Loading...
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 <xen/xen.h>
8#include "blk-crypto-internal.h"
9
10struct elevator_type;
11
12/* Max future timer expiry for timeouts */
13#define BLK_MAX_TIMEOUT (5 * HZ)
14
15extern struct dentry *blk_debugfs_root;
16
17struct blk_flush_queue {
18 spinlock_t mq_flush_lock;
19 unsigned int flush_pending_idx:1;
20 unsigned int flush_running_idx:1;
21 blk_status_t rq_status;
22 unsigned long flush_pending_since;
23 struct list_head flush_queue[2];
24 unsigned long flush_data_in_flight;
25 struct request *flush_rq;
26};
27
28bool is_flush_rq(struct request *req);
29
30struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
31 gfp_t flags);
32void blk_free_flush_queue(struct blk_flush_queue *q);
33
34void blk_freeze_queue(struct request_queue *q);
35void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
36void blk_queue_start_drain(struct request_queue *q);
37int __bio_queue_enter(struct request_queue *q, struct bio *bio);
38void submit_bio_noacct_nocheck(struct bio *bio);
39
40static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
41{
42 rcu_read_lock();
43 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
44 goto fail;
45
46 /*
47 * The code that increments the pm_only counter must ensure that the
48 * counter is globally visible before the queue is unfrozen.
49 */
50 if (blk_queue_pm_only(q) &&
51 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
52 goto fail_put;
53
54 rcu_read_unlock();
55 return true;
56
57fail_put:
58 blk_queue_exit(q);
59fail:
60 rcu_read_unlock();
61 return false;
62}
63
64static inline int bio_queue_enter(struct bio *bio)
65{
66 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
67
68 if (blk_try_enter_queue(q, false))
69 return 0;
70 return __bio_queue_enter(q, bio);
71}
72
73#define BIO_INLINE_VECS 4
74struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
75 gfp_t gfp_mask);
76void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
77
78bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
79 struct page *page, unsigned len, unsigned offset,
80 bool *same_page);
81
82static inline bool biovec_phys_mergeable(struct request_queue *q,
83 struct bio_vec *vec1, struct bio_vec *vec2)
84{
85 unsigned long mask = queue_segment_boundary(q);
86 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
87 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
88
89 /*
90 * Merging adjacent physical pages may not work correctly under KMSAN
91 * if their metadata pages aren't adjacent. Just disable merging.
92 */
93 if (IS_ENABLED(CONFIG_KMSAN))
94 return false;
95
96 if (addr1 + vec1->bv_len != addr2)
97 return false;
98 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
99 return false;
100 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
101 return false;
102 return true;
103}
104
105static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
106 struct bio_vec *bprv, unsigned int offset)
107{
108 return (offset & lim->virt_boundary_mask) ||
109 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
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(const struct queue_limits *lim,
117 struct bio_vec *bprv, unsigned int offset)
118{
119 if (!lim->virt_boundary_mask)
120 return false;
121 return __bvec_gap_to_prev(lim, bprv, offset);
122}
123
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
162static inline unsigned int blk_rq_get_max_segments(struct request *rq)
163{
164 if (req_op(rq) == REQ_OP_DISCARD)
165 return queue_max_discard_segments(rq->q);
166 return queue_max_segments(rq->q);
167}
168
169static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
170 enum req_op op)
171{
172 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
173 return min(q->limits.max_discard_sectors,
174 UINT_MAX >> SECTOR_SHIFT);
175
176 if (unlikely(op == REQ_OP_WRITE_ZEROES))
177 return q->limits.max_write_zeroes_sectors;
178
179 return q->limits.max_sectors;
180}
181
182#ifdef CONFIG_BLK_DEV_INTEGRITY
183void blk_flush_integrity(void);
184bool __bio_integrity_endio(struct bio *);
