with the range of driver
64 	 * private REQ_LB opcodes to differentiate what type of request this is
65 	 */
66 	REQ_TYPE_ATA_TASKFILE,
67 	REQ_TYPE_ATA_PC,
68 };
69 
70 /*
71  * For request of type REQ_TYPE_LINUX_BLOCK, rq->cmd[0] is the opcode being
72  * sent down (similar to how REQ_TYPE_BLOCK_PC means that ->cmd[] holds a
73  * SCSI cdb.
74  *
75  * 0x00 -> 0x3f are driver private, to be used for whatever purpose they need,
76  * typically to differentiate REQ_TYPE_SPECIAL requests.
77  *
78  */
79 enum {
80 	REQ_LB_OP_EJECT	= 0x40,		/* eject request */
81 	REQ_LB_OP_FLUSH = 0x41,		/* flush request */
82 	REQ_LB_OP_DISCARD = 0x42,	/* discard sectors */
83 };
84 
85 /*
86  * request type modified bits. first two bits match BIO_RW* bits, important
87  */
88 enum rq_flag_bits {
89 	__REQ_RW,		/* not set, read. set, write */
90 	__REQ_FAILFAST_DEV,	/* no driver retries of device errors */
91 	__REQ_FAILFAST_TRANSPORT, /* no driver retries of transport errors */
92 	__REQ_FAILFAST_DRIVER,	/* no driver retries of driver errors */
93 	__REQ_DISCARD,		/* request to discard sectors */
94 	__REQ_SORTED,		/* elevator knows about this request */
95 	__REQ_SOFTBARRIER,	/* may not be passed by ioscheduler */
96 	__REQ_HARDBARRIER,	/* may not be passed by drive either */
97 	__REQ_FUA,		/* forced unit access */
98 	__REQ_NOMERGE,		/* don't touch this for merging */
99 	__REQ_STARTED,		/* drive already may have started this one */
100 	__REQ_DONTPREP,		/* don't call prep for this one */
101 	__REQ_QUEUED,		/* uses queueing */
102 	__REQ_ELVPRIV,		/* elevator private data attached */
103 	__REQ_FAILED,		/* set if the request failed */
104 	__REQ_QUIET,		/* don't worry about errors */
105 	__REQ_PREEMPT,		/* set for "ide_preempt" requests */
106 	__REQ_ORDERED_COLOR,	/* is before or after barrier */
107 	__REQ_RW_SYNC,		/* request is sync (O_DIRECT) */
108 	__REQ_ALLOCED,		/* request came from our alloc pool */
109 	__REQ_RW_META,		/* metadata io request */
110 	__REQ_COPY_USER,	/* contains copies of user pages */
111 	__REQ_INTEGRITY,	/* integrity metadata has been remapped */
112 	__REQ_NR_BITS,		/* stops here */
113 };
114 
115 #define REQ_RW		(1 << __REQ_RW)
116 #define REQ_FAILFAST_DEV	(1 << __REQ_FAILFAST_DEV)
117 #define REQ_FAILFAST_TRANSPORT	(1 << __REQ_FAILFAST_TRANSPORT)
118 #define REQ_FAILFAST_DRIVER	(1 << __REQ_FAILFAST_DRIVER)
119 #define REQ_DISCARD	(1 << __REQ_DISCARD)
120 #define REQ_SORTED	(1 << __REQ_SORTED)
121 #define REQ_SOFTBARRIER	(1 << __REQ_SOFTBARRIER)
122 #define REQ_HARDBARRIER	(1 << __REQ_HARDBARRIER)
123 #define REQ_FUA		(1 << __REQ_FUA)
124 #define REQ_NOMERGE	(1 << __REQ_NOMERGE)
125 #define REQ_STARTED	(1 << __REQ_STARTED)
126 #define REQ_DONTPREP	(1 << __REQ_DONTPREP)
127 #define REQ_QUEUED	(1 << __REQ_QUEUED)
128 #define REQ_ELVPRIV	(1 << __REQ_ELVPRIV)
129 #define REQ_FAILED	(1 << __REQ_FAILED)
130 #define REQ_QUIET	(1 << __REQ_QUIET)
131 #define REQ_PREEMPT	(1 << __REQ_PREEMPT)
132 #define REQ_ORDERED_COLOR	(1 << __REQ_ORDERED_COLOR)
133 #define REQ_RW_SYNC	(1 << __REQ_RW_SYNC)
134 #define REQ_ALLOCED	(1 << __REQ_ALLOCED)
135 #define REQ_RW_META	(1 << __REQ_RW_META)
136 #define REQ_COPY_USER	(1 << __REQ_COPY_USER)
137 #define REQ_INTEGRITY	(1 << __REQ_INTEGRITY)
138 
139 #define BLK_MAX_CDB	16
140 
141 /*
142  * try to put the fields that are referenced together in the same cacheline.
143  * if you modify this structure, be sure to check block/blk-core.c:rq_init()
144  * as well!
145  */
146 struct request {
147 	struct list_head queuelist;
148 	struct call_single_data csd;
149 	int cpu;
150 
151 	struct request_queue *q;
152 
153 	unsigned int cmd_flags;
154 	enum rq_cmd_type_bits cmd_type;
155 	unsigned long atomic_flags;
156 
157 	/* Maintain bio traversal state for part by part I/O submission.
158 	 * hard_* are block layer internals, no driver should touch them!
