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 "blk-mq.h"
8
9/* Amount of time in which a process may batch requests */
10#define BLK_BATCH_TIME (HZ/50UL)
11
12/* Number of requests a "batching" process may submit */
13#define BLK_BATCH_REQ 32
14
15/* Max future timer expiry for timeouts */
16#define BLK_MAX_TIMEOUT (5 * HZ)
17
18#ifdef CONFIG_DEBUG_FS
19extern struct dentry *blk_debugfs_root;
20#endif
21
22struct blk_flush_queue {
23 unsigned int flush_queue_delayed:1;
24 unsigned int flush_pending_idx:1;
25 unsigned int flush_running_idx:1;
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 /*
32 * flush_rq shares tag with this rq, both can't be active
33 * at the same time
34 */
35 struct request *orig_rq;
36 spinlock_t mq_flush_lock;
37};
38
39extern struct kmem_cache *blk_requestq_cachep;
40extern struct kmem_cache *request_cachep;
41extern struct kobj_type blk_queue_ktype;
42extern struct ida blk_queue_ida;
43
44/*
45 * @q->queue_lock is set while a queue is being initialized. Since we know
46 * that no other threads access the queue object before @q->queue_lock has
47 * been set, it is safe to manipulate queue flags without holding the
48 * queue_lock if @q->queue_lock == NULL. See also blk_alloc_queue_node() and
49 * blk_init_allocated_queue().
50 */
51static inline void queue_lockdep_assert_held(struct request_queue *q)
52{
53 if (q->queue_lock)
54 lockdep_assert_held(q->queue_lock);
55}
56
57static inline void queue_flag_set_unlocked(unsigned int flag,
58 struct request_queue *q)
59{
60 if (test_bit(QUEUE_FLAG_INIT_DONE, &q->queue_flags) &&
61 kref_read(&q->kobj.kref))
62 lockdep_assert_held(q->queue_lock);
63 __set_bit(flag, &q->queue_flags);
64}
65
66static inline void queue_flag_clear_unlocked(unsigned int flag,
67 struct request_queue *q)
68{
69 if (test_bit(QUEUE_FLAG_INIT_DONE, &q->queue_flags) &&
70 kref_read(&q->kobj.kref))
71 lockdep_assert_held(q->queue_lock);
72 __clear_bit(flag, &q->queue_flags);
73}
74
75static inline int queue_flag_test_and_clear(unsigned int flag,
76 struct request_queue *q)
77{
78 queue_lockdep_assert_held(q);
79
80 if (test_bit(flag, &q->queue_flags)) {
81 __clear_bit(flag, &q->queue_flags);
82 return 1;
83 }
84
85 return 0;
86}
87
88static inline int queue_flag_test_and_set(unsigned int flag,
89 struct request_queue *q)
90{
91 queue_lockdep_assert_held(q);
92
93 if (!test_bit(flag, &q->queue_flags)) {
94 __set_bit(flag, &q->queue_flags);
95 return 0;
96 }
97
98 return 1;
99}
100
101static inline void queue_flag_set(unsigned int flag, struct request_queue *q)
102{
103 queue_lockdep_assert_held(q);
104 __set_bit(flag, &q->queue_flags);
105}
106
107static inline void queue_flag_clear(unsigned int flag, struct request_queue *q)
108{
109 queue_lockdep_assert_held(q);
110 __clear_bit(flag, &q->queue_flags);
111}
112
113static inline struct blk_flush_queue *blk_get_flush_queue(
114 struct request_queue *q, struct blk_mq_ctx *ctx)
115{
116 if (q->mq_ops)
117 return blk_mq_map_queue(q, ctx->cpu)->fq;
118 return q->fq;
119}
120
121static inline void __blk_get_queue(struct request_queue *q)
122{
123 kobject_get(&q->kobj);
124}
125
126struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q,
127 int node, int cmd_size);
128void blk_free_flush_queue(struct blk_flush_queue *q);
129
130int blk_init_rl(struct request_list *rl, struct request_queue *q,
131 gfp_t gfp_mask);
132void blk_exit_rl(struct request_queue *q, struct request_list *rl);
133void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
134 struct bio *bio);
135void blk_queue_bypass_start(struct request_queue *q);
136void blk_queue_bypass_end(struct request_queue *q);
137void __blk_queue_free_tags(struct request_queue *q);
138void blk_freeze_queue(struct request_queue *q);
139
140static inline void blk_queue_enter_live(struct request_queue *q)
141{
142 /*
143 * Given that running in generic_make_request() context
144 * guarantees that a live reference against q_usage_counter has
145 * been established, further references under that same context
146 * need not check that the queue has been frozen (marked dead).
