1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef BLK_INTERNAL_H
3 #define BLK_INTERNAL_H
4
5 #include <linux/blk-crypto.h>
6 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */
7 #include <linux/sched/sysctl.h>
8 #include <linux/timekeeping.h>
9 #include <xen/xen.h>
10 #include "blk-crypto-internal.h"
11
12 struct elevator_type;
13
14 /* Max future timer expiry for timeouts */
15 #define BLK_MAX_TIMEOUT (5 * HZ)
16
17 extern struct dentry *blk_debugfs_root;
18
19 struct blk_flush_queue {
20 spinlock_t mq_flush_lock;
21 unsigned int flush_pending_idx:1;
22 unsigned int flush_running_idx:1;
23 blk_status_t rq_status;
24 unsigned long flush_pending_since;
25 struct list_head flush_queue[2];
26 unsigned long flush_data_in_flight;
27 struct request *flush_rq;
28 };
29
30 bool is_flush_rq(struct request *req);
31
32 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
33 gfp_t flags);
34 void blk_free_flush_queue(struct blk_flush_queue *q);
35
36 void blk_freeze_queue(struct request_queue *q);
37 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
38 void blk_queue_start_drain(struct request_queue *q);
39 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
40 void submit_bio_noacct_nocheck(struct bio *bio);
41 void bio_await_chain(struct bio *bio);
42
blk_try_enter_queue(struct request_queue * q,bool pm)43 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
44 {
45 rcu_read_lock();
46 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
47 goto fail;
48
49 /*
50 * The code that increments the pm_only counter must ensure that the
51 * counter is globally visible before the queue is unfrozen.
52 */
53 if (blk_queue_pm_only(q) &&
54 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
55 goto fail_put;
56
57 rcu_read_unlock();
58 return true;
59
60 fail_put:
61 blk_queue_exit(q);
62 fail:
63 rcu_read_unlock();
64 return false;
65 }
66
bio_queue_enter(struct bio * bio)67 static inline int bio_queue_enter(struct bio *bio)
68 {
69 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
70
71 if (blk_try_enter_queue(q, false))
72 return 0;
73 return __bio_queue_enter(q, bio);
74 }
75
blk_wait_io(struct completion * done)76 static inline void blk_wait_io(struct completion *done)
77 {
78 /* Prevent hang_check timer from firing at us during very long I/O */
79 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
80
81 if (timeout)
82 while (!wait_for_completion_io_timeout(done, timeout))
83 ;
84 else
85 wait_for_completion_io(done);
86 }
87
88 #define BIO_INLINE_VECS 4
89 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
90 gfp_t gfp_mask);
91 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
92
93 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
94 struct page *page, unsigned len, unsigned offset,
95 bool *same_page);
96
biovec_phys_mergeable(struct request_queue * q,struct bio_vec * vec1,struct bio_vec * vec2)97 static inline bool biovec_phys_mergeable(struct request_queue *q,
98 struct bio_vec *vec1, struct bio_vec *vec2)
99 {
100 unsigned long mask = queue_segment_boundary(q);
101 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
102 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
103
104 /*
105 * Merging adjacent physical pages may not work correctly under KMSAN
106 * if their metadata pages aren't adjacent. Just disable merging.
107 */
108 if (IS_ENABLED(CONFIG_KMSAN))
109 return false;
110
111 if (addr1 + vec1->bv_len != addr2)
112 return false;
113 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
114 return false;
115 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
116 return false;
117 return true;
118 }
119
__bvec_gap_to_prev(const struct queue_limits * lim,struct bio_vec * bprv,unsigned int offset)120 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
121 struct bio_vec *bprv, unsigned int offset)
122 {
123 return (offset & lim->virt_boundary_mask) ||
124 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
125 }
126
127 /*
128 * Check if adding a bio_vec after bprv with offset would create a gap in
129 * the SG list. Most drivers don't care about this, but some do.
