Merge tag 'pci-v5.19-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaa...
[linux-block.git] / block / blk.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
8324aa91
JA
2#ifndef BLK_INTERNAL_H
3#define BLK_INTERNAL_H
4
a892c8d5 5#include <linux/blk-crypto.h>
9bb33f24 6#include <linux/memblock.h> /* for max_pfn/max_low_pfn */
c39ae60d 7#include <xen/xen.h>
a892c8d5 8#include "blk-crypto-internal.h"
a73f730d 9
2e9bc346
CH
10struct elevator_type;
11
0d2602ca
JA
12/* Max future timer expiry for timeouts */
13#define BLK_MAX_TIMEOUT (5 * HZ)
14
18fbda91 15extern struct dentry *blk_debugfs_root;
18fbda91 16
7c94e1c1 17struct blk_flush_queue {
7c94e1c1
ML
18 unsigned int flush_pending_idx:1;
19 unsigned int flush_running_idx:1;
8d699663 20 blk_status_t rq_status;
7c94e1c1
ML
21 unsigned long flush_pending_since;
22 struct list_head flush_queue[2];
23 struct list_head flush_data_in_flight;
24 struct request *flush_rq;
0048b483 25
7c94e1c1
ML
26 spinlock_t mq_flush_lock;
27};
28
8324aa91 29extern struct kmem_cache *blk_requestq_cachep;
704b914f 30extern struct kmem_cache *blk_requestq_srcu_cachep;
8324aa91 31extern struct kobj_type blk_queue_ktype;
a73f730d 32extern struct ida blk_queue_ida;
8324aa91 33
09ac46c4
TH
34static inline void __blk_get_queue(struct request_queue *q)
35{
36 kobject_get(&q->kobj);
37}
38
a9ed27a7 39bool is_flush_rq(struct request *req);
8d699663 40
754a1572
GJ
41struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
42 gfp_t flags);
f70ced09 43void blk_free_flush_queue(struct blk_flush_queue *q);
f3552655 44
3ef28e83 45void blk_freeze_queue(struct request_queue *q);
aec89dc5 46void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
8e141f9e 47void blk_queue_start_drain(struct request_queue *q);
c98cb5bb 48int __bio_queue_enter(struct request_queue *q, struct bio *bio);
3f98c753 49void submit_bio_noacct_nocheck(struct bio *bio);
c98cb5bb
JA
50
51static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
52{
53 rcu_read_lock();
54 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
55 goto fail;
56
57 /*
58 * The code that increments the pm_only counter must ensure that the
59 * counter is globally visible before the queue is unfrozen.
60 */
61 if (blk_queue_pm_only(q) &&
62 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
63 goto fail_put;
64
65 rcu_read_unlock();
66 return true;
67
68fail_put:
69 blk_queue_exit(q);
70fail:
71 rcu_read_unlock();
72 return false;
73}
74
75static inline int bio_queue_enter(struct bio *bio)
76{
77 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
78
79 if (blk_try_enter_queue(q, false))
80 return 0;
81 return __bio_queue_enter(q, bio);
82}
3ef28e83 83
dc0b8a57 84#define BIO_INLINE_VECS 4
7a800a20
CH
85struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
86 gfp_t gfp_mask);
87void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
eec716a1 88
3dccdae5
CH
89static inline bool biovec_phys_mergeable(struct request_queue *q,
90 struct bio_vec *vec1, struct bio_vec *vec2)
6a9f5f24 91{
3dccdae5 92 unsigned long mask = queue_segment_boundary(q);
6e768461
CH
93 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
94 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
3dccdae5
CH
95
96 if (addr1 + vec1->bv_len != addr2)
6a9f5f24 97 return false;
0383ad43 98 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
6a9f5f24 99 return false;
3dccdae5
CH
100 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
101 return false;
6a9f5f24
CH
102 return true;
103}
104
27ca1d4e
CH
105static inline bool __bvec_gap_to_prev(struct request_queue *q,
106 struct bio_vec *bprv, unsigned int offset)
107{
df376b2e 108 return (offset & queue_virt_boundary(q)) ||
27ca1d4e
CH
109 ((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
110}
111
112/*
113 * Check if adding a bio_vec after bprv with offset would create a gap in
114 * the SG list. Most drivers don't care about this, but some do.
