xref: /openbmc/linux/block/blk.h (revision e15a5365)
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 <xen/xen.h>
10 #include "blk-crypto-internal.h"
11 #include "blk-mq.h"
12 #include "blk-mq-sched.h"
13 
14 /* Max future timer expiry for timeouts */
15 #define BLK_MAX_TIMEOUT		(5 * HZ)
16 
17 extern struct dentry *blk_debugfs_root;
18 
19 struct blk_flush_queue {
20 	unsigned int		flush_pending_idx:1;
21 	unsigned int		flush_running_idx:1;
22 	blk_status_t 		rq_status;
23 	unsigned long		flush_pending_since;
24 	struct list_head	flush_queue[2];
25 	struct list_head	flush_data_in_flight;
26 	struct request		*flush_rq;
27 
28 	struct lock_class_key	key;
29 	spinlock_t		mq_flush_lock;
30 };
31 
32 extern struct kmem_cache *blk_requestq_cachep;
33 extern struct kobj_type blk_queue_ktype;
34 extern struct ida blk_queue_ida;
35 
36 static inline struct blk_flush_queue *
37 blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
38 {
39 	return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
40 }
41 
42 static inline void __blk_get_queue(struct request_queue *q)
43 {
44 	kobject_get(&q->kobj);
45 }
46 
47 static inline bool
48 is_flush_rq(struct request *req, struct blk_mq_hw_ctx *hctx)
49 {
50 	return hctx->fq->flush_rq == req;
51 }
52 
53 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
54 					      gfp_t flags);
55 void blk_free_flush_queue(struct blk_flush_queue *q);
56 
57 void blk_freeze_queue(struct request_queue *q);
58 
59 static inline bool biovec_phys_mergeable(struct request_queue *q,
60 		struct bio_vec *vec1, struct bio_vec *vec2)
61 {
62 	unsigned long mask = queue_segment_boundary(q);
63 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
64 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
65 
66 	if (addr1 + vec1->bv_len != addr2)
67 		return false;
68 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
69 		return false;
70 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
71 		return false;
72 	return true;
73 }
74 
75 static inline bool __bvec_gap_to_prev(struct request_queue *q,
76 		struct bio_vec *bprv, unsigned int offset)
77 {
78 	return (offset & queue_virt_boundary(q)) ||
79 		((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
80 }
81 
82 /*
83  * Check if adding a bio_vec after bprv with offset would create a gap in
84  * the SG list. Most drivers don't care about this, but some do.
85  */
86 static inline bool bvec_gap_to_prev(struct request_queue *q,
87 		struct bio_vec *bprv, unsigned int offset)
88 {
89 	if (!queue_virt_boundary(q))
90 		return false;
91 	return __bvec_gap_to_prev(q, bprv, offset);
92 }
93 
94 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
95 		unsigned int nr_segs)
96 {
97 	rq->nr_phys_segments = nr_segs;
98 	rq->__data_len = bio->bi_iter.bi_size;
99 	rq->bio = rq->biotail = bio;
100 	rq->ioprio = bio_prio(bio);
101 
102 	if (bio->bi_disk)
103 		rq->rq_disk = bio->bi_disk;
104 }
105 
106 #ifdef CONFIG_BLK_DEV_INTEGRITY
107 void blk_flush_integrity(void);
108 bool __bio_integrity_endio(struct bio *);
109 void bio_integrity_free(struct bio *bio);
110 static inline bool bio_integrity_endio(struct bio *bio)
111 {
112 	if (bio_integrity(bio))
113 		return __bio_integrity_endio(bio);
114 	return true;
115 }
116 
117 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
118 		struct request *);
119 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
120 		struct bio *);
121 
122 static inline bool integrity_req_gap_back_merge(struct request *req,
123 		struct bio *next)
124 {
125 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
126 	struct bio_integrity_payload *bip_next = bio_integrity(next);
127 
128 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
129 				bip_next->bip_vec[0].bv_offset);
130 }
131 
132 static inline bool integrity_req_gap_front_merge(struct request *req,
133 		struct bio *bio)
134 {
135 	struct bio_integrity_payload *bip = bio_integrity(bio);
136 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
137 
138 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
139 				bip_next->bip_vec[0].bv_offset);
140 }
141 
142 void blk_integrity_add(struct gendisk *);
143 void blk_integrity_del(struct gendisk *);
144 #else /* CONFIG_BLK_DEV_INTEGRITY */
145 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
146 		struct request *r1, struct request *r2)
147 {
148 	return true;