185void bio_integrity_free(struct bio *bio);
186static inline bool bio_integrity_endio(struct bio *bio)
187{
188 if (bio_integrity(bio))
189 return __bio_integrity_endio(bio);
190 return true;
191}
192
193bool blk_integrity_merge_rq(struct request_queue *, struct request *,
194 struct request *);
195bool blk_integrity_merge_bio(struct request_queue *, struct request *,
196 struct bio *);
197
198static inline bool integrity_req_gap_back_merge(struct request *req,
199 struct bio *next)
200{
201 struct bio_integrity_payload *bip = bio_integrity(req->bio);
202 struct bio_integrity_payload *bip_next = bio_integrity(next);
203
204 return bvec_gap_to_prev(&req->q->limits,
205 &bip->bip_vec[bip->bip_vcnt - 1],
206 bip_next->bip_vec[0].bv_offset);
207}
208
209static inline bool integrity_req_gap_front_merge(struct request *req,
210 struct bio *bio)
211{
212 struct bio_integrity_payload *bip = bio_integrity(bio);
213 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
214
215 return bvec_gap_to_prev(&req->q->limits,
216 &bip->bip_vec[bip->bip_vcnt - 1],
217 bip_next->bip_vec[0].bv_offset);
218}
219
220extern const struct attribute_group blk_integrity_attr_group;
221#else /* CONFIG_BLK_DEV_INTEGRITY */
222static inline bool blk_integrity_merge_rq(struct request_queue *rq,
223 struct request *r1, struct request *r2)
224{
225 return true;
226}
227static inline bool blk_integrity_merge_bio(struct request_queue *rq,
228 struct request *r, struct bio *b)
229{
230 return true;
231}
232static inline bool integrity_req_gap_back_merge(struct request *req,
233 struct bio *next)
234{
235 return false;
236}
237static inline bool integrity_req_gap_front_merge(struct request *req,
238 struct bio *bio)
239{
240 return false;
241}
242
243static inline void blk_flush_integrity(void)
244{
245}
246static inline bool bio_integrity_endio(struct bio *bio)
247{
248 return true;
249}
250static inline void bio_integrity_free(struct bio *bio)
251{
252}
253#endif /* CONFIG_BLK_DEV_INTEGRITY */
254
255unsigned long blk_rq_timeout(unsigned long timeout);
256void blk_add_timer(struct request *req);
257
258bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
259 unsigned int nr_segs);
260bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
261 struct bio *bio, unsigned int nr_segs);
262
263/*
264 * Plug flush limits
265 */
266#define BLK_MAX_REQUEST_COUNT 32
267#define BLK_PLUG_FLUSH_SIZE (128 * 1024)
268
269/*
270 * Internal elevator interface
271 */
272#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
273
274bool blk_insert_flush(struct request *rq);
275
276int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
277void elevator_disable(struct request_queue *q);
278void elevator_exit(struct request_queue *q);
279int elv_register_queue(struct request_queue *q, bool uevent);
280void elv_unregister_queue(struct request_queue *q);
281
282ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
283 char *buf);
284ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
285 char *buf);
286ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
287 char *buf);
288ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
289 char *buf);
290ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
291 const char *buf, size_t count);
292ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
293ssize_t part_timeout_store(struct device *, struct device_attribute *,
294 const char *, size_t);
295
296static inline bool bio_may_exceed_limits(struct bio *bio,
297 const struct queue_limits *lim)
298{
299 switch (bio_op(bio)) {
300 case REQ_OP_DISCARD:
301 case REQ_OP_SECURE_ERASE:
302 case REQ_OP_WRITE_ZEROES:
303 return true; /* non-trivial splitting decisions */
304 default:
305 break;
306 }
307
308 /*
309 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
310 * This is a quick and dirty check that relies on the fact that
311 * bi_io_vec[0] is always valid if a bio has data. The check might
312 * lead to occasional false negatives when bios are cloned, but compared
313 * to the performance impact of cloned bios themselves the loop below
314 * doesn't matter anyway.
