xref: /openbmc/linux/block/blk.h (revision c6fddb28)
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 <xen/xen.h>
9 #include "blk-mq.h"
10 #include "blk-mq-sched.h"
11 
12 /* Max future timer expiry for timeouts */
13 #define BLK_MAX_TIMEOUT		(5 * HZ)
14 
15 #ifdef CONFIG_DEBUG_FS
16 extern struct dentry *blk_debugfs_root;
17 #endif
18 
19 struct blk_flush_queue {
20 	unsigned int		flush_queue_delayed:1;
21 	unsigned int		flush_pending_idx:1;
22 	unsigned int		flush_running_idx:1;
23 	blk_status_t 		rq_status;
24 	unsigned long		flush_pending_since;
25 	struct list_head	flush_queue[2];
26 	struct list_head	flush_data_in_flight;
27 	struct request		*flush_rq;
28 
29 	/*
30 	 * flush_rq shares tag with this rq, both can't be active
31 	 * at the same time
32 	 */
33 	struct request		*orig_rq;
34 	struct lock_class_key	key;
35 	spinlock_t		mq_flush_lock;
36 };
37 
38 extern struct kmem_cache *blk_requestq_cachep;
39 extern struct kobj_type blk_queue_ktype;
40 extern struct ida blk_queue_ida;
41 
42 static inline struct blk_flush_queue *
43 blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
44 {
45 	return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
46 }
47 
48 static inline void __blk_get_queue(struct request_queue *q)
49 {
50 	kobject_get(&q->kobj);
51 }
52 
53 static inline bool
54 is_flush_rq(struct request *req, struct blk_mq_hw_ctx *hctx)
55 {
56 	return hctx->fq->flush_rq == req;
57 }
58 
59 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
60 					      gfp_t flags);
61 void blk_free_flush_queue(struct blk_flush_queue *q);
62 
63 void blk_freeze_queue(struct request_queue *q);
64 
65 static inline void blk_queue_enter_live(struct request_queue *q)
66 {
67 	/*
68 	 * Given that running in generic_make_request() context
69 	 * guarantees that a live reference against q_usage_counter has
70 	 * been established, further references under that same context
71 	 * need not check that the queue has been frozen (marked dead).
72 	 */
73 	percpu_ref_get(&q->q_usage_counter);
74 }
75 
76 static inline bool biovec_phys_mergeable(struct request_queue *q,
77 		struct bio_vec *vec1, struct bio_vec *vec2)
78 {
79 	unsigned long mask = queue_segment_boundary(q);
80 	phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
81 	phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
82 
83 	if (addr1 + vec1->bv_len != addr2)
84 		return false;
85 	if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
86 		return false;
87 	if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
88 		return false;
89 	return true;
90 }
91 
92 static inline bool __bvec_gap_to_prev(struct request_queue *q,
93 		struct bio_vec *bprv, unsigned int offset)
94 {
95 	return (offset & queue_virt_boundary(q)) ||
96 		((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
97 }
98 
99 /*
100  * Check if adding a bio_vec after bprv with offset would create a gap in
101  * the SG list. Most drivers don't care about this, but some do.