147 */
148 percpu_ref_get(&q->q_usage_counter);
149}
150
151#ifdef CONFIG_BLK_DEV_INTEGRITY
152void blk_flush_integrity(void);
153bool __bio_integrity_endio(struct bio *);
154static inline bool bio_integrity_endio(struct bio *bio)
155{
156 if (bio_integrity(bio))
157 return __bio_integrity_endio(bio);
158 return true;
159}
160#else
161static inline void blk_flush_integrity(void)
162{
163}
164static inline bool bio_integrity_endio(struct bio *bio)
165{
166 return true;
167}
168#endif
169
170void blk_timeout_work(struct work_struct *work);
171unsigned long blk_rq_timeout(unsigned long timeout);
172void blk_add_timer(struct request *req);
173void blk_delete_timer(struct request *);
174
175
176bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
177 struct bio *bio);
178bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
179 struct bio *bio);
180bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
181 struct bio *bio);
182bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
183 unsigned int *request_count,
184 struct request **same_queue_rq);
185unsigned int blk_plug_queued_count(struct request_queue *q);
186
187void blk_account_io_start(struct request *req, bool new_io);
188void blk_account_io_completion(struct request *req, unsigned int bytes);
189void blk_account_io_done(struct request *req);
190
191/*
192 * EH timer and IO completion will both attempt to 'grab' the request, make
193 * sure that only one of them succeeds. Steal the bottom bit of the
194 * __deadline field for this.
195 */
196static inline int blk_mark_rq_complete(struct request *rq)
197{
198 return test_and_set_bit(0, &rq->__deadline);
199}
200
201static inline void blk_clear_rq_complete(struct request *rq)
202{
203 clear_bit(0, &rq->__deadline);
204}
205
206static inline bool blk_rq_is_complete(struct request *rq)
207{
208 return test_bit(0, &rq->__deadline);
209}
210
211/*
212 * Internal elevator interface
213 */
214#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
215
216void blk_insert_flush(struct request *rq);
217
218static inline void elv_activate_rq(struct request_queue *q, struct request *rq)
219{
220 struct elevator_queue *e = q->elevator;
221
222 if (e->type->ops.sq.elevator_activate_req_fn)
223 e->type->ops.sq.elevator_activate_req_fn(q, rq);
224}
225
226static inline void elv_deactivate_rq(struct request_queue *q, struct request *rq)
227{
228 struct elevator_queue *e = q->elevator;
229
230 if (e->type->ops.sq.elevator_deactivate_req_fn)
231 e->type->ops.sq.elevator_deactivate_req_fn(q, rq);
232}
233
234int elv_register_queue(struct request_queue *q);
235void elv_unregister_queue(struct request_queue *q);
236
237struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
238
239#ifdef CONFIG_FAIL_IO_TIMEOUT
240int blk_should_fake_timeout(struct request_queue *);
241ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
242ssize_t part_timeout_store(struct device *, struct device_attribute *,
243 const char *, size_t);
244#else
245static inline int blk_should_fake_timeout(struct request_queue *q)
246{
247 return 0;
248}
249#endif
250
251int ll_back_merge_fn(struct request_queue *q, struct request *req,
252 struct bio *bio);
253int ll_front_merge_fn(struct request_queue *q, struct request *req,
254 struct bio *bio);
255struct request *attempt_back_merge(struct request_queue *q, struct request *rq);
256struct request *attempt_front_merge(struct request_queue *q, struct request *rq);
257int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
258 struct request *next);
259void blk_recalc_rq_segments(struct request *rq);
260void blk_rq_set_mixed_merge(struct request *rq);
261bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
262enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
263
264void blk_queue_congestion_threshold(struct request_queue *q);
265
266int blk_dev_init(void);
267
268
269/*
270 * Return the threshold (number of used requests) at which the queue is
271 * considered to be congested. It include a little hysteresis to keep the
272 * context switch rate down.