315 */
316 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
317 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
318}
319
320struct bio *__bio_split_to_limits(struct bio *bio,
321 const struct queue_limits *lim,
322 unsigned int *nr_segs);
323int ll_back_merge_fn(struct request *req, struct bio *bio,
324 unsigned int nr_segs);
325bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
326 struct request *next);
327unsigned int blk_recalc_rq_segments(struct request *rq);
328void blk_rq_set_mixed_merge(struct request *rq);
329bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
330enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
331
332void blk_set_default_limits(struct queue_limits *lim);
333int blk_dev_init(void);
334
335/*
336 * Contribute to IO statistics IFF:
337 *
338 * a) it's attached to a gendisk, and
339 * b) the queue had IO stats enabled when this request was started
340 */
341static inline bool blk_do_io_stat(struct request *rq)
342{
343 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
344}
345
346void update_io_ticks(struct block_device *part, unsigned long now, bool end);
347
348static inline void req_set_nomerge(struct request_queue *q, struct request *req)
349{
350 req->cmd_flags |= REQ_NOMERGE;
351 if (req == q->last_merge)
352 q->last_merge = NULL;
353}
354
355/*
356 * Internal io_context interface
357 */
358struct io_cq *ioc_find_get_icq(struct request_queue *q);
359struct io_cq *ioc_lookup_icq(struct request_queue *q);
360#ifdef CONFIG_BLK_ICQ
361void ioc_clear_queue(struct request_queue *q);
362#else
363static inline void ioc_clear_queue(struct request_queue *q)
364{
365}
366#endif /* CONFIG_BLK_ICQ */
367
368#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
369extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
370extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
371 const char *page, size_t count);
372extern void blk_throtl_bio_endio(struct bio *bio);
373extern void blk_throtl_stat_add(struct request *rq, u64 time);
374#else
375static inline void blk_throtl_bio_endio(struct bio *bio) { }
376static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
377#endif
378
379struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
380
381static inline bool blk_queue_may_bounce(struct request_queue *q)
382{
383 return IS_ENABLED(CONFIG_BOUNCE) &&
384 q->limits.bounce == BLK_BOUNCE_HIGH &&
385 max_low_pfn >= max_pfn;
386}
387
388static inline struct bio *blk_queue_bounce(struct bio *bio,
389 struct request_queue *q)
390{
391 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
392 return __blk_queue_bounce(bio, q);
393 return bio;
394}
395
396#ifdef CONFIG_BLK_DEV_ZONED
397void disk_free_zone_bitmaps(struct gendisk *disk);
398int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
399 unsigned long arg);
400int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
401 unsigned int cmd, unsigned long arg);
402#else /* CONFIG_BLK_DEV_ZONED */
403static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
404static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
405 unsigned int cmd, unsigned long arg)
406{
407 return -ENOTTY;
408}
409static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
410 blk_mode_t mode, unsigned int cmd, unsigned long arg)
411{
412 return -ENOTTY;
413}
414#endif /* CONFIG_BLK_DEV_ZONED */
415
416struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
417void bdev_add(struct block_device *bdev, dev_t dev);
418
419int blk_alloc_ext_minor(void);
420void blk_free_ext_minor(unsigned int minor);
421#define ADDPART_FLAG_NONE 0
422#define ADDPART_FLAG_RAID 1
423#define ADDPART_FLAG_WHOLEDISK 2
424int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
425 sector_t length);
426int bdev_del_partition(struct gendisk *disk, int partno);
427int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
428 sector_t length);
429void drop_partition(struct block_device *part);
430
431void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
432
433struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
434 struct lock_class_key *lkclass);
435
436int bio_add_hw_page(struct request_queue *q, struct bio *bio,
437 struct page *page, unsigned int len, unsigned int offset,
438 unsigned int max_sectors, bool *same_page);
439
440/*
441 * Clean up a page appropriately, where the page may be pinned, may have a
442 * ref taken on it or neither.
