Line data Source code
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 <xen/xen.h>
8 : #include "blk-crypto-internal.h"
9 :
10 : struct elevator_type;
11 :
12 : /* Max future timer expiry for timeouts */
13 : #define BLK_MAX_TIMEOUT (5 * HZ)
14 :
15 : extern struct dentry *blk_debugfs_root;
16 :
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;
25 :
26 : spinlock_t mq_flush_lock;
27 : };
28 :
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;
33 :
34 : static inline void __blk_get_queue(struct request_queue *q)
35 : {
36 0 : kobject_get(&q->kobj);
37 : }
38 :
39 : bool is_flush_rq(struct request *req);
40 :
41 : struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
42 : gfp_t flags);
43 : void blk_free_flush_queue(struct blk_flush_queue *q);
44 :
45 : void blk_freeze_queue(struct request_queue *q);
46 : void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
47 : void blk_queue_start_drain(struct request_queue *q);
48 : int __bio_queue_enter(struct request_queue *q, struct bio *bio);
49 : void submit_bio_noacct_nocheck(struct bio *bio);
50 :
51 0 : static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
52 : {
53 : rcu_read_lock();
54 0 : if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
55 : goto fail;
56 :
57 : /*
58 : * The code that increments the pm_only counter must ensure that the
59 : * counter is globally visible before the queue is unfrozen.
60 : */
61 0 : if (blk_queue_pm_only(q) &&
62 0 : (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
63 : goto fail_put;
64 :
65 : rcu_read_unlock();
66 0 : return true;
67 :
68 : fail_put:
69 0 : blk_queue_exit(q);
70 : fail:
71 : rcu_read_unlock();
72 0 : return false;
73 : }
74 :
75 0 : static inline int bio_queue_enter(struct bio *bio)
76 : {
77 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
78 :
79 0 : if (blk_try_enter_queue(q, false))
80 : return 0;
81 0 : return __bio_queue_enter(q, bio);
82 : }
83 :
84 : #define BIO_INLINE_VECS 4
85 : struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
86 : gfp_t gfp_mask);
87 : void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
88 :
89 0 : static inline bool biovec_phys_mergeable(struct request_queue *q,
90 : struct bio_vec *vec1, struct bio_vec *vec2)
91 : {
92 0 : unsigned long mask = queue_segment_boundary(q);
93 0 : phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
94 0 : phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
95 :
96 0 : if (addr1 + vec1->bv_len != addr2)
97 : return false;
98 : if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
99 : return false;
100 0 : if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
101 : return false;
102 0 : return true;
103 : }
104 :
105 : static inline bool __bvec_gap_to_prev(struct request_queue *q,
106 : struct bio_vec *bprv, unsigned int offset)
107 : {
108 0 : return (offset & queue_virt_boundary(q)) ||
109 0 : ((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
110 : }
111 :
112 : /*
113 : * Check if adding a bio_vec after bprv with offset would create a gap in
114 : * the SG list. Most drivers don't care about this, but some do.
115 : */
116 : static inline bool bvec_gap_to_prev(struct request_queue *q,
117 : struct bio_vec *bprv, unsigned int offset)
118 : {
119 0 : if (!queue_virt_boundary(q))
120 : return false;
121 0 : return __bvec_gap_to_prev(q, bprv, offset);
122 : }
123 :
124 : static inline bool rq_mergeable(struct request *rq)
125 : {
126 0 : if (blk_rq_is_passthrough(rq))
127 : return false;
128 :
129 0 : if (req_op(rq) == REQ_OP_FLUSH)
130 : return false;
131 :
132 0 : if (req_op(rq) == REQ_OP_WRITE_ZEROES)
133 : return false;
134 :
135 0 : if (req_op(rq) == REQ_OP_ZONE_APPEND)
136 : return false;
137 :
138 0 : if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
139 : return false;
140 0 : if (rq->rq_flags & RQF_NOMERGE_FLAGS)
141 : return false;
142 :
143 : return true;
144 : }
145 :
146 : /*
147 : * There are two different ways to handle DISCARD merges:
148 : * 1) If max_discard_segments > 1, the driver treats every bio as a range and
149 : * send the bios to controller together. The ranges don't need to be
150 : * contiguous.
151 : * 2) Otherwise, the request will be normal read/write requests. The ranges
152 : * need to be contiguous.
