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