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 unsigned int flush_pending_idx:1;
19 unsigned int flush_running_idx:1;
20 blk_status_t rq_status;
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;
25
26 spinlock_t mq_flush_lock;
27};
28
29bool is_flush_rq(struct request *req);
30
31struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
32 gfp_t flags);
33void blk_free_flush_queue(struct blk_flush_queue *q);
34
35void blk_freeze_queue(struct request_queue *q);
36void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
37void blk_queue_start_drain(struct request_queue *q);
38int __bio_queue_enter(struct request_queue *q, struct bio *bio);
39void submit_bio_noacct_nocheck(struct bio *bio);
40
41static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
42{
43 rcu_read_lock();
44 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
45 goto fail;
46
47 /*
48 * The code that increments the pm_only counter must ensure that the
49 * counter is globally visible before the queue is unfrozen.
50 */
51 if (blk_queue_pm_only(q) &&
52 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
53 goto fail_put;
54
55 rcu_read_unlock();
56 return true;
57
58fail_put:
59 blk_queue_exit(q);
60fail:
61 rcu_read_unlock();
62 return false;
63}
64
65static inline int bio_queue_enter(struct bio *bio)
66{
67 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
68
69 if (blk_try_enter_queue(q, false))
70 return 0;
71 return __bio_queue_enter(q, bio);
72}
73
74#define BIO_INLINE_VECS 4
75struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
76 gfp_t gfp_mask);
77void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
78
79static inline bool biovec_phys_mergeable(struct request_queue *q,
80 struct bio_vec *vec1, struct bio_vec *vec2)
81{
82 unsigned long mask = queue_segment_boundary(q);
83 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
84 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
85
86 /*
87 * Merging adjacent physical pages may not work correctly under KMSAN
88 * if their metadata pages aren't adjacent. Just disable merging.
89 */
90 if (IS_ENABLED(CONFIG_KMSAN))
91 return false;
92
93 if (addr1 + vec1->bv_len != addr2)
94 return false;
95 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
96 return false;
97 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
98 return false;
99 return true;
100}
101
102static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
103 struct bio_vec *bprv, unsigned int offset)
104{
105 return (offset & lim->virt_boundary_mask) ||
106 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
107}
108
109/*
110 * Check if adding a bio_vec after bprv with offset would create a gap in
111 * the SG list. Most drivers don't care about this, but some do.
112 */
113static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
114 struct bio_vec *bprv, unsigned int offset)
115{
116 if (!lim->virt_boundary_mask)
117 return false;
118 return __bvec_gap_to_prev(lim, bprv, offset);
119}
120
121static inline bool rq_mergeable(struct request *rq)
122{
123 if (blk_rq_is_passthrough(rq))
124 return false;
125
126 if (req_op(rq) == REQ_OP_FLUSH)
127 return false;
128
129 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
130 return false;
131
132 if (req_op(rq) == REQ_OP_ZONE_APPEND)
133 return false;
134
135 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
136 return false;
137 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
138 return false;
139
140 return true;
141}
142
143/*
144 * There are two different ways to handle DISCARD merges:
145 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
146 * send the bios to controller together. The ranges don't need to be
147 * contiguous.
148 * 2) Otherwise, the request will be normal read/write requests. The ranges
149 * need to be contiguous.
150 */
151static inline bool blk_discard_mergable(struct request *req)
152{
153 if (req_op(req) == REQ_OP_DISCARD &&
154 queue_max_discard_segments(req->q) > 1)
155 return true;
156 return false;
157}
158
159static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
160 enum req_op op)
161{
162 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
163 return min(q->limits.max_discard_sectors,
164 UINT_MAX >> SECTOR_SHIFT);
165
166 if (unlikely(op == REQ_OP_WRITE_ZEROES))
167 return q->limits.max_write_zeroes_sectors;
168
169 return q->limits.max_sectors;
170}
171
172#ifdef CONFIG_BLK_DEV_INTEGRITY
