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 | spinlock_t mq_flush_lock; |
19 | unsigned int flush_pending_idx:1; |
20 | unsigned int flush_running_idx:1; |
21 | blk_status_t rq_status; |
22 | unsigned long flush_pending_since; |
23 | struct list_head flush_queue[2]; |
24 | unsigned long flush_data_in_flight; |
25 | struct request *flush_rq; |
26 | }; |
27 | |
28 | bool is_flush_rq(struct request *req); |
29 | |
30 | struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size, |
31 | gfp_t flags); |
32 | void blk_free_flush_queue(struct blk_flush_queue *q); |
33 | |
34 | void blk_freeze_queue(struct request_queue *q); |
35 | void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic); |
36 | void blk_queue_start_drain(struct request_queue *q); |
37 | int __bio_queue_enter(struct request_queue *q, struct bio *bio); |
38 | void submit_bio_noacct_nocheck(struct bio *bio); |
39 | |
40 | static inline bool blk_try_enter_queue(struct request_queue *q, bool pm) |
41 | { |
42 | rcu_read_lock(); |
43 | if (!percpu_ref_tryget_live_rcu(ref: &q->q_usage_counter)) |
44 | goto fail; |
45 | |
46 | /* |
47 | * The code that increments the pm_only counter must ensure that the |
48 | * counter is globally visible before the queue is unfrozen. |
49 | */ |
50 | if (blk_queue_pm_only(q) && |
51 | (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) |
52 | goto fail_put; |
53 | |
54 | rcu_read_unlock(); |
55 | return true; |
56 | |
57 | fail_put: |
58 | blk_queue_exit(q); |
59 | fail: |
60 | rcu_read_unlock(); |
61 | return false; |
62 | } |
63 | |
64 | static inline int bio_queue_enter(struct bio *bio) |
65 | { |
66 | struct request_queue *q = bdev_get_queue(bdev: bio->bi_bdev); |
67 | |
68 | if (blk_try_enter_queue(q, pm: false)) |
69 | return 0; |
70 | return __bio_queue_enter(q, bio); |
71 | } |
72 | |
73 | #define BIO_INLINE_VECS 4 |
74 | struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, |
75 | gfp_t gfp_mask); |
76 | void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs); |
77 | |
78 | bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv, |
79 | struct page *page, unsigned len, unsigned offset, |
80 | bool *same_page); |
81 | |
82 | static inline bool biovec_phys_mergeable(struct request_queue *q, |
83 | struct bio_vec *vec1, struct bio_vec *vec2) |
84 | { |
85 | unsigned long mask = queue_segment_boundary(q); |
86 | phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset; |
87 | phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset; |
88 | |
89 | /* |
90 | * Merging adjacent physical pages may not work correctly under KMSAN |
91 | * if their metadata pages aren't adjacent. Just disable merging. |
92 | */ |
93 | if (IS_ENABLED(CONFIG_KMSAN)) |
94 | return false; |
95 | |
96 | if (addr1 + vec1->bv_len != addr2) |
97 | return false; |
98 | if (xen_domain() && !xen_biovec_phys_mergeable(vec1, page: vec2->bv_page)) |
99 | return false; |
100 | if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask)) |
101 | return false; |
102 | return true; |
103 | } |
104 | |
105 | static inline bool __bvec_gap_to_prev(const struct queue_limits *lim, |
106 | struct bio_vec *bprv, unsigned int offset) |
107 | { |
108 | return (offset & lim->virt_boundary_mask) || |
109 | ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask); |
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(const struct queue_limits *lim, |
117 | struct bio_vec *bprv, unsigned int offset) |
118 | { |
119 | if (!lim->virt_boundary_mask) |
120 | return false; |
121 | return __bvec_gap_to_prev(lim, bprv, offset); |
122 | } |
123 | |
124 | static inline bool rq_mergeable(struct request *rq) |
125 | { |
126 | if (blk_rq_is_passthrough(rq)) |
127 | return false; |
128 | |
129 | if (req_op(req: rq) == REQ_OP_FLUSH) |
130 | return false; |
131 | |
132 | if (req_op(req: rq) == REQ_OP_WRITE_ZEROES) |
133 | return false; |
134 | |
135 | if (req_op(req: rq) == REQ_OP_ZONE_APPEND) |
136 | return false; |
137 | |
138 | if (rq->cmd_flags & REQ_NOMERGE_FLAGS) |
139 | return false; |
140 | 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 | if (req_op(req) == REQ_OP_DISCARD && |
157 | queue_max_discard_segments(q: req->q) > 1) |
158 | return true; |
159 | return false; |
160 | } |
161 | |
162 | static inline unsigned int blk_rq_get_max_segments(struct request *rq) |
163 | { |
164 | if (req_op(req: rq) == REQ_OP_DISCARD) |