153 : */
154 : static inline bool blk_discard_mergable(struct request *req)
155 : {
156 0 : if (req_op(req) == REQ_OP_DISCARD &&
157 0 : queue_max_discard_segments(req->q) > 1)
158 : return true;
159 : return false;
160 : }
161 :
162 : #ifdef CONFIG_BLK_DEV_INTEGRITY
163 : void blk_flush_integrity(void);
164 : bool __bio_integrity_endio(struct bio *);
165 : void bio_integrity_free(struct bio *bio);
166 : static inline bool bio_integrity_endio(struct bio *bio)
167 : {
168 : if (bio_integrity(bio))
169 : return __bio_integrity_endio(bio);
170 : return true;
171 : }
172 :
173 : bool blk_integrity_merge_rq(struct request_queue *, struct request *,
174 : struct request *);
175 : bool blk_integrity_merge_bio(struct request_queue *, struct request *,
176 : struct bio *);
177 :
178 : static inline bool integrity_req_gap_back_merge(struct request *req,
179 : struct bio *next)
180 : {
181 : struct bio_integrity_payload *bip = bio_integrity(req->bio);
182 : struct bio_integrity_payload *bip_next = bio_integrity(next);
183 :
184 : return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
185 : bip_next->bip_vec[0].bv_offset);
186 : }
187 :
188 : static inline bool integrity_req_gap_front_merge(struct request *req,
189 : struct bio *bio)
190 : {
191 : struct bio_integrity_payload *bip = bio_integrity(bio);
192 : struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
193 :
194 : return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
195 : bip_next->bip_vec[0].bv_offset);
196 : }
197 :
198 : int blk_integrity_add(struct gendisk *disk);
199 : void blk_integrity_del(struct gendisk *);
200 : #else /* CONFIG_BLK_DEV_INTEGRITY */
201 : static inline bool blk_integrity_merge_rq(struct request_queue *rq,
202 : struct request *r1, struct request *r2)
203 : {
204 : return true;
205 : }
206 : static inline bool blk_integrity_merge_bio(struct request_queue *rq,
207 : struct request *r, struct bio *b)
208 : {
209 : return true;
210 : }
211 : static inline bool integrity_req_gap_back_merge(struct request *req,
212 : struct bio *next)
213 : {
214 : return false;
215 : }
216 : static inline bool integrity_req_gap_front_merge(struct request *req,
217 : struct bio *bio)
218 : {
219 : return false;
220 : }
221 :
222 : static inline void blk_flush_integrity(void)
223 : {
224 : }
225 : static inline bool bio_integrity_endio(struct bio *bio)
226 : {
227 : return true;
228 : }
229 : static inline void bio_integrity_free(struct bio *bio)
230 : {
231 : }
232 : static inline int blk_integrity_add(struct gendisk *disk)
233 : {
234 : return 0;
235 : }
236 : static inline void blk_integrity_del(struct gendisk *disk)
237 : {
238 : }
239 : #endif /* CONFIG_BLK_DEV_INTEGRITY */
240 :
241 : unsigned long blk_rq_timeout(unsigned long timeout);
242 : void blk_add_timer(struct request *req);
243 : const char *blk_status_to_str(blk_status_t status);
244 :
245 : bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
246 : unsigned int nr_segs);
247 : bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
248 : struct bio *bio, unsigned int nr_segs);
249 :
250 : /*
251 : * Plug flush limits
252 : */
253 : #define BLK_MAX_REQUEST_COUNT 32
254 : #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
255 :
256 : /*
257 : * Internal elevator interface
258 : */
259 : #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
260 :
261 : void blk_insert_flush(struct request *rq);
262 :
263 : int elevator_switch_mq(struct request_queue *q,
264 : struct elevator_type *new_e);
265 : void elevator_exit(struct request_queue *q);
266 : int elv_register_queue(struct request_queue *q, bool uevent);
267 : void elv_unregister_queue(struct request_queue *q);
268 :
269 : ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
270 : char *buf);
271 : ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
272 : char *buf);
273 : ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
274 : char *buf);
275 : ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
276 : char *buf);
277 : ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
278 : const char *buf, size_t count);
279 : ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
280 : ssize_t part_timeout_store(struct device *, struct device_attribute *,
281 : const char *, size_t);
282 :
283 0 : static inline bool blk_may_split(struct request_queue *q, struct bio *bio)
284 : {
285 0 : switch (bio_op(bio)) {
286 : case REQ_OP_DISCARD:
287 : case REQ_OP_SECURE_ERASE:
288 : case REQ_OP_WRITE_ZEROES:
289 : return true; /* non-trivial splitting decisions */
290 : default:
291 : break;
292 : }
293 :
294 : /*
295 : * All drivers must accept single-segments bios that are <= PAGE_SIZE.
