xref: /openbmc/linux/block/blk.h (revision 827beb77)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef BLK_INTERNAL_H
3 #define BLK_INTERNAL_H
4 
5 #include <linux/idr.h>
6 #include <linux/blk-mq.h>
7 #include <linux/part_stat.h>
8 #include <linux/blk-crypto.h>
9 #include <linux/memblock.h>	/* for max_pfn/max_low_pfn */
10 #include <xen/xen.h>
11 #include "blk-crypto-internal.h"
12 #include "blk-mq.h"
13 #include "blk-mq-sched.h"
14 
15 struct elevator_type;
16 
17 /* Max future timer expiry for timeouts */
18 #define BLK_MAX_TIMEOUT		(5 * HZ)
19 
20 extern struct dentry *blk_debugfs_root;
21 
22 struct blk_flush_queue {
23 	unsigned int		flush_pending_idx:1;
24 	unsigned int		flush_running_idx:1;
25 	blk_status_t 		rq_status;
26 	unsigned long		flush_pending_since;
27 	struct list_head	flush_queue[2];
28 	struct list_head	flush_data_in_flight;
29 	struct request		*flush_rq;
30 
31 	spinlock_t		mq_flush_lock;
32 };
33 
34 extern struct kmem_cache *blk_requestq_cachep;
35 extern struct kobj_type blk_queue_ktype;
36 extern struct ida blk_queue_ida;
37 
38 static inline struct blk_flush_queue *
39 blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
40 {
41 	return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
42 }
43 
44 static inline void __blk_get_queue(struct request_queue *q)
45 {
46 	kobject_get(&q->kobj);
47 }
48 
49 bool is_flush_rq(struct request *req);
50 
51 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
52 					      gfp_t flags);
53 void blk_free_flush_queue(struct blk_flush_queue *q);
54 
55 void blk_freeze_queue(struct request_queue *q);
56 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
57 void blk_queue_start_drain(struct request_queue *q);
58 
59 #define BIO_INLINE_VECS 4
60 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
61 		gfp_t gfp_mask);
62 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
63 
64 static inline bool biovec_phys_mergeable(struct request_queue *q,
65 		struct bio_vec *vec1, struct bio_vec *vec2)
66 {
67 	unsigned long mask = queue_segment_boundary(q);
68 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
69 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
70 
71 	if (addr1 + vec1->bv_len != addr2)
72 		return false;
73 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
74 		return false;
75 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
76 		return false;
77 	return true;
78 }
79 
80 static inline bool __bvec_gap_to_prev(struct request_queue *q,
81 		struct bio_vec *bprv, unsigned int offset)
82 {
83 	return (offset & queue_virt_boundary(q)) ||
84 		((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
85 }
86 
87 /*
88  * Check if adding a bio_vec after bprv with offset would create a gap in
89  * the SG list. Most drivers don't care about this, but some do.
90  */
91 static inline bool bvec_gap_to_prev(struct request_queue *q,
92 		struct bio_vec *bprv, unsigned int offset)
93 {
94 	if (!queue_virt_boundary(q))
95 		return false;
96 	return __bvec_gap_to_prev(q, bprv, offset);
97 }
98 
99 static inline bool rq_mergeable(struct request *rq)
100 {
101 	if (blk_rq_is_passthrough(rq))
102 		return false;
103 
104 	if (req_op(rq) == REQ_OP_FLUSH)
105 		return false;
106 
107 	if (req_op(rq) == REQ_OP_WRITE_ZEROES)
108 		return false;
109 
110 	if (req_op(rq) == REQ_OP_ZONE_APPEND)
111 		return false;
112 
113 	if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
114 		return false;
115 	if (rq->rq_flags & RQF_NOMERGE_FLAGS)
116 		return false;
117 
118 	return true;
119 }
120 
121 /*
122  * There are two different ways to handle DISCARD merges:
123  *  1) If max_discard_segments > 1, the driver treats every bio as a range and
124  *     send the bios to controller together. The ranges don't need to be
125  *     contiguous.
126  *  2) Otherwise, the request will be normal read/write requests.  The ranges
127  *     need to be contiguous.