159 	 */
160 
161 	sector_t sector;		/* next sector to submit */
162 	sector_t hard_sector;		/* next sector to complete */
163 	unsigned long nr_sectors;	/* no. of sectors left to submit */
164 	unsigned long hard_nr_sectors;	/* no. of sectors left to complete */
165 	/* no. of sectors left to submit in the current segment */
166 	unsigned int current_nr_sectors;
167 
168 	/* no. of sectors left to complete in the current segment */
169 	unsigned int hard_cur_sectors;
170 
171 	struct bio *bio;
172 	struct bio *biotail;
173 
174 	struct hlist_node hash;	/* merge hash */
175 	/*
176 	 * The rb_node is only used inside the io scheduler, requests
177 	 * are pruned when moved to the dispatch queue. So let the
178 	 * completion_data share space with the rb_node.
179 	 */
180 	union {
181 		struct rb_node rb_node;	/* sort/lookup */
182 		void *completion_data;
183 	};
184 
185 	/*
186 	 * two pointers are available for the IO schedulers, if they need
187 	 * more they have to dynamically allocate it.
188 	 */
189 	void *elevator_private;
190 	void *elevator_private2;
191 
192 	struct gendisk *rq_disk;
193 	unsigned long start_time;
194 
195 	/* Number of scatter-gather DMA addr+len pairs after
196 	 * physical address coalescing is performed.
197 	 */
198 	unsigned short nr_phys_segments;
199 
200 	unsigned short ioprio;
201 
202 	void *special;
203 	char *buffer;
204 
205 	int tag;
206 	int errors;
207 
208 	int ref_count;
209 
210 	/*
211 	 * when request is used as a packet command carrier
212 	 */
213 	unsigned short cmd_len;
214 	unsigned char __cmd[BLK_MAX_CDB];
215 	unsigned char *cmd;
216 
217 	unsigned int data_len;
218 	unsigned int extra_len;	/* length of alignment and padding */
219 	unsigned int sense_len;
220 	void *data;
221 	void *sense;
222 
223 	unsigned long deadline;
224 	struct list_head timeout_list;
225 	unsigned int timeout;
226 	int retries;
227 
228 	/*
229 	 * completion callback.
230 	 */
231 	rq_end_io_fn *end_io;
232 	void *end_io_data;
233 
234 	/* for bidi */
235 	struct request *next_rq;
236 };
237 
238 static inline unsigned short req_get_ioprio(struct request *req)
239 {
240 	return req->ioprio;
241 }
242 
243 /*
244  * State information carried for REQ_TYPE_PM_SUSPEND and REQ_TYPE_PM_RESUME
245  * requests. Some step values could eventually be made generic.
246  */
247 struct request_pm_state
248 {
249 	/* PM state machine step value, currently driver specific */
250 	int	pm_step;
251 	/* requested PM state value (S1, S2, S3, S4, ...) */
252 	u32	pm_state;
253 	void*	data;		/* for driver use */
254 };
255 
256 #include <linux/elevator.h>
257 
258 typedef void (request_fn_proc) (struct request_queue *q);
259 typedef int (make_request_fn) (struct request_queue *q, struct bio *bio);
260 typedef int (prep_rq_fn) (struct request_queue *, struct request *);
261 typedef void (unplug_fn) (struct request_queue *);
262 typedef int (prepare_discard_fn) (struct request_queue *, struct request *);
263 
264 struct bio_vec;
265 struct bvec_merge_data {
266 	struct block_device *bi_bdev;
267 	sector_t bi_sector;
268 	unsigned bi_size;
269 	unsigned long bi_rw;
270 };
271 typedef int (merge_bvec_fn) (struct request_queue *, struct bvec_merge_data *,
272 			     struct bio_vec *);
273 typedef void (prepare_flush_fn) (struct request_queue *, struct request *);
274 typedef void (softirq_done_fn)(struct request *);
275 typedef int (dma_drain_needed_fn)(struct request *);
276 typedef int (lld_busy_fn) (struct request_queue *q);
277 
278 enum blk_eh_timer_return {
279 	BLK_EH_NOT_HANDLED,
280 	BLK_EH_HANDLED,
281 	BLK_EH_RESET_TIMER,
282 };
283 
284 typedef enum blk_eh_timer_return (rq_timed_out_fn)(struct request *);
285 
286 enum blk_queue_state {
287 	Queue_down,
288 	Queue_up,
289 };
290 
291 struct blk_queue_tag {
292 	struct request **tag_index;	/* map of busy tags */
293 	unsigned long *tag_map;		/* bit map of free/busy tags */
294 	int busy;			/* current depth */
295 	int max_depth;			/* what we will send to device */
296 	int real_max_depth;		/* what the array can hold */
297 	atomic_t refcnt;		/* map can be shared */
298 };
299 
300 #define BLK_SCSI_MAX_CMDS	(256)
301 #define BLK_SCSI_CMD_PER_LONG	(BLK_SCSI_MAX_CMDS / (sizeof(long) * 8))
302 
303 struct blk_cmd_filter {
304 	unsigned long read_ok[BLK_SCSI_CMD_PER_LONG];
305 	unsigned long write_ok[BLK_SCSI_CMD_PER_LONG];
306 	struct kobject kobj;
307 };
308 
309 struct request_queue
310 {
311 	/*
312 	 * Together with queue_head for cacheline sharing
313 	 */
314 	struct list_head	queue_head;
315 	struct request		*last_merge;
316 	elevator_t		*elevator;
317 
318 	/*
319 	 * the queue request freelist, one for reads and one for writes
320 	 */
321 	struct request_list	rq;
322 
323 	request_fn_proc		*request_fn;
324 	make_request_fn		*make_request_fn;
325 	prep_rq_fn		*prep_rq_fn;
326 	unplug_fn		*unplug_fn;
327 	prepare_discard_fn	*prepare_discard_fn;
328 	merge_bvec_fn		*merge_bvec_fn;
329 	prepare_flush_fn	*prepare_flush_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 	/*
336 	 * Dispatch queue sorting
337 	 */
338 	sector_t		end_sector;
339 	struct request		*boundary_rq;
340 
341 	/*
342 	 * Auto-unplugging state
343 	 */
344 	struct timer_list	unplug_timer;
345 	int			unplug_thresh;	/* After this many requests */
346 	unsigned long		unplug_delay;	/* After this many jiffies */
347 	struct work_struct	unplug_work;
348 
349 	struct backing_dev_info	backing_dev_info;
350 
351 	/*
352 	 * The queue owner gets to use this for whatever they like.
