xref: /linux/block/blk.h (revision 2573c25e2c482b53b6e1142ff3cd28f6de13e659)
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 
42 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
43 {
44 	rcu_read_lock();
45 	if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
46 		goto fail;
47 
48 	/*
49 	 * The code that increments the pm_only counter must ensure that the
50 	 * counter is globally visible before the queue is unfrozen.
51 	 */
52 	if (blk_queue_pm_only(q) &&
53 	    (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
54 		goto fail_put;
55 
56 	rcu_read_unlock();
57 	return true;
58 
59 fail_put:
60 	blk_queue_exit(q);
61 fail:
62 	rcu_read_unlock();
63 	return false;
64 }
65 
66 static inline int bio_queue_enter(struct bio *bio)
67 {
68 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
69 
70 	if (blk_try_enter_queue(q, false))
71 		return 0;
72 	return __bio_queue_enter(q, bio);
73 }
74 
75 static inline void blk_wait_io(struct completion *done)
76 {
77 	/* Prevent hang_check timer from firing at us during very long I/O */
78 	unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
79 
80 	if (timeout)
81 		while (!wait_for_completion_io_timeout(done, timeout))
82 			;
83 	else
84 		wait_for_completion_io(done);
85 }
86 
87 #define BIO_INLINE_VECS 4
88 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
89 		gfp_t gfp_mask);
90 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
91 
92 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
93 		struct page *page, unsigned len, unsigned offset,
94 		bool *same_page);
95 
96 static inline bool biovec_phys_mergeable(struct request_queue *q,
97 		struct bio_vec *vec1, struct bio_vec *vec2)
98 {
99 	unsigned long mask = queue_segment_boundary(q);
100 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
101 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
102 
103 	/*
104 	 * Merging adjacent physical pages may not work correctly under KMSAN
105 	 * if their metadata pages aren't adjacent. Just disable merging.
106 	 */
107 	if (IS_ENABLED(CONFIG_KMSAN))
108 		return false;
109 
110 	if (addr1 + vec1->bv_len != addr2)
111 		return false;
112 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
113 		return false;
114 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
115 		return false;
116 	return true;
117 }
118 
119 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
120 		struct bio_vec *bprv, unsigned int offset)
121 {
122 	return (offset & lim->virt_boundary_mask) ||
123 		((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
124 }
125 
126 /*
127  * Check if adding a bio_vec after bprv with offset would create a gap in
128  * the SG list. Most drivers don't care about this, but some do.
129  */
130 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
131 		struct bio_vec *bprv, unsigned int offset)
132 {
133 	if (!lim->virt_boundary_mask)
134 		return false;
135 	return __bvec_gap_to_prev(lim, bprv, offset);
136 }
137 
138 static inline bool rq_mergeable(struct request *rq)
139 {
140 	if (blk_rq_is_passthrough(rq))
141 		return false;
142 
143 	if (req_op(rq) == REQ_OP_FLUSH)
144 		return false;
145 
146 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
147 		return false;
148 
149 	if (req_op(rq) == REQ_OP_ZONE_APPEND)
150 		return false;
151 
152 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
153 		return false;
154 	if (rq->rq_flags & RQF_NOMERGE_FLAGS)
155 		return false;
156 
157 	return true;
158 }
159 
160 /*
161  * There are two different ways to handle DISCARD merges:
162  *  1) If max_discard_segments > 1, the driver treats every bio as a range and
163  *     send the bios to controller together. The ranges don't need to be
164  *     contiguous.
165  *  2) Otherwise, the request will be normal read/write requests.  The ranges
166  *     need to be contiguous.