443 */
444static inline void bio_release_page(struct bio *bio, struct page *page)
445{
446 if (bio_flagged(bio, BIO_PAGE_PINNED))
447 unpin_user_page(page);
448}
449
450struct request_queue *blk_alloc_queue(int node_id);
451
452int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
453
454int disk_alloc_events(struct gendisk *disk);
455void disk_add_events(struct gendisk *disk);
456void disk_del_events(struct gendisk *disk);
457void disk_release_events(struct gendisk *disk);
458void disk_block_events(struct gendisk *disk);
459void disk_unblock_events(struct gendisk *disk);
460void disk_flush_events(struct gendisk *disk, unsigned int mask);
461extern struct device_attribute dev_attr_events;
462extern struct device_attribute dev_attr_events_async;
463extern struct device_attribute dev_attr_events_poll_msecs;
464
465extern struct attribute_group blk_trace_attr_group;
466
467blk_mode_t file_to_blk_mode(struct file *file);
468int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
469 loff_t lstart, loff_t lend);
470long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
471long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
472
473extern const struct address_space_operations def_blk_aops;
474
475int disk_register_independent_access_ranges(struct gendisk *disk);
476void disk_unregister_independent_access_ranges(struct gendisk *disk);
477
478#ifdef CONFIG_FAIL_MAKE_REQUEST
479bool should_fail_request(struct block_device *part, unsigned int bytes);
480#else /* CONFIG_FAIL_MAKE_REQUEST */
481static inline bool should_fail_request(struct block_device *part,
482 unsigned int bytes)
483{
484 return false;
485}
486#endif /* CONFIG_FAIL_MAKE_REQUEST */
487
488/*
489 * Optimized request reference counting. Ideally we'd make timeouts be more
490 * clever, as that's the only reason we need references at all... But until
491 * this happens, this is faster than using refcount_t. Also see:
492 *
493 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
494 */
495#define req_ref_zero_or_close_to_overflow(req) \
496 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
497
498static inline bool req_ref_inc_not_zero(struct request *req)
499{
500 return atomic_inc_not_zero(&req->ref);
501}
502
503static inline bool req_ref_put_and_test(struct request *req)
504{
505 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
506 return atomic_dec_and_test(&req->ref);
507}
508
509static inline void req_ref_set(struct request *req, int value)
510{
511 atomic_set(&req->ref, value);
512}
513
514static inline int req_ref_read(struct request *req)
515{
516 return atomic_read(&req->ref);
517}
518
519#endif /* BLK_INTERNAL_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef BLK_INTERNAL_H
3#define BLK_INTERNAL_H
4
5#include <linux/idr.h>
6#include <linux/blk-mq.h>
7#include <linux/part_stat.h>
8#include <linux/blk-crypto.h>
9#include <xen/xen.h>
10#include "blk-crypto-internal.h"
11#include "blk-mq.h"
12#include "blk-mq-sched.h"
13
14/* Max future timer expiry for timeouts */
15#define BLK_MAX_TIMEOUT (5 * HZ)
16
17extern struct dentry *blk_debugfs_root;
18
19struct blk_flush_queue {
20 unsigned int flush_pending_idx:1;
21 unsigned int flush_running_idx:1;
22 blk_status_t rq_status;
23 unsigned long flush_pending_since;
24 struct list_head flush_queue[2];
25 struct list_head flush_data_in_flight;
26 struct request *flush_rq;
27
28 struct lock_class_key key;
29 spinlock_t mq_flush_lock;
30};
31
32extern struct kmem_cache *blk_requestq_cachep;
33extern struct kobj_type blk_queue_ktype;
34extern struct ida blk_queue_ida;
35
36static inline struct blk_flush_queue *
37blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
38{
39 return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
40}
41
42static inline void __blk_get_queue(struct request_queue *q)
43{
44 kobject_get(&q->kobj);
45}
46
47static inline bool
48is_flush_rq(struct request *req, struct blk_mq_hw_ctx *hctx)
49{
50 return hctx->fq->flush_rq == req;
51}
52
53struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
54 gfp_t flags);
55void blk_free_flush_queue(struct blk_flush_queue *q);
56
57void blk_freeze_queue(struct request_queue *q);
58
59static inline bool biovec_phys_mergeable(struct request_queue *q,
60 struct bio_vec *vec1, struct bio_vec *vec2)
61{
62 unsigned long mask = queue_segment_boundary(q);
63 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
64 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
65
66 if (addr1 + vec1->bv_len != addr2)
67 return false;
68 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
69 return false;
70 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
71 return false;
72 return true;
73}
74
75static inline bool __bvec_gap_to_prev(struct request_queue *q,
76 struct bio_vec *bprv, unsigned int offset)
77{
78 return (offset & queue_virt_boundary(q)) ||
79 ((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
80}
81
82/*
83 * Check if adding a bio_vec after bprv with offset would create a gap in
84 * the SG list. Most drivers don't care about this, but some do.
