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
12struct elevator_type;
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 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
30bool is_flush_rq(struct request *req);
31
32struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
33 gfp_t flags);
34void blk_free_flush_queue(struct blk_flush_queue *q);
35
36void blk_freeze_queue(struct request_queue *q);
37void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
38void blk_queue_start_drain(struct request_queue *q);
39int __bio_queue_enter(struct request_queue *q, struct bio *bio);
40void submit_bio_noacct_nocheck(struct bio *bio);
41
42static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
43{
44 rcu_read_lock();
45 if (!percpu_ref_tryget_live_rcu(ref: &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
59fail_put:
60 blk_queue_exit(q);
61fail:
62 rcu_read_unlock();
63 return false;
64}
65
66static inline int bio_queue_enter(struct bio *bio)
67{
68 struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev);
69
70 if (blk_try_enter_queue(q, pm: false))
71 return 0;
72 return __bio_queue_enter(q, bio);
73}
74
75static 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(x: done, timeout))
82 ;
83 else
84 wait_for_completion_io(done);
85}
86
87#define BIO_INLINE_VECS 4
88struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
89 gfp_t gfp_mask);
90void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
91
92bool 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
96static 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, page: vec2->bv_page))
113 return false;
114 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
115 return false;
116 return true;
117}
118
119static 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 */
130static 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
138static inline bool rq_mergeable(struct request *rq)
139{
140 if (blk_rq_is_passthrough(rq))
141 return false;
142
143 if (req_op(req: rq) == REQ_OP_FLUSH)
144 return false;
145
146 if (req_op(req: rq) == REQ_OP_WRITE_ZEROES)
147 return false;
148
149 if (req_op(req: 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 */
168static inline bool blk_discard_mergable(struct request *req)
169{
170 if (req_op(req) == REQ_OP_DISCARD &&
171 queue_max_discard_segments(q: req->q) > 1)
172 return true;
173 return false;
174}
175
176static inline unsigned int blk_rq_get_max_segments(struct request *rq)
177{
178 if (req_op(req: rq) == REQ_OP_DISCARD)
179 return queue_max_discard_segments(q: rq->q);
180 return queue_max_segments(q: rq->q);
181}
182
183static 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
197void blk_flush_integrity(void);
198bool __bio_integrity_endio(struct bio *);
199void bio_integrity_free(struct bio *bio);
200static 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
207bool blk_integrity_merge_rq(struct request_queue *, struct request *,
208 struct request *);
209bool blk_integrity_merge_bio(struct request_queue *, struct request *,
210 struct bio *);
211
212static inline bool integrity_req_gap_back_merge(struct request *req,
213 struct bio *next)
214{
215 struct bio_integrity_payload *bip = bio_integrity(bio: req->bio);
216 struct bio_integrity_payload *bip_next = bio_integrity(bio: next);
217
218 return bvec_gap_to_prev(lim: &req->q->limits,
219 bprv: &bip->bip_vec[bip->bip_vcnt - 1],
220 offset: bip_next->bip_vec[0].bv_offset);
221}
222
223static 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(bio: req->bio);
228
229 return bvec_gap_to_prev(lim: &req->q->limits,
230 bprv: &bip->bip_vec[bip->bip_vcnt - 1],
231 offset: bip_next->bip_vec[0].bv_offset);
232}
233
234extern const struct attribute_group blk_integrity_attr_group;
235#else /* CONFIG_BLK_DEV_INTEGRITY */
236static inline bool blk_integrity_merge_rq(struct request_queue *rq,
237 struct request *r1, struct request *r2)
238{
239 return true;
240}
241static inline bool blk_integrity_merge_bio(struct request_queue *rq,
242 struct request *r, struct bio *b)
243{
244 return true;
245}
246static inline bool integrity_req_gap_back_merge(struct request *req,
247 struct bio *next)
248{
249 return false;
250}
251static inline bool integrity_req_gap_front_merge(struct request *req,
252 struct bio *bio)
253{
254 return false;
255}
256
257static inline void blk_flush_integrity(void)
258{
259}
260static inline bool bio_integrity_endio(struct bio *bio)
261{
262 return true;
263}
264static inline void bio_integrity_free(struct bio *bio)
265{
266}
267#endif /* CONFIG_BLK_DEV_INTEGRITY */
268
269unsigned long blk_rq_timeout(unsigned long timeout);
270void blk_add_timer(struct request *req);
271
272bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
273 unsigned int nr_segs);
274bool 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
288bool blk_insert_flush(struct request *rq);
289
290int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
291void elevator_disable(struct request_queue *q);