165 | return queue_max_discard_segments(q: rq->q); |
166 | return queue_max_segments(q: rq->q); |
167 | } |
168 | |
169 | static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q, |
170 | enum req_op op) |
171 | { |
172 | if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)) |
173 | return min(q->limits.max_discard_sectors, |
174 | UINT_MAX >> SECTOR_SHIFT); |
175 | |
176 | if (unlikely(op == REQ_OP_WRITE_ZEROES)) |
177 | return q->limits.max_write_zeroes_sectors; |
178 | |
179 | return q->limits.max_sectors; |
180 | } |
181 | |
182 | #ifdef CONFIG_BLK_DEV_INTEGRITY |
183 | void blk_flush_integrity(void); |
184 | bool __bio_integrity_endio(struct bio *); |
185 | void bio_integrity_free(struct bio *bio); |
186 | static inline bool bio_integrity_endio(struct bio *bio) |
187 | { |
188 | if (bio_integrity(bio)) |
189 | return __bio_integrity_endio(bio); |
190 | return true; |
191 | } |
192 | |
193 | bool blk_integrity_merge_rq(struct request_queue *, struct request *, |
194 | struct request *); |
195 | bool blk_integrity_merge_bio(struct request_queue *, struct request *, |
196 | struct bio *); |
197 | |
198 | static inline bool integrity_req_gap_back_merge(struct request *req, |
199 | struct bio *next) |
200 | { |
201 | struct bio_integrity_payload *bip = bio_integrity(bio: req->bio); |
202 | struct bio_integrity_payload *bip_next = bio_integrity(bio: next); |
203 | |
204 | return bvec_gap_to_prev(lim: &req->q->limits, |
205 | bprv: &bip->bip_vec[bip->bip_vcnt - 1], |
206 | offset: bip_next->bip_vec[0].bv_offset); |
207 | } |
208 | |
209 | static inline bool integrity_req_gap_front_merge(struct request *req, |
210 | struct bio *bio) |
211 | { |
212 | struct bio_integrity_payload *bip = bio_integrity(bio); |
213 | struct bio_integrity_payload *bip_next = bio_integrity(bio: req->bio); |
214 | |
215 | return bvec_gap_to_prev(lim: &req->q->limits, |
216 | bprv: &bip->bip_vec[bip->bip_vcnt - 1], |
217 | offset: bip_next->bip_vec[0].bv_offset); |
218 | } |
219 | |
220 | extern const struct attribute_group blk_integrity_attr_group; |
221 | #else /* CONFIG_BLK_DEV_INTEGRITY */ |
222 | static inline bool blk_integrity_merge_rq(struct request_queue *rq, |
223 | struct request *r1, struct request *r2) |
224 | { |
225 | return true; |
226 | } |
227 | static inline bool blk_integrity_merge_bio(struct request_queue *rq, |
228 | struct request *r, struct bio *b) |
229 | { |
230 | return true; |
231 | } |
232 | static inline bool integrity_req_gap_back_merge(struct request *req, |
233 | struct bio *next) |
234 | { |
235 | return false; |
236 | } |
237 | static inline bool integrity_req_gap_front_merge(struct request *req, |
238 | struct bio *bio) |
239 | { |
240 | return false; |
241 | } |
242 | |
243 | static inline void blk_flush_integrity(void) |
244 | { |
245 | } |
246 | static inline bool bio_integrity_endio(struct bio *bio) |
247 | { |
248 | return true; |
249 | } |
250 | static inline void bio_integrity_free(struct bio *bio) |
251 | { |
252 | } |
253 | #endif /* CONFIG_BLK_DEV_INTEGRITY */ |
254 | |
255 | unsigned long blk_rq_timeout(unsigned long timeout); |
256 | void blk_add_timer(struct request *req); |
257 | |
258 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
259 | unsigned int nr_segs); |
260 | bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, |
261 | struct bio *bio, unsigned int nr_segs); |
262 | |
263 | /* |
264 | * Plug flush limits |
265 | */ |
266 | #define BLK_MAX_REQUEST_COUNT 32 |
267 | #define BLK_PLUG_FLUSH_SIZE (128 * 1024) |
268 | |
269 | /* |
270 | * Internal elevator interface |
271 | */ |
272 | #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED) |
273 | |
274 | bool blk_insert_flush(struct request *rq); |
275 | |
276 | int elevator_switch(struct request_queue *q, struct elevator_type *new_e); |
277 | void elevator_disable(struct request_queue *q); |
278 | void elevator_exit(struct request_queue *q); |
279 | int elv_register_queue(struct request_queue *q, bool uevent); |
280 | void elv_unregister_queue(struct request_queue *q); |
281 | |
282 | ssize_t part_size_show(struct device *dev, struct device_attribute *attr, |
283 | char *buf); |
284 | ssize_t part_stat_show(struct device *dev, struct device_attribute *attr, |
285 | char *buf); |
286 | ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr, |
287 | char *buf); |