128  */
129 static inline bool blk_discard_mergable(struct request *req)
130 {
131 	if (req_op(req) == REQ_OP_DISCARD &&
132 	    queue_max_discard_segments(req->q) > 1)
133 		return true;
134 	return false;
135 }
136 
137 #ifdef CONFIG_BLK_DEV_INTEGRITY
138 void blk_flush_integrity(void);
139 bool __bio_integrity_endio(struct bio *);
140 void bio_integrity_free(struct bio *bio);
141 static inline bool bio_integrity_endio(struct bio *bio)
142 {
143 	if (bio_integrity(bio))
144 		return __bio_integrity_endio(bio);
145 	return true;
146 }
147 
148 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
149 		struct request *);
150 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
151 		struct bio *);
152 
153 static inline bool integrity_req_gap_back_merge(struct request *req,
154 		struct bio *next)
155 {
156 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
157 	struct bio_integrity_payload *bip_next = bio_integrity(next);
158 
159 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
160 				bip_next->bip_vec[0].bv_offset);
161 }
162 
163 static inline bool integrity_req_gap_front_merge(struct request *req,
164 		struct bio *bio)
165 {
166 	struct bio_integrity_payload *bip = bio_integrity(bio);
167 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
168 
169 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
170 				bip_next->bip_vec[0].bv_offset);
171 }
172 
173 int blk_integrity_add(struct gendisk *disk);
174 void blk_integrity_del(struct gendisk *);
175 #else /* CONFIG_BLK_DEV_INTEGRITY */
176 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
177 		struct request *r1, struct request *r2)
178 {
179 	return true;
180 }
181 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
182 		struct request *r, struct bio *b)
183 {
184 	return true;
185 }
186 static inline bool integrity_req_gap_back_merge(struct request *req,
187 		struct bio *next)
188 {
189 	return false;
190 }
191 static inline bool integrity_req_gap_front_merge(struct request *req,
192 		struct bio *bio)
193 {
194 	return false;
195 }
196 
197 static inline void blk_flush_integrity(void)
198 {
199 }
200 static inline bool bio_integrity_endio(struct bio *bio)
201 {
202 	return true;
203 }
204 static inline void bio_integrity_free(struct bio *bio)
205 {
206 }
207 static inline int blk_integrity_add(struct gendisk *disk)
208 {
209 	return 0;
210 }
211 static inline void blk_integrity_del(struct gendisk *disk)
212 {
213 }
214 #endif /* CONFIG_BLK_DEV_INTEGRITY */
215 
216 unsigned long blk_rq_timeout(unsigned long timeout);
217 void blk_add_timer(struct request *req);
218 void blk_print_req_error(struct request *req, blk_status_t status);
219 
220 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
221 		unsigned int nr_segs, bool *same_queue_rq);
222 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
223 			struct bio *bio, unsigned int nr_segs);
224 
225 void __blk_account_io_start(struct request *req);
226 void __blk_account_io_done(struct request *req, u64 now);
227 
228 /*
229  * Plug flush limits
230  */
231 #define BLK_MAX_REQUEST_COUNT	32
232 #define BLK_PLUG_FLUSH_SIZE	(128 * 1024)
233 
234 /*
235  * Internal elevator interface
236  */
237 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
238 
239 bool blk_insert_flush(struct request *rq);
240 
241 int elevator_switch_mq(struct request_queue *q,
242 			      struct elevator_type *new_e);
243 void __elevator_exit(struct request_queue *, struct elevator_queue *);
244 int elv_register_queue(struct request_queue *q, bool uevent);
245 void elv_unregister_queue(struct request_queue *q);
246 
247 static inline void elevator_exit(struct request_queue *q,
248 		struct elevator_queue *e)
249 {
250 	lockdep_assert_held(&q->sysfs_lock);
251 
252 	blk_mq_sched_free_rqs(q);
253 	__elevator_exit(q, e);
254 }
255 
256 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
257 		char *buf);
258 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
259 		char *buf);
260 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
261 		char *buf);
262 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
263 		char *buf);
264 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
265 		const char *buf, size_t count);
266 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
267 ssize_t part_timeout_store(struct device *, struct device_attribute *,
268 				const char *, size_t);
269 
270 static inline bool blk_may_split(struct request_queue *q, struct bio *bio)
271 {
272 	switch (bio_op(bio)) {
273 	case REQ_OP_DISCARD:
274 	case REQ_OP_SECURE_ERASE:
275 	case REQ_OP_WRITE_ZEROES:
276 	case REQ_OP_WRITE_SAME:
277 		return true; /* non-trivial splitting decisions */
278 	default:
279 		break;
280 	}
281 
282 	/*
283 	 * All drivers must accept single-segments bios that are <= PAGE_SIZE.
284 	 * This is a quick and dirty check that relies on the fact that
285 	 * bi_io_vec[0] is always valid if a bio has data.  The check might
286 	 * lead to occasional false negatives when bios are cloned, but compared
287 	 * to the performance impact of cloned bios themselves the loop below
288 	 * doesn't matter anyway.
289 	 */
290 	return q->limits.chunk_sectors || bio->bi_vcnt != 1 ||
291 		bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
292 }
293 
294 void __blk_queue_split(struct request_queue *q, struct bio **bio,
295 			unsigned int *nr_segs);
296 int ll_back_merge_fn(struct request *req, struct bio *bio,
297 		unsigned int nr_segs);
298 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
299 				struct request *next);
300 unsigned int blk_recalc_rq_segments(struct request *rq);
301 void blk_rq_set_mixed_merge(struct request *rq);
302 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
303 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
304 
305 int blk_dev_init(void);
306 
307 /*
308  * Contribute to IO statistics IFF:
309  *
310  *	a) it's attached to a gendisk, and
311  *	b) the queue had IO stats enabled when this request was started
312  */
313 static inline bool blk_do_io_stat(struct request *rq)
314 {
315 	return (rq->rq_flags & RQF_IO_STAT) && rq->rq_disk;
316 }
317 
318 static inline void blk_account_io_done(struct request *req, u64 now)
319 {
320 	/*
321 	 * Account IO completion.  flush_rq isn't accounted as a
322 	 * normal IO on queueing nor completion.  Accounting the
323 	 * containing request is enough.
