xref: /openbmc/linux/block/blk.h (revision 132db935)
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 #ifdef CONFIG_DEBUG_FS
18 extern struct dentry *blk_debugfs_root;
19 #endif
20 
21 struct blk_flush_queue {
22 	unsigned int		flush_pending_idx:1;
23 	unsigned int		flush_running_idx:1;
24 	blk_status_t 		rq_status;
25 	unsigned long		flush_pending_since;
26 	struct list_head	flush_queue[2];
27 	struct list_head	flush_data_in_flight;
28 	struct request		*flush_rq;
29 
30 	/*
31 	 * flush_rq shares tag with this rq, both can't be active
32 	 * at the same time
33 	 */
34 	struct request		*orig_rq;
35 	struct lock_class_key	key;
36 	spinlock_t		mq_flush_lock;
37 };
38 
39 extern struct kmem_cache *blk_requestq_cachep;
40 extern struct kobj_type blk_queue_ktype;
41 extern struct ida blk_queue_ida;
42 
43 static inline struct blk_flush_queue *
44 blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
45 {
46 	return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
47 }
48 
49 static inline void __blk_get_queue(struct request_queue *q)
50 {
51 	kobject_get(&q->kobj);
52 }
53 
54 static inline bool
55 is_flush_rq(struct request *req, struct blk_mq_hw_ctx *hctx)
56 {
57 	return hctx->fq->flush_rq == req;
58 }
59 
60 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
61 					      gfp_t flags);
62 void blk_free_flush_queue(struct blk_flush_queue *q);
63 
64 void blk_freeze_queue(struct request_queue *q);
65 
66 static inline bool biovec_phys_mergeable(struct request_queue *q,
67 		struct bio_vec *vec1, struct bio_vec *vec2)
68 {
69 	unsigned long mask = queue_segment_boundary(q);
70 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
71 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
72 
73 	if (addr1 + vec1->bv_len != addr2)
74 		return false;
75 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
76 		return false;
77 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
78 		return false;
79 	return true;
80 }
81 
82 static inline bool __bvec_gap_to_prev(struct request_queue *q,
83 		struct bio_vec *bprv, unsigned int offset)
84 {
85 	return (offset & queue_virt_boundary(q)) ||
86 		((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
87 }
88 
89 /*
90  * Check if adding a bio_vec after bprv with offset would create a gap in
91  * the SG list. Most drivers don't care about this, but some do.
92  */
93 static inline bool bvec_gap_to_prev(struct request_queue *q,
94 		struct bio_vec *bprv, unsigned int offset)
95 {
96 	if (!queue_virt_boundary(q))
97 		return false;
98 	return __bvec_gap_to_prev(q, bprv, offset);
99 }
100 
101 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
102 		unsigned int nr_segs)
103 {
104 	rq->nr_phys_segments = nr_segs;
105 	rq->__data_len = bio->bi_iter.bi_size;
106 	rq->bio = rq->biotail = bio;
107 	rq->ioprio = bio_prio(bio);
108 
109 	if (bio->bi_disk)
110 		rq->rq_disk = bio->bi_disk;
111 }
112 
113 #ifdef CONFIG_BLK_DEV_INTEGRITY
114 void blk_flush_integrity(void);
115 bool __bio_integrity_endio(struct bio *);
116 void bio_integrity_free(struct bio *bio);
117 static inline bool bio_integrity_endio(struct bio *bio)
118 {
119 	if (bio_integrity(bio))
120 		return __bio_integrity_endio(bio);
121 	return true;
122 }
123 
124 static inline bool integrity_req_gap_back_merge(struct request *req,
125 		struct bio *next)
126 {
127 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
128 	struct bio_integrity_payload *bip_next = bio_integrity(next);
129 
130 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
131 				bip_next->bip_vec[0].bv_offset);
132 }
133 
134 static inline bool integrity_req_gap_front_merge(struct request *req,
135 		struct bio *bio)
136 {
137 	struct bio_integrity_payload *bip = bio_integrity(bio);
138 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
139 
140 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
141 				bip_next->bip_vec[0].bv_offset);
142 }
143 
144 void blk_integrity_add(struct gendisk *);
145 void blk_integrity_del(struct gendisk *);
146 #else /* CONFIG_BLK_DEV_INTEGRITY */
147 static inline bool integrity_req_gap_back_merge(struct request *req,
148 		struct bio *next)
149 {
150 	return false;
151 }
152 static inline bool integrity_req_gap_front_merge(struct request *req,