288 | ssize_t part_fail_show(struct device *dev, struct device_attribute *attr, |
289 | char *buf); |
290 | ssize_t part_fail_store(struct device *dev, struct device_attribute *attr, |
291 | const char *buf, size_t count); |
292 | ssize_t part_timeout_show(struct device *, struct device_attribute *, char *); |
293 | ssize_t part_timeout_store(struct device *, struct device_attribute *, |
294 | const char *, size_t); |
295 | |
296 | static inline bool bio_may_exceed_limits(struct bio *bio, |
297 | const struct queue_limits *lim) |
298 | { |
299 | switch (bio_op(bio)) { |
300 | case REQ_OP_DISCARD: |
301 | case REQ_OP_SECURE_ERASE: |
302 | case REQ_OP_WRITE_ZEROES: |
303 | return true; /* non-trivial splitting decisions */ |
304 | default: |
305 | break; |
306 | } |
307 | |
308 | /* |
309 | * All drivers must accept single-segments bios that are <= PAGE_SIZE. |
310 | * This is a quick and dirty check that relies on the fact that |
311 | * bi_io_vec[0] is always valid if a bio has data. The check might |
312 | * lead to occasional false negatives when bios are cloned, but compared |
313 | * to the performance impact of cloned bios themselves the loop below |
314 | * doesn't matter anyway. |
315 | */ |
316 | return lim->chunk_sectors || bio->bi_vcnt != 1 || |
317 | bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE; |
318 | } |
319 | |
320 | struct bio *__bio_split_to_limits(struct bio *bio, |
321 | const struct queue_limits *lim, |
322 | unsigned int *nr_segs); |
323 | int ll_back_merge_fn(struct request *req, struct bio *bio, |
324 | unsigned int nr_segs); |
325 | bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, |
326 | struct request *next); |
327 | unsigned int blk_recalc_rq_segments(struct request *rq); |
328 | void blk_rq_set_mixed_merge(struct request *rq); |
329 | bool blk_rq_merge_ok(struct request *rq, struct bio *bio); |
330 | enum elv_merge blk_try_merge(struct request *rq, struct bio *bio); |
331 | |
332 | void blk_set_default_limits(struct queue_limits *lim); |
333 | int blk_dev_init(void); |
334 | |
335 | /* |
336 | * Contribute to IO statistics IFF: |
337 | * |
338 | * a) it's attached to a gendisk, and |
339 | * b) the queue had IO stats enabled when this request was started |
340 | */ |
341 | static inline bool blk_do_io_stat(struct request *rq) |
342 | { |
343 | return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq); |
344 | } |
345 | |
346 | void update_io_ticks(struct block_device *part, unsigned long now, bool end); |
347 | |
348 | static inline void req_set_nomerge(struct request_queue *q, struct request *req) |
349 | { |
350 | req->cmd_flags |= REQ_NOMERGE; |
351 | if (req == q->last_merge) |
352 | q->last_merge = NULL; |
353 | } |
354 | |
355 | /* |
356 | * Internal io_context interface |
357 | */ |
358 | struct io_cq *ioc_find_get_icq(struct request_queue *q); |
359 | struct io_cq *ioc_lookup_icq(struct request_queue *q); |
360 | #ifdef CONFIG_BLK_ICQ |
361 | void ioc_clear_queue(struct request_queue *q); |
362 | #else |
363 | static inline void ioc_clear_queue(struct request_queue *q) |
364 | { |
365 | } |
366 | #endif /* CONFIG_BLK_ICQ */ |
367 | |
368 | #ifdef CONFIG_BLK_DEV_THROTTLING_LOW |
369 | extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page); |
370 | extern ssize_t blk_throtl_sample_time_store(struct request_queue *q, |
371 | const char *page, size_t count); |
372 | extern void blk_throtl_bio_endio(struct bio *bio); |
373 | extern void blk_throtl_stat_add(struct request *rq, u64 time); |
374 | #else |
375 | static inline void blk_throtl_bio_endio(struct bio *bio) { } |
376 | static inline void blk_throtl_stat_add(struct request *rq, u64 time) { } |
377 | #endif |
378 | |
379 | struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q); |
380 | |
381 | static inline bool blk_queue_may_bounce(struct request_queue *q) |
382 | { |
383 | return IS_ENABLED(CONFIG_BOUNCE) && |
384 | q->limits.bounce == BLK_BOUNCE_HIGH && |
385 | max_low_pfn >= max_pfn; |
386 | } |
387 | |
388 | static inline struct bio *blk_queue_bounce(struct bio *bio, |
389 | struct request_queue *q) |
390 | { |
391 | if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio))) |
392 | return __blk_queue_bounce(bio, q); |
393 | return bio; |
394 | } |
395 | |
396 | #ifdef CONFIG_BLK_DEV_ZONED |
397 | void disk_free_zone_bitmaps(struct gendisk *disk); |
398 | void disk_clear_zone_settings(struct gendisk *disk); |