296 : * This is a quick and dirty check that relies on the fact that
297 : * bi_io_vec[0] is always valid if a bio has data. The check might
298 : * lead to occasional false negatives when bios are cloned, but compared
299 : * to the performance impact of cloned bios themselves the loop below
300 : * doesn't matter anyway.
301 : */
302 0 : return q->limits.chunk_sectors || bio->bi_vcnt != 1 ||
303 0 : bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
304 : }
305 :
306 : void __blk_queue_split(struct request_queue *q, struct bio **bio,
307 : unsigned int *nr_segs);
308 : int ll_back_merge_fn(struct request *req, struct bio *bio,
309 : unsigned int nr_segs);
310 : bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
311 : struct request *next);
312 : unsigned int blk_recalc_rq_segments(struct request *rq);
313 : void blk_rq_set_mixed_merge(struct request *rq);
314 : bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
315 : enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
316 :
317 : int blk_dev_init(void);
318 :
319 : /*
320 : * Contribute to IO statistics IFF:
321 : *
322 : * a) it's attached to a gendisk, and
323 : * b) the queue had IO stats enabled when this request was started
324 : */
325 : static inline bool blk_do_io_stat(struct request *rq)
326 : {
327 0 : return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
328 : }
329 :
330 : void update_io_ticks(struct block_device *part, unsigned long now, bool end);
331 :
332 : static inline void req_set_nomerge(struct request_queue *q, struct request *req)
333 : {
334 0 : req->cmd_flags |= REQ_NOMERGE;
335 0 : if (req == q->last_merge)
336 0 : q->last_merge = NULL;
337 : }
338 :
339 : /*
340 : * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
341 : * is defined as 'unsigned int', meantime it has to aligned to with logical
342 : * block size which is the minimum accepted unit by hardware.
343 : */
344 : static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
345 : {
346 0 : return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
347 : }
348 :
349 : /*
350 : * The max bio size which is aligned to q->limits.discard_granularity. This
351 : * is a hint to split large discard bio in generic block layer, then if device
352 : * driver needs to split the discard bio into smaller ones, their bi_size can
353 : * be very probably and easily aligned to discard_granularity of the device's
354 : * queue.
355 : */
356 : static inline unsigned int bio_aligned_discard_max_sectors(
357 : struct request_queue *q)
358 : {
359 0 : return round_down(UINT_MAX, q->limits.discard_granularity) >>
360 : SECTOR_SHIFT;
361 : }
362 :
363 : /*
364 : * Internal io_context interface
365 : */
366 : struct io_cq *ioc_find_get_icq(struct request_queue *q);
367 : struct io_cq *ioc_lookup_icq(struct request_queue *q);
368 : #ifdef CONFIG_BLK_ICQ
369 : void ioc_clear_queue(struct request_queue *q);
370 : #else
371 : static inline void ioc_clear_queue(struct request_queue *q)
372 : {
373 : }
374 : #endif /* CONFIG_BLK_ICQ */
375 :
376 : #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
377 : extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
378 : extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
379 : const char *page, size_t count);
380 : extern void blk_throtl_bio_endio(struct bio *bio);
381 : extern void blk_throtl_stat_add(struct request *rq, u64 time);
382 : #else
383 : static inline void blk_throtl_bio_endio(struct bio *bio) { }
384 : static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
385 : #endif
386 :
387 : void __blk_queue_bounce(struct request_queue *q, struct bio **bio);
388 :
389 : static inline bool blk_queue_may_bounce(struct request_queue *q)
390 : {
391 : return IS_ENABLED(CONFIG_BOUNCE) &&
392 : q->limits.bounce == BLK_BOUNCE_HIGH &&
393 : max_low_pfn >= max_pfn;
394 : }
395 :
396 : static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
397 : {
398 0 : if (unlikely(blk_queue_may_bounce(q) && bio_has_data(*bio)))
399 : __blk_queue_bounce(q, bio);
400 : }
401 :
402 : #ifdef CONFIG_BLK_CGROUP_IOLATENCY