153 		struct bio *bio)
154 {
155 	return false;
156 }
157 
158 static inline void blk_flush_integrity(void)
159 {
160 }
161 static inline bool bio_integrity_endio(struct bio *bio)
162 {
163 	return true;
164 }
165 static inline void bio_integrity_free(struct bio *bio)
166 {
167 }
168 static inline void blk_integrity_add(struct gendisk *disk)
169 {
170 }
171 static inline void blk_integrity_del(struct gendisk *disk)
172 {
173 }
174 #endif /* CONFIG_BLK_DEV_INTEGRITY */
175 
176 unsigned long blk_rq_timeout(unsigned long timeout);
177 void blk_add_timer(struct request *req);
178 
179 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
180 		unsigned int nr_segs);
181 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
182 		unsigned int nr_segs);
183 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
184 		struct bio *bio);
185 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
186 		unsigned int nr_segs, struct request **same_queue_rq);
187 
188 void blk_account_io_start(struct request *req);
189 void blk_account_io_done(struct request *req, u64 now);
190 
191 /*
192  * Internal elevator interface
193  */
194 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
195 
196 void blk_insert_flush(struct request *rq);
197 
198 void elevator_init_mq(struct request_queue *q);
199 int elevator_switch_mq(struct request_queue *q,
200 			      struct elevator_type *new_e);
201 void __elevator_exit(struct request_queue *, struct elevator_queue *);
202 int elv_register_queue(struct request_queue *q, bool uevent);
203 void elv_unregister_queue(struct request_queue *q);
204 
205 static inline void elevator_exit(struct request_queue *q,
206 		struct elevator_queue *e)
207 {
208 	lockdep_assert_held(&q->sysfs_lock);
209 
210 	blk_mq_sched_free_requests(q);
211 	__elevator_exit(q, e);
212 }
213 
214 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
215 
216 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
217 		char *buf);
218 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
219 		char *buf);
220 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
221 		char *buf);
222 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
223 		char *buf);
224 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
225 		const char *buf, size_t count);
226 
227 #ifdef CONFIG_FAIL_IO_TIMEOUT
228 int blk_should_fake_timeout(struct request_queue *);
229 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
230 ssize_t part_timeout_store(struct device *, struct device_attribute *,
231 				const char *, size_t);
232 #else
233 static inline int blk_should_fake_timeout(struct request_queue *q)
234 {
235 	return 0;
236 }
237 #endif
238 
239 void __blk_queue_split(struct request_queue *q, struct bio **bio,
240 		unsigned int *nr_segs);
241 int ll_back_merge_fn(struct request *req, struct bio *bio,
242 		unsigned int nr_segs);
243 int ll_front_merge_fn(struct request *req,  struct bio *bio,
244 		unsigned int nr_segs);
245 struct request *attempt_back_merge(struct request_queue *q, struct request *rq);
246 struct request *attempt_front_merge(struct request_queue *q, struct request *rq);
247 int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
248 				struct request *next);
249 unsigned int blk_recalc_rq_segments(struct request *rq);
250 void blk_rq_set_mixed_merge(struct request *rq);
251 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
252 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
253 
254 int blk_dev_init(void);
255 
256 /*
257  * Contribute to IO statistics IFF:
258  *
259  *	a) it's attached to a gendisk, and
260  *	b) the queue had IO stats enabled when this request was started
261  */
262 static inline bool blk_do_io_stat(struct request *rq)
263 {
264 	return rq->rq_disk && (rq->rq_flags & RQF_IO_STAT);
265 }
266 
267 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
268 {
269 	req->cmd_flags |= REQ_NOMERGE;
270 	if (req == q->last_merge)
271 		q->last_merge = NULL;
272 }
273 
274 /*
275  * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
276  * is defined as 'unsigned int', meantime it has to aligned to with logical
277  * block size which is the minimum accepted unit by hardware.
