Loading...
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 <xen/xen.h>
8#include "blk-mq.h"
9#include "blk-mq-sched.h"
10
11/* Max future timer expiry for timeouts */
12#define BLK_MAX_TIMEOUT (5 * HZ)
13
14#ifdef CONFIG_DEBUG_FS
15extern struct dentry *blk_debugfs_root;
16#endif
17
18struct blk_flush_queue {
19 unsigned int flush_queue_delayed:1;
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 /*
29 * flush_rq shares tag with this rq, both can't be active
30 * at the same time
31 */
32 struct request *orig_rq;
33 spinlock_t mq_flush_lock;
34};
35
36extern struct kmem_cache *blk_requestq_cachep;
37extern struct kobj_type blk_queue_ktype;
38extern struct ida blk_queue_ida;
39
40static inline struct blk_flush_queue *
41blk_get_flush_queue(struct request_queue *q, struct blk_mq_ctx *ctx)
42{
43 return blk_mq_map_queue(q, REQ_OP_FLUSH, ctx)->fq;
44}
45
46static inline void __blk_get_queue(struct request_queue *q)
47{
48 kobject_get(&q->kobj);
49}
50
51static inline bool
52is_flush_rq(struct request *req, struct blk_mq_hw_ctx *hctx)
53{
54 return hctx->fq->flush_rq == req;
55}
56
57struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q,
58 int node, int cmd_size, gfp_t flags);
59void blk_free_flush_queue(struct blk_flush_queue *q);
60
61void blk_freeze_queue(struct request_queue *q);
62
63static inline void blk_queue_enter_live(struct request_queue *q)
64{
65 /*
66 * Given that running in generic_make_request() context
67 * guarantees that a live reference against q_usage_counter has
68 * been established, further references under that same context
69 * need not check that the queue has been frozen (marked dead).
70 */
71 percpu_ref_get(&q->q_usage_counter);
72}
73
74static inline bool biovec_phys_mergeable(struct request_queue *q,
75 struct bio_vec *vec1, struct bio_vec *vec2)
76{
77 unsigned long mask = queue_segment_boundary(q);
78 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
79 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
80
81 if (addr1 + vec1->bv_len != addr2)
82 return false;
83 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
84 return false;
85 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
86 return false;
87 return true;
88}
89
90static inline bool __bvec_gap_to_prev(struct request_queue *q,
91 struct bio_vec *bprv, unsigned int offset)
92{
93 return (offset & queue_virt_boundary(q)) ||
94 ((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
95}
96
97/*
98 * Check if adding a bio_vec after bprv with offset would create a gap in
99 * the SG list. Most drivers don't care about this, but some do.
100 */
101static inline bool bvec_gap_to_prev(struct request_queue *q,
102 struct bio_vec *bprv, unsigned int offset)
103{
104 if (!queue_virt_boundary(q))
105 return false;
106 return __bvec_gap_to_prev(q, bprv, offset);
107}
108
109static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
110 unsigned int nr_segs)
111{
112 rq->nr_phys_segments = nr_segs;
113 rq->__data_len = bio->bi_iter.bi_size;
114 rq->bio = rq->biotail = bio;
115 rq->ioprio = bio_prio(bio);
116
117 if (bio->bi_disk)
118 rq->rq_disk = bio->bi_disk;
119}
120
121#ifdef CONFIG_BLK_DEV_INTEGRITY
122void blk_flush_integrity(void);
123bool __bio_integrity_endio(struct bio *);
124static inline bool bio_integrity_endio(struct bio *bio)
125{
126 if (bio_integrity(bio))
127 return __bio_integrity_endio(bio);
128 return true;
129}
130
131static inline bool integrity_req_gap_back_merge(struct request *req,
132 struct bio *next)
133{
134 struct bio_integrity_payload *bip = bio_integrity(req->bio);
135 struct bio_integrity_payload *bip_next = bio_integrity(next);
136
137 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
138 bip_next->bip_vec[0].bv_offset);
139}
140
141static inline bool integrity_req_gap_front_merge(struct request *req,
142 struct bio *bio)
143{
144 struct bio_integrity_payload *bip = bio_integrity(bio);
145 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
146
147 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
148 bip_next->bip_vec[0].bv_offset);
149}
150#else /* CONFIG_BLK_DEV_INTEGRITY */
151static inline bool integrity_req_gap_back_merge(struct request *req,
152 struct bio *next)
153{
154 return false;
155}
156static inline bool integrity_req_gap_front_merge(struct request *req,
157 struct bio *bio)
158{
159 return false;
160}
161
162static inline void blk_flush_integrity(void)