173void blk_flush_integrity(void);
174bool __bio_integrity_endio(struct bio *);
175void bio_integrity_free(struct bio *bio);
176static inline bool bio_integrity_endio(struct bio *bio)
177{
178 if (bio_integrity(bio))
179 return __bio_integrity_endio(bio);
180 return true;
181}
182
183bool blk_integrity_merge_rq(struct request_queue *, struct request *,
184 struct request *);
185bool blk_integrity_merge_bio(struct request_queue *, struct request *,
186 struct bio *);
187
188static inline bool integrity_req_gap_back_merge(struct request *req,
189 struct bio *next)
190{
191 struct bio_integrity_payload *bip = bio_integrity(req->bio);
192 struct bio_integrity_payload *bip_next = bio_integrity(next);
193
194 return bvec_gap_to_prev(&req->q->limits,
195 &bip->bip_vec[bip->bip_vcnt - 1],
196 bip_next->bip_vec[0].bv_offset);
197}
198
199static inline bool integrity_req_gap_front_merge(struct request *req,
200 struct bio *bio)
201{
202 struct bio_integrity_payload *bip = bio_integrity(bio);
203 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
204
205 return bvec_gap_to_prev(&req->q->limits,
206 &bip->bip_vec[bip->bip_vcnt - 1],
207 bip_next->bip_vec[0].bv_offset);
208}
209
210int blk_integrity_add(struct gendisk *disk);
211void blk_integrity_del(struct gendisk *);
212#else /* CONFIG_BLK_DEV_INTEGRITY */
213static inline bool blk_integrity_merge_rq(struct request_queue *rq,
214 struct request *r1, struct request *r2)
215{
216 return true;
217}
218static inline bool blk_integrity_merge_bio(struct request_queue *rq,
219 struct request *r, struct bio *b)
220{
221 return true;
222}
223static inline bool integrity_req_gap_back_merge(struct request *req,
224 struct bio *next)
225{
226 return false;
227}
228static inline bool integrity_req_gap_front_merge(struct request *req,
229 struct bio *bio)
230{
231 return false;
232}
233
234static inline void blk_flush_integrity(void)
235{
236}
237static inline bool bio_integrity_endio(struct bio *bio)
238{
239 return true;
240}
241static inline void bio_integrity_free(struct bio *bio)
242{
243}
244static inline int blk_integrity_add(struct gendisk *disk)
245{
246 return 0;
247}
248static inline void blk_integrity_del(struct gendisk *disk)
249{
250}
251#endif /* CONFIG_BLK_DEV_INTEGRITY */
252
253unsigned long blk_rq_timeout(unsigned long timeout);
254void blk_add_timer(struct request *req);
255const char *blk_status_to_str(blk_status_t status);
256
257bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
258 unsigned int nr_segs);
259bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
260 struct bio *bio, unsigned int nr_segs);
261
262/*
263 * Plug flush limits
264 */
265#define BLK_MAX_REQUEST_COUNT 32
266#define BLK_PLUG_FLUSH_SIZE (128 * 1024)
267
268/*
269 * Internal elevator interface
270 */
271#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
272
273void blk_insert_flush(struct request *rq);
274
275int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
276void elevator_disable(struct request_queue *q);
277void elevator_exit(struct request_queue *q);
278int elv_register_queue(struct request_queue *q, bool uevent);
279void elv_unregister_queue(struct request_queue *q);
280
281ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
282 char *buf);
283ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
284 char *buf);
285ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
286 char *buf);
287ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
288 char *buf);
289ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
290 const char *buf, size_t count);
291ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
292ssize_t part_timeout_store(struct device *, struct device_attribute *,
293 const char *, size_t);
294
295static inline bool bio_may_exceed_limits(struct bio *bio,
296 const struct queue_limits *lim)
297{
298 switch (bio_op(bio)) {
299 case REQ_OP_DISCARD:
300 case REQ_OP_SECURE_ERASE:
301 case REQ_OP_WRITE_ZEROES:
302 return true; /* non-trivial splitting decisions */
303 default:
304 break;
305 }
306
307 /*
308 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
309 * This is a quick and dirty check that relies on the fact that
310 * bi_io_vec[0] is always valid if a bio has data. The check might
311 * lead to occasional false negatives when bios are cloned, but compared
312 * to the performance impact of cloned bios themselves the loop below
313 * doesn't matter anyway.