273 */
274static inline int queue_congestion_on_threshold(struct request_queue *q)
275{
276 return q->nr_congestion_on;
277}
278
279/*
280 * The threshold at which a queue is considered to be uncongested
281 */
282static inline int queue_congestion_off_threshold(struct request_queue *q)
283{
284 return q->nr_congestion_off;
285}
286
287extern int blk_update_nr_requests(struct request_queue *, unsigned int);
288
289/*
290 * Contribute to IO statistics IFF:
291 *
292 * a) it's attached to a gendisk, and
293 * b) the queue had IO stats enabled when this request was started, and
294 * c) it's a file system request
295 */
296static inline int blk_do_io_stat(struct request *rq)
297{
298 return rq->rq_disk &&
299 (rq->rq_flags & RQF_IO_STAT) &&
300 !blk_rq_is_passthrough(rq);
301}
302
303static inline void req_set_nomerge(struct request_queue *q, struct request *req)
304{
305 req->cmd_flags |= REQ_NOMERGE;
306 if (req == q->last_merge)
307 q->last_merge = NULL;
308}
309
310/*
311 * Steal a bit from this field for legacy IO path atomic IO marking. Note that
312 * setting the deadline clears the bottom bit, potentially clearing the
313 * completed bit. The user has to be OK with this (current ones are fine).
314 */
315static inline void blk_rq_set_deadline(struct request *rq, unsigned long time)
316{
317 rq->__deadline = time & ~0x1UL;
318}
319
320static inline unsigned long blk_rq_deadline(struct request *rq)
321{
322 return rq->__deadline & ~0x1UL;
323}
324
325/*
326 * Internal io_context interface
327 */
328void get_io_context(struct io_context *ioc);
329struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
330struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
331 gfp_t gfp_mask);
332void ioc_clear_queue(struct request_queue *q);
333
334int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
335
336/**
337 * rq_ioc - determine io_context for request allocation
338 * @bio: request being allocated is for this bio (can be %NULL)
339 *
340 * Determine io_context to use for request allocation for @bio. May return
341 * %NULL if %current->io_context doesn't exist.
342 */
343static inline struct io_context *rq_ioc(struct bio *bio)
344{
345#ifdef CONFIG_BLK_CGROUP
346 if (bio && bio->bi_ioc)
347 return bio->bi_ioc;
348#endif
349 return current->io_context;
350}
351
352/**
353 * create_io_context - try to create task->io_context
354 * @gfp_mask: allocation mask
355 * @node: allocation node
356 *
357 * If %current->io_context is %NULL, allocate a new io_context and install
358 * it. Returns the current %current->io_context which may be %NULL if
359 * allocation failed.
360 *
361 * Note that this function can't be called with IRQ disabled because
362 * task_lock which protects %current->io_context is IRQ-unsafe.
363 */
364static inline struct io_context *create_io_context(gfp_t gfp_mask, int node)
365{
366 WARN_ON_ONCE(irqs_disabled());
367 if (unlikely(!current->io_context))
368 create_task_io_context(current, gfp_mask, node);
369 return current->io_context;
370}
371
372/*
373 * Internal throttling interface
374 */
375#ifdef CONFIG_BLK_DEV_THROTTLING
376extern void blk_throtl_drain(struct request_queue *q);
377extern int blk_throtl_init(struct request_queue *q);
378extern void blk_throtl_exit(struct request_queue *q);
379extern void blk_throtl_register_queue(struct request_queue *q);
380#else /* CONFIG_BLK_DEV_THROTTLING */
381static inline void blk_throtl_drain(struct request_queue *q) { }
382static inline int blk_throtl_init(struct request_queue *q) { return 0; }
383static inline void blk_throtl_exit(struct request_queue *q) { }
384static inline void blk_throtl_register_queue(struct request_queue *q) { }
385#endif /* CONFIG_BLK_DEV_THROTTLING */
386#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
387extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
388extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
389 const char *page, size_t count);
390extern void blk_throtl_bio_endio(struct bio *bio);
391extern void blk_throtl_stat_add(struct request *rq, u64 time);
392#else
393static inline void blk_throtl_bio_endio(struct bio *bio) { }
394static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
395#endif
396
397#ifdef CONFIG_BOUNCE
398extern int init_emergency_isa_pool(void);
399extern void blk_queue_bounce(struct request_queue *q, struct bio **bio);
400#else
401static inline int init_emergency_isa_pool(void)
402{
403 return 0;
404}
405static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
406{
407}
408#endif /* CONFIG_BOUNCE */
409
410extern void blk_drain_queue(struct request_queue *q);
411
412#endif /* BLK_INTERNAL_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef BLK_INTERNAL_H
3#define BLK_INTERNAL_H
4
5#include <linux/bio-integrity.h>
6#include <linux/blk-crypto.h>
7#include <linux/lockdep.h>
8#include <linux/memblock.h> /* for max_pfn/max_low_pfn */
9#include <linux/sched/sysctl.h>
10#include <linux/timekeeping.h>
11#include <xen/xen.h>
12#include "blk-crypto-internal.h"
13
14struct elevator_type;
15
16/* Max future timer expiry for timeouts */
17#define BLK_MAX_TIMEOUT (5 * HZ)
18
19extern struct dentry *blk_debugfs_root;
20
21struct blk_flush_queue {
22 spinlock_t mq_flush_lock;
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 unsigned long flush_data_in_flight;
29 struct request *flush_rq;
30};
31
32bool is_flush_rq(struct request *req);
33
34struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
35 gfp_t flags);
36void blk_free_flush_queue(struct blk_flush_queue *q);
37
38bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
39bool blk_queue_start_drain(struct request_queue *q);
40bool __blk_freeze_queue_start(struct request_queue *q,
41 struct task_struct *owner);
42int __bio_queue_enter(struct request_queue *q, struct bio *bio);
43void submit_bio_noacct_nocheck(struct bio *bio);
44void bio_await_chain(struct bio *bio);
45
46static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
47{
48 rcu_read_lock();
49 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
50 goto fail;
51
52 /*
53 * The code that increments the pm_only counter must ensure that the
54 * counter is globally visible before the queue is unfrozen.