102  */
103 static inline bool bvec_gap_to_prev(struct request_queue *q,
104 		struct bio_vec *bprv, unsigned int offset)
105 {
106 	if (!queue_virt_boundary(q))
107 		return false;
108 	return __bvec_gap_to_prev(q, bprv, offset);
109 }
110 
111 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
112 		unsigned int nr_segs)
113 {
114 	rq->nr_phys_segments = nr_segs;
115 	rq->__data_len = bio->bi_iter.bi_size;
116 	rq->bio = rq->biotail = bio;
117 	rq->ioprio = bio_prio(bio);
118 
119 	if (bio->bi_disk)
120 		rq->rq_disk = bio->bi_disk;
121 }
122 
123 #ifdef CONFIG_BLK_DEV_INTEGRITY
124 void blk_flush_integrity(void);
125 bool __bio_integrity_endio(struct bio *);
126 void bio_integrity_free(struct bio *bio);
127 static inline bool bio_integrity_endio(struct bio *bio)
128 {
129 	if (bio_integrity(bio))
130 		return __bio_integrity_endio(bio);
131 	return true;
132 }
133 
134 static inline bool integrity_req_gap_back_merge(struct request *req,
135 		struct bio *next)
136 {
137 	struct bio_integrity_payload *bip = bio_integrity(req->bio);
138 	struct bio_integrity_payload *bip_next = bio_integrity(next);
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 static inline bool integrity_req_gap_front_merge(struct request *req,
145 		struct bio *bio)
146 {
147 	struct bio_integrity_payload *bip = bio_integrity(bio);
148 	struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
149 
150 	return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
151 				bip_next->bip_vec[0].bv_offset);
152 }
153 
154 void blk_integrity_add(struct gendisk *);
155 void blk_integrity_del(struct gendisk *);
156 #else /* CONFIG_BLK_DEV_INTEGRITY */
157 static inline bool integrity_req_gap_back_merge(struct request *req,
158 		struct bio *next)
159 {
160 	return false;
161 }
162 static inline bool integrity_req_gap_front_merge(struct request *req,
163 		struct bio *bio)
164 {
165 	return false;
166 }
167 
168 static inline void blk_flush_integrity(void)
169 {
170 }
171 static inline bool bio_integrity_endio(struct bio *bio)
172 {
173 	return true;
174 }
175 static inline void bio_integrity_free(struct bio *bio)
176 {
177 }
178 static inline void blk_integrity_add(struct gendisk *disk)
179 {
180 }
181 static inline void blk_integrity_del(struct gendisk *disk)
182 {
183 }
184 #endif /* CONFIG_BLK_DEV_INTEGRITY */
185 
186 unsigned long blk_rq_timeout(unsigned long timeout);
187 void blk_add_timer(struct request *req);
188 
189 bool bio_attempt_front_merge(struct request *req, struct bio *bio,
190 		unsigned int nr_segs);
191 bool bio_attempt_back_merge(struct request *req, struct bio *bio,
192 		unsigned int nr_segs);
193 bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
194 		struct bio *bio);
195 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
196 		unsigned int nr_segs, struct request **same_queue_rq);
197 
198 void blk_account_io_start(struct request *req, bool new_io);
199 void blk_account_io_completion(struct request *req, unsigned int bytes);
200 void blk_account_io_done(struct request *req, u64 now);
201 
202 /*
203  * Internal elevator interface
204  */
205 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
206 
207 void blk_insert_flush(struct request *rq);
208 
209 void elevator_init_mq(struct request_queue *q);
210 int elevator_switch_mq(struct request_queue *q,
211 			      struct elevator_type *new_e);
212 void __elevator_exit(struct request_queue *, struct elevator_queue *);
213 int elv_register_queue(struct request_queue *q, bool uevent);
214 void elv_unregister_queue(struct request_queue *q);
215 
216 static inline void elevator_exit(struct request_queue *q,
217 		struct elevator_queue *e)
218 {
219 	lockdep_assert_held(&q->sysfs_lock);
220 
221 	blk_mq_sched_free_requests(q);
222 	__elevator_exit(q, e);
223 }
224 
225 struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
226 
227 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
228 		char *buf);
229 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
230 		char *buf);
231 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
232 		char *buf);
233 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
234 		char *buf);
235 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
236 		const char *buf, size_t count);
237 
238 #ifdef CONFIG_FAIL_IO_TIMEOUT
239 int blk_should_fake_timeout(struct request_queue *);
240 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
241 ssize_t part_timeout_store(struct device *, struct device_attribute *,
242 				const char *, size_t);
243 #else
244 static inline int blk_should_fake_timeout(struct request_queue *q)
245 {
246 	return 0;
247 }
248 #endif
249 
250 void __blk_queue_split(struct request_queue *q, struct bio **bio,
251 		unsigned int *nr_segs);
252 int ll_back_merge_fn(struct request *req, struct bio *bio,
253 		unsigned int nr_segs);
254 int ll_front_merge_fn(struct request *req,  struct bio *bio,
255 		unsigned int nr_segs);
256 struct request *attempt_back_merge(struct request_queue *q, struct request *rq);
257 struct request *attempt_front_merge(struct request_queue *q, struct request *rq);
258 int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
259 				struct request *next);
260 unsigned int blk_recalc_rq_segments(struct request *rq);
261 void blk_rq_set_mixed_merge(struct request *rq);
262 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
263 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
264 
265 int blk_dev_init(void);
266 
267 /*
268  * Contribute to IO statistics IFF:
269  *
270  *	a) it's attached to a gendisk, and
271  *	b) the queue had IO stats enabled when this request was started
272  */
273 static inline bool blk_do_io_stat(struct request *rq)
274 {
275 	return rq->rq_disk && (rq->rq_flags & RQF_IO_STAT);
276 }
277 
278 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
279 {
280 	req->cmd_flags |= REQ_NOMERGE;
281 	if (req == q->last_merge)
282 		q->last_merge = NULL;
283 }
284 
285 /*
286  * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
287  * is defined as 'unsigned int', meantime it has to aligned to with logical
288  * block size which is the minimum accepted unit by hardware.