167  */
168 static inline bool blk_discard_mergable(struct request *req)
169 {
170 	if (req_op(req) == REQ_OP_DISCARD &&
171 	    queue_max_discard_segments(req->q) > 1)
172 		return true;
173 	return false;
174 }
175 
176 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
177 {
178 	if (req_op(rq) == REQ_OP_DISCARD)
179 		return queue_max_discard_segments(rq->q);
180 	return queue_max_segments(rq->q);
181 }
182 
183 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
184 						     enum req_op op)
185 {
186 	if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
187 		return min(q->limits.max_discard_sectors,
188 			   UINT_MAX >> SECTOR_SHIFT);
189 
190 	if (unlikely(op == REQ_OP_WRITE_ZEROES))
191 		return q->limits.max_write_zeroes_sectors;
192 
193 	return q->limits.max_sectors;
194 }
195 
196 #ifdef CONFIG_BLK_DEV_INTEGRITY
197 void blk_flush_integrity(void);
198 bool __bio_integrity_endio(struct bio *);
199 void bio_integrity_free(struct bio *bio);
200 static inline bool bio_integrity_endio(struct bio *bio)
201 {
202 	if (bio_integrity(bio))
203 		return __bio_integrity_endio(bio);
204 	return true;
205 }
206 
207 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
208 		struct request *);
209 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
210 		struct bio *);
211 
212 static inline bool integrity_req_gap_back_merge(struct request *req,
213 		struct bio *next)
214 {
215 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
216 	struct bio_integrity_payload *bip_next = bio_integrity(next);
217 
218 	return bvec_gap_to_prev(&req->q->limits,
219 				&bip->bip_vec[bip->bip_vcnt - 1],
220 				bip_next->bip_vec[0].bv_offset);
221 }
222 
223 static inline bool integrity_req_gap_front_merge(struct request *req,
224 		struct bio *bio)
225 {
226 	struct bio_integrity_payload *bip = bio_integrity(bio);
227 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
228 
229 	return bvec_gap_to_prev(&req->q->limits,
230 				&bip->bip_vec[bip->bip_vcnt - 1],
231 				bip_next->bip_vec[0].bv_offset);
232 }
233 
234 extern const struct attribute_group blk_integrity_attr_group;
235 #else /* CONFIG_BLK_DEV_INTEGRITY */
236 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
237 		struct request *r1, struct request *r2)
238 {
239 	return true;
240 }
241 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
242 		struct request *r, struct bio *b)
243 {
244 	return true;
245 }
246 static inline bool integrity_req_gap_back_merge(struct request *req,
247 		struct bio *next)
248 {
249 	return false;
250 }
251 static inline bool integrity_req_gap_front_merge(struct request *req,
252 		struct bio *bio)
253 {
254 	return false;
255 }
256 
257 static inline void blk_flush_integrity(void)
258 {
259 }
260 static inline bool bio_integrity_endio(struct bio *bio)
261 {
262 	return true;
263 }
264 static inline void bio_integrity_free(struct bio *bio)
265 {
266 }
267 #endif /* CONFIG_BLK_DEV_INTEGRITY */
268 
269 unsigned long blk_rq_timeout(unsigned long timeout);
270 void blk_add_timer(struct request *req);
271 
272 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
273 		unsigned int nr_segs);
274 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
275 			struct bio *bio, unsigned int nr_segs);
276 
277 /*
278  * Plug flush limits
279  */
280 #define BLK_MAX_REQUEST_COUNT	32
281 #define BLK_PLUG_FLUSH_SIZE	(128 * 1024)
282 
283 /*
284  * Internal elevator interface
285  */
286 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
287 
288 bool blk_insert_flush(struct request *rq);
289 
290 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
291 void elevator_disable(struct request_queue *q);
292 void elevator_exit(struct request_queue *q);
293 int elv_register_queue(struct request_queue *q, bool uevent);
294 void elv_unregister_queue(struct request_queue *q);
295 
296 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
297 		char *buf);
298 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
299 		char *buf);
300 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
301 		char *buf);
302 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
303 		char *buf);
304 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
305 		const char *buf, size_t count);
306 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
307 ssize_t part_timeout_store(struct device *, struct device_attribute *,
308 				const char *, size_t);
309 
310 static inline bool bio_may_exceed_limits(struct bio *bio,
311 					 const struct queue_limits *lim)
312 {
313 	switch (bio_op(bio)) {
314 	case REQ_OP_DISCARD:
315 	case REQ_OP_SECURE_ERASE:
316 	case REQ_OP_WRITE_ZEROES:
317 		return true; /* non-trivial splitting decisions */
318 	default:
319 		break;
320 	}
321 
322 	/*
323 	 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
324 	 * This is a quick and dirty check that relies on the fact that
325 	 * bi_io_vec[0] is always valid if a bio has data.  The check might
326 	 * lead to occasional false negatives when bios are cloned, but compared
327 	 * to the performance impact of cloned bios themselves the loop below
328 	 * doesn't matter anyway.