353 	 * ll_rw_blk doesn't touch it.
354 	 */
355 	void			*queuedata;
356 
357 	/*
358 	 * queue needs bounce pages for pages above this limit
359 	 */
360 	unsigned long		bounce_pfn;
361 	gfp_t			bounce_gfp;
362 
363 	/*
364 	 * various queue flags, see QUEUE_* below
365 	 */
366 	unsigned long		queue_flags;
367 
368 	/*
369 	 * protects queue structures from reentrancy. ->__queue_lock should
370 	 * _never_ be used directly, it is queue private. always use
371 	 * ->queue_lock.
372 	 */
373 	spinlock_t		__queue_lock;
374 	spinlock_t		*queue_lock;
375 
376 	/*
377 	 * queue kobject
378 	 */
379 	struct kobject kobj;
380 
381 	/*
382 	 * queue settings
383 	 */
384 	unsigned long		nr_requests;	/* Max # of requests */
385 	unsigned int		nr_congestion_on;
386 	unsigned int		nr_congestion_off;
387 	unsigned int		nr_batching;
388 
389 	unsigned int		max_sectors;
390 	unsigned int		max_hw_sectors;
391 	unsigned short		max_phys_segments;
392 	unsigned short		max_hw_segments;
393 	unsigned short		hardsect_size;
394 	unsigned int		max_segment_size;
395 
396 	unsigned long		seg_boundary_mask;
397 	void			*dma_drain_buffer;
398 	unsigned int		dma_drain_size;
399 	unsigned int		dma_pad_mask;
400 	unsigned int		dma_alignment;
401 
402 	struct blk_queue_tag	*queue_tags;
403 	struct list_head	tag_busy_list;
404 
405 	unsigned int		nr_sorted;
406 	unsigned int		in_flight;
407 
408 	unsigned int		rq_timeout;
409 	struct timer_list	timeout;
410 	struct list_head	timeout_list;
411 
412 	/*
413 	 * sg stuff
414 	 */
415 	unsigned int		sg_timeout;
416 	unsigned int		sg_reserved_size;
417 	int			node;
418 #ifdef CONFIG_BLK_DEV_IO_TRACE
419 	struct blk_trace	*blk_trace;
420 #endif
421 	/*
422 	 * reserved for flush operations
423 	 */
424 	unsigned int		ordered, next_ordered, ordseq;
425 	int			orderr, ordcolor;
426 	struct request		pre_flush_rq, bar_rq, post_flush_rq;
427 	struct request		*orig_bar_rq;
428 
429 	struct mutex		sysfs_lock;
430 
431 #if defined(CONFIG_BLK_DEV_BSG)
432 	struct bsg_class_device bsg_dev;
433 #endif
434 	struct blk_cmd_filter cmd_filter;
435 };
436 
437 #define QUEUE_FLAG_CLUSTER	0	/* cluster several segments into 1 */
438 #define QUEUE_FLAG_QUEUED	1	/* uses generic tag queueing */
439 #define QUEUE_FLAG_STOPPED	2	/* queue is stopped */
440 #define	QUEUE_FLAG_READFULL	3	/* read queue has been filled */
441 #define QUEUE_FLAG_WRITEFULL	4	/* write queue has been filled */
442 #define QUEUE_FLAG_DEAD		5	/* queue being torn down */
443 #define QUEUE_FLAG_REENTER	6	/* Re-entrancy avoidance */
444 #define QUEUE_FLAG_PLUGGED	7	/* queue is plugged */
445 #define QUEUE_FLAG_ELVSWITCH	8	/* don't use elevator, just do FIFO */
446 #define QUEUE_FLAG_BIDI		9	/* queue supports bidi requests */
447 #define QUEUE_FLAG_NOMERGES    10	/* disable merge attempts */
448 #define QUEUE_FLAG_SAME_COMP   11	/* force complete on same CPU */
449 #define QUEUE_FLAG_FAIL_IO     12	/* fake timeout */
450 #define QUEUE_FLAG_STACKABLE   13	/* supports request stacking */
451 #define QUEUE_FLAG_NONROT      14	/* non-rotational device (SSD) */
452 
453 static inline int queue_is_locked(struct request_queue *q)
454 {
455 #ifdef CONFIG_SMP
456 	spinlock_t *lock = q->queue_lock;
457 	return lock && spin_is_locked(lock);
458 #else
459 	return 1;
460 #endif
461 }
462 
463 static inline void queue_flag_set_unlocked(unsigned int flag,
464 					   struct request_queue *q)
465 {
466 	__set_bit(flag, &q->queue_flags);
467 }
468 
469 static inline int queue_flag_test_and_clear(unsigned int flag,
470 					    struct request_queue *q)
471 {
472 	WARN_ON_ONCE(!queue_is_locked(q));
473 
474 	if (test_bit(flag, &q->queue_flags)) {
475 		__clear_bit(flag, &q->queue_flags);
476 		return 1;
477 	}
478 
479 	return 0;
480 }
481 
482 static inline int queue_flag_test_and_set(unsigned int flag,
483 					  struct request_queue *q)
484 {
485 	WARN_ON_ONCE(!queue_is_locked(q));
486 
487 	if (!test_bit(flag, &q->queue_flags)) {
488 		__set_bit(flag, &q->queue_flags);
489 		return 0;
490 	}
491 
492 	return 1;
493 }
494 
495 static inline void queue_flag_set(unsigned int flag, struct request_queue *q)
496 {
497 	WARN_ON_ONCE(!queue_is_locked(q));
498 	__set_bit(flag, &q->queue_flags);
499 }
500 
501 static inline void queue_flag_clear_unlocked(unsigned int flag,
502 					     struct request_queue *q)
503 {
504 	__clear_bit(flag, &q->queue_flags);
505 }
506 
507 static inline void queue_flag_clear(unsigned int flag, struct request_queue *q)
508 {
509 	WARN_ON_ONCE(!queue_is_locked(q));
510 	__clear_bit(flag, &q->queue_flags);
511 }
512 
513 enum {
514 	/*
515 	 * Hardbarrier is supported with one of the following methods.