55 */
56 if (blk_queue_pm_only(q) &&
57 (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
58 goto fail_put;
59
60 rcu_read_unlock();
61 return true;
62
63fail_put:
64 blk_queue_exit(q);
65fail:
66 rcu_read_unlock();
67 return false;
68}
69
70static inline int bio_queue_enter(struct bio *bio)
71{
72 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
73
74 if (blk_try_enter_queue(q, false)) {
75 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_);
76 rwsem_release(&q->io_lockdep_map, _RET_IP_);
77 return 0;
78 }
79 return __bio_queue_enter(q, bio);
80}
81
82static inline void blk_wait_io(struct completion *done)
83{
84 /* Prevent hang_check timer from firing at us during very long I/O */
85 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
86
87 if (timeout)
88 while (!wait_for_completion_io_timeout(done, timeout))
89 ;
90 else
91 wait_for_completion_io(done);
92}
93
94#define BIO_INLINE_VECS 4
95struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
96 gfp_t gfp_mask);
97void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
98
99bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
100 struct page *page, unsigned len, unsigned offset,
101 bool *same_page);
102
103static inline bool biovec_phys_mergeable(struct request_queue *q,
104 struct bio_vec *vec1, struct bio_vec *vec2)
105{
106 unsigned long mask = queue_segment_boundary(q);
107 phys_addr_t addr1 = bvec_phys(vec1);
108 phys_addr_t addr2 = bvec_phys(vec2);
109
110 /*
111 * Merging adjacent physical pages may not work correctly under KMSAN
112 * if their metadata pages aren't adjacent. Just disable merging.
113 */
114 if (IS_ENABLED(CONFIG_KMSAN))
115 return false;
116
117 if (addr1 + vec1->bv_len != addr2)
118 return false;
119 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
120 return false;
121 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
122 return false;
123 return true;
124}
125
126static inline bool __bvec_gap_to_prev(const struct queue_limits *lim,
127 struct bio_vec *bprv, unsigned int offset)
128{
129 return (offset & lim->virt_boundary_mask) ||
130 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
131}
132
133/*
134 * Check if adding a bio_vec after bprv with offset would create a gap in
135 * the SG list. Most drivers don't care about this, but some do.
136 */
137static inline bool bvec_gap_to_prev(const struct queue_limits *lim,
138 struct bio_vec *bprv, unsigned int offset)
139{
140 if (!lim->virt_boundary_mask)
141 return false;
142 return __bvec_gap_to_prev(lim, bprv, offset);
143}
144
145static inline bool rq_mergeable(struct request *rq)
146{
147 if (blk_rq_is_passthrough(rq))
148 return false;
149
150 if (req_op(rq) == REQ_OP_FLUSH)
151 return false;
152
153 if (req_op(rq) == REQ_OP_WRITE_ZEROES)
154 return false;
155
156 if (req_op(rq) == REQ_OP_ZONE_APPEND)
157 return false;
158
159 if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
160 return false;
161 if (rq->rq_flags & RQF_NOMERGE_FLAGS)
162 return false;
163
164 return true;
165}
166
167/*
168 * There are two different ways to handle DISCARD merges:
169 * 1) If max_discard_segments > 1, the driver treats every bio as a range and
170 * send the bios to controller together. The ranges don't need to be
171 * contiguous.