85 */
86static inline bool bvec_gap_to_prev(struct request_queue *q,
87 struct bio_vec *bprv, unsigned int offset)
88{
89 if (!queue_virt_boundary(q))
90 return false;
91 return __bvec_gap_to_prev(q, bprv, offset);
92}
93
94static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
95 unsigned int nr_segs)
96{
97 rq->nr_phys_segments = nr_segs;
98 rq->__data_len = bio->bi_iter.bi_size;
99 rq->bio = rq->biotail = bio;
100 rq->ioprio = bio_prio(bio);
101
102 if (bio->bi_disk)
103 rq->rq_disk = bio->bi_disk;
104}
105
106#ifdef CONFIG_BLK_DEV_INTEGRITY
107void blk_flush_integrity(void);
108bool __bio_integrity_endio(struct bio *);
109void bio_integrity_free(struct bio *bio);
110static inline bool bio_integrity_endio(struct bio *bio)
111{
112 if (bio_integrity(bio))
113 return __bio_integrity_endio(bio);
114 return true;
115}
116
117static inline bool integrity_req_gap_back_merge(struct request *req,
118 struct bio *next)
119{
120 struct bio_integrity_payload *bip = bio_integrity(req->bio);
121 struct bio_integrity_payload *bip_next = bio_integrity(next);
122
123 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
124 bip_next->bip_vec[0].bv_offset);
125}
126
127static inline bool integrity_req_gap_front_merge(struct request *req,
128 struct bio *bio)
129{
130 struct bio_integrity_payload *bip = bio_integrity(bio);
131 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
132
133 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
134 bip_next->bip_vec[0].bv_offset);
135}
136
137void blk_integrity_add(struct gendisk *);
138void blk_integrity_del(struct gendisk *);
139#else /* CONFIG_BLK_DEV_INTEGRITY */
140static inline bool integrity_req_gap_back_merge(struct request *req,
141 struct bio *next)
142{
143 return false;
144}
145static inline bool integrity_req_gap_front_merge(struct request *req,
146 struct bio *bio)
147{
148 return false;
149}
150
151static inline void blk_flush_integrity(void)
152{
153}
154static inline bool bio_integrity_endio(struct bio *bio)
155{
156 return true;
157}
158static inline void bio_integrity_free(struct bio *bio)
159{
160}
161static inline void blk_integrity_add(struct gendisk *disk)
162{
163}
164static inline void blk_integrity_del(struct gendisk *disk)
165{
166}
167#endif /* CONFIG_BLK_DEV_INTEGRITY */
168
169unsigned long blk_rq_timeout(unsigned long timeout);
170void blk_add_timer(struct request *req);
171
172bool bio_attempt_front_merge(struct request *req, struct bio *bio,
173 unsigned int nr_segs);
174bool bio_attempt_back_merge(struct request *req, struct bio *bio,
175 unsigned int nr_segs);
176bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
177 struct bio *bio);
178bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
179 unsigned int nr_segs, struct request **same_queue_rq);
180
181void blk_account_io_start(struct request *req);
182void blk_account_io_done(struct request *req, u64 now);
183
184/*
185 * Internal elevator interface
186 */
187#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
188
189void blk_insert_flush(struct request *rq);
190
191void elevator_init_mq(struct request_queue *q);
192int elevator_switch_mq(struct request_queue *q,
193 struct elevator_type *new_e);
194void __elevator_exit(struct request_queue *, struct elevator_queue *);
195int elv_register_queue(struct request_queue *q, bool uevent);
196void elv_unregister_queue(struct request_queue *q);
197
198static inline void elevator_exit(struct request_queue *q,
199 struct elevator_queue *e)
200{
201 lockdep_assert_held(&q->sysfs_lock);
202
203 blk_mq_sched_free_requests(q);
204 __elevator_exit(q, e);
205}
206
207struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
208
209ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
210 char *buf);
211ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
212 char *buf);
213ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
214 char *buf);
215ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