115 */
116static inline bool bvec_gap_to_prev(struct request_queue *q,
117 struct bio_vec *bprv, unsigned int offset)
118{
119 if (!queue_virt_boundary(q))
120 return false;
121 return __bvec_gap_to_prev(q, bprv, offset);
122}
123
badf7f64
CH
124static inline bool rq_mergeable(struct request *rq)
125{
126 if (blk_rq_is_passthrough(rq))
127 return false;
128
129 if (req_op(rq) == REQ_OP_FLUSH)
130 return false;
131
132 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
133 return false;
134
135 if (req_op(rq) == REQ_OP_ZONE_APPEND)
136 return false;
137
138 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
139 return false;
140 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
141 return false;
142
143 return true;
144}
145
146/*
147 * There are two different ways to handle DISCARD merges:
148 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
149 * send the bios to controller together. The ranges don't need to be
150 * contiguous.
151 * 2) Otherwise, the request will be normal read/write requests. The ranges
152 * need to be contiguous.
153 */
154static inline bool blk_discard_mergable(struct request *req)
155{
156 if (req_op(req) == REQ_OP_DISCARD &&
157 queue_max_discard_segments(req->q) > 1)
158 return true;
159 return false;
160}
161
5a48fc14
DW
162#ifdef CONFIG_BLK_DEV_INTEGRITY
163void blk_flush_integrity(void);
7c20f116 164bool __bio_integrity_endio(struct bio *);
ece841ab 165void bio_integrity_free(struct bio *bio);
7c20f116
CH
166static inline bool bio_integrity_endio(struct bio *bio)
167{
168 if (bio_integrity(bio))
169 return __bio_integrity_endio(bio);
170 return true;
171}
43b729bf 172
92cf2fd1
CH
173bool blk_integrity_merge_rq(struct request_queue *, struct request *,
174 struct request *);
d59da419
CH
175bool blk_integrity_merge_bio(struct request_queue *, struct request *,
176 struct bio *);
92cf2fd1 177
43b729bf
CH
178static inline bool integrity_req_gap_back_merge(struct request *req,
179 struct bio *next)
180{
181 struct bio_integrity_payload *bip = bio_integrity(req->bio);
182 struct bio_integrity_payload *bip_next = bio_integrity(next);
183
184 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
185 bip_next->bip_vec[0].bv_offset);
186}
187
188static inline bool integrity_req_gap_front_merge(struct request *req,
189 struct bio *bio)
190{
191 struct bio_integrity_payload *bip = bio_integrity(bio);
192 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
193
194 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
195 bip_next->bip_vec[0].bv_offset);
196}
581e2600 197
614310c9 198int blk_integrity_add(struct gendisk *disk);
581e2600 199void blk_integrity_del(struct gendisk *);
43b729bf 200#else /* CONFIG_BLK_DEV_INTEGRITY */
92cf2fd1
CH
201static inline bool blk_integrity_merge_rq(struct request_queue *rq,
202 struct request *r1, struct request *r2)
203{
204 return true;
205}
d59da419
CH
206static inline bool blk_integrity_merge_bio(struct request_queue *rq,
207 struct request *r, struct bio *b)
208{
209 return true;
210}
43b729bf
CH
211static inline bool integrity_req_gap_back_merge(struct request *req,
212 struct bio *next)
213{
214 return false;
215}
216static inline bool integrity_req_gap_front_merge(struct request *req,
217 struct bio *bio)
218{
219 return false;
220}
221
5a48fc14
DW
222static inline void blk_flush_integrity(void)
223{
224}
7c20f116
CH
225static inline bool bio_integrity_endio(struct bio *bio)
226{
227 return true;
228}
ece841ab
JT
229static inline void bio_integrity_free(struct bio *bio)
230{
231}
614310c9 232static inline int blk_integrity_add(struct gendisk *disk)
581e2600 233{
614310c9 234 return 0;
581e2600
CH
235}
236static inline void blk_integrity_del(struct gendisk *disk)
237{
238}
43b729bf 239#endif /* CONFIG_BLK_DEV_INTEGRITY */
8324aa91 240
0d2602ca 241unsigned long blk_rq_timeout(unsigned long timeout);
87ee7b11 242void blk_add_timer(struct request *req);