163{
164}
165static inline bool bio_integrity_endio(struct bio *bio)
166{
167 return true;
168}
169#endif /* CONFIG_BLK_DEV_INTEGRITY */
170
171unsigned long blk_rq_timeout(unsigned long timeout);
172void blk_add_timer(struct request *req);
173
174bool bio_attempt_front_merge(struct request *req, struct bio *bio,
175 unsigned int nr_segs);
176bool bio_attempt_back_merge(struct request *req, struct bio *bio,
177 unsigned int nr_segs);
178bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
179 struct bio *bio);
180bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
181 unsigned int nr_segs, struct request **same_queue_rq);
182
183void blk_account_io_start(struct request *req, bool new_io);
184void blk_account_io_completion(struct request *req, unsigned int bytes);
185void blk_account_io_done(struct request *req, u64 now);
186
187/*
188 * Internal elevator interface
189 */
190#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
191
192void blk_insert_flush(struct request *rq);
193
194void elevator_init_mq(struct request_queue *q);
195int elevator_switch_mq(struct request_queue *q,
196 struct elevator_type *new_e);
197void __elevator_exit(struct request_queue *, struct elevator_queue *);
198int elv_register_queue(struct request_queue *q, bool uevent);
199void elv_unregister_queue(struct request_queue *q);
200
201static inline void elevator_exit(struct request_queue *q,
202 struct elevator_queue *e)
203{
204 lockdep_assert_held(&q->sysfs_lock);
205
206 blk_mq_sched_free_requests(q);
207 __elevator_exit(q, e);
208}
209
210struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
211
212#ifdef CONFIG_FAIL_IO_TIMEOUT
213int blk_should_fake_timeout(struct request_queue *);
214ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
215ssize_t part_timeout_store(struct device *, struct device_attribute *,
216 const char *, size_t);
217#else
218static inline int blk_should_fake_timeout(struct request_queue *q)
219{
220 return 0;
221}
222#endif
223
224void __blk_queue_split(struct request_queue *q, struct bio **bio,
225 unsigned int *nr_segs);
226int ll_back_merge_fn(struct request *req, struct bio *bio,
227 unsigned int nr_segs);
228int ll_front_merge_fn(struct request *req, struct bio *bio,
229 unsigned int nr_segs);
230struct request *attempt_back_merge(struct request_queue *q, struct request *rq);
231struct request *attempt_front_merge(struct request_queue *q, struct request *rq);
232int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
233 struct request *next);
234unsigned int blk_recalc_rq_segments(struct request *rq);
235void blk_rq_set_mixed_merge(struct request *rq);
236bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
237enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
238
239int blk_dev_init(void);
240
241/*
242 * Contribute to IO statistics IFF:
243 *
244 * a) it's attached to a gendisk, and
245 * b) the queue had IO stats enabled when this request was started, and
246 * c) it's a file system request
247 */
248static inline bool blk_do_io_stat(struct request *rq)
249{
250 return rq->rq_disk &&
251 (rq->rq_flags & RQF_IO_STAT) &&
252 !blk_rq_is_passthrough(rq);
253}
254
255static inline void req_set_nomerge(struct request_queue *q, struct request *req)
256{
257 req->cmd_flags |= REQ_NOMERGE;
258 if (req == q->last_merge)
259 q->last_merge = NULL;
260}
261
262/*
263 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
264 * is defined as 'unsigned int', meantime it has to aligned to with logical
265 * block size which is the minimum accepted unit by hardware.
266 */
267static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
268{
269 return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
270}
271
272/*
273 * Internal io_context interface
274 */
275void get_io_context(struct io_context *ioc);
276struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
277struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
278 gfp_t gfp_mask);
279void ioc_clear_queue(struct request_queue *q);
280
281int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
282
283/**
284 * create_io_context - try to create task->io_context
285 * @gfp_mask: allocation mask
286 * @node: allocation node
287 *
288 * If %current->io_context is %NULL, allocate a new io_context and install
289 * it. Returns the current %current->io_context which may be %NULL if
290 * allocation failed.
