1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/blk-crypto.h>
6 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */
8 #include "blk-crypto-internal.h"
12 /* Max future timer expiry for timeouts */
13 #define BLK_MAX_TIMEOUT (5 * HZ)
15 extern struct dentry *blk_debugfs_root;
17 struct 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;
26 spinlock_t mq_flush_lock;
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 static inline bool biovec_phys_mergeable(struct request_queue *q,
80 struct bio_vec *vec1, struct bio_vec *vec2)
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;
87 * Merging adjacent physical pages may not work correctly under KMSAN
88 * if their metadata pages aren't adjacent. Just disable merging.
90 if (IS_ENABLED(CONFIG_KMSAN))
93 if (addr1 + vec1->bv_len != addr2)
95 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
97 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
102 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
103 struct bio_vec *bprv, unsigned int offset)
105 return (offset & lim->virt_boundary_mask) ||
106 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
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.
113 static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
114 struct bio_vec *bprv, unsigned int offset)
116 if (!lim->virt_boundary_mask)
118 return __bvec_gap_to_prev(lim, bprv, offset);
121 static inline bool rq_mergeable(struct request *rq)
123 if (blk_rq_is_passthrough(rq))
126 if (req_op(rq) == REQ_OP_FLUSH)
129 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
132 if (req_op(rq) == REQ_OP_ZONE_APPEND)
135 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
137 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
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
148 * 2) Otherwise, the request will be normal read/write requests. The ranges
149 * need to be contiguous.
151 static inline bool blk_discard_mergable(struct request *req)
153 if (req_op(req) == REQ_OP_DISCARD &&
154 queue_max_discard_segments(req->q) > 1)
159 static inline unsigned int blk_rq_get_max_segments(struct request *rq)
161 if (req_op(rq) == REQ_OP_DISCARD)
162 return queue_max_discard_segments(rq->q);
163 return queue_max_segments(rq->q);
166 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
169 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
170 return min(q->limits.max_discard_sectors,
171 UINT_MAX >> SECTOR_SHIFT);
173 if (unlikely(op == REQ_OP_WRITE_ZEROES))
174 return q->limits.max_write_zeroes_sectors;
176 return q->limits.max_sectors;
179 #ifdef CONFIG_BLK_DEV_INTEGRITY
180 void blk_flush_integrity(void);
181 bool __bio_integrity_endio(struct bio *);
182 void bio_integrity_free(struct bio *bio);
183 static inline bool bio_integrity_endio(struct bio *bio)
185 if (bio_integrity(bio))
186 return __bio_integrity_endio(bio);
190 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
192 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
195 static inline bool integrity_req_gap_back_merge(struct request *req,
198 struct bio_integrity_payload *bip = bio_integrity(req->bio);
199 struct bio_integrity_payload *bip_next = bio_integrity(next);
201 return bvec_gap_to_prev(&req->q->limits,
202 &bip->bip_vec[bip->bip_vcnt - 1],
203 bip_next->bip_vec[0].bv_offset);
206 static inline bool integrity_req_gap_front_merge(struct request *req,
209 struct bio_integrity_payload *bip = bio_integrity(bio);
210 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
212 return bvec_gap_to_prev(&req->q->limits,
213 &bip->bip_vec[bip->bip_vcnt - 1],
214 bip_next->bip_vec[0].bv_offset);
217 extern const struct attribute_group blk_integrity_attr_group;
218 #else /* CONFIG_BLK_DEV_INTEGRITY */
219 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
220 struct request *r1, struct request *r2)
224 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
225 struct request *r, struct bio *b)
229 static inline bool integrity_req_gap_back_merge(struct request *req,
234 static inline bool integrity_req_gap_front_merge(struct request *req,
240 static inline void blk_flush_integrity(void)
243 static inline bool bio_integrity_endio(struct bio *bio)
247 static inline void bio_integrity_free(struct bio *bio)
250 #endif /* CONFIG_BLK_DEV_INTEGRITY */
252 unsigned long blk_rq_timeout(unsigned long timeout);
253 void blk_add_timer(struct request *req);
254 const char *blk_status_to_str(blk_status_t status);
256 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
257 unsigned int nr_segs);
258 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
259 struct bio *bio, unsigned int nr_segs);
264 #define BLK_MAX_REQUEST_COUNT 32
265 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
268 * Internal elevator interface
270 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
272 bool blk_insert_flush(struct request *rq);
274 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
275 void elevator_disable(struct request_queue *q);
276 void elevator_exit(struct request_queue *q);
277 int elv_register_queue(struct request_queue *q, bool uevent);
278 void elv_unregister_queue(struct request_queue *q);
280 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
282 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
284 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
286 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
288 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
289 const char *buf, size_t count);
290 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
291 ssize_t part_timeout_store(struct device *, struct device_attribute *,
292 const char *, size_t);
294 static inline bool bio_may_exceed_limits(struct bio *bio,
295 const struct queue_limits *lim)
297 switch (bio_op(bio)) {
299 case REQ_OP_SECURE_ERASE:
300 case REQ_OP_WRITE_ZEROES:
301 return true; /* non-trivial splitting decisions */
307 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
308 * This is a quick and dirty check that relies on the fact that
309 * bi_io_vec[0] is always valid if a bio has data. The check might
310 * lead to occasional false negatives when bios are cloned, but compared
311 * to the performance impact of cloned bios themselves the loop below
312 * doesn't matter anyway.
