xref: /linux/block/blk.h (revision 23b0f90ba871f096474e1c27c3d14f455189d2d9)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef BLK_INTERNAL_H
3 #define BLK_INTERNAL_H
4 
5 #include <linux/bio-integrity.h>
6 #include <linux/blk-crypto.h>
7 #include <linux/lockdep.h>
8 #include <linux/memblock.h>	/* for max_pfn/max_low_pfn */
9 #include <linux/sched/sysctl.h>
10 #include <linux/timekeeping.h>
11 #include <xen/xen.h>
12 #include "blk-crypto-internal.h"
13 
14 struct elv_change_ctx;
15 
16 /*
17  * Default upper limit for the software max_sectors limit used for regular I/Os.
18  * This can be increased through sysfs.
19  *
20  * This should not be confused with the max_hw_sector limit that is entirely
21  * controlled by the block device driver, usually based on hardware limits.
22  */
23 #define BLK_DEF_MAX_SECTORS_CAP	(SZ_4M >> SECTOR_SHIFT)
24 
25 #define	BLK_DEV_MAX_SECTORS	(LLONG_MAX >> 9)
26 #define	BLK_MIN_SEGMENT_SIZE	4096
27 
28 /* Max future timer expiry for timeouts */
29 #define BLK_MAX_TIMEOUT		(5 * HZ)
30 
31 extern const struct kobj_type blk_queue_ktype;
32 extern struct dentry *blk_debugfs_root;
33 
34 struct blk_flush_queue {
35 	spinlock_t		mq_flush_lock;
36 	unsigned int		flush_pending_idx:1;
37 	unsigned int		flush_running_idx:1;
38 	blk_status_t 		rq_status;
39 	unsigned long		flush_pending_since;
40 	struct list_head	flush_queue[2];
41 	unsigned long		flush_data_in_flight;
42 	struct request		*flush_rq;
43 	struct rcu_head		rcu_head;
44 };
45 
46 bool is_flush_rq(struct request *req);
47 
48 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
49 					      gfp_t flags);
50 void blk_free_flush_queue(struct blk_flush_queue *q);
51 
52 bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
53 bool blk_queue_start_drain(struct request_queue *q);
54 bool __blk_freeze_queue_start(struct request_queue *q,
55 			      struct task_struct *owner);
56 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
57 void submit_bio_noacct_nocheck(struct bio *bio, bool split);
58 void bio_await_chain(struct bio *bio);
59 
60 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
61 {
62 	rcu_read_lock();
63 	if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
64 		goto fail;
65 
66 	/*
67 	 * The code that increments the pm_only counter must ensure that the
68 	 * counter is globally visible before the queue is unfrozen.
69 	 */
70 	if (blk_queue_pm_only(q) &&
71 	    (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
72 		goto fail_put;
73 
74 	rcu_read_unlock();
75 	return true;
76 
77 fail_put:
78 	blk_queue_exit(q);
79 fail:
80 	rcu_read_unlock();
81 	return false;
82 }
83 
84 static inline int bio_queue_enter(struct bio *bio)
85 {
86 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
87 
88 	if (blk_try_enter_queue(q, false)) {
89 		rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_);
90 		rwsem_release(&q->io_lockdep_map, _RET_IP_);
91 		return 0;
92 	}
93 	return __bio_queue_enter(q, bio);
94 }
95 
96 static inline void blk_wait_io(struct completion *done)
97 {
98 	/* Prevent hang_check timer from firing at us during very long I/O */
99 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
100 
101 	if (timeout)
102 		while (!wait_for_completion_io_timeout(done, timeout))
103 			;
104 	else
105 		wait_for_completion_io(done);
106 }
107 
108 struct block_device *blkdev_get_no_open(dev_t dev, bool autoload);
109 void blkdev_put_no_open(struct block_device *bdev);
110 
111 #define BIO_INLINE_VECS 4
112 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
113 		gfp_t gfp_mask);
114 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
115 
116 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
117 		struct page *page, unsigned len, unsigned offset);
118 
119 static inline bool biovec_phys_mergeable(struct request_queue *q,
120 		struct bio_vec *vec1, struct bio_vec *vec2)
121 {
122 	unsigned long mask = queue_segment_boundary(q);
123 	phys_addr_t addr1 = bvec_phys(vec1);
124 	phys_addr_t addr2 = bvec_phys(vec2);
125 
126 	/*
127 	 * Merging adjacent physical pages may not work correctly under KMSAN
128 	 * if their metadata pages aren't adjacent. Just disable merging.