278  */
279 static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
280 {
281 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
282 }
283 
284 /*
285  * Internal io_context interface
286  */
287 void get_io_context(struct io_context *ioc);
288 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
289 struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
290 			     gfp_t gfp_mask);
291 void ioc_clear_queue(struct request_queue *q);
292 
293 int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
294 
295 /*
296  * Internal throttling interface
297  */
298 #ifdef CONFIG_BLK_DEV_THROTTLING
299 extern int blk_throtl_init(struct request_queue *q);
300 extern void blk_throtl_exit(struct request_queue *q);
301 extern void blk_throtl_register_queue(struct request_queue *q);
302 #else /* CONFIG_BLK_DEV_THROTTLING */
303 static inline int blk_throtl_init(struct request_queue *q) { return 0; }
304 static inline void blk_throtl_exit(struct request_queue *q) { }
305 static inline void blk_throtl_register_queue(struct request_queue *q) { }
306 #endif /* CONFIG_BLK_DEV_THROTTLING */
307 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
308 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
309 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
310 	const char *page, size_t count);
311 extern void blk_throtl_bio_endio(struct bio *bio);
312 extern void blk_throtl_stat_add(struct request *rq, u64 time);
313 #else
314 static inline void blk_throtl_bio_endio(struct bio *bio) { }
315 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
316 #endif
317 
318 #ifdef CONFIG_BOUNCE
319 extern int init_emergency_isa_pool(void);
320 extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
321 #else
322 static inline int init_emergency_isa_pool(void)
323 {
324 	return 0;
325 }
326 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
327 {
328 }
329 #endif /* CONFIG_BOUNCE */
330 
331 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
332 extern int blk_iolatency_init(struct request_queue *q);
333 #else
334 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
335 #endif
336 
337 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
338 
339 #ifdef CONFIG_BLK_DEV_ZONED
340 void blk_queue_free_zone_bitmaps(struct request_queue *q);
341 #else
342 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
343 #endif
344 
345 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
346 
347 int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
348 void blk_free_devt(dev_t devt);
349 void blk_invalidate_devt(dev_t devt);
350 char *disk_name(struct gendisk *hd, int partno, char *buf);
351 #define ADDPART_FLAG_NONE	0
352 #define ADDPART_FLAG_RAID	1
353 #define ADDPART_FLAG_WHOLEDISK	2
354 void delete_partition(struct gendisk *disk, struct hd_struct *part);
355 int bdev_add_partition(struct block_device *bdev, int partno,
356 		sector_t start, sector_t length);
357 int bdev_del_partition(struct block_device *bdev, int partno);
358 int bdev_resize_partition(struct block_device *bdev, int partno,
359 		sector_t start, sector_t length);
360 int disk_expand_part_tbl(struct gendisk *disk, int target);
361 int hd_ref_init(struct hd_struct *part);
362 
363 /* no need to get/put refcount of part0 */
364 static inline int hd_struct_try_get(struct hd_struct *part)
365 {
366 	if (part->partno)
367 		return percpu_ref_tryget_live(&part->ref);
368 	return 1;
369 }
370 
371 static inline void hd_struct_put(struct hd_struct *part)
372 {
373 	if (part->partno)
374 		percpu_ref_put(&part->ref);
375 }
376 
377 static inline void hd_free_part(struct hd_struct *part)
378 {
379 	free_percpu(part->dkstats);
380 	kfree(part->info);
381 	percpu_ref_exit(&part->ref);
382 }
383 
384 /*
385  * Any access of part->nr_sects which is not protected by partition
386  * bd_mutex or gendisk bdev bd_mutex, should be done using this
387  * accessor function.
388  *
389  * Code written along the lines of i_size_read() and i_size_write().
390  * CONFIG_PREEMPTION case optimizes the case of UP kernel with preemption
391  * on.
392  */
393 static inline sector_t part_nr_sects_read(struct hd_struct *part)
394 {
395 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
396 	sector_t nr_sects;
397 	unsigned seq;
398 	do {
399 		seq = read_seqcount_begin(&part->nr_sects_seq);
400 		nr_sects = part->nr_sects;
401 	} while (read_seqcount_retry(&part->nr_sects_seq, seq));
402 	return nr_sects;
403 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
404 	sector_t nr_sects;
405 
406 	preempt_disable();
407 	nr_sects = part->nr_sects;
408 	preempt_enable();
409 	return nr_sects;
410 #else
411 	return part->nr_sects;
412 #endif
413 }
414 
415 /*
416  * Should be called with mutex lock held (typically bd_mutex) of partition
417  * to provide mutual exlusion among writers otherwise seqcount might be
418  * left in wrong state leaving the readers spinning infinitely.
419  */
420 static inline void part_nr_sects_write(struct hd_struct *part, sector_t size)
421 {
422 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
423 	preempt_disable();
424 	write_seqcount_begin(&part->nr_sects_seq);
425 	part->nr_sects = size;
426 	write_seqcount_end(&part->nr_sects_seq);
427 	preempt_enable();
428 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
429 	preempt_disable();
430 	part->nr_sects = size;
431 	preempt_enable();
432 #else
433 	part->nr_sects = size;
434 #endif
435 }
436 
437 struct request_queue *__blk_alloc_queue(int node_id);
438 
439 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
440 		struct page *page, unsigned int len, unsigned int offset,
441 		unsigned int max_sectors, bool *same_page);
442 
443 #endif /* BLK_INTERNAL_H */
444