516 	 *
517 	 * NONE		: hardbarrier unsupported
518 	 * DRAIN	: ordering by draining is enough
519 	 * DRAIN_FLUSH	: ordering by draining w/ pre and post flushes
520 	 * DRAIN_FUA	: ordering by draining w/ pre flush and FUA write
521 	 * TAG		: ordering by tag is enough
522 	 * TAG_FLUSH	: ordering by tag w/ pre and post flushes
523 	 * TAG_FUA	: ordering by tag w/ pre flush and FUA write
524 	 */
525 	QUEUE_ORDERED_NONE	= 0x00,
526 	QUEUE_ORDERED_DRAIN	= 0x01,
527 	QUEUE_ORDERED_TAG	= 0x02,
528 
529 	QUEUE_ORDERED_PREFLUSH	= 0x10,
530 	QUEUE_ORDERED_POSTFLUSH	= 0x20,
531 	QUEUE_ORDERED_FUA	= 0x40,
532 
533 	QUEUE_ORDERED_DRAIN_FLUSH = QUEUE_ORDERED_DRAIN |
534 			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH,
535 	QUEUE_ORDERED_DRAIN_FUA	= QUEUE_ORDERED_DRAIN |
536 			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_FUA,
537 	QUEUE_ORDERED_TAG_FLUSH	= QUEUE_ORDERED_TAG |
538 			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH,
539 	QUEUE_ORDERED_TAG_FUA	= QUEUE_ORDERED_TAG |
540 			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_FUA,
541 
542 	/*
543 	 * Ordered operation sequence
544 	 */
545 	QUEUE_ORDSEQ_STARTED	= 0x01,	/* flushing in progress */
546 	QUEUE_ORDSEQ_DRAIN	= 0x02,	/* waiting for the queue to be drained */
547 	QUEUE_ORDSEQ_PREFLUSH	= 0x04,	/* pre-flushing in progress */
548 	QUEUE_ORDSEQ_BAR	= 0x08,	/* original barrier req in progress */
549 	QUEUE_ORDSEQ_POSTFLUSH	= 0x10,	/* post-flushing in progress */
550 	QUEUE_ORDSEQ_DONE	= 0x20,
551 };
552 
553 #define blk_queue_plugged(q)	test_bit(QUEUE_FLAG_PLUGGED, &(q)->queue_flags)
554 #define blk_queue_tagged(q)	test_bit(QUEUE_FLAG_QUEUED, &(q)->queue_flags)
555 #define blk_queue_stopped(q)	test_bit(QUEUE_FLAG_STOPPED, &(q)->queue_flags)
556 #define blk_queue_nomerges(q)	test_bit(QUEUE_FLAG_NOMERGES, &(q)->queue_flags)
557 #define blk_queue_nonrot(q)	test_bit(QUEUE_FLAG_NONROT, &(q)->queue_flags)
558 #define blk_queue_flushing(q)	((q)->ordseq)
559 #define blk_queue_stackable(q)	\
560 	test_bit(QUEUE_FLAG_STACKABLE, &(q)->queue_flags)
561 
562 #define blk_fs_request(rq)	((rq)->cmd_type == REQ_TYPE_FS)
563 #define blk_pc_request(rq)	((rq)->cmd_type == REQ_TYPE_BLOCK_PC)
564 #define blk_special_request(rq)	((rq)->cmd_type == REQ_TYPE_SPECIAL)
565 #define blk_sense_request(rq)	((rq)->cmd_type == REQ_TYPE_SENSE)
566 
567 #define blk_failfast_dev(rq)	((rq)->cmd_flags & REQ_FAILFAST_DEV)
568 #define blk_failfast_transport(rq) ((rq)->cmd_flags & REQ_FAILFAST_TRANSPORT)
569 #define blk_failfast_driver(rq)	((rq)->cmd_flags & REQ_FAILFAST_DRIVER)
570 #define blk_noretry_request(rq)	(blk_failfast_dev(rq) ||	\
571 				 blk_failfast_transport(rq) ||	\
572 				 blk_failfast_driver(rq))
573 #define blk_rq_started(rq)	((rq)->cmd_flags & REQ_STARTED)
574 
575 #define blk_account_rq(rq)	(blk_rq_started(rq) && (blk_fs_request(rq) || blk_discard_rq(rq)))
576 
577 #define blk_pm_suspend_request(rq)	((rq)->cmd_type == REQ_TYPE_PM_SUSPEND)
578 #define blk_pm_resume_request(rq)	((rq)->cmd_type == REQ_TYPE_PM_RESUME)
579 #define blk_pm_request(rq)	\
580 	(blk_pm_suspend_request(rq) || blk_pm_resume_request(rq))
581 
582 #define blk_rq_cpu_valid(rq)	((rq)->cpu != -1)
583 #define blk_sorted_rq(rq)	((rq)->cmd_flags & REQ_SORTED)
584 #define blk_barrier_rq(rq)	((rq)->cmd_flags & REQ_HARDBARRIER)
585 #define blk_fua_rq(rq)		((rq)->cmd_flags & REQ_FUA)
586 #define blk_discard_rq(rq)	((rq)->cmd_flags & REQ_DISCARD)
587 #define blk_bidi_rq(rq)		((rq)->next_rq != NULL)
588 #define blk_empty_barrier(rq)	(blk_barrier_rq(rq) && blk_fs_request(rq) && !(rq)->hard_nr_sectors)
589 /* rq->queuelist of dequeued request must be list_empty() */
590 #define blk_queued_rq(rq)	(!list_empty(&(rq)->queuelist))
591 
592 #define list_entry_rq(ptr)	list_entry((ptr), struct request, queuelist)
593 
594 #define rq_data_dir(rq)		((rq)->cmd_flags & 1)
595 
596 /*
597  * We regard a request as sync, if it's a READ or a SYNC write.