129 	 */
130 	if (IS_ENABLED(CONFIG_KMSAN))
131 		return false;
132 
133 	if (addr1 + vec1->bv_len != addr2)
134 		return false;
135 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
136 		return false;
137 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
138 		return false;
139 	return true;
140 }
141 
142 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
143 		struct bio_vec *bprv, unsigned int offset)
144 {
145 	return (offset & lim->virt_boundary_mask) ||
146 		((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
147 }
148 
149 /*
150  * Check if adding a bio_vec after bprv with offset would create a gap in
151  * the SG list. Most drivers don't care about this, but some do.
152  */
153 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
154 		struct bio_vec *bprv, unsigned int offset)
155 {
156 	if (!lim->virt_boundary_mask)
157 		return false;
158 	return __bvec_gap_to_prev(lim, bprv, offset);
159 }
160 
161 static inline bool rq_mergeable(struct request *rq)
162 {
163 	if (blk_rq_is_passthrough(rq))
164 		return false;
165 
166 	if (req_op(rq) == REQ_OP_FLUSH)
167 		return false;
168 
169 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
170 		return false;
171 
172 	if (req_op(rq) == REQ_OP_ZONE_APPEND)
173 		return false;
174 
175 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
176 		return false;
177 	if (rq->rq_flags & RQF_NOMERGE_FLAGS)
178 		return false;
179 
180 	return true;
181 }
182 
183 /*
184  * There are two different ways to handle DISCARD merges:
185  *  1) If max_discard_segments > 1, the driver treats every bio as a range and
186  *     send the bios to controller together. The ranges don't need to be
187  *     contiguous.
188  *  2) Otherwise, the request will be normal read/write requests.  The ranges
189  *     need to be contiguous.
190  */
191 static inline bool blk_discard_mergable(struct request *req)
192 {
193 	if (req_op(req) == REQ_OP_DISCARD &&
194 	    queue_max_discard_segments(req->q) > 1)
195 		return true;
196 	return false;
197 }
198 
199 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
200 {
201 	if (req_op(rq) == REQ_OP_DISCARD)
202 		return queue_max_discard_segments(rq->q);
203 	return queue_max_segments(rq->q);
204 }
205 
206 static inline unsigned int blk_queue_get_max_sectors(struct request *rq)
207 {
208 	struct request_queue *q = rq->q;
209 	enum req_op op = req_op(rq);
210 
211 	if (unlikely(op == REQ_OP_DISCARD))
212 		return min(q->limits.max_discard_sectors,
213 			   UINT_MAX >> SECTOR_SHIFT);
214 
215 	if (unlikely(op == REQ_OP_SECURE_ERASE))
216 		return min(q->limits.max_secure_erase_sectors,
217 			   UINT_MAX >> SECTOR_SHIFT);
218 
219 	if (unlikely(op == REQ_OP_WRITE_ZEROES))
220 		return q->limits.max_write_zeroes_sectors;
221 
222 	if (rq->cmd_flags & REQ_ATOMIC)
223 		return q->limits.atomic_write_max_sectors;
224 
225 	return q->limits.max_sectors;
226 }
227 
228 #ifdef CONFIG_BLK_DEV_INTEGRITY
229 void blk_flush_integrity(void);
230 void bio_integrity_free(struct bio *bio);
231 
232 /*
233  * Integrity payloads can either be owned by the submitter, in which case
234  * bio_uninit will free them, or owned and generated by the block layer,
235  * in which case we'll verify them here (for reads) and free them before
236  * the bio is handed back to the submitted.