329 	 */
330 	return lim->chunk_sectors || bio->bi_vcnt != 1 ||
331 		bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
332 }
333 
334 struct bio *__bio_split_to_limits(struct bio *bio,
335 				  const struct queue_limits *lim,
336 				  unsigned int *nr_segs);
337 int ll_back_merge_fn(struct request *req, struct bio *bio,
338 		unsigned int nr_segs);
339 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
340 				struct request *next);
341 unsigned int blk_recalc_rq_segments(struct request *rq);
342 void blk_rq_set_mixed_merge(struct request *rq);
343 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
344 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
345 
346 int blk_set_default_limits(struct queue_limits *lim);
347 int blk_dev_init(void);
348 
349 /*
350  * Contribute to IO statistics IFF:
351  *
352  *	a) it's attached to a gendisk, and
353  *	b) the queue had IO stats enabled when this request was started
354  */
355 static inline bool blk_do_io_stat(struct request *rq)
356 {
357 	return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
358 }
359 
360 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
361 
362 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
363 {
364 	req->cmd_flags |= REQ_NOMERGE;
365 	if (req == q->last_merge)
366 		q->last_merge = NULL;
367 }
368 
369 /*
370  * Internal io_context interface
371  */
372 struct io_cq *ioc_find_get_icq(struct request_queue *q);
373 struct io_cq *ioc_lookup_icq(struct request_queue *q);
374 #ifdef CONFIG_BLK_ICQ
375 void ioc_clear_queue(struct request_queue *q);
376 #else
377 static inline void ioc_clear_queue(struct request_queue *q)
378 {
379 }
380 #endif /* CONFIG_BLK_ICQ */
381 
382 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
383 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
384 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
385 	const char *page, size_t count);
386 extern void blk_throtl_bio_endio(struct bio *bio);
387 extern void blk_throtl_stat_add(struct request *rq, u64 time);
388 #else
389 static inline void blk_throtl_bio_endio(struct bio *bio) { }
390 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
391 #endif
392 
393 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
394 
395 static inline bool blk_queue_may_bounce(struct request_queue *q)
396 {
397 	return IS_ENABLED(CONFIG_BOUNCE) &&
398 		q->limits.bounce == BLK_BOUNCE_HIGH &&
399 		max_low_pfn >= max_pfn;
400 }
401 
402 static inline struct bio *blk_queue_bounce(struct bio *bio,
403 		struct request_queue *q)
404 {
405 	if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
406 		return __blk_queue_bounce(bio, q);
407 	return bio;
408 }
409 
410 #ifdef CONFIG_BLK_DEV_ZONED
411 void disk_free_zone_bitmaps(struct gendisk *disk);
412 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
413 		unsigned long arg);
414 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
415 		unsigned int cmd, unsigned long arg);
416 #else /* CONFIG_BLK_DEV_ZONED */
417 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
418 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
419 		unsigned int cmd, unsigned long arg)
420 {
421 	return -ENOTTY;
422 }
423 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
424 		blk_mode_t mode, unsigned int cmd, unsigned long arg)
425 {
426 	return -ENOTTY;
427 }
428 #endif /* CONFIG_BLK_DEV_ZONED */
429 
430 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
431 void bdev_add(struct block_device *bdev, dev_t dev);
432 
433 int blk_alloc_ext_minor(void);
434 void blk_free_ext_minor(unsigned int minor);
435 #define ADDPART_FLAG_NONE	0
436 #define ADDPART_FLAG_RAID	1
437 #define ADDPART_FLAG_WHOLEDISK	2
438 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
439 		sector_t length);
440 int bdev_del_partition(struct gendisk *disk, int partno);
441 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
442 		sector_t length);
443 void drop_partition(struct block_device *part);
444 
445 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
446 
447 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
448 		struct lock_class_key *lkclass);
449 
450 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
451 		struct page *page, unsigned int len, unsigned int offset,
452 		unsigned int max_sectors, bool *same_page);
453 
454 /*
455  * Clean up a page appropriately, where the page may be pinned, may have a
456  * ref taken on it or neither.
