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 extern struct kmem_cache *blk_requestq_cachep;
30 extern struct kmem_cache *blk_requestq_srcu_cachep;
31 extern struct kobj_type blk_queue_ktype;
32 extern struct ida blk_queue_ida;
34 bool is_flush_rq(struct request *req);
36 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
38 void blk_free_flush_queue(struct blk_flush_queue *q);
40 void blk_freeze_queue(struct request_queue *q);
41 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
42 void blk_queue_start_drain(struct request_queue *q);
43 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
44 void submit_bio_noacct_nocheck(struct bio *bio);
46 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
49 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
53 * The code that increments the pm_only counter must ensure that the
54 * counter is globally visible before the queue is unfrozen.
56 if (blk_queue_pm_only(q) &&
57 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
70 static inline int bio_queue_enter(struct bio *bio)
72 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
74 if (blk_try_enter_queue(q, false))
76 return __bio_queue_enter(q, bio);
79 #define BIO_INLINE_VECS 4
80 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
82 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
84 static inline bool biovec_phys_mergeable(struct request_queue *q,
85 struct bio_vec *vec1, struct bio_vec *vec2)
87 unsigned long mask = queue_segment_boundary(q);
88 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
89 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
92 * Merging adjacent physical pages may not work correctly under KMSAN
93 * if their metadata pages aren't adjacent. Just disable merging.
95 if (IS_ENABLED(CONFIG_KMSAN))
98 if (addr1 + vec1->bv_len != addr2)
100 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
102 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
107 static inline bool __bvec_gap_to_prev(struct queue_limits *lim,
108 struct bio_vec *bprv, unsigned int offset)
110 return (offset & lim->virt_boundary_mask) ||
111 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
115 * Check if adding a bio_vec after bprv with offset would create a gap in
116 * the SG list. Most drivers don't care about this, but some do.
118 static inline bool bvec_gap_to_prev(struct queue_limits *lim,
119 struct bio_vec *bprv, unsigned int offset)
121 if (!lim->virt_boundary_mask)
123 return __bvec_gap_to_prev(lim, bprv, offset);
126 static inline bool rq_mergeable(struct request *rq)
128 if (blk_rq_is_passthrough(rq))
131 if (req_op(rq) == REQ_OP_FLUSH)
134 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
137 if (req_op(rq) == REQ_OP_ZONE_APPEND)
140 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
142 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
149 * There are two different ways to handle DISCARD merges:
150 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
151 * send the bios to controller together. The ranges don't need to be
153 * 2) Otherwise, the request will be normal read/write requests. The ranges
154 * need to be contiguous.
156 static inline bool blk_discard_mergable(struct request *req)
158 if (req_op(req) == REQ_OP_DISCARD &&
159 queue_max_discard_segments(req->q) > 1)
164 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
167 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
168 return min(q->limits.max_discard_sectors,
169 UINT_MAX >> SECTOR_SHIFT);
171 if (unlikely(op == REQ_OP_WRITE_ZEROES))
172 return q->limits.max_write_zeroes_sectors;
174 return q->limits.max_sectors;
177 #ifdef CONFIG_BLK_DEV_INTEGRITY
178 void blk_flush_integrity(void);
179 bool __bio_integrity_endio(struct bio *);
180 void bio_integrity_free(struct bio *bio);
181 static inline bool bio_integrity_endio(struct bio *bio)
183 if (bio_integrity(bio))
184 return __bio_integrity_endio(bio);
188 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
190 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
193 static inline bool integrity_req_gap_back_merge(struct request *req,
196 struct bio_integrity_payload *bip = bio_integrity(req->bio);
197 struct bio_integrity_payload *bip_next = bio_integrity(next);
199 return bvec_gap_to_prev(&req->q->limits,
200 &bip->bip_vec[bip->bip_vcnt - 1],
201 bip_next->bip_vec[0].bv_offset);
204 static inline bool integrity_req_gap_front_merge(struct request *req,
207 struct bio_integrity_payload *bip = bio_integrity(bio);
208 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
210 return bvec_gap_to_prev(&req->q->limits,
211 &bip->bip_vec[bip->bip_vcnt - 1],
212 bip_next->bip_vec[0].bv_offset);
215 int blk_integrity_add(struct gendisk *disk);
216 void blk_integrity_del(struct gendisk *);
217 #else /* CONFIG_BLK_DEV_INTEGRITY */
218 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
219 struct request *r1, struct request *r2)
223 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
224 struct request *r, struct bio *b)
228 static inline bool integrity_req_gap_back_merge(struct request *req,
233 static inline bool integrity_req_gap_front_merge(struct request *req,
239 static inline void blk_flush_integrity(void)
242 static inline bool bio_integrity_endio(struct bio *bio)
246 static inline void bio_integrity_free(struct bio *bio)
249 static inline int blk_integrity_add(struct gendisk *disk)
253 static inline void blk_integrity_del(struct gendisk *disk)
256 #endif /* CONFIG_BLK_DEV_INTEGRITY */
258 unsigned long blk_rq_timeout(unsigned long timeout);
259 void blk_add_timer(struct request *req);
260 const char *blk_status_to_str(blk_status_t status);
262 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
263 unsigned int nr_segs);
264 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
265 struct bio *bio, unsigned int nr_segs);
270 #define BLK_MAX_REQUEST_COUNT 32
271 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
274 * Internal elevator interface
276 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
278 void blk_insert_flush(struct request *rq);
280 int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
281 void elevator_exit(struct request_queue *q);
282 int elv_register_queue(struct request_queue *q, bool uevent);
283 void elv_unregister_queue(struct request_queue *q);
285 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
287 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
289 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
291 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
293 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
294 const char *buf, size_t count);
295 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
296 ssize_t part_timeout_store(struct device *, struct device_attribute *,
297 const char *, size_t);
299 static inline bool bio_may_exceed_limits(struct bio *bio,
300 struct queue_limits *lim)
302 switch (bio_op(bio)) {
304 case REQ_OP_SECURE_ERASE:
305 case REQ_OP_WRITE_ZEROES:
306 return true; /* non-trivial splitting decisions */
312 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
313 * This is a quick and dirty check that relies on the fact that
314 * bi_io_vec[0] is always valid if a bio has data. The check might
315 * lead to occasional false negatives when bios are cloned, but compared
316 * to the performance impact of cloned bios themselves the loop below
317 * doesn't matter anyway.