130 */
bvec_gap_to_prev(const struct queue_limits * lim,struct bio_vec * bprv,unsigned int offset)131 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
132 struct bio_vec *bprv, unsigned int offset)
133 {
134 if (!lim->virt_boundary_mask)
135 return false;
136 return __bvec_gap_to_prev(lim, bprv, offset);
137 }
138
rq_mergeable(struct request * rq)139 static inline bool rq_mergeable(struct request *rq)
140 {
141 if (blk_rq_is_passthrough(rq))
142 return false;
143
144 if (req_op(rq) == REQ_OP_FLUSH)
145 return false;
146
147 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
148 return false;
149
150 if (req_op(rq) == REQ_OP_ZONE_APPEND)
151 return false;
152
153 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
154 return false;
155 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
156 return false;
157
158 return true;
159 }
160
161 /*
162 * There are two different ways to handle DISCARD merges:
163 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
164 * send the bios to controller together. The ranges don't need to be
165 * contiguous.
166 * 2) Otherwise, the request will be normal read/write requests. The ranges
167 * need to be contiguous.
168 */
blk_discard_mergable(struct request * req)169 static inline bool blk_discard_mergable(struct request *req)
170 {
171 if (req_op(req) == REQ_OP_DISCARD &&
172 queue_max_discard_segments(req->q) > 1)
173 return true;
174 return false;
175 }
176
blk_rq_get_max_segments(struct request * rq)177 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
178 {
179 if (req_op(rq) == REQ_OP_DISCARD)
180 return queue_max_discard_segments(rq->q);
181 return queue_max_segments(rq->q);
182 }
183
blk_queue_get_max_sectors(struct request_queue * q,enum req_op op)184 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
185 enum req_op op)
186 {
187 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
188 return min(q->limits.max_discard_sectors,
189 UINT_MAX >> SECTOR_SHIFT);
190
191 if (unlikely(op == REQ_OP_WRITE_ZEROES))
192 return q->limits.max_write_zeroes_sectors;
193
194 return q->limits.max_sectors;
195 }
196
197 #ifdef CONFIG_BLK_DEV_INTEGRITY
198 void blk_flush_integrity(void);
199 bool __bio_integrity_endio(struct bio *);
200 void bio_integrity_free(struct bio *bio);
bio_integrity_endio(struct bio * bio)201 static inline bool bio_integrity_endio(struct bio *bio)
202 {
203 if (bio_integrity(bio))
204 return __bio_integrity_endio(bio);
205 return true;
206 }
207
208 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
209 struct request *);
210 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
211 struct bio *);
212
integrity_req_gap_back_merge(struct request * req,struct bio * next)213 static inline bool integrity_req_gap_back_merge(struct request *req,
214 struct bio *next)
215 {
216 struct bio_integrity_payload *bip = bio_integrity(req->bio);
217 struct bio_integrity_payload *bip_next = bio_integrity(next);
218
219 return bvec_gap_to_prev(&req->q->limits,
220 &bip->bip_vec[bip->bip_vcnt - 1],
221 bip_next->bip_vec[0].bv_offset);
222 }
223
integrity_req_gap_front_merge(struct request * req,struct bio * bio)224 static inline bool integrity_req_gap_front_merge(struct request *req,
225 struct bio *bio)
226 {
227 struct bio_integrity_payload *bip = bio_integrity(bio);
228 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
229
230 return bvec_gap_to_prev(&req->q->limits,
231 &bip->bip_vec[bip->bip_vcnt - 1],
232 bip_next->bip_vec[0].bv_offset);
233 }
234
235 extern const struct attribute_group blk_integrity_attr_group;
236 #else /* CONFIG_BLK_DEV_INTEGRITY */