0d7a29a2 243const char *blk_status_to_str(blk_status_t status);
320ae51f 244
320ae51f 245bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
0c5bcc92 246 unsigned int nr_segs);
bdc6a287
BW
247bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
248 struct bio *bio, unsigned int nr_segs);
320ae51f 249
ba0ffdd8
JA
250/*
251 * Plug flush limits
252 */
253#define BLK_MAX_REQUEST_COUNT 32
254#define BLK_PLUG_FLUSH_SIZE (128 * 1024)
255
158dbda0
TH
256/*
257 * Internal elevator interface
258 */
e8064021 259#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
158dbda0 260
2b504bd4 261void blk_insert_flush(struct request *rq);
dd831006 262
d48ece20
JW
263int elevator_switch_mq(struct request_queue *q,
264 struct elevator_type *new_e);
0c6cb3a2 265void elevator_exit(struct request_queue *q);
cecf5d87 266int elv_register_queue(struct request_queue *q, bool uevent);
83d016ac
BVA
267void elv_unregister_queue(struct request_queue *q);
268
3ad5cee5
CH
269ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
270 char *buf);
271ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
272 char *buf);
273ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
274 char *buf);
275ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
276 char *buf);
277ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
278 const char *buf, size_t count);
581d4e28
JA
279ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
280ssize_t part_timeout_store(struct device *, struct device_attribute *,
281 const char *, size_t);
581d4e28 282
abd45c15
JA
283static inline bool blk_may_split(struct request_queue *q, struct bio *bio)
284{
285 switch (bio_op(bio)) {
286 case REQ_OP_DISCARD:
287 case REQ_OP_SECURE_ERASE:
288 case REQ_OP_WRITE_ZEROES:
abd45c15
JA
289 return true; /* non-trivial splitting decisions */
290 default:
291 break;
292 }
293
294 /*
295 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
296 * This is a quick and dirty check that relies on the fact that
297 * bi_io_vec[0] is always valid if a bio has data. The check might
298 * lead to occasional false negatives when bios are cloned, but compared
299 * to the performance impact of cloned bios themselves the loop below
300 * doesn't matter anyway.
301 */
302 return q->limits.chunk_sectors || bio->bi_vcnt != 1 ||
303 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
304}
305
306void __blk_queue_split(struct request_queue *q, struct bio **bio,
307 unsigned int *nr_segs);
14ccb66b
CH
308int ll_back_merge_fn(struct request *req, struct bio *bio,
309 unsigned int nr_segs);
fd2ef39c 310bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
5e84ea3a 311 struct request *next);
e9cd19c0 312unsigned int blk_recalc_rq_segments(struct request *rq);
80a761fd 313void blk_rq_set_mixed_merge(struct request *rq);
050c8ea8 314bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
34fe7c05 315enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
d6d48196 316
ff88972c
AB
317int blk_dev_init(void);
318
c2553b58
JA
319/*
320 * Contribute to IO statistics IFF:
321 *
322 * a) it's attached to a gendisk, and
48d9b0d4 323 * b) the queue had IO stats enabled when this request was started
c2553b58 324 */
599d067d 325static inline bool blk_do_io_stat(struct request *rq)
fb8ec18c 326{
41fa7222 327 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
be6bfe36
PB
328}
329
450b7879 330void update_io_ticks(struct block_device *part, unsigned long now, bool end);
fb8ec18c 331
6cf7677f
CH
332static inline void req_set_nomerge(struct request_queue *q, struct request *req)
333{
334 req->cmd_flags |= REQ_NOMERGE;
335 if (req == q->last_merge)
336 q->last_merge = NULL;
337}
338
1adfc5e4
ML
339/*
340 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
341 * is defined as 'unsigned int', meantime it has to aligned to with logical
342 * block size which is the minimum accepted unit by hardware.
