1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/bio-integrity.h>
6 #include <linux/blk-crypto.h>
7 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */
8 #include <linux/sched/sysctl.h>
9 #include <linux/timekeeping.h>
11 #include "blk-crypto-internal.h"
15 /* Max future timer expiry for timeouts */
16 #define BLK_MAX_TIMEOUT (5 * HZ)
18 extern struct dentry *blk_debugfs_root;
20 struct blk_flush_queue {
21 spinlock_t mq_flush_lock;
22 unsigned int flush_pending_idx:1;
23 unsigned int flush_running_idx:1;
24 blk_status_t rq_status;
25 unsigned long flush_pending_since;
26 struct list_head flush_queue[2];
27 unsigned long flush_data_in_flight;
28 struct request *flush_rq;
31 bool is_flush_rq(struct request *req);
33 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
35 void blk_free_flush_queue(struct blk_flush_queue *q);
37 void blk_freeze_queue(struct request_queue *q);
38 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
39 void blk_queue_start_drain(struct request_queue *q);
40 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
41 void submit_bio_noacct_nocheck(struct bio *bio);
42 void bio_await_chain(struct bio *bio);
44 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
47 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
51 * The code that increments the pm_only counter must ensure that the
52 * counter is globally visible before the queue is unfrozen.
54 if (blk_queue_pm_only(q) &&
55 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
68 static inline int bio_queue_enter(struct bio *bio)
70 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
72 if (blk_try_enter_queue(q, false))
74 return __bio_queue_enter(q, bio);
77 static inline void blk_wait_io(struct completion *done)
79 /* Prevent hang_check timer from firing at us during very long I/O */
80 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
83 while (!wait_for_completion_io_timeout(done, timeout))
86 wait_for_completion_io(done);
89 #define BIO_INLINE_VECS 4
90 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
92 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
94 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
95 struct page *page, unsigned len, unsigned offset,
98 static inline bool biovec_phys_mergeable(struct request_queue *q,
99 struct bio_vec *vec1, struct bio_vec *vec2)
101 unsigned long mask = queue_segment_boundary(q);
102 phys_addr_t addr1 = bvec_phys(vec1);
103 phys_addr_t addr2 = bvec_phys(vec2);
106 * Merging adjacent physical pages may not work correctly under KMSAN
107 * if their metadata pages aren't adjacent. Just disable merging.
109 if (IS_ENABLED(CONFIG_KMSAN))
112 if (addr1 + vec1->bv_len != addr2)
114 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
116 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
121 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
122 struct bio_vec *bprv, unsigned int offset)
124 return (offset & lim->virt_boundary_mask) ||
125 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
129 * Check if adding a bio_vec after bprv with offset would create a gap in
130 * the SG list. Most drivers don't care about this, but some do.
132 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
133 struct bio_vec *bprv, unsigned int offset)
135 if (!lim->virt_boundary_mask)
137 return __bvec_gap_to_prev(lim, bprv, offset);
140 static inline bool rq_mergeable(struct request *rq)
142 if (blk_rq_is_passthrough(rq))
145 if (req_op(rq) == REQ_OP_FLUSH)
148 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
151 if (req_op(rq) == REQ_OP_ZONE_APPEND)
154 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
156 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
163 * There are two different ways to handle DISCARD merges:
164 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
165 * send the bios to controller together. The ranges don't need to be
167 * 2) Otherwise, the request will be normal read/write requests. The ranges
168 * need to be contiguous.
170 static inline bool blk_discard_mergable(struct request *req)
172 if (req_op(req) == REQ_OP_DISCARD &&
173 queue_max_discard_segments(req->q) > 1)
178 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
180 if (req_op(rq) == REQ_OP_DISCARD)
181 return queue_max_discard_segments(rq->q);
182 return queue_max_segments(rq->q);
185 static inline unsigned int blk_queue_get_max_sectors(struct request *rq)
187 struct request_queue *q = rq->q;
188 enum req_op op = req_op(rq);
190 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
191 return min(q->limits.max_discard_sectors,
192 UINT_MAX >> SECTOR_SHIFT);
194 if (unlikely(op == REQ_OP_WRITE_ZEROES))
195 return q->limits.max_write_zeroes_sectors;
197 if (rq->cmd_flags & REQ_ATOMIC)
198 return q->limits.atomic_write_max_sectors;
200 return q->limits.max_sectors;
203 #ifdef CONFIG_BLK_DEV_INTEGRITY
204 void blk_flush_integrity(void);
205 void bio_integrity_free(struct bio *bio);
208 * Integrity payloads can either be owned by the submitter, in which case
209 * bio_uninit will free them, or owned and generated by the block layer,
210 * in which case we'll verify them here (for reads) and free them before
211 * the bio is handed back to the submitted.