457  */
458 static inline void bio_release_page(struct bio *bio, struct page *page)
459 {
460 	if (bio_flagged(bio, BIO_PAGE_PINNED))
461 		unpin_user_page(page);
462 }
463 
464 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
465 
466 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
467 
468 int disk_alloc_events(struct gendisk *disk);
469 void disk_add_events(struct gendisk *disk);
470 void disk_del_events(struct gendisk *disk);
471 void disk_release_events(struct gendisk *disk);
472 void disk_block_events(struct gendisk *disk);
473 void disk_unblock_events(struct gendisk *disk);
474 void disk_flush_events(struct gendisk *disk, unsigned int mask);
475 extern struct device_attribute dev_attr_events;
476 extern struct device_attribute dev_attr_events_async;
477 extern struct device_attribute dev_attr_events_poll_msecs;
478 
479 extern struct attribute_group blk_trace_attr_group;
480 
481 blk_mode_t file_to_blk_mode(struct file *file);
482 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
483 		loff_t lstart, loff_t lend);
484 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
485 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
486 
487 extern const struct address_space_operations def_blk_aops;
488 
489 int disk_register_independent_access_ranges(struct gendisk *disk);
490 void disk_unregister_independent_access_ranges(struct gendisk *disk);
491 
492 #ifdef CONFIG_FAIL_MAKE_REQUEST
493 bool should_fail_request(struct block_device *part, unsigned int bytes);
494 #else /* CONFIG_FAIL_MAKE_REQUEST */
495 static inline bool should_fail_request(struct block_device *part,
496 					unsigned int bytes)
497 {
498 	return false;
499 }
500 #endif /* CONFIG_FAIL_MAKE_REQUEST */
501 
502 /*
503  * Optimized request reference counting. Ideally we'd make timeouts be more
504  * clever, as that's the only reason we need references at all... But until
505  * this happens, this is faster than using refcount_t. Also see:
506  *
507  * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
508  */
509 #define req_ref_zero_or_close_to_overflow(req)	\
510 	((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
511 
512 static inline bool req_ref_inc_not_zero(struct request *req)
513 {
514 	return atomic_inc_not_zero(&req->ref);
515 }
516 
517 static inline bool req_ref_put_and_test(struct request *req)
518 {
519 	WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
520 	return atomic_dec_and_test(&req->ref);
521 }
522 
523 static inline void req_ref_set(struct request *req, int value)
524 {
525 	atomic_set(&req->ref, value);
526 }
527 
528 static inline int req_ref_read(struct request *req)
529 {
530 	return atomic_read(&req->ref);
531 }
532 
533 static inline u64 blk_time_get_ns(void)
534 {
535 	struct blk_plug *plug = current->plug;
536 
537 	if (!plug || !in_task())
538 		return ktime_get_ns();
539 
540 	/*
541 	 * 0 could very well be a valid time, but rather than flag "this is
542 	 * a valid timestamp" separately, just accept that we'll do an extra
543 	 * ktime_get_ns() if we just happen to get 0 as the current time.
544 	 */
545 	if (!plug->cur_ktime) {
546 		plug->cur_ktime = ktime_get_ns();
547 		current->flags |= PF_BLOCK_TS;
548 	}
549 	return plug->cur_ktime;
550 }
551 
552 static inline ktime_t blk_time_get(void)
553 {
554 	return ns_to_ktime(blk_time_get_ns());
555 }
556 
557 /*
558  * From most significant bit:
559  * 1 bit: reserved for other usage, see below
560  * 12 bits: original size of bio
561  * 51 bits: issue time of bio
562  */
563 #define BIO_ISSUE_RES_BITS      1
564 #define BIO_ISSUE_SIZE_BITS     12
565 #define BIO_ISSUE_RES_SHIFT     (64 - BIO_ISSUE_RES_BITS)
566 #define BIO_ISSUE_SIZE_SHIFT    (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
567 #define BIO_ISSUE_TIME_MASK     ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
568 #define BIO_ISSUE_SIZE_MASK     \
569 	(((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
570 #define BIO_ISSUE_RES_MASK      (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
571 
572 /* Reserved bit for blk-throtl */
573 #define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
574 
575 static inline u64 __bio_issue_time(u64 time)
576 {
577 	return time & BIO_ISSUE_TIME_MASK;
578 }
579 
580 static inline u64 bio_issue_time(struct bio_issue *issue)
581 {
582 	return __bio_issue_time(issue->value);
583 }
584 
585 static inline sector_t bio_issue_size(struct bio_issue *issue)
586 {
587 	return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
588 }
589 
590 static inline void bio_issue_init(struct bio_issue *issue,
591 				       sector_t size)
592 {
593 	size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
594 	issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
595 			(blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
596 			((u64)size << BIO_ISSUE_SIZE_SHIFT));
597 }
598 
599 void bdev_release(struct file *bdev_file);
600 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
601 	      const struct blk_holder_ops *hops, struct file *bdev_file);
602 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
603 
604 #endif /* BLK_INTERNAL_H */
605