403 : extern int blk_iolatency_init(struct request_queue *q);
404 : #else
405 : static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
406 : #endif
407 :
408 : #ifdef CONFIG_BLK_DEV_ZONED
409 : void blk_queue_free_zone_bitmaps(struct request_queue *q);
410 : void blk_queue_clear_zone_settings(struct request_queue *q);
411 : #else
412 : static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
413 : static inline void blk_queue_clear_zone_settings(struct request_queue *q) {}
414 : #endif
415 :
416 : int blk_alloc_ext_minor(void);
417 : void blk_free_ext_minor(unsigned int minor);
418 : #define ADDPART_FLAG_NONE 0
419 : #define ADDPART_FLAG_RAID 1
420 : #define ADDPART_FLAG_WHOLEDISK 2
421 : int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
422 : sector_t length);
423 : int bdev_del_partition(struct gendisk *disk, int partno);
424 : int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
425 : sector_t length);
426 : void blk_drop_partitions(struct gendisk *disk);
427 :
428 : int bio_add_hw_page(struct request_queue *q, struct bio *bio,
429 : struct page *page, unsigned int len, unsigned int offset,
430 : unsigned int max_sectors, bool *same_page);
431 :
432 : static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
433 : {
434 0 : if (srcu)
435 0 : return blk_requestq_srcu_cachep;
436 0 : return blk_requestq_cachep;
437 : }
438 : struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
439 :
440 : int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
441 :
442 : int disk_alloc_events(struct gendisk *disk);
443 : void disk_add_events(struct gendisk *disk);
444 : void disk_del_events(struct gendisk *disk);
445 : void disk_release_events(struct gendisk *disk);
446 : void disk_block_events(struct gendisk *disk);
447 : void disk_unblock_events(struct gendisk *disk);
448 : void disk_flush_events(struct gendisk *disk, unsigned int mask);
449 : extern struct device_attribute dev_attr_events;
450 : extern struct device_attribute dev_attr_events_async;
451 : extern struct device_attribute dev_attr_events_poll_msecs;
452 :
453 : static inline void bio_clear_polled(struct bio *bio)
454 : {
455 : /* can't support alloc cache if we turn off polling */
456 0 : bio_clear_flag(bio, BIO_PERCPU_CACHE);
457 0 : bio->bi_opf &= ~REQ_POLLED;
458 : }
459 :
460 : long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
461 : long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
462 :
463 : extern const struct address_space_operations def_blk_aops;
464 :
465 : int disk_register_independent_access_ranges(struct gendisk *disk,
466 : struct blk_independent_access_ranges *new_iars);
467 : void disk_unregister_independent_access_ranges(struct gendisk *disk);
468 :
469 : #ifdef CONFIG_FAIL_MAKE_REQUEST
470 : bool should_fail_request(struct block_device *part, unsigned int bytes);
471 : #else /* CONFIG_FAIL_MAKE_REQUEST */
472 : static inline bool should_fail_request(struct block_device *part,
473 : unsigned int bytes)
474 : {
475 : return false;
476 : }
477 : #endif /* CONFIG_FAIL_MAKE_REQUEST */
478 :
479 : /*
480 : * Optimized request reference counting. Ideally we'd make timeouts be more
481 : * clever, as that's the only reason we need references at all... But until
482 : * this happens, this is faster than using refcount_t. Also see:
483 : *
484 : * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
485 : */
486 : #define req_ref_zero_or_close_to_overflow(req) \
487 : ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
488 :
489 : static inline bool req_ref_inc_not_zero(struct request *req)
490 : {
491 0 : return atomic_inc_not_zero(&req->ref);
492 : }
493 :
494 0 : static inline bool req_ref_put_and_test(struct request *req)
495 : {
496 0 : WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
497 0 : return atomic_dec_and_test(&req->ref);
498 : }
499 :
500 : static inline void req_ref_set(struct request *req, int value)
501 : {
502 0 : atomic_set(&req->ref, value);
503 : }
504 :
505 : static inline int req_ref_read(struct request *req)
506 : {
507 0 : return atomic_read(&req->ref);
508 : }
509 :
510 : #endif /* BLK_INTERNAL_H */
|