598  */
599 #define rq_is_sync(rq)		(rq_data_dir((rq)) == READ || (rq)->cmd_flags & REQ_RW_SYNC)
600 #define rq_is_meta(rq)		((rq)->cmd_flags & REQ_RW_META)
601 
602 static inline int blk_queue_full(struct request_queue *q, int rw)
603 {
604 	if (rw == READ)
605 		return test_bit(QUEUE_FLAG_READFULL, &q->queue_flags);
606 	return test_bit(QUEUE_FLAG_WRITEFULL, &q->queue_flags);
607 }
608 
609 static inline void blk_set_queue_full(struct request_queue *q, int rw)
610 {
611 	if (rw == READ)
612 		queue_flag_set(QUEUE_FLAG_READFULL, q);
613 	else
614 		queue_flag_set(QUEUE_FLAG_WRITEFULL, q);
615 }
616 
617 static inline void blk_clear_queue_full(struct request_queue *q, int rw)
618 {
619 	if (rw == READ)
620 		queue_flag_clear(QUEUE_FLAG_READFULL, q);
621 	else
622 		queue_flag_clear(QUEUE_FLAG_WRITEFULL, q);
623 }
624 
625 
626 /*
627  * mergeable request must not have _NOMERGE or _BARRIER bit set, nor may
628  * it already be started by driver.
629  */
630 #define RQ_NOMERGE_FLAGS	\
631 	(REQ_NOMERGE | REQ_STARTED | REQ_HARDBARRIER | REQ_SOFTBARRIER)
632 #define rq_mergeable(rq)	\
633 	(!((rq)->cmd_flags & RQ_NOMERGE_FLAGS) && \
634 	 (blk_discard_rq(rq) || blk_fs_request((rq))))
635 
636 /*
637  * q->prep_rq_fn return values
638  */
639 #define BLKPREP_OK		0	/* serve it */
640 #define BLKPREP_KILL		1	/* fatal error, kill */
641 #define BLKPREP_DEFER		2	/* leave on queue */
642 
643 extern unsigned long blk_max_low_pfn, blk_max_pfn;
644 
645 /*
646  * standard bounce addresses:
647  *
648  * BLK_BOUNCE_HIGH	: bounce all highmem pages
649  * BLK_BOUNCE_ANY	: don't bounce anything
650  * BLK_BOUNCE_ISA	: bounce pages above ISA DMA boundary
651  */
652 
653 #if BITS_PER_LONG == 32
654 #define BLK_BOUNCE_HIGH		((u64)blk_max_low_pfn << PAGE_SHIFT)
655 #else
656 #define BLK_BOUNCE_HIGH		-1ULL
657 #endif
658 #define BLK_BOUNCE_ANY		(-1ULL)
659 #define BLK_BOUNCE_ISA		(ISA_DMA_THRESHOLD)
660 
661 /*
662  * default timeout for SG_IO if none specified
663  */
664 #define BLK_DEFAULT_SG_TIMEOUT	(60 * HZ)
665 
666 #ifdef CONFIG_BOUNCE
667 extern int init_emergency_isa_pool(void);
668 extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
669 #else
670 static inline int init_emergency_isa_pool(void)
671 {
672 	return 0;
673 }
674 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
675 {
676 }
677 #endif /* CONFIG_MMU */
678 
679 struct rq_map_data {
680 	struct page **pages;
681 	int page_order;
682 	int nr_entries;
683 };
684 
685 struct req_iterator {
686 	int i;
687 	struct bio *bio;
688 };
689 
690 /* This should not be used directly - use rq_for_each_segment */
691 #define __rq_for_each_bio(_bio, rq)	\
692 	if ((rq->bio))			\
693 		for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
694 
695 #define rq_for_each_segment(bvl, _rq, _iter)			\
696 	__rq_for_each_bio(_iter.bio, _rq)			\
697 		bio_for_each_segment(bvl, _iter.bio, _iter.i)
698 
699 #define rq_iter_last(rq, _iter)					\
700 		(_iter.bio->bi_next == NULL && _iter.i == _iter.bio->bi_vcnt-1)
701 
702 extern int blk_register_queue(struct gendisk *disk);
703 extern void blk_unregister_queue(struct gendisk *disk);
704 extern void register_disk(struct gendisk *dev);
705 extern void generic_make_request(struct bio *bio);
706 extern void blk_rq_init(struct request_queue *q, struct request *rq);
707 extern void blk_put_request(struct request *);
708 extern void __blk_put_request(struct request_queue *, struct request *);
709 extern struct request *blk_get_request(struct request_queue *, int, gfp_t);
710 extern void blk_insert_request(struct request_queue *, struct request *, int, void *);
711 extern void blk_requeue_request(struct request_queue *, struct request *);
712 extern int blk_rq_check_limits(struct request_queue *q, struct request *rq);
713 extern int blk_lld_busy(struct request_queue *q);
714 extern int blk_insert_cloned_request(struct request_queue *q,
715 				     struct request *rq);
716 extern void blk_plug_device(struct request_queue *);
717 extern void blk_plug_device_unlocked(struct request_queue *);
718 extern int blk_remove_plug(struct request_queue *);
719 extern void blk_recount_segments(struct request_queue *, struct bio *);
720 extern int scsi_cmd_ioctl(struct file *, struct request_queue *,
721 			  struct gendisk *, unsigned int, void __user *);
722 extern int sg_scsi_ioctl(struct file *, struct request_queue *,
723 		struct gendisk *, struct scsi_ioctl_command __user *);
724 
725 /*
726  * Temporary export, until SCSI gets fixed up.
