xref: /openbmc/linux/block/blk.h (revision 8e74a48d)
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 	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 bool submit_bio_checks(struct bio *bio);
50 
51 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
52 {
53 	rcu_read_lock();
54 	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 	if (blk_queue_pm_only(q) &&
62 	    (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
63 		goto fail_put;
64 
65 	rcu_read_unlock();
66 	return true;
67 
68 fail_put:
69 	blk_queue_exit(q);
70 fail:
71 	rcu_read_unlock();
72 	return false;
73 }
74 
75 static inline int bio_queue_enter(struct bio *bio)
76 {
77 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
78 
79 	if (blk_try_enter_queue(q, false))
80 		return 0;
81 	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 static inline bool biovec_phys_mergeable(struct request_queue *q,
90 		struct bio_vec *vec1, struct bio_vec *vec2)
91 {
92 	unsigned long mask = queue_segment_boundary(q);
93 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
94 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
95 
96 	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 	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(struct request_queue *q,
106 		struct bio_vec *bprv, unsigned int offset)
107 {
108 	return (offset & queue_virt_boundary(q)) ||
109 		((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 	if (!queue_virt_boundary(q))
120 		return false;
121 	return __bvec_gap_to_prev(q, 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(rq) == REQ_OP_FLUSH)
130 		return false;
131 
132 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
133 		return false;
134 
135 	if (req_op(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(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 static inline bool blk_may_split(struct request_queue *q, struct bio *bio)
284 {
285 	switch (bio_op(bio)) {
286 	case REQ_OP_DISCARD:
287 	case REQ_OP_SECURE_ERASE:
288 	case REQ_OP_WRITE_ZEROES:
289 	case REQ_OP_WRITE_SAME:
290 		return true; /* non-trivial splitting decisions */
291 	default:
292 		break;
293 	}
294 
295 	/*
296 	 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
297 	 * This is a quick and dirty check that relies on the fact that
298 	 * bi_io_vec[0] is always valid if a bio has data.  The check might
299 	 * lead to occasional false negatives when bios are cloned, but compared
300 	 * to the performance impact of cloned bios themselves the loop below
301 	 * doesn't matter anyway.
302 	 */
303 	return q->limits.chunk_sectors || bio->bi_vcnt != 1 ||
304 		bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
305 }
306 
307 void __blk_queue_split(struct request_queue *q, struct bio **bio,
308 			unsigned int *nr_segs);
309 int ll_back_merge_fn(struct request *req, struct bio *bio,
310 		unsigned int nr_segs);
311 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
312 				struct request *next);
313 unsigned int blk_recalc_rq_segments(struct request *rq);
314 void blk_rq_set_mixed_merge(struct request *rq);
315 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
316 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
317 
318 int blk_dev_init(void);
319 
320 /*
321  * Contribute to IO statistics IFF:
322  *
323  *	a) it's attached to a gendisk, and
324  *	b) the queue had IO stats enabled when this request was started
325  */
326 static inline bool blk_do_io_stat(struct request *rq)
327 {
328 	return (rq->rq_flags & RQF_IO_STAT) && rq->q->disk;
329 }
330 
331 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
332 
333 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
334 {
335 	req->cmd_flags |= REQ_NOMERGE;
336 	if (req == q->last_merge)
337 		q->last_merge = NULL;
338 }
339 
340 /*
341  * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
342  * is defined as 'unsigned int', meantime it has to aligned to with logical
343  * block size which is the minimum accepted unit by hardware.
344  */
345 static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
346 {
347 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
348 }
349 
350 /*
351  * The max bio size which is aligned to q->limits.discard_granularity. This
352  * is a hint to split large discard bio in generic block layer, then if device
353  * driver needs to split the discard bio into smaller ones, their bi_size can
354  * be very probably and easily aligned to discard_granularity of the device's
355  * queue.