237  */
238 bool __bio_integrity_endio(struct bio *bio);
239 static inline bool bio_integrity_endio(struct bio *bio)
240 {
241 	struct bio_integrity_payload *bip = bio_integrity(bio);
242 
243 	if (bip && (bip->bip_flags & BIP_BLOCK_INTEGRITY))
244 		return __bio_integrity_endio(bio);
245 	return true;
246 }
247 
248 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
249 		struct request *);
250 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
251 		struct bio *);
252 
253 static inline bool integrity_req_gap_back_merge(struct request *req,
254 		struct bio *next)
255 {
256 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
257 	struct bio_integrity_payload *bip_next = bio_integrity(next);
258 
259 	return bvec_gap_to_prev(&req->q->limits,
260 				&bip->bip_vec[bip->bip_vcnt - 1],
261 				bip_next->bip_vec[0].bv_offset);
262 }
263 
264 static inline bool integrity_req_gap_front_merge(struct request *req,
265 		struct bio *bio)
266 {
267 	struct bio_integrity_payload *bip = bio_integrity(bio);
268 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
269 
270 	return bvec_gap_to_prev(&req->q->limits,
271 				&bip->bip_vec[bip->bip_vcnt - 1],
272 				bip_next->bip_vec[0].bv_offset);
273 }
274 
275 extern const struct attribute_group blk_integrity_attr_group;
276 #else /* CONFIG_BLK_DEV_INTEGRITY */
277 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
278 		struct request *r1, struct request *r2)
279 {
280 	return true;
281 }
282 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
283 		struct request *r, struct bio *b)
284 {
285 	return true;
286 }
287 static inline bool integrity_req_gap_back_merge(struct request *req,
288 		struct bio *next)
289 {
290 	return false;
291 }
292 static inline bool integrity_req_gap_front_merge(struct request *req,
293 		struct bio *bio)
294 {
295 	return false;
296 }
297 
298 static inline void blk_flush_integrity(void)
299 {
300 }
301 static inline bool bio_integrity_endio(struct bio *bio)
302 {
303 	return true;
304 }
305 static inline void bio_integrity_free(struct bio *bio)
306 {
307 }
308 #endif /* CONFIG_BLK_DEV_INTEGRITY */
309 
310 unsigned long blk_rq_timeout(unsigned long timeout);
311 void blk_add_timer(struct request *req);
312 
313 enum bio_merge_status {
314 	BIO_MERGE_OK,
315 	BIO_MERGE_NONE,
316 	BIO_MERGE_FAILED,
317 };
318 
319 enum bio_merge_status bio_attempt_back_merge(struct request *req,
320 		struct bio *bio, unsigned int nr_segs);
321 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
322 		unsigned int nr_segs);
323 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
324 			struct bio *bio, unsigned int nr_segs);
325 
326 /*
327  * Plug flush limits
328  */
329 #define BLK_MAX_REQUEST_COUNT	32
330 #define BLK_PLUG_FLUSH_SIZE	(128 * 1024)
331 
332 /*
333  * Internal elevator interface
334  */
335 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
336 
337 bool blk_insert_flush(struct request *rq);
338 
339 void elv_update_nr_hw_queues(struct request_queue *q,
340 		struct elv_change_ctx *ctx);
341 void elevator_set_default(struct request_queue *q);
342 void elevator_set_none(struct request_queue *q);
343 
344 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
345 		char *buf);
346 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
347 		char *buf);
348 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
349 		char *buf);
350 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
351 		char *buf);
352 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
353 		const char *buf, size_t count);
354 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
355 ssize_t part_timeout_store(struct device *, struct device_attribute *,
356 				const char *, size_t);
357 
358 struct bio *bio_split_discard(struct bio *bio, const struct queue_limits *lim,
359 		unsigned *nsegs);
360 struct bio *bio_split_write_zeroes(struct bio *bio,
361 		const struct queue_limits *lim, unsigned *nsegs);
362 struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim,
363 		unsigned *nr_segs);
364 struct bio *bio_split_zone_append(struct bio *bio,
365 		const struct queue_limits *lim, unsigned *nr_segs);
366 
367 /*
368  * All drivers must accept single-segments bios that are smaller than PAGE_SIZE.
369  *
370  * This is a quick and dirty check that relies on the fact that bi_io_vec[0] is
371  * always valid if a bio has data.  The check might lead to occasional false
372  * positives when bios are cloned, but compared to the performance impact of
373  * cloned bios themselves the loop below doesn't matter anyway.
374  */
375 static inline bool bio_may_need_split(struct bio *bio,
376 		const struct queue_limits *lim)
377 {
378 	const struct bio_vec *bv;
379 
380 	if (lim->chunk_sectors)
381 		return true;
382 
383 	if (!bio->bi_io_vec)
384 		return true;
385 
386 	bv = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
387 	if (bio->bi_iter.bi_size > bv->bv_len - bio->bi_iter.bi_bvec_done)
388 		return true;
389 	return bv->bv_len + bv->bv_offset > lim->max_fast_segment_size;
390 }
391 
392 /**
393  * __bio_split_to_limits - split a bio to fit the queue limits
394  * @bio:     bio to be split
395  * @lim:     queue limits to split based on
396  * @nr_segs: returns the number of segments in the returned bio
397  *
398  * Check if @bio needs splitting based on the queue limits, and if so split off
399  * a bio fitting the limits from the beginning of @bio and return it.  @bio is
400  * shortened to the remainder and re-submitted.