blk_integrity_merge_rq(struct request_queue * rq,struct request * r1,struct request * r2)237 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
238 struct request *r1, struct request *r2)
239 {
240 return true;
241 }
blk_integrity_merge_bio(struct request_queue * rq,struct request * r,struct bio * b)242 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
243 struct request *r, struct bio *b)
244 {
245 return true;
246 }
integrity_req_gap_back_merge(struct request * req,struct bio * next)247 static inline bool integrity_req_gap_back_merge(struct request *req,
248 struct bio *next)
249 {
250 return false;
251 }
integrity_req_gap_front_merge(struct request * req,struct bio * bio)252 static inline bool integrity_req_gap_front_merge(struct request *req,
253 struct bio *bio)
254 {
255 return false;
256 }
257
blk_flush_integrity(void)258 static inline void blk_flush_integrity(void)
259 {
260 }
bio_integrity_endio(struct bio * bio)261 static inline bool bio_integrity_endio(struct bio *bio)
262 {
263 return true;
264 }
bio_integrity_free(struct bio * bio)265 static inline void bio_integrity_free(struct bio *bio)
266 {
267 }
268 #endif /* CONFIG_BLK_DEV_INTEGRITY */
269
270 unsigned long blk_rq_timeout(unsigned long timeout);
271 void blk_add_timer(struct request *req);
272
273 enum bio_merge_status {
274 BIO_MERGE_OK,
275 BIO_MERGE_NONE,
276 BIO_MERGE_FAILED,
277 };
278
279 enum bio_merge_status bio_attempt_back_merge(struct request *req,
280 struct bio *bio, unsigned int nr_segs);
281 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
282 unsigned int nr_segs);
283 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
284 struct bio *bio, unsigned int nr_segs);
285
286 /*
287 * Plug flush limits
288 */
289 #define BLK_MAX_REQUEST_COUNT 32
290 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
291
292 /*
293 * Internal elevator interface
294 */
295 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
296
297 bool blk_insert_flush(struct request *rq);
298
299 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
300 void elevator_disable(struct request_queue *q);
301 void elevator_exit(struct request_queue *q);
302 int elv_register_queue(struct request_queue *q, bool uevent);
303 void elv_unregister_queue(struct request_queue *q);
304
305 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
306 char *buf);
307 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
308 char *buf);
309 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
310 char *buf);
311 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
312 char *buf);
313 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
314 const char *buf, size_t count);
315 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
316 ssize_t part_timeout_store(struct device *, struct device_attribute *,
317 const char *, size_t);
318
bio_may_exceed_limits(struct bio * bio,const struct queue_limits * lim)319 static inline bool bio_may_exceed_limits(struct bio *bio,
320 const struct queue_limits *lim)
321 {
322 switch (bio_op(bio)) {
323 case REQ_OP_DISCARD:
324 case REQ_OP_SECURE_ERASE:
325 case REQ_OP_WRITE_ZEROES:
326 return true; /* non-trivial splitting decisions */
327 default:
328 break;
329 }
330
331 /*
332 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
333 * This is a quick and dirty check that relies on the fact that
334 * bi_io_vec[0] is always valid if a bio has data. The check might
335 * lead to occasional false negatives when bios are cloned, but compared
336 * to the performance impact of cloned bios themselves the loop below
337 * doesn't matter anyway.