289  */
290 static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
291 {
292 	return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
293 }
294 
295 /*
296  * Internal io_context interface
297  */
298 void get_io_context(struct io_context *ioc);
299 struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
300 struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
301 			     gfp_t gfp_mask);
302 void ioc_clear_queue(struct request_queue *q);
303 
304 int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
305 
306 /**
307  * create_io_context - try to create task->io_context
308  * @gfp_mask: allocation mask
309  * @node: allocation node
310  *
311  * If %current->io_context is %NULL, allocate a new io_context and install
312  * it.  Returns the current %current->io_context which may be %NULL if
313  * allocation failed.
314  *
315  * Note that this function can't be called with IRQ disabled because
316  * task_lock which protects %current->io_context is IRQ-unsafe.
317  */
318 static inline struct io_context *create_io_context(gfp_t gfp_mask, int node)
319 {
320 	WARN_ON_ONCE(irqs_disabled());
321 	if (unlikely(!current->io_context))
322 		create_task_io_context(current, gfp_mask, node);
323 	return current->io_context;
324 }
325 
326 /*
327  * Internal throttling interface
328  */
329 #ifdef CONFIG_BLK_DEV_THROTTLING
330 extern void blk_throtl_drain(struct request_queue *q);
331 extern int blk_throtl_init(struct request_queue *q);
332 extern void blk_throtl_exit(struct request_queue *q);
333 extern void blk_throtl_register_queue(struct request_queue *q);
334 #else /* CONFIG_BLK_DEV_THROTTLING */
335 static inline void blk_throtl_drain(struct request_queue *q) { }
336 static inline int blk_throtl_init(struct request_queue *q) { return 0; }
337 static inline void blk_throtl_exit(struct request_queue *q) { }
338 static inline void blk_throtl_register_queue(struct request_queue *q) { }
339 #endif /* CONFIG_BLK_DEV_THROTTLING */
340 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
341 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
342 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
343 	const char *page, size_t count);
344 extern void blk_throtl_bio_endio(struct bio *bio);
345 extern void blk_throtl_stat_add(struct request *rq, u64 time);
346 #else
347 static inline void blk_throtl_bio_endio(struct bio *bio) { }
348 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
349 #endif
350 
351 #ifdef CONFIG_BOUNCE
352 extern int init_emergency_isa_pool(void);
353 extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
354 #else
355 static inline int init_emergency_isa_pool(void)
356 {
357 	return 0;
358 }
359 static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
360 {
361 }
362 #endif /* CONFIG_BOUNCE */
363 
364 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
365 extern int blk_iolatency_init(struct request_queue *q);
366 #else
367 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
368 #endif
369 
370 struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
371 
372 #ifdef CONFIG_BLK_DEV_ZONED
373 void blk_queue_free_zone_bitmaps(struct request_queue *q);
374 #else
375 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
376 #endif
377 
378 void part_dec_in_flight(struct request_queue *q, struct hd_struct *part,
379 			int rw);
380 void part_inc_in_flight(struct request_queue *q, struct hd_struct *part,
381 			int rw);
382 void update_io_ticks(struct hd_struct *part, unsigned long now, bool end);