291 *
292 * Note that this function can't be called with IRQ disabled because
293 * task_lock which protects %current->io_context is IRQ-unsafe.
294 */
295static inline struct io_context *create_io_context(gfp_t gfp_mask, int node)
296{
297 WARN_ON_ONCE(irqs_disabled());
298 if (unlikely(!current->io_context))
299 create_task_io_context(current, gfp_mask, node);
300 return current->io_context;
301}
302
303/*
304 * Internal throttling interface
305 */
306#ifdef CONFIG_BLK_DEV_THROTTLING
307extern void blk_throtl_drain(struct request_queue *q);
308extern int blk_throtl_init(struct request_queue *q);
309extern void blk_throtl_exit(struct request_queue *q);
310extern void blk_throtl_register_queue(struct request_queue *q);
311#else /* CONFIG_BLK_DEV_THROTTLING */
312static inline void blk_throtl_drain(struct request_queue *q) { }
313static inline int blk_throtl_init(struct request_queue *q) { return 0; }
314static inline void blk_throtl_exit(struct request_queue *q) { }
315static inline void blk_throtl_register_queue(struct request_queue *q) { }
316#endif /* CONFIG_BLK_DEV_THROTTLING */
317#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
318extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
319extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
320 const char *page, size_t count);
321extern void blk_throtl_bio_endio(struct bio *bio);
322extern void blk_throtl_stat_add(struct request *rq, u64 time);
323#else
324static inline void blk_throtl_bio_endio(struct bio *bio) { }
325static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
326#endif
327
328#ifdef CONFIG_BOUNCE
329extern int init_emergency_isa_pool(void);
330extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
331#else
332static inline int init_emergency_isa_pool(void)
333{
334 return 0;
335}
336static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
337{
338}
339#endif /* CONFIG_BOUNCE */
340
341#ifdef CONFIG_BLK_CGROUP_IOLATENCY
342extern int blk_iolatency_init(struct request_queue *q);
343#else
344static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
345#endif
346
347struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
348
349#ifdef CONFIG_BLK_DEV_ZONED
350void blk_queue_free_zone_bitmaps(struct request_queue *q);
351#else
352static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
353#endif
354
355#endif /* BLK_INTERNAL_H */
1#ifndef BLK_INTERNAL_H
2#define BLK_INTERNAL_H
3
4#include <linux/idr.h>
5#include <linux/blk-mq.h>
6#include "blk-mq.h"
7
8/* Amount of time in which a process may batch requests */
9#define BLK_BATCH_TIME (HZ/50UL)
10
11/* Number of requests a "batching" process may submit */
12#define BLK_BATCH_REQ 32
13
14/* Max future timer expiry for timeouts */
15#define BLK_MAX_TIMEOUT (5 * HZ)
16
17struct blk_flush_queue {
18 unsigned int flush_queue_delayed:1;
19 unsigned int flush_pending_idx:1;
20 unsigned int flush_running_idx:1;
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 /*
27 * flush_rq shares tag with this rq, both can't be active
28 * at the same time
29 */
30 struct request *orig_rq;
31 spinlock_t mq_flush_lock;
32};
33
34extern struct kmem_cache *blk_requestq_cachep;
35extern struct kmem_cache *request_cachep;
36extern struct kobj_type blk_queue_ktype;
37extern struct ida blk_queue_ida;
38
39static inline struct blk_flush_queue *blk_get_flush_queue(
40 struct request_queue *q, struct blk_mq_ctx *ctx)
41{
42 struct blk_mq_hw_ctx *hctx;
43
44 if (!q->mq_ops)
45 return q->fq;
46
47 hctx = q->mq_ops->map_queue(q, ctx->cpu);
48
49 return hctx->fq;
50}
51
52static inline void __blk_get_queue(struct request_queue *q)
53{
54 kobject_get(&q->kobj);
55}
56
57struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q,
58 int node, int cmd_size);
59void blk_free_flush_queue(struct blk_flush_queue *q);
60
61int blk_init_rl(struct request_list *rl, struct request_queue *q,
62 gfp_t gfp_mask);
63void blk_exit_rl(struct request_list *rl);
64void init_request_from_bio(struct request *req, struct bio *bio);
65void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
66 struct bio *bio);
67int blk_rq_append_bio(struct request_queue *q, struct request *rq,
68 struct bio *bio);
69void blk_queue_bypass_start(struct request_queue *q);
70void blk_queue_bypass_end(struct request_queue *q);
71void blk_dequeue_request(struct request *rq);
72void __blk_queue_free_tags(struct request_queue *q);
73bool __blk_end_bidi_request(struct request *rq, int error,
74 unsigned int nr_bytes, unsigned int bidi_bytes);
75void blk_freeze_queue(struct request_queue *q);
76
77static inline void blk_queue_enter_live(struct request_queue *q)
78{
79 /*
80 * Given that running in generic_make_request() context
81 * guarantees that a live reference against q_usage_counter has
82 * been established, further references under that same context
83 * need not check that the queue has been frozen (marked dead).