314 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
315 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
318 struct bio *__bio_split_to_limits(struct bio *bio,
319 const struct queue_limits *lim,
320 unsigned int *nr_segs);
321 int ll_back_merge_fn(struct request *req, struct bio *bio,
322 unsigned int nr_segs);
323 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
324 struct request *next);
325 unsigned int blk_recalc_rq_segments(struct request *rq);
326 void blk_rq_set_mixed_merge(struct request *rq);
327 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
328 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
330 void blk_set_default_limits(struct queue_limits *lim);
331 int blk_dev_init(void);
334 * Contribute to IO statistics IFF:
336 * a) it's attached to a gendisk, and
337 * b) the queue had IO stats enabled when this request was started
339 static inline bool blk_do_io_stat(struct request *rq)
341 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
344 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
346 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
348 req->cmd_flags |= REQ_NOMERGE;
349 if (req == q->last_merge)
350 q->last_merge = NULL;
354 * Internal io_context interface
356 struct io_cq *ioc_find_get_icq(struct request_queue *q);
357 struct io_cq *ioc_lookup_icq(struct request_queue *q);
358 #ifdef CONFIG_BLK_ICQ
359 void ioc_clear_queue(struct request_queue *q);
361 static inline void ioc_clear_queue(struct request_queue *q)
364 #endif /* CONFIG_BLK_ICQ */
366 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
367 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
368 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
369 const char *page, size_t count);
370 extern void blk_throtl_bio_endio(struct bio *bio);
371 extern void blk_throtl_stat_add(struct request *rq, u64 time);
373 static inline void blk_throtl_bio_endio(struct bio *bio) { }
374 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
377 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
379 static inline bool blk_queue_may_bounce(struct request_queue *q)
381 return IS_ENABLED(CONFIG_BOUNCE) &&
382 q->limits.bounce == BLK_BOUNCE_HIGH &&
383 max_low_pfn >= max_pfn;
386 static inline struct bio *blk_queue_bounce(struct bio *bio,
387 struct request_queue *q)
389 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
390 return __blk_queue_bounce(bio, q);
394 #ifdef CONFIG_BLK_DEV_ZONED
395 void disk_free_zone_bitmaps(struct gendisk *disk);
396 void disk_clear_zone_settings(struct gendisk *disk);
397 int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode,
398 unsigned int cmd, unsigned long arg);
399 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode,
400 unsigned int cmd, unsigned long arg);
401 #else /* CONFIG_BLK_DEV_ZONED */
402 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
403 static inline void disk_clear_zone_settings(struct gendisk *disk) {}
404 static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
405 fmode_t mode, unsigned int cmd, unsigned long arg)
409 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
410 fmode_t mode, unsigned int cmd, unsigned long arg)
414 #endif /* CONFIG_BLK_DEV_ZONED */
416 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
417 void bdev_add(struct block_device *bdev, dev_t dev);
419 int blk_alloc_ext_minor(void);
420 void 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
424 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
426 int bdev_del_partition(struct gendisk *disk, int partno);
427 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
429 void drop_partition(struct block_device *part);
431 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
433 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
434 struct lock_class_key *lkclass);
436 int 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);
441 * Clean up a page appropriately, where the page may be pinned, may have a
442 * ref taken on it or neither.
444 static inline void bio_release_page(struct bio *bio, struct page *page)
446 if (bio_flagged(bio, BIO_PAGE_PINNED))
447 unpin_user_page(page);
448 else if (bio_flagged(bio, BIO_PAGE_REFFED))
452 struct request_queue *blk_alloc_queue(int node_id);
454 int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
456 int disk_alloc_events(struct gendisk *disk);
457 void disk_add_events(struct gendisk *disk);
458 void disk_del_events(struct gendisk *disk);
459 void disk_release_events(struct gendisk *disk);
460 void disk_block_events(struct gendisk *disk);
461 void disk_unblock_events(struct gendisk *disk);
462 void disk_flush_events(struct gendisk *disk, unsigned int mask);
463 extern struct device_attribute dev_attr_events;
464 extern struct device_attribute dev_attr_events_async;
465 extern struct device_attribute dev_attr_events_poll_msecs;
467 extern struct attribute_group blk_trace_attr_group;
469 int truncate_bdev_range(struct block_device *bdev, fmode_t mode, loff_t lstart,
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 #endif /* BLK_INTERNAL_H */