1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/blk-crypto.h>
6 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */
7 #include <linux/timekeeping.h>
9 #include "blk-crypto-internal.h"
13 /* Max future timer expiry for timeouts */
14 #define BLK_MAX_TIMEOUT (5 * HZ)
16 extern struct dentry *blk_debugfs_root;
18 struct blk_flush_queue {
19 spinlock_t mq_flush_lock;
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 unsigned long flush_data_in_flight;
26 struct request *flush_rq;
29 bool is_flush_rq(struct request *req);
31 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
33 void blk_free_flush_queue(struct blk_flush_queue *q);
35 void blk_freeze_queue(struct request_queue *q);
36 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
37 void blk_queue_start_drain(struct request_queue *q);
38 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
39 void submit_bio_noacct_nocheck(struct bio *bio);
41 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
44 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
48 * The code that increments the pm_only counter must ensure that the
49 * counter is globally visible before the queue is unfrozen.
51 if (blk_queue_pm_only(q) &&
52 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
65 static inline int bio_queue_enter(struct bio *bio)
67 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
69 if (blk_try_enter_queue(q, false))
71 return __bio_queue_enter(q, bio);
74 #define BIO_INLINE_VECS 4
75 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
77 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
79 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
80 struct page *page, unsigned len, unsigned offset,
83 static inline bool biovec_phys_mergeable(struct request_queue *q,
84 struct bio_vec *vec1, struct bio_vec *vec2)
86 unsigned long mask = queue_segment_boundary(q);
87 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
88 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
91 * Merging adjacent physical pages may not work correctly under KMSAN
92 * if their metadata pages aren't adjacent. Just disable merging.
94 if (IS_ENABLED(CONFIG_KMSAN))
97 if (addr1 + vec1->bv_len != addr2)
99 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
101 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
106 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
107 struct bio_vec *bprv, unsigned int offset)
109 return (offset & lim->virt_boundary_mask) ||
110 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
114 * Check if adding a bio_vec after bprv with offset would create a gap in
115 * the SG list. Most drivers don't care about this, but some do.
117 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
118 struct bio_vec *bprv, unsigned int offset)
120 if (!lim->virt_boundary_mask)
122 return __bvec_gap_to_prev(lim, bprv, offset);
125 static inline bool rq_mergeable(struct request *rq)
127 if (blk_rq_is_passthrough(rq))
130 if (req_op(rq) == REQ_OP_FLUSH)
133 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
136 if (req_op(rq) == REQ_OP_ZONE_APPEND)
139 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
141 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
148 * There are two different ways to handle DISCARD merges:
149 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
150 * send the bios to controller together. The ranges don't need to be
152 * 2) Otherwise, the request will be normal read/write requests. The ranges
153 * need to be contiguous.
155 static inline bool blk_discard_mergable(struct request *req)
157 if (req_op(req) == REQ_OP_DISCARD &&
158 queue_max_discard_segments(req->q) > 1)
163 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
165 if (req_op(rq) == REQ_OP_DISCARD)
166 return queue_max_discard_segments(rq->q);
167 return queue_max_segments(rq->q);
170 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
173 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
174 return min(q->limits.max_discard_sectors,
175 UINT_MAX >> SECTOR_SHIFT);
177 if (unlikely(op == REQ_OP_WRITE_ZEROES))
178 return q->limits.max_write_zeroes_sectors;
180 return q->limits.max_sectors;
183 #ifdef CONFIG_BLK_DEV_INTEGRITY
184 void blk_flush_integrity(void);
185 bool __bio_integrity_endio(struct bio *);
186 void bio_integrity_free(struct bio *bio);
187 static inline bool bio_integrity_endio(struct bio *bio)
189 if (bio_integrity(bio))
190 return __bio_integrity_endio(bio);
194 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
196 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
199 static inline bool integrity_req_gap_back_merge(struct request *req,
202 struct bio_integrity_payload *bip = bio_integrity(req->bio);
203 struct bio_integrity_payload *bip_next = bio_integrity(next);
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);
210 static inline bool integrity_req_gap_front_merge(struct request *req,
213 struct bio_integrity_payload *bip = bio_integrity(bio);
214 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
216 return bvec_gap_to_prev(&req->q->limits,
217 &bip->bip_vec[bip->bip_vcnt - 1],
218 bip_next->bip_vec[0].bv_offset);
221 extern const struct attribute_group blk_integrity_attr_group;
222 #else /* CONFIG_BLK_DEV_INTEGRITY */
223 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
224 struct request *r1, struct request *r2)
228 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
229 struct request *r, struct bio *b)
233 static inline bool integrity_req_gap_back_merge(struct request *req,
238 static inline bool integrity_req_gap_front_merge(struct request *req,
244 static inline void blk_flush_integrity(void)
247 static inline bool bio_integrity_endio(struct bio *bio)
251 static inline void bio_integrity_free(struct bio *bio)
254 #endif /* CONFIG_BLK_DEV_INTEGRITY */
256 unsigned long blk_rq_timeout(unsigned long timeout);