216 char *buf);
217ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
218 const char *buf, size_t count);
219ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
220ssize_t part_timeout_store(struct device *, struct device_attribute *,
221 const char *, size_t);
222
223void __blk_queue_split(struct bio **bio, unsigned int *nr_segs);
224int ll_back_merge_fn(struct request *req, struct bio *bio,
225 unsigned int nr_segs);
226int ll_front_merge_fn(struct request *req, struct bio *bio,
227 unsigned int nr_segs);
228struct request *attempt_back_merge(struct request_queue *q, struct request *rq);
229struct request *attempt_front_merge(struct request_queue *q, struct request *rq);
230int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
231 struct request *next);
232unsigned int blk_recalc_rq_segments(struct request *rq);
233void blk_rq_set_mixed_merge(struct request *rq);
234bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
235enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
236
237int blk_dev_init(void);
238
239/*
240 * Contribute to IO statistics IFF:
241 *
242 * a) it's attached to a gendisk, and
243 * b) the queue had IO stats enabled when this request was started
244 */
245static inline bool blk_do_io_stat(struct request *rq)
246{
247 return rq->rq_disk && (rq->rq_flags & RQF_IO_STAT);
248}
249
250static inline void req_set_nomerge(struct request_queue *q, struct request *req)
251{
252 req->cmd_flags |= REQ_NOMERGE;
253 if (req == q->last_merge)
254 q->last_merge = NULL;
255}
256
257/*
258 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
259 * is defined as 'unsigned int', meantime it has to aligned to with logical
260 * block size which is the minimum accepted unit by hardware.
261 */
262static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
263{
264 return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
265}
266
267/*
268 * The max bio size which is aligned to q->limits.discard_granularity. This
269 * is a hint to split large discard bio in generic block layer, then if device
270 * driver needs to split the discard bio into smaller ones, their bi_size can
271 * be very probably and easily aligned to discard_granularity of the device's
272 * queue.
273 */
274static inline unsigned int bio_aligned_discard_max_sectors(
275 struct request_queue *q)
276{
277 return round_down(UINT_MAX, q->limits.discard_granularity) >>
278 SECTOR_SHIFT;
279}
280
281/*
282 * Internal io_context interface
283 */
284void get_io_context(struct io_context *ioc);
285struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
286struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
287 gfp_t gfp_mask);
288void ioc_clear_queue(struct request_queue *q);
289
290int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
291
292/*
293 * Internal throttling interface
294 */
295#ifdef CONFIG_BLK_DEV_THROTTLING
296extern int blk_throtl_init(struct request_queue *q);
297extern void blk_throtl_exit(struct request_queue *q);
298extern void blk_throtl_register_queue(struct request_queue *q);
299bool blk_throtl_bio(struct bio *bio);
300#else /* CONFIG_BLK_DEV_THROTTLING */
301static inline int blk_throtl_init(struct request_queue *q) { return 0; }
302static inline void blk_throtl_exit(struct request_queue *q) { }
303static inline void blk_throtl_register_queue(struct request_queue *q) { }
304static inline bool blk_throtl_bio(struct bio *bio) { return false; }
305#endif /* CONFIG_BLK_DEV_THROTTLING */
306#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
307extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
308extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
309 const char *page, size_t count);
310extern void blk_throtl_bio_endio(struct bio *bio);
311extern void blk_throtl_stat_add(struct request *rq, u64 time);
312#else
313static inline void blk_throtl_bio_endio(struct bio *bio) { }
314static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