401  *
402  * The split bio is allocated from @q->bio_split, which is provided by the
403  * block layer.
404  */
405 static inline struct bio *__bio_split_to_limits(struct bio *bio,
406 		const struct queue_limits *lim, unsigned int *nr_segs)
407 {
408 	switch (bio_op(bio)) {
409 	case REQ_OP_READ:
410 	case REQ_OP_WRITE:
411 		if (bio_may_need_split(bio, lim))
412 			return bio_split_rw(bio, lim, nr_segs);
413 		*nr_segs = 1;
414 		return bio;
415 	case REQ_OP_ZONE_APPEND:
416 		return bio_split_zone_append(bio, lim, nr_segs);
417 	case REQ_OP_DISCARD:
418 	case REQ_OP_SECURE_ERASE:
419 		return bio_split_discard(bio, lim, nr_segs);
420 	case REQ_OP_WRITE_ZEROES:
421 		return bio_split_write_zeroes(bio, lim, nr_segs);
422 	default:
423 		/* other operations can't be split */
424 		*nr_segs = 0;
425 		return bio;
426 	}
427 }
428 
429 /**
430  * get_max_segment_size() - maximum number of bytes to add as a single segment
431  * @lim: Request queue limits.
432  * @paddr: address of the range to add
433  * @len: maximum length available to add at @paddr
434  *
435  * Returns the maximum number of bytes of the range starting at @paddr that can
436  * be added to a single segment.
437  */
438 static inline unsigned get_max_segment_size(const struct queue_limits *lim,
439 		phys_addr_t paddr, unsigned int len)
440 {
441 	/*
442 	 * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1
443 	 * after having calculated the minimum.
444 	 */
445 	return min_t(unsigned long, len,
446 		min(lim->seg_boundary_mask - (lim->seg_boundary_mask & paddr),
447 		    (unsigned long)lim->max_segment_size - 1) + 1);
448 }
449 
450 int ll_back_merge_fn(struct request *req, struct bio *bio,
451 		unsigned int nr_segs);
452 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
453 				struct request *next);
454 unsigned int blk_recalc_rq_segments(struct request *rq);
455 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
456 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
457 
458 int blk_set_default_limits(struct queue_limits *lim);
459 void blk_apply_bdi_limits(struct backing_dev_info *bdi,
460 		struct queue_limits *lim);
461 int blk_dev_init(void);
462 
463 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
464 
465 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
466 {
467 	req->cmd_flags |= REQ_NOMERGE;
468 	if (req == q->last_merge)
469 		q->last_merge = NULL;
470 }
471 
472 /*
473  * Internal io_context interface
474  */
475 struct io_cq *ioc_find_get_icq(struct request_queue *q);
476 struct io_cq *ioc_lookup_icq(struct request_queue *q);
477 #ifdef CONFIG_BLK_ICQ
478 void ioc_clear_queue(struct request_queue *q);
479 #else
480 static inline void ioc_clear_queue(struct request_queue *q)
481 {
482 }
483 #endif /* CONFIG_BLK_ICQ */
484 
485 #ifdef CONFIG_BLK_DEV_ZONED
486 void disk_init_zone_resources(struct gendisk *disk);
487 void disk_free_zone_resources(struct gendisk *disk);
488 static inline bool bio_zone_write_plugging(struct bio *bio)
489 {
490 	return bio_flagged(bio, BIO_ZONE_WRITE_PLUGGING);
491 }
492 static inline bool blk_req_bio_is_zone_append(struct request *rq,
493 					      struct bio *bio)
494 {
495 	return req_op(rq) == REQ_OP_ZONE_APPEND ||
496 	       bio_flagged(bio, BIO_EMULATES_ZONE_APPEND);
497 }
498 void blk_zone_write_plug_bio_merged(struct bio *bio);
499 void blk_zone_write_plug_init_request(struct request *rq);
500 void blk_zone_append_update_request_bio(struct request *rq, struct bio *bio);
501 void blk_zone_mgmt_bio_endio(struct bio *bio);
502 void blk_zone_write_plug_bio_endio(struct bio *bio);
503 static inline void blk_zone_bio_endio(struct bio *bio)
504 {
505 	/*
506 	 * Zone management BIOs may impact zone write plugs (e.g. a zone reset
507 	 * changes a zone write plug zone write pointer offset), but these
508 	 * operation do not go through zone write plugging as they may operate
509 	 * on zones that do not have a zone write
510 	 * plug. blk_zone_mgmt_bio_endio() handles the potential changes to zone
511 	 * write plugs that are present.