257 void blk_add_timer(struct request *req);
259 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
260 unsigned int nr_segs);
261 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
262 struct bio *bio, unsigned int nr_segs);
267 #define BLK_MAX_REQUEST_COUNT 32
268 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
271 * Internal elevator interface
273 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
275 bool blk_insert_flush(struct request *rq);
277 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
278 void elevator_disable(struct request_queue *q);
279 void elevator_exit(struct request_queue *q);
280 int elv_register_queue(struct request_queue *q, bool uevent);
281 void elv_unregister_queue(struct request_queue *q);
283 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
285 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
287 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
289 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
291 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
292 const char *buf, size_t count);
293 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
294 ssize_t part_timeout_store(struct device *, struct device_attribute *,
295 const char *, size_t);
297 static inline bool bio_may_exceed_limits(struct bio *bio,
298 const struct queue_limits *lim)
300 switch (bio_op(bio)) {
302 case REQ_OP_SECURE_ERASE:
303 case REQ_OP_WRITE_ZEROES:
304 return true; /* non-trivial splitting decisions */
310 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
311 * This is a quick and dirty check that relies on the fact that
312 * bi_io_vec[0] is always valid if a bio has data. The check might
313 * lead to occasional false negatives when bios are cloned, but compared
314 * to the performance impact of cloned bios themselves the loop below
315 * doesn't matter anyway.
317 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
318 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
321 struct bio *__bio_split_to_limits(struct bio *bio,
322 const struct queue_limits *lim,
323 unsigned int *nr_segs);
324 int ll_back_merge_fn(struct request *req, struct bio *bio,
325 unsigned int nr_segs);
326 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
327 struct request *next);
328 unsigned int blk_recalc_rq_segments(struct request *rq);
329 void blk_rq_set_mixed_merge(struct request *rq);
330 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
331 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
333 int blk_set_default_limits(struct queue_limits *lim);
334 int blk_dev_init(void);
337 * Contribute to IO statistics IFF:
339 * a) it's attached to a gendisk, and
340 * b) the queue had IO stats enabled when this request was started
342 static inline bool blk_do_io_stat(struct request *rq)
344 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
347 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
349 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
351 req->cmd_flags |= REQ_NOMERGE;
352 if (req == q->last_merge)
353 q->last_merge = NULL;
357 * Internal io_context interface
359 struct io_cq *ioc_find_get_icq(struct request_queue *q);
360 struct io_cq *ioc_lookup_icq(struct request_queue *q);
361 #ifdef CONFIG_BLK_ICQ
362 void ioc_clear_queue(struct request_queue *q);
364 static inline void ioc_clear_queue(struct request_queue *q)
367 #endif /* CONFIG_BLK_ICQ */
369 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
370 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
371 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
372 const char *page, size_t count);
373 extern void blk_throtl_bio_endio(struct bio *bio);
374 extern void blk_throtl_stat_add(struct request *rq, u64 time);
376 static inline void blk_throtl_bio_endio(struct bio *bio) { }
377 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
380 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
382 static inline bool blk_queue_may_bounce(struct request_queue *q)
384 return IS_ENABLED(CONFIG_BOUNCE) &&
385 q->limits.bounce == BLK_BOUNCE_HIGH &&
386 max_low_pfn >= max_pfn;
389 static inline struct bio *blk_queue_bounce(struct bio *bio,
390 struct request_queue *q)
392 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
393 return __blk_queue_bounce(bio, q);
397 #ifdef CONFIG_BLK_DEV_ZONED
398 void disk_free_zone_bitmaps(struct gendisk *disk);
399 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
401 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
402 unsigned int cmd, unsigned long arg);
403 #else /* CONFIG_BLK_DEV_ZONED */
404 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
405 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
406 unsigned int cmd, unsigned long arg)
410 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
411 blk_mode_t mode, unsigned int cmd, unsigned long arg)
415 #endif /* CONFIG_BLK_DEV_ZONED */
417 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
418 void bdev_add(struct block_device *bdev, dev_t dev);
420 int blk_alloc_ext_minor(void);
421 void blk_free_ext_minor(unsigned int minor);
422 #define ADDPART_FLAG_NONE 0
423 #define ADDPART_FLAG_RAID 1
424 #define ADDPART_FLAG_WHOLEDISK 2
425 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
427 int bdev_del_partition(struct gendisk *disk, int partno);
428 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
430 void drop_partition(struct block_device *part);
432 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
434 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
435 struct lock_class_key *lkclass);
437 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
438 struct page *page, unsigned int len, unsigned int offset,
439 unsigned int max_sectors, bool *same_page);
442 * Clean up a page appropriately, where the page may be pinned, may have a
443 * ref taken on it or neither.