383 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
384 
385 int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
386 void blk_free_devt(dev_t devt);
387 void blk_invalidate_devt(dev_t devt);
388 char *disk_name(struct gendisk *hd, int partno, char *buf);
389 #define ADDPART_FLAG_NONE	0
390 #define ADDPART_FLAG_RAID	1
391 #define ADDPART_FLAG_WHOLEDISK	2
392 struct hd_struct *__must_check add_partition(struct gendisk *disk, int partno,
393 		sector_t start, sector_t len, int flags,
394 		struct partition_meta_info *info);
395 void __delete_partition(struct percpu_ref *ref);
396 void delete_partition(struct gendisk *disk, int partno);
397 int disk_expand_part_tbl(struct gendisk *disk, int target);
398 
399 static inline int hd_ref_init(struct hd_struct *part)
400 {
401 	if (percpu_ref_init(&part->ref, __delete_partition, 0,
402 				GFP_KERNEL))
403 		return -ENOMEM;
404 	return 0;
405 }
406 
407 static inline void hd_struct_get(struct hd_struct *part)
408 {
409 	percpu_ref_get(&part->ref);
410 }
411 
412 static inline int hd_struct_try_get(struct hd_struct *part)
413 {
414 	return percpu_ref_tryget_live(&part->ref);
415 }
416 
417 static inline void hd_struct_put(struct hd_struct *part)
418 {
419 	percpu_ref_put(&part->ref);
420 }
421 
422 static inline void hd_struct_kill(struct hd_struct *part)
423 {
424 	percpu_ref_kill(&part->ref);
425 }
426 
427 static inline void hd_free_part(struct hd_struct *part)
428 {
429 	free_part_stats(part);
430 	kfree(part->info);
431 	percpu_ref_exit(&part->ref);
432 }
433 
434 /*
435  * Any access of part->nr_sects which is not protected by partition
436  * bd_mutex or gendisk bdev bd_mutex, should be done using this
437  * accessor function.
438  *
439  * Code written along the lines of i_size_read() and i_size_write().
440  * CONFIG_PREEMPTION case optimizes the case of UP kernel with preemption
441  * on.
442  */
443 static inline sector_t part_nr_sects_read(struct hd_struct *part)
444 {
445 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
446 	sector_t nr_sects;
447 	unsigned seq;
448 	do {
449 		seq = read_seqcount_begin(&part->nr_sects_seq);
450 		nr_sects = part->nr_sects;
451 	} while (read_seqcount_retry(&part->nr_sects_seq, seq));
452 	return nr_sects;
453 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
454 	sector_t nr_sects;
455 
456 	preempt_disable();
457 	nr_sects = part->nr_sects;
458 	preempt_enable();
459 	return nr_sects;
460 #else
461 	return part->nr_sects;
462 #endif
463 }
464 
465 /*
466  * Should be called with mutex lock held (typically bd_mutex) of partition
467  * to provide mutual exlusion among writers otherwise seqcount might be
468  * left in wrong state leaving the readers spinning infinitely.
469  */
470 static inline void part_nr_sects_write(struct hd_struct *part, sector_t size)
471 {
472 #if BITS_PER_LONG==32 && defined(CONFIG_SMP)
473 	write_seqcount_begin(&part->nr_sects_seq);
474 	part->nr_sects = size;
475 	write_seqcount_end(&part->nr_sects_seq);
476 #elif BITS_PER_LONG==32 && defined(CONFIG_PREEMPTION)
477 	preempt_disable();
478 	part->nr_sects = size;
479 	preempt_enable();
480 #else
481 	part->nr_sects = size;
482 #endif
483 }
484 
485 struct request_queue *__blk_alloc_queue(int node_id);
486 
487 int __bio_add_pc_page(struct request_queue *q, struct bio *bio,
488 		struct page *page, unsigned int len, unsigned int offset,
489 		bool *same_page);
490 
491 #endif /* BLK_INTERNAL_H */
492