213 bool __bio_integrity_endio(struct bio *bio);
214 static inline bool bio_integrity_endio(struct bio *bio)
216 struct bio_integrity_payload *bip = bio_integrity(bio);
218 if (bip && (bip->bip_flags & BIP_BLOCK_INTEGRITY))
219 return __bio_integrity_endio(bio);
223 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
225 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
228 static inline bool integrity_req_gap_back_merge(struct request *req,
231 struct bio_integrity_payload *bip = bio_integrity(req->bio);
232 struct bio_integrity_payload *bip_next = bio_integrity(next);
234 return bvec_gap_to_prev(&req->q->limits,
235 &bip->bip_vec[bip->bip_vcnt - 1],
236 bip_next->bip_vec[0].bv_offset);
239 static inline bool integrity_req_gap_front_merge(struct request *req,
242 struct bio_integrity_payload *bip = bio_integrity(bio);
243 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
245 return bvec_gap_to_prev(&req->q->limits,
246 &bip->bip_vec[bip->bip_vcnt - 1],
247 bip_next->bip_vec[0].bv_offset);
250 extern const struct attribute_group blk_integrity_attr_group;
251 #else /* CONFIG_BLK_DEV_INTEGRITY */
252 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
253 struct request *r1, struct request *r2)
257 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
258 struct request *r, struct bio *b)
262 static inline bool integrity_req_gap_back_merge(struct request *req,
267 static inline bool integrity_req_gap_front_merge(struct request *req,
273 static inline void blk_flush_integrity(void)
276 static inline bool bio_integrity_endio(struct bio *bio)
280 static inline void bio_integrity_free(struct bio *bio)
283 #endif /* CONFIG_BLK_DEV_INTEGRITY */
285 unsigned long blk_rq_timeout(unsigned long timeout);
286 void blk_add_timer(struct request *req);
288 enum bio_merge_status {
294 enum bio_merge_status bio_attempt_back_merge(struct request *req,
295 struct bio *bio, unsigned int nr_segs);
296 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
297 unsigned int nr_segs);
298 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
299 struct bio *bio, unsigned int nr_segs);
304 #define BLK_MAX_REQUEST_COUNT 32
305 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
308 * Internal elevator interface
310 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
312 bool blk_insert_flush(struct request *rq);
314 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
315 void elevator_disable(struct request_queue *q);
316 void elevator_exit(struct request_queue *q);
317 int elv_register_queue(struct request_queue *q, bool uevent);
318 void elv_unregister_queue(struct request_queue *q);
320 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
322 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
324 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
326 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
328 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
329 const char *buf, size_t count);
330 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
331 ssize_t part_timeout_store(struct device *, struct device_attribute *,
332 const char *, size_t);
334 static inline bool bio_may_exceed_limits(struct bio *bio,
335 const struct queue_limits *lim)
337 switch (bio_op(bio)) {
339 case REQ_OP_SECURE_ERASE:
340 case REQ_OP_WRITE_ZEROES:
341 return true; /* non-trivial splitting decisions */
347 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
348 * This is a quick and dirty check that relies on the fact that
349 * bi_io_vec[0] is always valid if a bio has data. The check might
350 * lead to occasional false negatives when bios are cloned, but compared
351 * to the performance impact of cloned bios themselves the loop below
352 * doesn't matter anyway.