343 */
344static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
345{
346 return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
347}
348
f2dbd76a
TH
349/*
350 * Internal io_context interface
351 */
87dd1d63 352struct io_cq *ioc_find_get_icq(struct request_queue *q);
eca5892a 353struct io_cq *ioc_lookup_icq(struct request_queue *q);
5ef16305 354#ifdef CONFIG_BLK_ICQ
7e5a8794 355void ioc_clear_queue(struct request_queue *q);
5ef16305
CH
356#else
357static inline void ioc_clear_queue(struct request_queue *q)
358{
359}
360#endif /* CONFIG_BLK_ICQ */
f2dbd76a 361
297e3d85
SL
362#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
363extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
364extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
365 const char *page, size_t count);
9e234eea 366extern void blk_throtl_bio_endio(struct bio *bio);
b9147dd1 367extern void blk_throtl_stat_add(struct request *rq, u64 time);
9e234eea
SL
368#else
369static inline void blk_throtl_bio_endio(struct bio *bio) { }
b9147dd1 370static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
297e3d85 371#endif
bc9fcbf9 372
9bb33f24
CH
373void __blk_queue_bounce(struct request_queue *q, struct bio **bio);
374
375static inline bool blk_queue_may_bounce(struct request_queue *q)
376{
377 return IS_ENABLED(CONFIG_BOUNCE) &&
378 q->limits.bounce == BLK_BOUNCE_HIGH &&
379 max_low_pfn >= max_pfn;
380}
381
3bce016a
CH
382static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
383{
9bb33f24
CH
384 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(*bio)))
385 __blk_queue_bounce(q, bio);
3bce016a 386}
3bce016a 387
d7067512
JB
388#ifdef CONFIG_BLK_CGROUP_IOLATENCY
389extern int blk_iolatency_init(struct request_queue *q);
390#else
391static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
392#endif
393
bf505456
DLM
394#ifdef CONFIG_BLK_DEV_ZONED
395void blk_queue_free_zone_bitmaps(struct request_queue *q);
508aebb8 396void blk_queue_clear_zone_settings(struct request_queue *q);
bf505456
DLM
397#else
398static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
508aebb8 399static inline void blk_queue_clear_zone_settings(struct request_queue *q) {}
bf505456
DLM
400#endif
401
7c3f828b
CH
402int blk_alloc_ext_minor(void);
403void blk_free_ext_minor(unsigned int minor);
581e2600
CH
404#define ADDPART_FLAG_NONE 0
405#define ADDPART_FLAG_RAID 1
406#define ADDPART_FLAG_WHOLEDISK 2
7f6be376
CH
407int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
408 sector_t length);
926fbb16 409int bdev_del_partition(struct gendisk *disk, int partno);
3d2e7989
CH
410int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
411 sector_t length);
e7243285 412void blk_drop_partitions(struct gendisk *disk);
581e2600 413
e4581105 414int bio_add_hw_page(struct request_queue *q, struct bio *bio,
130879f1 415 struct page *page, unsigned int len, unsigned int offset,
e4581105 416 unsigned int max_sectors, bool *same_page);
130879f1 417
704b914f
ML
418static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
419{
420 if (srcu)
421 return blk_requestq_srcu_cachep;
422 return blk_requestq_cachep;
423}
424struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
425
e16e506c 426int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
da7ba729 427
92e7755e 428int disk_alloc_events(struct gendisk *disk);
d5870edf
CH
429void disk_add_events(struct gendisk *disk);
430void disk_del_events(struct gendisk *disk);
431void disk_release_events(struct gendisk *disk);
926597ff
CH
432void disk_block_events(struct gendisk *disk);
433void disk_unblock_events(struct gendisk *disk);
434void disk_flush_events(struct gendisk *disk, unsigned int mask);
2bc8cda5
CH
435extern struct device_attribute dev_attr_events;
436extern struct device_attribute dev_attr_events_async;
437extern struct device_attribute dev_attr_events_poll_msecs;
d5870edf 438
8a709512 439long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
84b8514b
CH
440long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
441
cd82cca7
CH
442extern const struct address_space_operations def_blk_aops;
443
a2247f19
DLM
444int disk_register_independent_access_ranges(struct gendisk *disk,
445 struct blk_independent_access_ranges *new_iars);
446void disk_unregister_independent_access_ranges(struct gendisk *disk);
447
06c8c691
CH
448#ifdef CONFIG_FAIL_MAKE_REQUEST
449bool should_fail_request(struct block_device *part, unsigned int bytes);
450#else /* CONFIG_FAIL_MAKE_REQUEST */
451static inline bool should_fail_request(struct block_device *part,
452 unsigned int bytes)
453{
454 return false;
455}
456#endif /* CONFIG_FAIL_MAKE_REQUEST */
457
0a467d0f
JA
458/*
459 * Optimized request reference counting. Ideally we'd make timeouts be more
460 * clever, as that's the only reason we need references at all... But until
461 * this happens, this is faster than using refcount_t. Also see:
462 *
463 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
464 */
465#define req_ref_zero_or_close_to_overflow(req) \
466 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
467
468static inline bool req_ref_inc_not_zero(struct request *req)
469{
470 return atomic_inc_not_zero(&req->ref);
471}
472
473static inline bool req_ref_put_and_test(struct request *req)
474{
475 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
476 return atomic_dec_and_test(&req->ref);
477}
478
479static inline void req_ref_set(struct request *req, int value)
480{
481 atomic_set(&req->ref, value);
482}
483
484static inline int req_ref_read(struct request *req)
485{
486 return atomic_read(&req->ref);
487}
488
bc9fcbf9 489#endif /* BLK_INTERNAL_H */