512 	 */
513 	if (op_is_zone_mgmt(bio_op(bio))) {
514 		blk_zone_mgmt_bio_endio(bio);
515 		return;
516 	}
517 
518 	/*
519 	 * For write BIOs to zoned devices, signal the completion of the BIO so
520 	 * that the next write BIO can be submitted by zone write plugging.
521 	 */
522 	if (bio_zone_write_plugging(bio))
523 		blk_zone_write_plug_bio_endio(bio);
524 }
525 
526 void blk_zone_write_plug_finish_request(struct request *rq);
527 static inline void blk_zone_finish_request(struct request *rq)
528 {
529 	if (rq->rq_flags & RQF_ZONE_WRITE_PLUGGING)
530 		blk_zone_write_plug_finish_request(rq);
531 }
532 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
533 		unsigned long arg);
534 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
535 		unsigned int cmd, unsigned long arg);
536 #else /* CONFIG_BLK_DEV_ZONED */
537 static inline void disk_init_zone_resources(struct gendisk *disk)
538 {
539 }
540 static inline void disk_free_zone_resources(struct gendisk *disk)
541 {
542 }
543 static inline bool bio_zone_write_plugging(struct bio *bio)
544 {
545 	return false;
546 }
547 static inline bool blk_req_bio_is_zone_append(struct request *req,
548 					      struct bio *bio)
549 {
550 	return false;
551 }
552 static inline void blk_zone_write_plug_bio_merged(struct bio *bio)
553 {
554 }
555 static inline void blk_zone_write_plug_init_request(struct request *rq)
556 {
557 }
558 static inline void blk_zone_append_update_request_bio(struct request *rq,
559 						      struct bio *bio)
560 {
561 }
562 static inline void blk_zone_bio_endio(struct bio *bio)
563 {
564 }
565 static inline void blk_zone_finish_request(struct request *rq)
566 {
567 }
568 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
569 		unsigned int cmd, unsigned long arg)
570 {
571 	return -ENOTTY;
572 }
573 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
574 		blk_mode_t mode, unsigned int cmd, unsigned long arg)
575 {
576 	return -ENOTTY;
577 }
578 #endif /* CONFIG_BLK_DEV_ZONED */
579 
580 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
581 void bdev_add(struct block_device *bdev, dev_t dev);
582 void bdev_unhash(struct block_device *bdev);
583 void bdev_drop(struct block_device *bdev);
584 
585 int blk_alloc_ext_minor(void);
586 void blk_free_ext_minor(unsigned int minor);
587 #define ADDPART_FLAG_NONE	0
588 #define ADDPART_FLAG_RAID	1
589 #define ADDPART_FLAG_WHOLEDISK	2
590 #define ADDPART_FLAG_READONLY	4
591 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
592 		sector_t length);
593 int bdev_del_partition(struct gendisk *disk, int partno);
594 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
595 		sector_t length);
596 void drop_partition(struct block_device *part);
597 
598 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
599 
600 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
601 		struct lock_class_key *lkclass);
602 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
603 
604 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
605 
606 int disk_alloc_events(struct gendisk *disk);
607 void disk_add_events(struct gendisk *disk);
608 void disk_del_events(struct gendisk *disk);
609 void disk_release_events(struct gendisk *disk);
610 void disk_block_events(struct gendisk *disk);
611 void disk_unblock_events(struct gendisk *disk);
612 void disk_flush_events(struct gendisk *disk, unsigned int mask);
613 extern struct device_attribute dev_attr_events;
614 extern struct device_attribute dev_attr_events_async;
615 extern struct device_attribute dev_attr_events_poll_msecs;
616 
617 extern struct attribute_group blk_trace_attr_group;
618 
619 blk_mode_t file_to_blk_mode(struct file *file);
620 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
621 		loff_t lstart, loff_t lend);
622 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
623 int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags);
624 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
625 
626 extern const struct address_space_operations def_blk_aops;
627 
628 int disk_register_independent_access_ranges(struct gendisk *disk);
629 void disk_unregister_independent_access_ranges(struct gendisk *disk);
630 
631 int should_fail_bio(struct bio *bio);