445 static inline void bio_release_page(struct bio *bio, struct page *page)
447 if (bio_flagged(bio, BIO_PAGE_PINNED))
448 unpin_user_page(page);
451 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
453 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
455 int disk_alloc_events(struct gendisk *disk);
456 void disk_add_events(struct gendisk *disk);
457 void disk_del_events(struct gendisk *disk);
458 void disk_release_events(struct gendisk *disk);
459 void disk_block_events(struct gendisk *disk);
460 void disk_unblock_events(struct gendisk *disk);
461 void disk_flush_events(struct gendisk *disk, unsigned int mask);
462 extern struct device_attribute dev_attr_events;
463 extern struct device_attribute dev_attr_events_async;
464 extern struct device_attribute dev_attr_events_poll_msecs;
466 extern struct attribute_group blk_trace_attr_group;
468 blk_mode_t file_to_blk_mode(struct file *file);
469 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
470 loff_t lstart, loff_t lend);
471 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
472 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
474 extern const struct address_space_operations def_blk_aops;
476 int disk_register_independent_access_ranges(struct gendisk *disk);
477 void disk_unregister_independent_access_ranges(struct gendisk *disk);
479 #ifdef CONFIG_FAIL_MAKE_REQUEST
480 bool should_fail_request(struct block_device *part, unsigned int bytes);
481 #else /* CONFIG_FAIL_MAKE_REQUEST */
482 static inline bool should_fail_request(struct block_device *part,
487 #endif /* CONFIG_FAIL_MAKE_REQUEST */
490 * Optimized request reference counting. Ideally we'd make timeouts be more
491 * clever, as that's the only reason we need references at all... But until
492 * this happens, this is faster than using refcount_t. Also see:
494 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
496 #define req_ref_zero_or_close_to_overflow(req) \
497 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
499 static inline bool req_ref_inc_not_zero(struct request *req)
501 return atomic_inc_not_zero(&req->ref);
504 static inline bool req_ref_put_and_test(struct request *req)
506 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
507 return atomic_dec_and_test(&req->ref);
510 static inline void req_ref_set(struct request *req, int value)
512 atomic_set(&req->ref, value);
515 static inline int req_ref_read(struct request *req)
517 return atomic_read(&req->ref);
520 static inline u64 blk_time_get_ns(void)
522 struct blk_plug *plug = current->plug;
525 return ktime_get_ns();
528 * 0 could very well be a valid time, but rather than flag "this is
529 * a valid timestamp" separately, just accept that we'll do an extra
530 * ktime_get_ns() if we just happen to get 0 as the current time.
532 if (!plug->cur_ktime) {
533 plug->cur_ktime = ktime_get_ns();
534 current->flags |= PF_BLOCK_TS;
536 return plug->cur_ktime;
539 static inline ktime_t blk_time_get(void)
541 return ns_to_ktime(blk_time_get_ns());
545 * From most significant bit:
546 * 1 bit: reserved for other usage, see below
547 * 12 bits: original size of bio
548 * 51 bits: issue time of bio
550 #define BIO_ISSUE_RES_BITS 1
551 #define BIO_ISSUE_SIZE_BITS 12
552 #define BIO_ISSUE_RES_SHIFT (64 - BIO_ISSUE_RES_BITS)
553 #define BIO_ISSUE_SIZE_SHIFT (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
554 #define BIO_ISSUE_TIME_MASK ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
555 #define BIO_ISSUE_SIZE_MASK \
556 (((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
557 #define BIO_ISSUE_RES_MASK (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
559 /* Reserved bit for blk-throtl */
560 #define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
562 static inline u64 __bio_issue_time(u64 time)
564 return time & BIO_ISSUE_TIME_MASK;
567 static inline u64 bio_issue_time(struct bio_issue *issue)
569 return __bio_issue_time(issue->value);
572 static inline sector_t bio_issue_size(struct bio_issue *issue)
574 return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
577 static inline void bio_issue_init(struct bio_issue *issue,
580 size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
581 issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
582 (blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
583 ((u64)size << BIO_ISSUE_SIZE_SHIFT));
586 #endif /* BLK_INTERNAL_H */