149 }
150 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
151 		struct request *r, struct bio *b)
152 {
153 	return true;
154 }
155 static inline bool integrity_req_gap_back_merge(struct request *req,
156 		struct bio *next)
157 {
158 	return false;
159 }
160 static inline bool integrity_req_gap_front_merge(struct request *req,
161 		struct bio *bio)
162 {
163 	return false;
164 }
165 
166 static inline void blk_flush_integrity(void)
167 {
168 }
169 static inline bool bio_integrity_endio(struct bio *bio)
170 {
171 	return true;
172 }
173 static inline void bio_integrity_free(struct bio *bio)
174 {
175 }
176 static inline void blk_integrity_add(struct gendisk *disk)
177 {
178 }
179 static inline void blk_integrity_del(struct gendisk *disk)
180 {
181 }
182 #endif /* CONFIG_BLK_DEV_INTEGRITY */
183 
184 unsigned long blk_rq_timeout(unsigned long timeout);
185 void blk_add_timer(struct request *req);
186 
187 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
188 		unsigned int nr_segs, struct request **same_queue_rq);
189 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
190 			struct bio *bio, unsigned int nr_segs);
191 
192 void blk_account_io_start(struct request *req);
193 void blk_account_io_done(struct request *req, u64 now);
194 
195 /*
196  * Internal elevator interface
197  */
198 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
199 
200 void blk_insert_flush(struct request *rq);
201 
202 void elevator_init_mq(struct request_queue *q);
203 int elevator_switch_mq(struct request_queue *q,
204 			      struct elevator_type *new_e);
205 void __elevator_exit(struct request_queue *, struct elevator_queue *);
206 int elv_register_queue(struct request_queue *q, bool uevent);
207 void elv_unregister_queue(struct request_queue *q);
208 
209 static inline void elevator_exit(struct request_queue *q,
210 		struct elevator_queue *e)
211 {
212 	lockdep_assert_held(&q->sysfs_lock);
213 
214 	blk_mq_sched_free_requests(q);
215 	__elevator_exit(q, e);
216 }
217 
218 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
219 
220 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
221 		char *buf);
222 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
223 		char *buf);
224 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
225 		char *buf);
226 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
227 		char *buf);
228 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
229 		const char *buf, size_t count);
230 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
231 ssize_t part_timeout_store(struct device *, struct device_attribute *,
232 				const char *, size_t);
233 
234 void __blk_queue_split(struct bio **bio, unsigned int *nr_segs);
235 int ll_back_merge_fn(struct request *req, struct bio *bio,
236 		unsigned int nr_segs);
237 int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
238 				struct request *next);
239 unsigned int blk_recalc_rq_segments(struct request *rq);
240 void blk_rq_set_mixed_merge(struct request *rq);
241 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
242 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
243 
244 int blk_dev_init(void);
245 
246 /*
247  * Contribute to IO statistics IFF:
248  *
249  *	a) it's attached to a gendisk, and
250  *	b) the queue had IO stats enabled when this request was started
251  */
252 static inline bool blk_do_io_stat(struct request *rq)
253 {
254 	return rq->rq_disk && (rq->rq_flags & RQF_IO_STAT);
255 }
256 
257 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
258 {
259 	req->cmd_flags |= REQ_NOMERGE;
260 	if (req == q->last_merge)
261 		q->last_merge = NULL;
262 }
263 
264 /*
265  * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
266  * is defined as 'unsigned int', meantime it has to aligned to with logical
267  * block size which is the minimum accepted unit by hardware.
268  */
269 static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
270 {
271 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
272 }
273 
274 /*
275  * The max bio size which is aligned to q->limits.discard_granularity. This
276  * is a hint to split large discard bio in generic block layer, then if device
277  * driver needs to split the discard bio into smaller ones, their bi_size can
278  * be very probably and easily aligned to discard_granularity of the device's
279  * queue.