84 */
85 percpu_ref_get(&q->q_usage_counter);
86}
87
88#ifdef CONFIG_BLK_DEV_INTEGRITY
89void blk_flush_integrity(void);
90#else
91static inline void blk_flush_integrity(void)
92{
93}
94#endif
95
96void blk_timeout_work(struct work_struct *work);
97unsigned long blk_rq_timeout(unsigned long timeout);
98void blk_add_timer(struct request *req);
99void blk_delete_timer(struct request *);
100
101
102bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
103 struct bio *bio);
104bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
105 struct bio *bio);
106bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
107 unsigned int *request_count,
108 struct request **same_queue_rq);
109unsigned int blk_plug_queued_count(struct request_queue *q);
110
111void blk_account_io_start(struct request *req, bool new_io);
112void blk_account_io_completion(struct request *req, unsigned int bytes);
113void blk_account_io_done(struct request *req);
114
115/*
116 * Internal atomic flags for request handling
117 */
118enum rq_atomic_flags {
119 REQ_ATOM_COMPLETE = 0,
120 REQ_ATOM_STARTED,
121};
122
123/*
124 * EH timer and IO completion will both attempt to 'grab' the request, make
125 * sure that only one of them succeeds
126 */
127static inline int blk_mark_rq_complete(struct request *rq)
128{
129 return test_and_set_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
130}
131
132static inline void blk_clear_rq_complete(struct request *rq)
133{
134 clear_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
135}
136
137/*
138 * Internal elevator interface
139 */
140#define ELV_ON_HASH(rq) ((rq)->cmd_flags & REQ_HASHED)
141
142void blk_insert_flush(struct request *rq);
143
144static inline struct request *__elv_next_request(struct request_queue *q)
145{
146 struct request *rq;
147 struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
148
149 while (1) {
150 if (!list_empty(&q->queue_head)) {
151 rq = list_entry_rq(q->queue_head.next);
152 return rq;
153 }
154
155 /*
156 * Flush request is running and flush request isn't queueable
157 * in the drive, we can hold the queue till flush request is
158 * finished. Even we don't do this, driver can't dispatch next
159 * requests and will requeue them. And this can improve
160 * throughput too. For example, we have request flush1, write1,
161 * flush 2. flush1 is dispatched, then queue is hold, write1
162 * isn't inserted to queue. After flush1 is finished, flush2
163 * will be dispatched. Since disk cache is already clean,
164 * flush2 will be finished very soon, so looks like flush2 is
165 * folded to flush1.
166 * Since the queue is hold, a flag is set to indicate the queue
167 * should be restarted later. Please see flush_end_io() for
168 * details.