338 */
339 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
340 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
341 }
342
343 struct bio *__bio_split_to_limits(struct bio *bio,
344 const struct queue_limits *lim,
345 unsigned int *nr_segs);
346 int ll_back_merge_fn(struct request *req, struct bio *bio,
347 unsigned int nr_segs);
348 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
349 struct request *next);
350 unsigned int blk_recalc_rq_segments(struct request *rq);
351 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
352 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
353
354 int blk_set_default_limits(struct queue_limits *lim);
355 int blk_dev_init(void);
356
357 /*
358 * Contribute to IO statistics IFF:
359 *
360 * a) it's attached to a gendisk, and
361 * b) the queue had IO stats enabled when this request was started
362 */
blk_do_io_stat(struct request * rq)363 static inline bool blk_do_io_stat(struct request *rq)
364 {
365 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
366 }
367
368 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
369 unsigned int part_in_flight(struct block_device *part);
370
req_set_nomerge(struct request_queue * q,struct request * req)371 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
372 {
373 req->cmd_flags |= REQ_NOMERGE;
374 if (req == q->last_merge)
375 q->last_merge = NULL;
376 }
377
378 /*
379 * Internal io_context interface
380 */
381 struct io_cq *ioc_find_get_icq(struct request_queue *q);
382 struct io_cq *ioc_lookup_icq(struct request_queue *q);
383 #ifdef CONFIG_BLK_ICQ
384 void ioc_clear_queue(struct request_queue *q);
385 #else
ioc_clear_queue(struct request_queue * q)386 static inline void ioc_clear_queue(struct request_queue *q)
387 {
388 }
389 #endif /* CONFIG_BLK_ICQ */
390
391 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
392
blk_queue_may_bounce(struct request_queue * q)393 static inline bool blk_queue_may_bounce(struct request_queue *q)
394 {
395 return IS_ENABLED(CONFIG_BOUNCE) &&
396 q->limits.bounce == BLK_BOUNCE_HIGH &&
397 max_low_pfn >= max_pfn;
398 }
399
blk_queue_bounce(struct bio * bio,struct request_queue * q)400 static inline struct bio *blk_queue_bounce(struct bio *bio,
401 struct request_queue *q)
402 {
403 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
404 return __blk_queue_bounce(bio, q);
405 return bio;
406 }
407
408 #ifdef CONFIG_BLK_DEV_ZONED
409 void disk_init_zone_resources(struct gendisk *disk);
410 void disk_free_zone_resources(struct gendisk *disk);
bio_zone_write_plugging(struct bio * bio)411 static inline bool bio_zone_write_plugging(struct bio *bio)
412 {
413 return bio_flagged(bio, BIO_ZONE_WRITE_PLUGGING);
414 }
bio_is_zone_append(struct bio * bio)415 static inline bool bio_is_zone_append(struct bio *bio)
416 {
417 return bio_op(bio) == REQ_OP_ZONE_APPEND ||
418 bio_flagged(bio, BIO_EMULATES_ZONE_APPEND);
419 }
420 void blk_zone_write_plug_bio_merged(struct bio *bio);
421 void blk_zone_write_plug_init_request(struct request *rq);
blk_zone_update_request_bio(struct request * rq,struct bio * bio)422 static inline void blk_zone_update_request_bio(struct request *rq,
423 struct bio *bio)
424 {
425 /*
426 * For zone append requests, the request sector indicates the location
427 * at which the BIO data was written. Return this value to the BIO
428 * issuer through the BIO iter sector.
429 * For plugged zone writes, which include emulated zone append, we need
430 * the original BIO sector so that blk_zone_write_plug_bio_endio() can
431 * lookup the zone write plug.
432 */
433 if (req_op(rq) == REQ_OP_ZONE_APPEND || bio_zone_write_plugging(bio))
434 bio->bi_iter.bi_sector = rq->__sector;
435 }
436 void blk_zone_write_plug_bio_endio(struct bio *bio);
blk_zone_bio_endio(struct bio * bio)437 static inline void blk_zone_bio_endio(struct bio *bio)
438 {
439 /*
440 * For write BIOs to zoned devices, signal the completion of the BIO so
441 * that the next write BIO can be submitted by zone write plugging.