324 	 */
325 	if (blk_do_io_stat(req) && req->part &&
326 	    !(req->rq_flags & RQF_FLUSH_SEQ))
327 		__blk_account_io_done(req, now);
328 }
329 
330 static inline void blk_account_io_start(struct request *req)
331 {
332 	if (blk_do_io_stat(req))
333 		__blk_account_io_start(req);
334 }
335 
336 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
337 {
338 	req->cmd_flags |= REQ_NOMERGE;
339 	if (req == q->last_merge)
340 		q->last_merge = NULL;
341 }
342 
343 /*
344  * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
345  * is defined as 'unsigned int', meantime it has to aligned to with logical
346  * block size which is the minimum accepted unit by hardware.
347  */
348 static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
349 {
350 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
351 }
352 
353 /*
354  * The max bio size which is aligned to q->limits.discard_granularity. This
355  * is a hint to split large discard bio in generic block layer, then if device
356  * driver needs to split the discard bio into smaller ones, their bi_size can
357  * be very probably and easily aligned to discard_granularity of the device's
358  * queue.
359  */
360 static inline unsigned int bio_aligned_discard_max_sectors(
361 					struct request_queue *q)
362 {
363 	return round_down(UINT_MAX, q->limits.discard_granularity) >>
364 			SECTOR_SHIFT;
365 }
366 
367 /*
368  * Internal io_context interface
369  */
370 void get_io_context(struct io_context *ioc);
371 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
372 struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
373 			     gfp_t gfp_mask);
374 void ioc_clear_queue(struct request_queue *q);
375 
376 int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
377 
378 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
379 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
380 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
381 	const char *page, size_t count);
382 extern void blk_throtl_bio_endio(struct bio *bio);
383 extern void blk_throtl_stat_add(struct request *rq, u64 time);
384 #else
385 static inline void blk_throtl_bio_endio(struct bio *bio) { }
386 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
387 #endif
388 
389 void __blk_queue_bounce(struct request_queue *q, struct bio **bio);
390 
391 static inline bool blk_queue_may_bounce(struct request_queue *q)
392 {
393 	return IS_ENABLED(CONFIG_BOUNCE) &&
394 		q->limits.bounce == BLK_BOUNCE_HIGH &&
395 		max_low_pfn >= max_pfn;
396 }
397 
398 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
399 {
400 	if (unlikely(blk_queue_may_bounce(q) && bio_has_data(*bio)))
401 		__blk_queue_bounce(q, bio);
402 }
403 
404 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
405 extern int blk_iolatency_init(struct request_queue *q);
406 #else
407 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
408 #endif
409 
410 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
411 
412 #ifdef CONFIG_BLK_DEV_ZONED
413 void blk_queue_free_zone_bitmaps(struct request_queue *q);
414 void blk_queue_clear_zone_settings(struct request_queue *q);
415 #else
416 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
417 static inline void blk_queue_clear_zone_settings(struct request_queue *q) {}
418 #endif
419 
420 int blk_alloc_ext_minor(void);
421 void blk_free_ext_minor(unsigned int minor);
422 #define ADDPART_FLAG_NONE	0
423 #define ADDPART_FLAG_RAID	1
424 #define ADDPART_FLAG_WHOLEDISK	2
425 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
426 		sector_t length);
427 int bdev_del_partition(struct gendisk *disk, int partno);
428 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
429 		sector_t length);
430 
431 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
432 		struct page *page, unsigned int len, unsigned int offset,
433 		unsigned int max_sectors, bool *same_page);
434 
435 struct request_queue *blk_alloc_queue(int node_id);
436 
437 int disk_alloc_events(struct gendisk *disk);
438 void disk_add_events(struct gendisk *disk);
439 void disk_del_events(struct gendisk *disk);
440 void disk_release_events(struct gendisk *disk);
441 extern struct device_attribute dev_attr_events;
442 extern struct device_attribute dev_attr_events_async;
443 extern struct device_attribute dev_attr_events_poll_msecs;
444 
445 static inline void bio_clear_polled(struct bio *bio)
446 {
447 	/* can't support alloc cache if we turn off polling */
448 	bio_clear_flag(bio, BIO_PERCPU_CACHE);
449 	bio->bi_opf &= ~REQ_POLLED;
450 }
451 
452 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
453 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
454 
455 extern const struct address_space_operations def_blk_aops;
456 
457 int disk_register_independent_access_ranges(struct gendisk *disk,
458 				struct blk_independent_access_ranges *new_iars);
459 void disk_unregister_independent_access_ranges(struct gendisk *disk);
460 
461 #endif /* BLK_INTERNAL_H */
462