169 */
170 if (fq->flush_pending_idx != fq->flush_running_idx &&
171 !queue_flush_queueable(q)) {
172 fq->flush_queue_delayed = 1;
173 return NULL;
174 }
175 if (unlikely(blk_queue_bypass(q)) ||
176 !q->elevator->type->ops.elevator_dispatch_fn(q, 0))
177 return NULL;
178 }
179}
180
181static inline void elv_activate_rq(struct request_queue *q, struct request *rq)
182{
183 struct elevator_queue *e = q->elevator;
184
185 if (e->type->ops.elevator_activate_req_fn)
186 e->type->ops.elevator_activate_req_fn(q, rq);
187}
188
189static inline void elv_deactivate_rq(struct request_queue *q, struct request *rq)
190{
191 struct elevator_queue *e = q->elevator;
192
193 if (e->type->ops.elevator_deactivate_req_fn)
194 e->type->ops.elevator_deactivate_req_fn(q, rq);
195}
196
197#ifdef CONFIG_FAIL_IO_TIMEOUT
198int blk_should_fake_timeout(struct request_queue *);
199ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
200ssize_t part_timeout_store(struct device *, struct device_attribute *,
201 const char *, size_t);
202#else
203static inline int blk_should_fake_timeout(struct request_queue *q)
204{
205 return 0;
206}
207#endif
208
209int ll_back_merge_fn(struct request_queue *q, struct request *req,
210 struct bio *bio);
211int ll_front_merge_fn(struct request_queue *q, struct request *req,
212 struct bio *bio);
213int attempt_back_merge(struct request_queue *q, struct request *rq);
214int attempt_front_merge(struct request_queue *q, struct request *rq);
215int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
216 struct request *next);
217void blk_recalc_rq_segments(struct request *rq);
218void blk_rq_set_mixed_merge(struct request *rq);
219bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
220int blk_try_merge(struct request *rq, struct bio *bio);
221
222void blk_queue_congestion_threshold(struct request_queue *q);
223
224int blk_dev_init(void);
225
226
227/*
228 * Return the threshold (number of used requests) at which the queue is
229 * considered to be congested. It include a little hysteresis to keep the
230 * context switch rate down.
231 */
232static inline int queue_congestion_on_threshold(struct request_queue *q)
233{
234 return q->nr_congestion_on;
235}
236
237/*
238 * The threshold at which a queue is considered to be uncongested
239 */
240static inline int queue_congestion_off_threshold(struct request_queue *q)
241{
242 return q->nr_congestion_off;
243}
244
245extern int blk_update_nr_requests(struct request_queue *, unsigned int);
246
247/*
248 * Contribute to IO statistics IFF:
249 *
250 * a) it's attached to a gendisk, and
251 * b) the queue had IO stats enabled when this request was started, and
252 * c) it's a file system request
253 */
254static inline int blk_do_io_stat(struct request *rq)
255{
256 return rq->rq_disk &&
257 (rq->cmd_flags & REQ_IO_STAT) &&
258 (rq->cmd_type == REQ_TYPE_FS);
259}
260
261/*
262 * Internal io_context interface
263 */
264void get_io_context(struct io_context *ioc);
265struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
266struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
267 gfp_t gfp_mask);
268void ioc_clear_queue(struct request_queue *q);
269
270int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
271
272/**
273 * create_io_context - try to create task->io_context
274 * @gfp_mask: allocation mask
275 * @node: allocation node
276 *
277 * If %current->io_context is %NULL, allocate a new io_context and install
278 * it. Returns the current %current->io_context which may be %NULL if
279 * allocation failed.
280 *
281 * Note that this function can't be called with IRQ disabled because
282 * task_lock which protects %current->io_context is IRQ-unsafe.
283 */
284static inline struct io_context *create_io_context(gfp_t gfp_mask, int node)
285{
286 WARN_ON_ONCE(irqs_disabled());
287 if (unlikely(!current->io_context))
288 create_task_io_context(current, gfp_mask, node);
289 return current->io_context;
290}
291
292/*
293 * Internal throttling interface
294 */
295#ifdef CONFIG_BLK_DEV_THROTTLING
296extern void blk_throtl_drain(struct request_queue *q);
297extern int blk_throtl_init(struct request_queue *q);
298extern void blk_throtl_exit(struct request_queue *q);
299#else /* CONFIG_BLK_DEV_THROTTLING */
300static inline void blk_throtl_drain(struct request_queue *q) { }
301static inline int blk_throtl_init(struct request_queue *q) { return 0; }
302static inline void blk_throtl_exit(struct request_queue *q) { }
303#endif /* CONFIG_BLK_DEV_THROTTLING */
304
305#endif /* BLK_INTERNAL_H */