727  */
728 extern int blk_rq_append_bio(struct request_queue *q, struct request *rq,
729 			     struct bio *bio);
730 
731 /*
732  * A queue has just exitted congestion.  Note this in the global counter of
733  * congested queues, and wake up anyone who was waiting for requests to be
734  * put back.
735  */
736 static inline void blk_clear_queue_congested(struct request_queue *q, int rw)
737 {
738 	clear_bdi_congested(&q->backing_dev_info, rw);
739 }
740 
741 /*
742  * A queue has just entered congestion.  Flag that in the queue's VM-visible
743  * state flags and increment the global gounter of congested queues.
744  */
745 static inline void blk_set_queue_congested(struct request_queue *q, int rw)
746 {
747 	set_bdi_congested(&q->backing_dev_info, rw);
748 }
749 
750 extern void blk_start_queue(struct request_queue *q);
751 extern void blk_stop_queue(struct request_queue *q);
752 extern void blk_sync_queue(struct request_queue *q);
753 extern void __blk_stop_queue(struct request_queue *q);
754 extern void __blk_run_queue(struct request_queue *);
755 extern void blk_run_queue(struct request_queue *);
756 extern void blk_start_queueing(struct request_queue *);
757 extern int blk_rq_map_user(struct request_queue *, struct request *,
758 			   struct rq_map_data *, void __user *, unsigned long,
759 			   gfp_t);
760 extern int blk_rq_unmap_user(struct bio *);
761 extern int blk_rq_map_kern(struct request_queue *, struct request *, void *, unsigned int, gfp_t);
762 extern int blk_rq_map_user_iov(struct request_queue *, struct request *,
763 			       struct rq_map_data *, struct sg_iovec *, int,
764 			       unsigned int, gfp_t);
765 extern int blk_execute_rq(struct request_queue *, struct gendisk *,
766 			  struct request *, int);
767 extern void blk_execute_rq_nowait(struct request_queue *, struct gendisk *,
768 				  struct request *, int, rq_end_io_fn *);
769 extern void blk_unplug(struct request_queue *q);
770 
771 static inline struct request_queue *bdev_get_queue(struct block_device *bdev)
772 {
773 	return bdev->bd_disk->queue;
774 }
775 
776 static inline void blk_run_backing_dev(struct backing_dev_info *bdi,
777 				       struct page *page)
778 {
779 	if (bdi && bdi->unplug_io_fn)
780 		bdi->unplug_io_fn(bdi, page);
781 }
782 
783 static inline void blk_run_address_space(struct address_space *mapping)
784 {
785 	if (mapping)
786 		blk_run_backing_dev(mapping->backing_dev_info, NULL);
787 }
788 
789 /*
790  * blk_end_request() and friends.
791  * __blk_end_request() and end_request() must be called with
792  * the request queue spinlock acquired.
793  *
794  * Several drivers define their own end_request and call
795  * blk_end_request() for parts of the original function.
796  * This prevents code duplication in drivers.
797  */
798 extern int blk_end_request(struct request *rq, int error,
799 				unsigned int nr_bytes);
800 extern int __blk_end_request(struct request *rq, int error,
801 				unsigned int nr_bytes);
802 extern int blk_end_bidi_request(struct request *rq, int error,
803 				unsigned int nr_bytes, unsigned int bidi_bytes);
804 extern void end_request(struct request *, int);
805 extern int blk_end_request_callback(struct request *rq, int error,
806 				unsigned int nr_bytes,
807 				int (drv_callback)(struct request *));
808 extern void blk_complete_request(struct request *);
809 extern void __blk_complete_request(struct request *);
810 extern void blk_abort_request(struct request *);
811 extern void blk_abort_queue(struct request_queue *);
812 extern void blk_update_request(struct request *rq, int error,
813 			       unsigned int nr_bytes);
814 
815 /*
816  * blk_end_request() takes bytes instead of sectors as a complete size.
817  * blk_rq_bytes() returns bytes left to complete in the entire request.
818  * blk_rq_cur_bytes() returns bytes left to complete in the current segment.