319 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
320 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
323 struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim,
324 unsigned int *nr_segs);
325 int ll_back_merge_fn(struct request *req, struct bio *bio,
326 unsigned int nr_segs);
327 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
328 struct request *next);
329 unsigned int blk_recalc_rq_segments(struct request *rq);
330 void blk_rq_set_mixed_merge(struct request *rq);
331 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
332 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
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_CGROUP_IOLATENCY
398 int blk_iolatency_init(struct gendisk *disk);
400 static inline int blk_iolatency_init(struct gendisk *disk) { return 0; };
403 #ifdef CONFIG_BLK_DEV_ZONED
404 void disk_free_zone_bitmaps(struct gendisk *disk);
405 void disk_clear_zone_settings(struct gendisk *disk);
407 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
408 static inline void disk_clear_zone_settings(struct gendisk *disk) {}
411 int blk_alloc_ext_minor(void);
412 void blk_free_ext_minor(unsigned int minor);
413 #define ADDPART_FLAG_NONE 0
414 #define ADDPART_FLAG_RAID 1
415 #define ADDPART_FLAG_WHOLEDISK 2
416 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
418 int bdev_del_partition(struct gendisk *disk, int partno);
419 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
421 void blk_drop_partitions(struct gendisk *disk);
423 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
424 struct lock_class_key *lkclass);
426 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
427 struct page *page, unsigned int len, unsigned int offset,
428 unsigned int max_sectors, bool *same_page);
430 static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
433 return blk_requestq_srcu_cachep;
434 return blk_requestq_cachep;
436 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
438 int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
440 int disk_alloc_events(struct gendisk *disk);
441 void disk_add_events(struct gendisk *disk);
442 void disk_del_events(struct gendisk *disk);
443 void disk_release_events(struct gendisk *disk);
444 void disk_block_events(struct gendisk *disk);
445 void disk_unblock_events(struct gendisk *disk);
446 void disk_flush_events(struct gendisk *disk, unsigned int mask);
447 extern struct device_attribute dev_attr_events;
448 extern struct device_attribute dev_attr_events_async;
449 extern struct device_attribute dev_attr_events_poll_msecs;
451 extern struct attribute_group blk_trace_attr_group;
453 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
454 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
456 extern const struct address_space_operations def_blk_aops;
458 int disk_register_independent_access_ranges(struct gendisk *disk);
459 void disk_unregister_independent_access_ranges(struct gendisk *disk);
461 #ifdef CONFIG_FAIL_MAKE_REQUEST
462 bool should_fail_request(struct block_device *part, unsigned int bytes);
463 #else /* CONFIG_FAIL_MAKE_REQUEST */
464 static inline bool should_fail_request(struct block_device *part,
469 #endif /* CONFIG_FAIL_MAKE_REQUEST */
472 * Optimized request reference counting. Ideally we'd make timeouts be more
473 * clever, as that's the only reason we need references at all... But until
474 * this happens, this is faster than using refcount_t. Also see:
476 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
478 #define req_ref_zero_or_close_to_overflow(req) \
479 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
481 static inline bool req_ref_inc_not_zero(struct request *req)
483 return atomic_inc_not_zero(&req->ref);
486 static inline bool req_ref_put_and_test(struct request *req)
488 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
489 return atomic_dec_and_test(&req->ref);
492 static inline void req_ref_set(struct request *req, int value)
494 atomic_set(&req->ref, value);
497 static inline int req_ref_read(struct request *req)
499 return atomic_read(&req->ref);
502 #endif /* BLK_INTERNAL_H */