632 #ifdef CONFIG_FAIL_MAKE_REQUEST
633 bool should_fail_request(struct block_device *part, unsigned int bytes);
634 #else /* CONFIG_FAIL_MAKE_REQUEST */
635 static inline bool should_fail_request(struct block_device *part,
636 					unsigned int bytes)
637 {
638 	return false;
639 }
640 #endif /* CONFIG_FAIL_MAKE_REQUEST */
641 
642 /*
643  * Optimized request reference counting. Ideally we'd make timeouts be more
644  * clever, as that's the only reason we need references at all... But until
645  * this happens, this is faster than using refcount_t. Also see:
646  *
647  * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
648  */
649 #define req_ref_zero_or_close_to_overflow(req)	\
650 	((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
651 
652 static inline bool req_ref_inc_not_zero(struct request *req)
653 {
654 	return atomic_inc_not_zero(&req->ref);
655 }
656 
657 static inline bool req_ref_put_and_test(struct request *req)
658 {
659 	WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
660 	return atomic_dec_and_test(&req->ref);
661 }
662 
663 static inline void req_ref_set(struct request *req, int value)
664 {
665 	atomic_set(&req->ref, value);
666 }
667 
668 static inline int req_ref_read(struct request *req)
669 {
670 	return atomic_read(&req->ref);
671 }
672 
673 static inline u64 blk_time_get_ns(void)
674 {
675 	struct blk_plug *plug = current->plug;
676 
677 	if (!plug || !in_task())
678 		return ktime_get_ns();
679 
680 	/*
681 	 * 0 could very well be a valid time, but rather than flag "this is
682 	 * a valid timestamp" separately, just accept that we'll do an extra
683 	 * ktime_get_ns() if we just happen to get 0 as the current time.
684 	 */
685 	if (!plug->cur_ktime) {
686 		plug->cur_ktime = ktime_get_ns();
687 		current->flags |= PF_BLOCK_TS;
688 	}
689 	return plug->cur_ktime;
690 }
691 
692 static inline ktime_t blk_time_get(void)
693 {
694 	return ns_to_ktime(blk_time_get_ns());
695 }
696 
697 void bdev_release(struct file *bdev_file);
698 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
699 	      const struct blk_holder_ops *hops, struct file *bdev_file);
700 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
701 
702 void blk_integrity_generate(struct bio *bio);
703 void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter);
704 void blk_integrity_prepare(struct request *rq);
705 void blk_integrity_complete(struct request *rq, unsigned int nr_bytes);
706 
707 #ifdef CONFIG_LOCKDEP
708 static inline void blk_freeze_acquire_lock(struct request_queue *q)
709 {
710 	if (!q->mq_freeze_disk_dead)
711 		rwsem_acquire(&q->io_lockdep_map, 0, 1, _RET_IP_);
712 	if (!q->mq_freeze_queue_dying)
713 		rwsem_acquire(&q->q_lockdep_map, 0, 1, _RET_IP_);
714 }
715 
716 static inline void blk_unfreeze_release_lock(struct request_queue *q)
717 {
718 	if (!q->mq_freeze_queue_dying)
719 		rwsem_release(&q->q_lockdep_map, _RET_IP_);
720 	if (!q->mq_freeze_disk_dead)
721 		rwsem_release(&q->io_lockdep_map, _RET_IP_);
722 }
723 #else
724 static inline void blk_freeze_acquire_lock(struct request_queue *q)
725 {
726 }
727 static inline void blk_unfreeze_release_lock(struct request_queue *q)
728 {
729 }
730 #endif
731 
732 /*
733  * debugfs directory and file creation can trigger fs reclaim, which can enter
734  * back into the block layer request_queue. This can cause deadlock if the
735  * queue is frozen. Use NOIO context together with debugfs_mutex to prevent fs
736  * reclaim from triggering block I/O.
737  */
738 static inline void blk_debugfs_lock_nomemsave(struct request_queue *q)
739 {
740 	mutex_lock(&q->debugfs_mutex);
741 }
742 
743 static inline void blk_debugfs_unlock_nomemrestore(struct request_queue *q)
744 {
745 	mutex_unlock(&q->debugfs_mutex);
746 }
747 
748 static inline unsigned int __must_check blk_debugfs_lock(struct request_queue *q)
749 {
750 	unsigned int memflags = memalloc_noio_save();
751 
752 	blk_debugfs_lock_nomemsave(q);
753 	return memflags;
754 }
755 
756 static inline void blk_debugfs_unlock(struct request_queue *q,
757 				      unsigned int memflags)
758 {
759 	blk_debugfs_unlock_nomemrestore(q);
760 	memalloc_noio_restore(memflags);
761 }
762 
763 #endif /* BLK_INTERNAL_H */
764