819  */
820 extern unsigned int blk_rq_bytes(struct request *rq);
821 extern unsigned int blk_rq_cur_bytes(struct request *rq);
822 
823 static inline void blkdev_dequeue_request(struct request *req)
824 {
825 	elv_dequeue_request(req->q, req);
826 }
827 
828 /*
829  * Access functions for manipulating queue properties
830  */
831 extern struct request_queue *blk_init_queue_node(request_fn_proc *rfn,
832 					spinlock_t *lock, int node_id);
833 extern struct request_queue *blk_init_queue(request_fn_proc *, spinlock_t *);
834 extern void blk_cleanup_queue(struct request_queue *);
835 extern void blk_queue_make_request(struct request_queue *, make_request_fn *);
836 extern void blk_queue_bounce_limit(struct request_queue *, u64);
837 extern void blk_queue_max_sectors(struct request_queue *, unsigned int);
838 extern void blk_queue_max_phys_segments(struct request_queue *, unsigned short);
839 extern void blk_queue_max_hw_segments(struct request_queue *, unsigned short);
840 extern void blk_queue_max_segment_size(struct request_queue *, unsigned int);
841 extern void blk_queue_hardsect_size(struct request_queue *, unsigned short);
842 extern void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b);
843 extern void blk_queue_dma_pad(struct request_queue *, unsigned int);
844 extern void blk_queue_update_dma_pad(struct request_queue *, unsigned int);
845 extern int blk_queue_dma_drain(struct request_queue *q,
846 			       dma_drain_needed_fn *dma_drain_needed,
847 			       void *buf, unsigned int size);
848 extern void blk_queue_lld_busy(struct request_queue *q, lld_busy_fn *fn);
849 extern void blk_queue_segment_boundary(struct request_queue *, unsigned long);
850 extern void blk_queue_prep_rq(struct request_queue *, prep_rq_fn *pfn);
851 extern void blk_queue_merge_bvec(struct request_queue *, merge_bvec_fn *);
852 extern void blk_queue_dma_alignment(struct request_queue *, int);
853 extern void blk_queue_update_dma_alignment(struct request_queue *, int);
854 extern void blk_queue_softirq_done(struct request_queue *, softirq_done_fn *);
855 extern void blk_queue_set_discard(struct request_queue *, prepare_discard_fn *);
856 extern void blk_queue_rq_timed_out(struct request_queue *, rq_timed_out_fn *);
857 extern void blk_queue_rq_timeout(struct request_queue *, unsigned int);
858 extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev);
859 extern int blk_queue_ordered(struct request_queue *, unsigned, prepare_flush_fn *);
860 extern int blk_do_ordered(struct request_queue *, struct request **);
861 extern unsigned blk_ordered_cur_seq(struct request_queue *);
862 extern unsigned blk_ordered_req_seq(struct request *);
863 extern void blk_ordered_complete_seq(struct request_queue *, unsigned, int);
864 
865 extern int blk_rq_map_sg(struct request_queue *, struct request *, struct scatterlist *);
866 extern void blk_dump_rq_flags(struct request *, char *);
867 extern void generic_unplug_device(struct request_queue *);
868 extern long nr_blockdev_pages(void);
869 
870 int blk_get_queue(struct request_queue *);
871 struct request_queue *blk_alloc_queue(gfp_t);
872 struct request_queue *blk_alloc_queue_node(gfp_t, int);
873 extern void blk_put_queue(struct request_queue *);
874 
875 /*
876  * tag stuff
877  */
878 #define blk_rq_tagged(rq)		((rq)->cmd_flags & REQ_QUEUED)
879 extern int blk_queue_start_tag(struct request_queue *, struct request *);
880 extern struct request *blk_queue_find_tag(struct request_queue *, int);
881 extern void blk_queue_end_tag(struct request_queue *, struct request *);
882 extern int blk_queue_init_tags(struct request_queue *, int, struct blk_queue_tag *);
883 extern void blk_queue_free_tags(struct request_queue *);
884 extern int blk_queue_resize_tags(struct request_queue *, int);
885 extern void blk_queue_invalidate_tags(struct request_queue *);
886 extern struct blk_queue_tag *blk_init_tags(int);
887 extern void blk_free_tags(struct blk_queue_tag *);
888 
889 static inline struct request *blk_map_queue_find_tag(struct blk_queue_tag *bqt,
890 						int tag)
891 {
892 	if (unlikely(bqt == NULL || tag >= bqt->real_max_depth))
893 		return NULL;
894 	return bqt->tag_index[tag];
895 }
896 
897 extern int blkdev_issue_flush(struct block_device *, sector_t *);
898 extern int blkdev_issue_discard(struct block_device *,
899 				sector_t sector, sector_t nr_sects, gfp_t);
900 
901 static inline int sb_issue_discard(struct super_block *sb,
902 				   sector_t block, sector_t nr_blocks)
903 {
904 	block <<= (sb->s_blocksize_bits - 9);
905 	nr_blocks <<= (sb->s_blocksize_bits - 9);
906 	return blkdev_issue_discard(sb->s_bdev, block, nr_blocks, GFP_KERNEL);
907 }
908 
909 /*
910 * command filter functions
911 */
912 extern int blk_verify_command(struct blk_cmd_filter *filter,