292void elevator_exit(struct request_queue *q);
293int elv_register_queue(struct request_queue *q, bool uevent);
294void elv_unregister_queue(struct request_queue *q);
295
296ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
297 char *buf);
298ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
299 char *buf);
300ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
301 char *buf);
302ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
303 char *buf);
304ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
305 const char *buf, size_t count);
306ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
307ssize_t part_timeout_store(struct device *, struct device_attribute *,
308 const char *, size_t);
309
310static 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
334struct bio *__bio_split_to_limits(struct bio *bio,
335 const struct queue_limits *lim,
336 unsigned int *nr_segs);
337int ll_back_merge_fn(struct request *req, struct bio *bio,
338 unsigned int nr_segs);
339bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
340 struct request *next);
341unsigned int blk_recalc_rq_segments(struct request *rq);
342bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
343enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
344
345int blk_set_default_limits(struct queue_limits *lim);
346int blk_dev_init(void);
347
348/*
349 * Contribute to IO statistics IFF:
350 *
351 * a) it's attached to a gendisk, and
352 * b) the queue had IO stats enabled when this request was started
353 */
354static inline bool blk_do_io_stat(struct request *rq)
355{
356 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
357}
358
359void update_io_ticks(struct block_device *part, unsigned long now, bool end);
360
361static inline void req_set_nomerge(struct request_queue *q, struct request *req)
362{
363 req->cmd_flags |= REQ_NOMERGE;
364 if (req == q->last_merge)
365 q->last_merge = NULL;
366}
367
368/*
369 * Internal io_context interface
370 */
371struct io_cq *ioc_find_get_icq(struct request_queue *q);
372struct io_cq *ioc_lookup_icq(struct request_queue *q);
373#ifdef CONFIG_BLK_ICQ
374void ioc_clear_queue(struct request_queue *q);
375#else
376static inline void ioc_clear_queue(struct request_queue *q)
377{
378}
379#endif /* CONFIG_BLK_ICQ */
380
381#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
382extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
383extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
384 const char *page, size_t count);
385extern void blk_throtl_bio_endio(struct bio *bio);
386extern void blk_throtl_stat_add(struct request *rq, u64 time);
387#else
388static inline void blk_throtl_bio_endio(struct bio *bio) { }
389static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
390#endif
391
392struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
393
394static inline bool blk_queue_may_bounce(struct request_queue *q)
395{
396 return IS_ENABLED(CONFIG_BOUNCE) &&
397 q->limits.bounce == BLK_BOUNCE_HIGH &&
398 max_low_pfn >= max_pfn;
399}
400
401static inline struct bio *blk_queue_bounce(struct bio *bio,
402 struct request_queue *q)
403{
404 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
405 return __blk_queue_bounce(bio, q);
406 return bio;
407}
408
409#ifdef CONFIG_BLK_DEV_ZONED
410void disk_free_zone_bitmaps(struct gendisk *disk);
411int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
412 unsigned long arg);
413int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
414 unsigned int cmd, unsigned long arg);
415#else /* CONFIG_BLK_DEV_ZONED */
416static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
417static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
418 unsigned int cmd, unsigned long arg)
419{
420 return -ENOTTY;
421}
422static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
423 blk_mode_t mode, unsigned int cmd, unsigned long arg)
424{
425 return -ENOTTY;
426}
427#endif /* CONFIG_BLK_DEV_ZONED */
428
429struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
430void bdev_add(struct block_device *bdev, dev_t dev);
431
432int blk_alloc_ext_minor(void);
433void blk_free_ext_minor(unsigned int minor);
434#define ADDPART_FLAG_NONE 0
435#define ADDPART_FLAG_RAID 1
436#define ADDPART_FLAG_WHOLEDISK 2
437int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
438 sector_t length);
439int bdev_del_partition(struct gendisk *disk, int partno);
440int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
441 sector_t length);
442void drop_partition(struct block_device *part);
443
444void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
445
446struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
447 struct lock_class_key *lkclass);
448
449int bio_add_hw_page(struct request_queue *q, struct bio *bio,
450 struct page *page, unsigned int len, unsigned int offset,
451 unsigned int max_sectors, bool *same_page);
452
453/*
454 * Clean up a page appropriately, where the page may be pinned, may have a
455 * ref taken on it or neither.