314 */
315 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
316 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
317}
318
319struct bio *__bio_split_to_limits(struct bio *bio,
320 const struct queue_limits *lim,
321 unsigned int *nr_segs);
322int ll_back_merge_fn(struct request *req, struct bio *bio,
323 unsigned int nr_segs);
324bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
325 struct request *next);
326unsigned int blk_recalc_rq_segments(struct request *rq);
327void blk_rq_set_mixed_merge(struct request *rq);
328bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
329enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
330
331void blk_set_default_limits(struct queue_limits *lim);
332int blk_dev_init(void);
333
334/*
335 * Contribute to IO statistics IFF:
336 *
337 * a) it's attached to a gendisk, and
338 * b) the queue had IO stats enabled when this request was started
339 */
340static inline bool blk_do_io_stat(struct request *rq)
341{
342 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
343}
344
345void update_io_ticks(struct block_device *part, unsigned long now, bool end);
346
347static inline void req_set_nomerge(struct request_queue *q, struct request *req)
348{
349 req->cmd_flags |= REQ_NOMERGE;
350 if (req == q->last_merge)
351 q->last_merge = NULL;
352}
353
354/*
355 * Internal io_context interface
356 */
357struct io_cq *ioc_find_get_icq(struct request_queue *q);
358struct io_cq *ioc_lookup_icq(struct request_queue *q);
359#ifdef CONFIG_BLK_ICQ
360void ioc_clear_queue(struct request_queue *q);
361#else
362static inline void ioc_clear_queue(struct request_queue *q)
363{
364}
365#endif /* CONFIG_BLK_ICQ */
366
367#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
368extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
369extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
370 const char *page, size_t count);
371extern void blk_throtl_bio_endio(struct bio *bio);
372extern void blk_throtl_stat_add(struct request *rq, u64 time);
373#else
374static inline void blk_throtl_bio_endio(struct bio *bio) { }
375static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
376#endif
377
378struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
379
380static inline bool blk_queue_may_bounce(struct request_queue *q)
381{
382 return IS_ENABLED(CONFIG_BOUNCE) &&
383 q->limits.bounce == BLK_BOUNCE_HIGH &&
384 max_low_pfn >= max_pfn;
385}
386
387static inline struct bio *blk_queue_bounce(struct bio *bio,
388 struct request_queue *q)
389{
390 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
391 return __blk_queue_bounce(bio, q);
392 return bio;
393}
394
395#ifdef CONFIG_BLK_CGROUP_IOLATENCY
396int blk_iolatency_init(struct gendisk *disk);
397#else
398static inline int blk_iolatency_init(struct gendisk *disk) { return 0; };
399#endif
400
401#ifdef CONFIG_BLK_DEV_ZONED
402void disk_free_zone_bitmaps(struct gendisk *disk);
403void disk_clear_zone_settings(struct gendisk *disk);
404#else
405static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
406static inline void disk_clear_zone_settings(struct gendisk *disk) {}
407#endif
408
409int blk_alloc_ext_minor(void);
410void blk_free_ext_minor(unsigned int minor);
411#define ADDPART_FLAG_NONE 0
412#define ADDPART_FLAG_RAID 1
413#define ADDPART_FLAG_WHOLEDISK 2
414int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
415 sector_t length);
416int bdev_del_partition(struct gendisk *disk, int partno);
417int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
418 sector_t length);
419void blk_drop_partitions(struct gendisk *disk);
420
421struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
422 struct lock_class_key *lkclass);
423
424int bio_add_hw_page(struct request_queue *q, struct bio *bio,
425 struct page *page, unsigned int len, unsigned int offset,
426 unsigned int max_sectors, bool *same_page);
427
428struct request_queue *blk_alloc_queue(int node_id);
429
430int disk_scan_partitions(struct gendisk *disk, fmode_t mode, void *owner);
431
432int disk_alloc_events(struct gendisk *disk);
433void disk_add_events(struct gendisk *disk);
434void disk_del_events(struct gendisk *disk);
435void disk_release_events(struct gendisk *disk);
436void disk_block_events(struct gendisk *disk);
437void disk_unblock_events(struct gendisk *disk);
438void disk_flush_events(struct gendisk *disk, unsigned int mask);
439extern struct device_attribute dev_attr_events;
440extern struct device_attribute dev_attr_events_async;
441extern struct device_attribute dev_attr_events_poll_msecs;
442
443extern struct attribute_group blk_trace_attr_group;
444
445long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
446long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
447
448extern const struct address_space_operations def_blk_aops;
449
450int disk_register_independent_access_ranges(struct gendisk *disk);
451void disk_unregister_independent_access_ranges(struct gendisk *disk);
452
453#ifdef CONFIG_FAIL_MAKE_REQUEST
454bool should_fail_request(struct block_device *part, unsigned int bytes);
455#else /* CONFIG_FAIL_MAKE_REQUEST */
456static inline bool should_fail_request(struct block_device *part,
457 unsigned int bytes)
458{
459 return false;
460}
461#endif /* CONFIG_FAIL_MAKE_REQUEST */
462
463/*
464 * Optimized request reference counting. Ideally we'd make timeouts be more
465 * clever, as that's the only reason we need references at all... But until
466 * this happens, this is faster than using refcount_t. Also see:
467 *
468 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
469 */
470#define req_ref_zero_or_close_to_overflow(req) \
471 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
472
473static inline bool req_ref_inc_not_zero(struct request *req)
474{
475 return atomic_inc_not_zero(&req->ref);
476}
477
478static inline bool req_ref_put_and_test(struct request *req)
479{
480 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
481 return atomic_dec_and_test(&req->ref);
482}
483
484static inline void req_ref_set(struct request *req, int value)
485{
486 atomic_set(&req->ref, value);
487}
488
489static inline int req_ref_read(struct request *req)
490{
491 return atomic_read(&req->ref);
492}
493
494#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 */