280  */
281 static inline unsigned int bio_aligned_discard_max_sectors(
282 					struct request_queue *q)
283 {
284 	return round_down(UINT_MAX, q->limits.discard_granularity) >>
285 			SECTOR_SHIFT;
286 }
287 
288 /*
289  * Internal io_context interface
290  */
291 void get_io_context(struct io_context *ioc);
292 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
293 struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
294 			     gfp_t gfp_mask);
295 void ioc_clear_queue(struct request_queue *q);
296 
297 int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
298 
299 /*
300  * Internal throttling interface
301  */
302 #ifdef CONFIG_BLK_DEV_THROTTLING
303 extern int blk_throtl_init(struct request_queue *q);
304 extern void blk_throtl_exit(struct request_queue *q);
305 extern void blk_throtl_register_queue(struct request_queue *q);
306 bool blk_throtl_bio(struct bio *bio);
307 #else /* CONFIG_BLK_DEV_THROTTLING */
308 static inline int blk_throtl_init(struct request_queue *q) { return 0; }
309 static inline void blk_throtl_exit(struct request_queue *q) { }
310 static inline void blk_throtl_register_queue(struct request_queue *q) { }
311 static inline bool blk_throtl_bio(struct bio *bio) { return false; }
312 #endif /* CONFIG_BLK_DEV_THROTTLING */
313 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
314 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
315 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
316 	const char *page, size_t count);
317 extern void blk_throtl_bio_endio(struct bio *bio);
318 extern void blk_throtl_stat_add(struct request *rq, u64 time);
319 #else
320 static inline void blk_throtl_bio_endio(struct bio *bio) { }
321 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
322 #endif
323 
324 #ifdef CONFIG_BOUNCE
325 extern int init_emergency_isa_pool(void);
326 extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
327 #else
328 static inline int init_emergency_isa_pool(void)
329 {
330 	return 0;
331 }
332 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
333 {
334 }
335 #endif /* CONFIG_BOUNCE */
336 
337 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
338 extern int blk_iolatency_init(struct request_queue *q);
339 #else
340 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
341 #endif
342 
343 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
344 
345 #ifdef CONFIG_BLK_DEV_ZONED
346 void blk_queue_free_zone_bitmaps(struct request_queue *q);
347 #else
348 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
349 #endif
350 
351 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
352 
353 int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
354 void blk_free_devt(dev_t devt);
355 void blk_invalidate_devt(dev_t devt);
356 char *disk_name(struct gendisk *hd, int partno, char *buf);
357 #define ADDPART_FLAG_NONE	0
358 #define ADDPART_FLAG_RAID	1
359 #define ADDPART_FLAG_WHOLEDISK	2
360 void delete_partition(struct hd_struct *part);
361 int bdev_add_partition(struct block_device *bdev, int partno,
362 		sector_t start, sector_t length);
363 int bdev_del_partition(struct block_device *bdev, int partno);
364 int bdev_resize_partition(struct block_device *bdev, int partno,
365 		sector_t start, sector_t length);
366 int disk_expand_part_tbl(struct gendisk *disk, int target);
367 int hd_ref_init(struct hd_struct *part);
368 
369 /* no need to get/put refcount of part0 */
370 static inline int hd_struct_try_get(struct hd_struct *part)
371 {
372 	if (part->partno)
373 		return percpu_ref_tryget_live(&part->ref);
374 	return 1;
375 }
376 
377 static inline void hd_struct_put(struct hd_struct *part)
378 {
379 	if (part->partno)
380 		percpu_ref_put(&part->ref);
381 }
382 
383 static inline void hd_free_part(struct hd_struct *part)
384 {
385 	free_percpu(part->dkstats);
386 	kfree(part->info);
387 	percpu_ref_exit(&part->ref);
388 }
389 
390 /*
391  * Any access of part->nr_sects which is not protected by partition
392  * bd_mutex or gendisk bdev bd_mutex, should be done using this
393  * accessor function.
394  *
395  * Code written along the lines of i_size_read() and i_size_write().
396  * CONFIG_PREEMPTION case optimizes the case of UP kernel with preemption
397  * on.
398  */
399 static inline sector_t part_nr_sects_read(struct hd_struct *part)
400 {
401 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
402 	sector_t nr_sects;
403 	unsigned seq;
404 	do {
405 		seq = read_seqcount_begin(&part->nr_sects_seq);
406 		nr_sects = part->nr_sects;
407 	} while (read_seqcount_retry(&part->nr_sects_seq, seq));
408 	return nr_sects;
409 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
410 	sector_t nr_sects;
411 
412 	preempt_disable();
413 	nr_sects = part->nr_sects;
414 	preempt_enable();
415 	return nr_sects;
416 #else
417 	return part->nr_sects;
418 #endif
419 }
420 
421 /*
422  * Should be called with mutex lock held (typically bd_mutex) of partition
423  * to provide mutual exlusion among writers otherwise seqcount might be
424  * left in wrong state leaving the readers spinning infinitely.
425  */
426 static inline void part_nr_sects_write(struct hd_struct *part, sector_t size)
427 {
428 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
429 	preempt_disable();
430 	write_seqcount_begin(&part->nr_sects_seq);
431 	part->nr_sects = size;
432 	write_seqcount_end(&part->nr_sects_seq);
433 	preempt_enable();
434 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
435 	preempt_disable();
436 	part->nr_sects = size;
437 	preempt_enable();
438 #else
439 	part->nr_sects = size;
440 #endif
441 }
442 
443 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
444 		struct page *page, unsigned int len, unsigned int offset,
445 		unsigned int max_sectors, bool *same_page);
446 
447 #endif /* BLK_INTERNAL_H */
448