442 */
443 if (bio_zone_write_plugging(bio))
444 blk_zone_write_plug_bio_endio(bio);
445 }
446
447 void blk_zone_write_plug_finish_request(struct request *rq);
blk_zone_finish_request(struct request * rq)448 static inline void blk_zone_finish_request(struct request *rq)
449 {
450 if (rq->rq_flags & RQF_ZONE_WRITE_PLUGGING)
451 blk_zone_write_plug_finish_request(rq);
452 }
453 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
454 unsigned long arg);
455 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
456 unsigned int cmd, unsigned long arg);
457 #else /* CONFIG_BLK_DEV_ZONED */
disk_init_zone_resources(struct gendisk * disk)458 static inline void disk_init_zone_resources(struct gendisk *disk)
459 {
460 }
disk_free_zone_resources(struct gendisk * disk)461 static inline void disk_free_zone_resources(struct gendisk *disk)
462 {
463 }
bio_zone_write_plugging(struct bio * bio)464 static inline bool bio_zone_write_plugging(struct bio *bio)
465 {
466 return false;
467 }
bio_is_zone_append(struct bio * bio)468 static inline bool bio_is_zone_append(struct bio *bio)
469 {
470 return false;
471 }
blk_zone_write_plug_bio_merged(struct bio * bio)472 static inline void blk_zone_write_plug_bio_merged(struct bio *bio)
473 {
474 }
blk_zone_write_plug_init_request(struct request * rq)475 static inline void blk_zone_write_plug_init_request(struct request *rq)
476 {
477 }
blk_zone_update_request_bio(struct request * rq,struct bio * bio)478 static inline void blk_zone_update_request_bio(struct request *rq,
479 struct bio *bio)
480 {
481 }
blk_zone_bio_endio(struct bio * bio)482 static inline void blk_zone_bio_endio(struct bio *bio)
483 {
484 }
blk_zone_finish_request(struct request * rq)485 static inline void blk_zone_finish_request(struct request *rq)
486 {
487 }
blkdev_report_zones_ioctl(struct block_device * bdev,unsigned int cmd,unsigned long arg)488 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
489 unsigned int cmd, unsigned long arg)
490 {
491 return -ENOTTY;
492 }
blkdev_zone_mgmt_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)493 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
494 blk_mode_t mode, unsigned int cmd, unsigned long arg)
495 {
496 return -ENOTTY;
497 }
498 #endif /* CONFIG_BLK_DEV_ZONED */
499
500 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
501 void bdev_add(struct block_device *bdev, dev_t dev);
502 void bdev_unhash(struct block_device *bdev);
503 void bdev_drop(struct block_device *bdev);
504
505 int blk_alloc_ext_minor(void);
506 void blk_free_ext_minor(unsigned int minor);
507 #define ADDPART_FLAG_NONE 0
508 #define ADDPART_FLAG_RAID 1
509 #define ADDPART_FLAG_WHOLEDISK 2
510 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
511 sector_t length);
512 int bdev_del_partition(struct gendisk *disk, int partno);
513 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
514 sector_t length);
515 void drop_partition(struct block_device *part);
516
517 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
518
519 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
520 struct lock_class_key *lkclass);
521
522 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
523 struct page *page, unsigned int len, unsigned int offset,
524 unsigned int max_sectors, bool *same_page);
525
526 /*
527 * Clean up a page appropriately, where the page may be pinned, may have a
528 * ref taken on it or neither.
529 */
bio_release_page(struct bio * bio,struct page * page)530 static inline void bio_release_page(struct bio *bio, struct page *page)
531 {
532 if (bio_flagged(bio, BIO_PAGE_PINNED))
533 unpin_user_page(page);
534 }
535
536 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
537
538 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
539
540 int disk_alloc_events(struct gendisk *disk);
541 void disk_add_events(struct gendisk *disk);
542 void disk_del_events(struct gendisk *disk);
543 void disk_release_events(struct gendisk *disk);
544 void disk_block_events(struct gendisk *disk);
545 void disk_unblock_events(struct gendisk *disk);
546 void disk_flush_events(struct gendisk *disk, unsigned int mask);
547 extern struct device_attribute dev_attr_events;
548 extern struct device_attribute dev_attr_events_async;
549 extern struct device_attribute dev_attr_events_poll_msecs;
550
551 extern struct attribute_group blk_trace_attr_group;
552
553 blk_mode_t file_to_blk_mode(struct file *file);
554 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