913 			      unsigned char *cmd, int has_write_perm);
914 extern void blk_set_cmd_filter_defaults(struct blk_cmd_filter *filter);
915 
916 #define MAX_PHYS_SEGMENTS 128
917 #define MAX_HW_SEGMENTS 128
918 #define SAFE_MAX_SECTORS 255
919 #define BLK_DEF_MAX_SECTORS 1024
920 
921 #define MAX_SEGMENT_SIZE	65536
922 
923 #define blkdev_entry_to_request(entry) list_entry((entry), struct request, queuelist)
924 
925 static inline int queue_hardsect_size(struct request_queue *q)
926 {
927 	int retval = 512;
928 
929 	if (q && q->hardsect_size)
930 		retval = q->hardsect_size;
931 
932 	return retval;
933 }
934 
935 static inline int bdev_hardsect_size(struct block_device *bdev)
936 {
937 	return queue_hardsect_size(bdev_get_queue(bdev));
938 }
939 
940 static inline int queue_dma_alignment(struct request_queue *q)
941 {
942 	return q ? q->dma_alignment : 511;
943 }
944 
945 static inline int blk_rq_aligned(struct request_queue *q, void *addr,
946 				 unsigned int len)
947 {
948 	unsigned int alignment = queue_dma_alignment(q) | q->dma_pad_mask;
949 	return !((unsigned long)addr & alignment) && !(len & alignment);
950 }
951 
952 /* assumes size > 256 */
953 static inline unsigned int blksize_bits(unsigned int size)
954 {
955 	unsigned int bits = 8;
956 	do {
957 		bits++;
958 		size >>= 1;
959 	} while (size > 256);
960 	return bits;
961 }
962 
963 static inline unsigned int block_size(struct block_device *bdev)
964 {
965 	return bdev->bd_block_size;
966 }
967 
968 typedef struct {struct page *v;} Sector;
969 
970 unsigned char *read_dev_sector(struct block_device *, sector_t, Sector *);
971 
972 static inline void put_dev_sector(Sector p)
973 {
974 	page_cache_release(p.v);
975 }
976 
977 struct work_struct;
978 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work);
979 void kblockd_flush_work(struct work_struct *work);
980 
981 #define MODULE_ALIAS_BLOCKDEV(major,minor) \
982 	MODULE_ALIAS("block-major-" __stringify(major) "-" __stringify(minor))
983 #define MODULE_ALIAS_BLOCKDEV_MAJOR(major) \
984 	MODULE_ALIAS("block-major-" __stringify(major) "-*")
985 
986 #if defined(CONFIG_BLK_DEV_INTEGRITY)
987 
988 #define INTEGRITY_FLAG_READ	2	/* verify data integrity on read */
989 #define INTEGRITY_FLAG_WRITE	4	/* generate data integrity on write */
990 
991 struct blk_integrity_exchg {
992 	void			*prot_buf;
993 	void			*data_buf;
994 	sector_t		sector;
995 	unsigned int		data_size;
996 	unsigned short		sector_size;
997 	const char		*disk_name;
998 };
999 
1000 typedef void (integrity_gen_fn) (struct blk_integrity_exchg *);
1001 typedef int (integrity_vrfy_fn) (struct blk_integrity_exchg *);
1002 typedef void (integrity_set_tag_fn) (void *, void *, unsigned int);
1003 typedef void (integrity_get_tag_fn) (void *, void *, unsigned int);
1004 
1005 struct blk_integrity {
1006 	integrity_gen_fn	*generate_fn;
1007 	integrity_vrfy_fn	*verify_fn;
1008 	integrity_set_tag_fn	*set_tag_fn;
1009 	integrity_get_tag_fn	*get_tag_fn;
1010 
1011 	unsigned short		flags;
1012 	unsigned short		tuple_size;
1013 	unsigned short		sector_size;
1014 	unsigned short		tag_size;
1015 
1016 	const char		*name;
1017 
1018 	struct kobject		kobj;
1019 };
1020 
1021 extern int blk_integrity_register(struct gendisk *, struct blk_integrity *);
1022 extern void blk_integrity_unregister(struct gendisk *);
1023 extern int blk_integrity_compare(struct gendisk *, struct gendisk *);
1024 extern int blk_rq_map_integrity_sg(struct request *, struct scatterlist *);
1025 extern int blk_rq_count_integrity_sg(struct request *);
1026 
1027 static inline
1028 struct blk_integrity *bdev_get_integrity(struct block_device *bdev)
1029 {
1030 	return bdev->bd_disk->integrity;
1031 }
1032 
1033 static inline struct blk_integrity *blk_get_integrity(struct gendisk *disk)
1034 {
1035 	return disk->integrity;
1036 }
1037 
1038 static inline int blk_integrity_rq(struct request *rq)
1039 {
1040 	if (rq->bio == NULL)
1041 		return 0;
1042 
1043 	return bio_integrity(rq->bio);
1044 }
1045 
1046 #else /* CONFIG_BLK_DEV_INTEGRITY */
1047 
1048 #define blk_integrity_rq(rq)			(0)
1049 #define blk_rq_count_integrity_sg(a)		(0)
1050 #define blk_rq_map_integrity_sg(a, b)		(0)
1051 #define bdev_get_integrity(a)			(0)
1052 #define blk_get_integrity(a)			(0)
1053 #define blk_integrity_compare(a, b)		(0)
1054 #define blk_integrity_register(a, b)		(0)
1055 #define blk_integrity_unregister(a)		do { } while (0);
1056 
1057 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1058 
1059 #else /* CONFIG_BLOCK */
1060 /*
1061  * stubs for when the block layer is configured out
1062  */
1063 #define buffer_heads_over_limit 0
1064 
1065 static inline long nr_blockdev_pages(void)
1066 {
1067 	return 0;
1068 }
1069 
1070 #endif /* CONFIG_BLOCK */
1071 
1072 #endif
1073