354 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
355 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
358 struct bio *__bio_split_to_limits(struct bio *bio,
359 const struct queue_limits *lim,
360 unsigned int *nr_segs);
361 int ll_back_merge_fn(struct request *req, struct bio *bio,
362 unsigned int nr_segs);
363 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
364 struct request *next);
365 unsigned int blk_recalc_rq_segments(struct request *rq);
366 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
367 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
369 int blk_set_default_limits(struct queue_limits *lim);
370 void blk_apply_bdi_limits(struct backing_dev_info *bdi,
371 struct queue_limits *lim);
372 int blk_dev_init(void);
375 * Contribute to IO statistics IFF:
377 * a) it's attached to a gendisk, and
378 * b) the queue had IO stats enabled when this request was started
380 static inline bool blk_do_io_stat(struct request *rq)
382 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
385 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
386 unsigned int part_in_flight(struct block_device *part);
388 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
390 req->cmd_flags |= REQ_NOMERGE;
391 if (req == q->last_merge)
392 q->last_merge = NULL;
396 * Internal io_context interface
398 struct io_cq *ioc_find_get_icq(struct request_queue *q);
399 struct io_cq *ioc_lookup_icq(struct request_queue *q);
400 #ifdef CONFIG_BLK_ICQ
401 void ioc_clear_queue(struct request_queue *q);
403 static inline void ioc_clear_queue(struct request_queue *q)
406 #endif /* CONFIG_BLK_ICQ */
408 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
410 static inline bool blk_queue_may_bounce(struct request_queue *q)
412 return IS_ENABLED(CONFIG_BOUNCE) &&
413 (q->limits.features & BLK_FEAT_BOUNCE_HIGH) &&
414 max_low_pfn >= max_pfn;
417 static inline struct bio *blk_queue_bounce(struct bio *bio,
418 struct request_queue *q)
420 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
421 return __blk_queue_bounce(bio, q);
425 #ifdef CONFIG_BLK_DEV_ZONED
426 void disk_init_zone_resources(struct gendisk *disk);
427 void disk_free_zone_resources(struct gendisk *disk);
428 static inline bool bio_zone_write_plugging(struct bio *bio)
430 return bio_flagged(bio, BIO_ZONE_WRITE_PLUGGING);
432 static inline bool bio_is_zone_append(struct bio *bio)
434 return bio_op(bio) == REQ_OP_ZONE_APPEND ||
435 bio_flagged(bio, BIO_EMULATES_ZONE_APPEND);
437 void blk_zone_write_plug_bio_merged(struct bio *bio);
438 void blk_zone_write_plug_init_request(struct request *rq);
439 static inline void blk_zone_update_request_bio(struct request *rq,
443 * For zone append requests, the request sector indicates the location
444 * at which the BIO data was written. Return this value to the BIO
445 * issuer through the BIO iter sector.
446 * For plugged zone writes, which include emulated zone append, we need
447 * the original BIO sector so that blk_zone_write_plug_bio_endio() can
448 * lookup the zone write plug.
450 if (req_op(rq) == REQ_OP_ZONE_APPEND || bio_zone_write_plugging(bio))
451 bio->bi_iter.bi_sector = rq->__sector;
453 void blk_zone_write_plug_bio_endio(struct bio *bio);
454 static inline void blk_zone_bio_endio(struct bio *bio)
457 * For write BIOs to zoned devices, signal the completion of the BIO so
458 * that the next write BIO can be submitted by zone write plugging.
460 if (bio_zone_write_plugging(bio))
461 blk_zone_write_plug_bio_endio(bio);
464 void blk_zone_write_plug_finish_request(struct request *rq);
465 static inline void blk_zone_finish_request(struct request *rq)
467 if (rq->rq_flags & RQF_ZONE_WRITE_PLUGGING)
468 blk_zone_write_plug_finish_request(rq);
470 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
472 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
473 unsigned int cmd, unsigned long arg);
474 #else /* CONFIG_BLK_DEV_ZONED */
475 static inline void disk_init_zone_resources(struct gendisk *disk)
478 static inline void disk_free_zone_resources(struct gendisk *disk)
481 static inline bool bio_zone_write_plugging(struct bio *bio)
485 static inline bool bio_is_zone_append(struct bio *bio)
489 static inline void blk_zone_write_plug_bio_merged(struct bio *bio)
492 static inline void blk_zone_write_plug_init_request(struct request *rq)
495 static inline void blk_zone_update_request_bio(struct request *rq,
499 static inline void blk_zone_bio_endio(struct bio *bio)
502 static inline void blk_zone_finish_request(struct request *rq)
505 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
506 unsigned int cmd, unsigned long arg)
510 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
511 blk_mode_t mode, unsigned int cmd, unsigned long arg)
515 #endif /* CONFIG_BLK_DEV_ZONED */
517 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
518 void bdev_add(struct block_device *bdev, dev_t dev);
519 void bdev_unhash(struct block_device *bdev);
520 void bdev_drop(struct block_device *bdev);
522 int blk_alloc_ext_minor(void);
523 void blk_free_ext_minor(unsigned int minor);
524 #define ADDPART_FLAG_NONE 0
525 #define ADDPART_FLAG_RAID 1
526 #define ADDPART_FLAG_WHOLEDISK 2
527 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
529 int bdev_del_partition(struct gendisk *disk, int partno);
530 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
532 void drop_partition(struct block_device *part);
534 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
536 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
537 struct lock_class_key *lkclass);
539 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
540 struct page *page, unsigned int len, unsigned int offset,
541 unsigned int max_sectors, bool *same_page);
544 * Clean up a page appropriately, where the page may be pinned, may have a
545 * ref taken on it or neither.