399 | int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd, |
400 | unsigned long arg); |
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 void disk_clear_zone_settings(struct gendisk *disk) {} |
406 | static inline int blkdev_report_zones_ioctl(struct block_device *bdev, |
407 | unsigned int cmd, unsigned long arg) |
408 | { |
409 | return -ENOTTY; |
410 | } |
411 | static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev, |
412 | blk_mode_t mode, unsigned int cmd, unsigned long arg) |
413 | { |
414 | return -ENOTTY; |
415 | } |
416 | #endif /* CONFIG_BLK_DEV_ZONED */ |
417 | |
418 | struct block_device *bdev_alloc(struct gendisk *disk, u8 partno); |
419 | void bdev_add(struct block_device *bdev, dev_t dev); |
420 | |
421 | int blk_alloc_ext_minor(void); |
422 | void blk_free_ext_minor(unsigned int minor); |
423 | #define ADDPART_FLAG_NONE 0 |
424 | #define ADDPART_FLAG_RAID 1 |
425 | #define ADDPART_FLAG_WHOLEDISK 2 |
426 | int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, |
427 | sector_t length); |
428 | int bdev_del_partition(struct gendisk *disk, int partno); |
429 | int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, |
430 | sector_t length); |
431 | void drop_partition(struct block_device *part); |
432 | |
433 | void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors); |
434 | |
435 | struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id, |
436 | struct lock_class_key *lkclass); |
437 | |
438 | int bio_add_hw_page(struct request_queue *q, struct bio *bio, |
439 | struct page *page, unsigned int len, unsigned int offset, |
440 | unsigned int max_sectors, bool *same_page); |
441 | |
442 | /* |
443 | * Clean up a page appropriately, where the page may be pinned, may have a |
444 | * ref taken on it or neither. |
445 | */ |
446 | static inline void bio_release_page(struct bio *bio, struct page *page) |
447 | { |
448 | if (bio_flagged(bio, bit: BIO_PAGE_PINNED)) |
449 | unpin_user_page(page); |
450 | } |
451 | |
452 | struct request_queue *blk_alloc_queue(int node_id); |
453 | |
454 | int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode); |
455 | |
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; |
466 | |
467 | extern struct attribute_group blk_trace_attr_group; |
468 | |
469 | blk_mode_t file_to_blk_mode(struct file *file); |
470 | int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode, |
471 | loff_t lstart, loff_t lend); |
472 | long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); |
473 | long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); |
474 | |
475 | extern const struct address_space_operations def_blk_aops; |
476 | |
477 | int disk_register_independent_access_ranges(struct gendisk *disk); |
478 | void disk_unregister_independent_access_ranges(struct gendisk *disk); |
479 | |
480 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
481 | bool should_fail_request(struct block_device *part, unsigned int bytes); |
482 | #else /* CONFIG_FAIL_MAKE_REQUEST */ |
483 | static inline bool should_fail_request(struct block_device *part, |
484 | unsigned int bytes) |
485 | { |
486 | return false; |
487 | } |
488 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
489 | |
490 | /* |
491 | * Optimized request reference counting. Ideally we'd make timeouts be more |
492 | * clever, as that's the only reason we need references at all... But until |
493 | * this happens, this is faster than using refcount_t. Also see: |
494 | * |
495 | * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count") |
496 | */ |
497 | #define req_ref_zero_or_close_to_overflow(req) \ |
498 | ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u) |
499 | |
500 | static inline bool req_ref_inc_not_zero(struct request *req) |
501 | { |
502 | return atomic_inc_not_zero(v: &req->ref); |
503 | } |
504 | |
505 | static inline bool req_ref_put_and_test(struct request *req) |
506 | { |
507 | WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); |
508 | return atomic_dec_and_test(v: &req->ref); |
509 | } |
510 | |
511 | static inline void req_ref_set(struct request *req, int value) |
512 | { |
513 | atomic_set(v: &req->ref, i: value); |
514 | } |
515 | |
516 | static inline int req_ref_read(struct request *req) |
517 | { |
518 | return atomic_read(v: &req->ref); |
519 | } |
520 | |
521 | #endif /* BLK_INTERNAL_H */ |
522 | |