456 */
457static inline void bio_release_page(struct bio *bio, struct page *page)
458{
459 if (bio_flagged(bio, bit: BIO_PAGE_PINNED))
460 unpin_user_page(page);
461}
462
463struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
464
465int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
466
467int disk_alloc_events(struct gendisk *disk);
468void disk_add_events(struct gendisk *disk);
469void disk_del_events(struct gendisk *disk);
470void disk_release_events(struct gendisk *disk);
471void disk_block_events(struct gendisk *disk);
472void disk_unblock_events(struct gendisk *disk);
473void disk_flush_events(struct gendisk *disk, unsigned int mask);
474extern struct device_attribute dev_attr_events;
475extern struct device_attribute dev_attr_events_async;
476extern struct device_attribute dev_attr_events_poll_msecs;
477
478extern struct attribute_group blk_trace_attr_group;
479
480blk_mode_t file_to_blk_mode(struct file *file);
481int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
482 loff_t lstart, loff_t lend);
483long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
484long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
485
486extern const struct address_space_operations def_blk_aops;
487
488int disk_register_independent_access_ranges(struct gendisk *disk);
489void disk_unregister_independent_access_ranges(struct gendisk *disk);
490
491#ifdef CONFIG_FAIL_MAKE_REQUEST
492bool should_fail_request(struct block_device *part, unsigned int bytes);
493#else /* CONFIG_FAIL_MAKE_REQUEST */
494static inline bool should_fail_request(struct block_device *part,
495 unsigned int bytes)
496{
497 return false;
498}
499#endif /* CONFIG_FAIL_MAKE_REQUEST */
500
501/*
502 * Optimized request reference counting. Ideally we'd make timeouts be more
503 * clever, as that's the only reason we need references at all... But until
504 * this happens, this is faster than using refcount_t. Also see:
505 *
506 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
507 */
508#define req_ref_zero_or_close_to_overflow(req) \
509 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
510
511static inline bool req_ref_inc_not_zero(struct request *req)
512{
513 return atomic_inc_not_zero(v: &req->ref);
514}
515
516static inline bool req_ref_put_and_test(struct request *req)
517{
518 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
519 return atomic_dec_and_test(v: &req->ref);
520}
521
522static inline void req_ref_set(struct request *req, int value)
523{
524 atomic_set(v: &req->ref, i: value);
525}
526
527static inline int req_ref_read(struct request *req)
528{
529 return atomic_read(v: &req->ref);
530}
531
532static inline u64 blk_time_get_ns(void)
533{
534 struct blk_plug *plug = current->plug;
535
536 if (!plug || !in_task())
537 return ktime_get_ns();
538
539 /*
540 * 0 could very well be a valid time, but rather than flag "this is
541 * a valid timestamp" separately, just accept that we'll do an extra
542 * ktime_get_ns() if we just happen to get 0 as the current time.
543 */
544 if (!plug->cur_ktime) {
545 plug->cur_ktime = ktime_get_ns();
546 current->flags |= PF_BLOCK_TS;
547 }
548 return plug->cur_ktime;
549}
550
551static inline ktime_t blk_time_get(void)
552{
553 return ns_to_ktime(ns: blk_time_get_ns());
554}
555
556/*
557 * From most significant bit:
558 * 1 bit: reserved for other usage, see below
559 * 12 bits: original size of bio
560 * 51 bits: issue time of bio
561 */
562#define BIO_ISSUE_RES_BITS 1
563#define BIO_ISSUE_SIZE_BITS 12
564#define BIO_ISSUE_RES_SHIFT (64 - BIO_ISSUE_RES_BITS)
565#define BIO_ISSUE_SIZE_SHIFT (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
566#define BIO_ISSUE_TIME_MASK ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
567#define BIO_ISSUE_SIZE_MASK \
568 (((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
569#define BIO_ISSUE_RES_MASK (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
570
571/* Reserved bit for blk-throtl */
572#define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
573
574static inline u64 __bio_issue_time(u64 time)
575{
576 return time & BIO_ISSUE_TIME_MASK;
577}
578
579static inline u64 bio_issue_time(struct bio_issue *issue)
580{
581 return __bio_issue_time(time: issue->value);
582}
583
584static inline sector_t bio_issue_size(struct bio_issue *issue)
585{
586 return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
587}
588
589static inline void bio_issue_init(struct bio_issue *issue,
590 sector_t size)
591{
592 size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
593 issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
594 (blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
595 ((u64)size << BIO_ISSUE_SIZE_SHIFT));
596}
597
598void bdev_release(struct file *bdev_file);
599int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
600 const struct blk_holder_ops *hops, struct file *bdev_file);
601int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
602
603#endif /* BLK_INTERNAL_H */
604

source code of linux/block/blk.h