356  */
357 static inline unsigned int bio_aligned_discard_max_sectors(
358 					struct request_queue *q)
359 {
360 	return round_down(UINT_MAX, q->limits.discard_granularity) >>
361 			SECTOR_SHIFT;
362 }
363 
364 /*
365  * Internal io_context interface
366  */
367 struct io_cq *ioc_find_get_icq(struct request_queue *q);
368 struct io_cq *ioc_lookup_icq(struct request_queue *q);
369 #ifdef CONFIG_BLK_ICQ
370 void ioc_clear_queue(struct request_queue *q);
371 #else
372 static inline void ioc_clear_queue(struct request_queue *q)
373 {
374 }
375 #endif /* CONFIG_BLK_ICQ */
376 
377 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
378 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
379 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
380 	const char *page, size_t count);
381 extern void blk_throtl_bio_endio(struct bio *bio);
382 extern void blk_throtl_stat_add(struct request *rq, u64 time);
383 #else
384 static inline void blk_throtl_bio_endio(struct bio *bio) { }
385 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
386 #endif
387 
388 void __blk_queue_bounce(struct request_queue *q, struct bio **bio);
389 
390 static inline bool blk_queue_may_bounce(struct request_queue *q)
391 {
392 	return IS_ENABLED(CONFIG_BOUNCE) &&
393 		q->limits.bounce == BLK_BOUNCE_HIGH &&
394 		max_low_pfn >= max_pfn;
395 }
396 
397 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
398 {
399 	if (unlikely(blk_queue_may_bounce(q) && bio_has_data(*bio)))
400 		__blk_queue_bounce(q, bio);
401 }
402 
403 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
404 extern int blk_iolatency_init(struct request_queue *q);
405 #else
406 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
407 #endif
408 
409 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
410 
411 #ifdef CONFIG_BLK_DEV_ZONED
412 void blk_queue_free_zone_bitmaps(struct request_queue *q);
413 void blk_queue_clear_zone_settings(struct request_queue *q);
414 #else
415 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
416 static inline void blk_queue_clear_zone_settings(struct request_queue *q) {}
417 #endif
418 
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,
425 		sector_t length);
426 int bdev_del_partition(struct gendisk *disk, int partno);
427 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
428 		sector_t length);
429 
430 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
431 		struct page *page, unsigned int len, unsigned int offset,
432 		unsigned int max_sectors, bool *same_page);
433 
434 static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
435 {
436 	if (srcu)
437 		return blk_requestq_srcu_cachep;
438 	return blk_requestq_cachep;
439 }
440 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
441 
442 int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
443 
444 int disk_alloc_events(struct gendisk *disk);
445 void disk_add_events(struct gendisk *disk);
446 void disk_del_events(struct gendisk *disk);
447 void disk_release_events(struct gendisk *disk);
448 extern struct device_attribute dev_attr_events;
449 extern struct device_attribute dev_attr_events_async;
450 extern struct device_attribute dev_attr_events_poll_msecs;
451 
452 static inline void bio_clear_polled(struct bio *bio)
453 {
454 	/* can't support alloc cache if we turn off polling */
455 	bio_clear_flag(bio, BIO_PERCPU_CACHE);
456 	bio->bi_opf &= ~REQ_POLLED;
457 }
458 
459 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
460 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
461 
462 extern const struct address_space_operations def_blk_aops;
463 
464 int disk_register_independent_access_ranges(struct gendisk *disk,
465 				struct blk_independent_access_ranges *new_iars);
466 void disk_unregister_independent_access_ranges(struct gendisk *disk);
467 
468 #ifdef CONFIG_FAIL_MAKE_REQUEST
469 bool should_fail_request(struct block_device *part, unsigned int bytes);
470 #else /* CONFIG_FAIL_MAKE_REQUEST */
471 static inline bool should_fail_request(struct block_device *part,
472 					unsigned int bytes)
473 {
474 	return false;
475 }
476 #endif /* CONFIG_FAIL_MAKE_REQUEST */
477 
478 /*
479  * Optimized request reference counting. Ideally we'd make timeouts be more
480  * clever, as that's the only reason we need references at all... But until
481  * this happens, this is faster than using refcount_t. Also see:
482  *
483  * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
484  */
485 #define req_ref_zero_or_close_to_overflow(req)	\
486 	((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
487 
488 static inline bool req_ref_inc_not_zero(struct request *req)
489 {
490 	return atomic_inc_not_zero(&req->ref);
491 }
492 
493 static inline bool req_ref_put_and_test(struct request *req)
494 {
495 	WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
496 	return atomic_dec_and_test(&req->ref);
497 }
498 
499 static inline void req_ref_set(struct request *req, int value)
500 {
501 	atomic_set(&req->ref, value);
502 }
503 
504 static inline int req_ref_read(struct request *req)
505 {
506 	return atomic_read(&req->ref);
507 }
508 
509 #endif /* BLK_INTERNAL_H */
510