547 static inline void bio_release_page(struct bio *bio, struct page *page)
549 if (bio_flagged(bio, BIO_PAGE_PINNED))
550 unpin_user_page(page);
553 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
555 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
557 int disk_alloc_events(struct gendisk *disk);
558 void disk_add_events(struct gendisk *disk);
559 void disk_del_events(struct gendisk *disk);
560 void disk_release_events(struct gendisk *disk);
561 void disk_block_events(struct gendisk *disk);
562 void disk_unblock_events(struct gendisk *disk);
563 void disk_flush_events(struct gendisk *disk, unsigned int mask);
564 extern struct device_attribute dev_attr_events;
565 extern struct device_attribute dev_attr_events_async;
566 extern struct device_attribute dev_attr_events_poll_msecs;
568 extern struct attribute_group blk_trace_attr_group;
570 blk_mode_t file_to_blk_mode(struct file *file);
571 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
572 loff_t lstart, loff_t lend);
573 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
574 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
576 extern const struct address_space_operations def_blk_aops;
578 int disk_register_independent_access_ranges(struct gendisk *disk);
579 void disk_unregister_independent_access_ranges(struct gendisk *disk);
581 #ifdef CONFIG_FAIL_MAKE_REQUEST
582 bool should_fail_request(struct block_device *part, unsigned int bytes);
583 #else /* CONFIG_FAIL_MAKE_REQUEST */
584 static inline bool should_fail_request(struct block_device *part,
589 #endif /* CONFIG_FAIL_MAKE_REQUEST */
592 * Optimized request reference counting. Ideally we'd make timeouts be more
593 * clever, as that's the only reason we need references at all... But until
594 * this happens, this is faster than using refcount_t. Also see:
596 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
598 #define req_ref_zero_or_close_to_overflow(req) \
599 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
601 static inline bool req_ref_inc_not_zero(struct request *req)
603 return atomic_inc_not_zero(&req->ref);
606 static inline bool req_ref_put_and_test(struct request *req)
608 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
609 return atomic_dec_and_test(&req->ref);
612 static inline void req_ref_set(struct request *req, int value)
614 atomic_set(&req->ref, value);
617 static inline int req_ref_read(struct request *req)
619 return atomic_read(&req->ref);
622 static inline u64 blk_time_get_ns(void)
624 struct blk_plug *plug = current->plug;
626 if (!plug || !in_task())
627 return ktime_get_ns();
630 * 0 could very well be a valid time, but rather than flag "this is
631 * a valid timestamp" separately, just accept that we'll do an extra
632 * ktime_get_ns() if we just happen to get 0 as the current time.
634 if (!plug->cur_ktime) {
635 plug->cur_ktime = ktime_get_ns();
636 current->flags |= PF_BLOCK_TS;
638 return plug->cur_ktime;
641 static inline ktime_t blk_time_get(void)
643 return ns_to_ktime(blk_time_get_ns());
647 * From most significant bit:
648 * 1 bit: reserved for other usage, see below
649 * 12 bits: original size of bio
650 * 51 bits: issue time of bio
652 #define BIO_ISSUE_RES_BITS 1
653 #define BIO_ISSUE_SIZE_BITS 12
654 #define BIO_ISSUE_RES_SHIFT (64 - BIO_ISSUE_RES_BITS)
655 #define BIO_ISSUE_SIZE_SHIFT (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
656 #define BIO_ISSUE_TIME_MASK ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
657 #define BIO_ISSUE_SIZE_MASK \
658 (((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
659 #define BIO_ISSUE_RES_MASK (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
661 /* Reserved bit for blk-throtl */
662 #define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
664 static inline u64 __bio_issue_time(u64 time)
666 return time & BIO_ISSUE_TIME_MASK;
669 static inline u64 bio_issue_time(struct bio_issue *issue)
671 return __bio_issue_time(issue->value);
674 static inline sector_t bio_issue_size(struct bio_issue *issue)
676 return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
679 static inline void bio_issue_init(struct bio_issue *issue,
682 size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
683 issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
684 (blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
685 ((u64)size << BIO_ISSUE_SIZE_SHIFT));
688 void bdev_release(struct file *bdev_file);
689 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
690 const struct blk_holder_ops *hops, struct file *bdev_file);
691 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
693 void blk_integrity_generate(struct bio *bio);
694 void blk_integrity_verify(struct bio *bio);
695 void blk_integrity_prepare(struct request *rq);
696 void blk_integrity_complete(struct request *rq, unsigned int nr_bytes);
698 #endif /* BLK_INTERNAL_H */