315#endif
316
317#ifdef CONFIG_BOUNCE
318extern int init_emergency_isa_pool(void);
319extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
320#else
321static inline int init_emergency_isa_pool(void)
322{
323 return 0;
324}
325static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
326{
327}
328#endif /* CONFIG_BOUNCE */
329
330#ifdef CONFIG_BLK_CGROUP_IOLATENCY
331extern int blk_iolatency_init(struct request_queue *q);
332#else
333static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
334#endif
335
336struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
337
338#ifdef CONFIG_BLK_DEV_ZONED
339void blk_queue_free_zone_bitmaps(struct request_queue *q);
340#else
341static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
342#endif
343
344struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
345
346int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
347void blk_free_devt(dev_t devt);
348void blk_invalidate_devt(dev_t devt);
349char *disk_name(struct gendisk *hd, int partno, char *buf);
350#define ADDPART_FLAG_NONE 0
351#define ADDPART_FLAG_RAID 1
352#define ADDPART_FLAG_WHOLEDISK 2
353void delete_partition(struct gendisk *disk, struct hd_struct *part);
354int bdev_add_partition(struct block_device *bdev, int partno,
355 sector_t start, sector_t length);
356int bdev_del_partition(struct block_device *bdev, int partno);
357int bdev_resize_partition(struct block_device *bdev, int partno,
358 sector_t start, sector_t length);
359int disk_expand_part_tbl(struct gendisk *disk, int target);
360int hd_ref_init(struct hd_struct *part);
361
362/* no need to get/put refcount of part0 */
363static inline int hd_struct_try_get(struct hd_struct *part)
364{
365 if (part->partno)
366 return percpu_ref_tryget_live(&part->ref);
367 return 1;
368}
369
370static inline void hd_struct_put(struct hd_struct *part)
371{
372 if (part->partno)
373 percpu_ref_put(&part->ref);
374}
375
376static inline void hd_free_part(struct hd_struct *part)
377{
378 free_percpu(part->dkstats);
379 kfree(part->info);
380 percpu_ref_exit(&part->ref);
381}
382
383/*
384 * Any access of part->nr_sects which is not protected by partition
385 * bd_mutex or gendisk bdev bd_mutex, should be done using this
386 * accessor function.
387 *
388 * Code written along the lines of i_size_read() and i_size_write().
389 * CONFIG_PREEMPTION case optimizes the case of UP kernel with preemption
390 * on.
391 */
392static inline sector_t part_nr_sects_read(struct hd_struct *part)
393{
394#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
395 sector_t nr_sects;
396 unsigned seq;
397 do {
398 seq = read_seqcount_begin(&part->nr_sects_seq);
399 nr_sects = part->nr_sects;
400 } while (read_seqcount_retry(&part->nr_sects_seq, seq));
401 return nr_sects;
402#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
403 sector_t nr_sects;
404
405 preempt_disable();
406 nr_sects = part->nr_sects;
407 preempt_enable();
408 return nr_sects;
409#else
410 return part->nr_sects;
411#endif
412}
413
414/*
415 * Should be called with mutex lock held (typically bd_mutex) of partition
416 * to provide mutual exlusion among writers otherwise seqcount might be
417 * left in wrong state leaving the readers spinning infinitely.
418 */
419static inline void part_nr_sects_write(struct hd_struct *part, sector_t size)
420{
421#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
422 preempt_disable();
423 write_seqcount_begin(&part->nr_sects_seq);
424 part->nr_sects = size;
425 write_seqcount_end(&part->nr_sects_seq);
426 preempt_enable();
427#elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
428 preempt_disable();
429 part->nr_sects = size;
430 preempt_enable();
431#else
432 part->nr_sects = size;
433#endif
434}
435
436int bio_add_hw_page(struct request_queue *q, struct bio *bio,
437 struct page *page, unsigned int len, unsigned int offset,
438 unsigned int max_sectors, bool *same_page);
439
440#endif /* BLK_INTERNAL_H */