555 loff_t lstart, loff_t lend);
556 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
557 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
558
559 extern const struct address_space_operations def_blk_aops;
560
561 int disk_register_independent_access_ranges(struct gendisk *disk);
562 void disk_unregister_independent_access_ranges(struct gendisk *disk);
563
564 #ifdef CONFIG_FAIL_MAKE_REQUEST
565 bool should_fail_request(struct block_device *part, unsigned int bytes);
566 #else /* CONFIG_FAIL_MAKE_REQUEST */
should_fail_request(struct block_device * part,unsigned int bytes)567 static inline bool should_fail_request(struct block_device *part,
568 unsigned int bytes)
569 {
570 return false;
571 }
572 #endif /* CONFIG_FAIL_MAKE_REQUEST */
573
574 /*
575 * Optimized request reference counting. Ideally we'd make timeouts be more
576 * clever, as that's the only reason we need references at all... But until
577 * this happens, this is faster than using refcount_t. Also see:
578 *
579 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
580 */
581 #define req_ref_zero_or_close_to_overflow(req) \
582 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
583
req_ref_inc_not_zero(struct request * req)584 static inline bool req_ref_inc_not_zero(struct request *req)
585 {
586 return atomic_inc_not_zero(&req->ref);
587 }
588
req_ref_put_and_test(struct request * req)589 static inline bool req_ref_put_and_test(struct request *req)
590 {
591 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
592 return atomic_dec_and_test(&req->ref);
593 }
594
req_ref_set(struct request * req,int value)595 static inline void req_ref_set(struct request *req, int value)
596 {
597 atomic_set(&req->ref, value);
598 }
599
req_ref_read(struct request * req)600 static inline int req_ref_read(struct request *req)
601 {
602 return atomic_read(&req->ref);
603 }
604
blk_time_get_ns(void)605 static inline u64 blk_time_get_ns(void)
606 {
607 struct blk_plug *plug = current->plug;
608
609 if (!plug || !in_task())
610 return ktime_get_ns();
611
612 /*
613 * 0 could very well be a valid time, but rather than flag "this is
614 * a valid timestamp" separately, just accept that we'll do an extra
615 * ktime_get_ns() if we just happen to get 0 as the current time.
616 */
617 if (!plug->cur_ktime) {
618 plug->cur_ktime = ktime_get_ns();
619 current->flags |= PF_BLOCK_TS;
620 }
621 return plug->cur_ktime;
622 }
623
blk_time_get(void)624 static inline ktime_t blk_time_get(void)
625 {
626 return ns_to_ktime(blk_time_get_ns());
627 }
628
629 /*
630 * From most significant bit:
631 * 1 bit: reserved for other usage, see below
632 * 12 bits: original size of bio
633 * 51 bits: issue time of bio
634 */
635 #define BIO_ISSUE_RES_BITS 1
636 #define BIO_ISSUE_SIZE_BITS 12
637 #define BIO_ISSUE_RES_SHIFT (64 - BIO_ISSUE_RES_BITS)
638 #define BIO_ISSUE_SIZE_SHIFT (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
639 #define BIO_ISSUE_TIME_MASK ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
640 #define BIO_ISSUE_SIZE_MASK \
641 (((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
642 #define BIO_ISSUE_RES_MASK (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
643
644 /* Reserved bit for blk-throtl */
645 #define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
646
__bio_issue_time(u64 time)647 static inline u64 __bio_issue_time(u64 time)
648 {
649 return time & BIO_ISSUE_TIME_MASK;
650 }
651
bio_issue_time(struct bio_issue * issue)652 static inline u64 bio_issue_time(struct bio_issue *issue)
653 {
654 return __bio_issue_time(issue->value);
655 }
656
bio_issue_size(struct bio_issue * issue)657 static inline sector_t bio_issue_size(struct bio_issue *issue)
658 {
659 return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
660 }
661
bio_issue_init(struct bio_issue * issue,sector_t size)662 static inline void bio_issue_init(struct bio_issue *issue,
663 sector_t size)
664 {
665 size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
666 issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
667 (blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
668 ((u64)size << BIO_ISSUE_SIZE_SHIFT));
669 }
670
671 void bdev_release(struct file *bdev_file);
672 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
673 const struct blk_holder_ops *hops, struct file *bdev_file);
674 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
675
676 #endif /* BLK_INTERNAL_H */
677