172 * 2) Otherwise, the request will be normal read/write requests. The ranges
173 * need to be contiguous.
174 */
175static inline bool blk_discard_mergable(struct request *req)
176{
177 if (req_op(req) == REQ_OP_DISCARD &&
178 queue_max_discard_segments(req->q) > 1)
179 return true;
180 return false;
181}
182
183static inline unsigned int blk_rq_get_max_segments(struct request *rq)
184{
185 if (req_op(rq) == REQ_OP_DISCARD)
186 return queue_max_discard_segments(rq->q);
187 return queue_max_segments(rq->q);
188}
189
190static inline unsigned int blk_queue_get_max_sectors(struct request *rq)
191{
192 struct request_queue *q = rq->q;
193 enum req_op op = req_op(rq);
194
195 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
196 return min(q->limits.max_discard_sectors,
197 UINT_MAX >> SECTOR_SHIFT);
198
199 if (unlikely(op == REQ_OP_WRITE_ZEROES))
200 return q->limits.max_write_zeroes_sectors;
201
202 if (rq->cmd_flags & REQ_ATOMIC)
203 return q->limits.atomic_write_max_sectors;
204
205 return q->limits.max_sectors;
206}
207
208#ifdef CONFIG_BLK_DEV_INTEGRITY
209void blk_flush_integrity(void);
210void bio_integrity_free(struct bio *bio);
211
212/*
213 * Integrity payloads can either be owned by the submitter, in which case
214 * bio_uninit will free them, or owned and generated by the block layer,
215 * in which case we'll verify them here (for reads) and free them before
216 * the bio is handed back to the submitted.
217 */
218bool __bio_integrity_endio(struct bio *bio);
219static inline bool bio_integrity_endio(struct bio *bio)
220{
221 struct bio_integrity_payload *bip = bio_integrity(bio);
222
223 if (bip && (bip->bip_flags & BIP_BLOCK_INTEGRITY))
224 return __bio_integrity_endio(bio);
225 return true;
226}
227
228bool blk_integrity_merge_rq(struct request_queue *, struct request *,
229 struct request *);
230bool blk_integrity_merge_bio(struct request_queue *, struct request *,
231 struct bio *);
232
233static inline bool integrity_req_gap_back_merge(struct request *req,
234 struct bio *next)
235{
236 struct bio_integrity_payload *bip = bio_integrity(req->bio);
237 struct bio_integrity_payload *bip_next = bio_integrity(next);
238
239 return bvec_gap_to_prev(&req->q->limits,
240 &bip->bip_vec[bip->bip_vcnt - 1],
241 bip_next->bip_vec[0].bv_offset);
242}
243
244static inline bool integrity_req_gap_front_merge(struct request *req,
245 struct bio *bio)
246{
247 struct bio_integrity_payload *bip = bio_integrity(bio);
248 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
249
250 return bvec_gap_to_prev(&req->q->limits,
251 &bip->bip_vec[bip->bip_vcnt - 1],
252 bip_next->bip_vec[0].bv_offset);
253}
254
255extern const struct attribute_group blk_integrity_attr_group;
256#else /* CONFIG_BLK_DEV_INTEGRITY */
257static inline bool blk_integrity_merge_rq(struct request_queue *rq,
258 struct request *r1, struct request *r2)
259{
260 return true;
261}
262static inline bool blk_integrity_merge_bio(struct request_queue *rq,
263 struct request *r, struct bio *b)
264{
265 return true;
266}
267static inline bool integrity_req_gap_back_merge(struct request *req,
268 struct bio *next)
269{
270 return false;
271}
272static inline bool integrity_req_gap_front_merge(struct request *req,
273 struct bio *bio)
274{
275 return false;
276}
277
278static inline void blk_flush_integrity(void)
279{
280}
281static inline bool bio_integrity_endio(struct bio *bio)
282{
283 return true;
284}
285static inline void bio_integrity_free(struct bio *bio)
286{
287}
288#endif /* CONFIG_BLK_DEV_INTEGRITY */
289
290unsigned long blk_rq_timeout(unsigned long timeout);
291void blk_add_timer(struct request *req);
292
293enum bio_merge_status {
294 BIO_MERGE_OK,
295 BIO_MERGE_NONE,
296 BIO_MERGE_FAILED,
297};
298
299enum bio_merge_status bio_attempt_back_merge(struct request *req,
300 struct bio *bio, unsigned int nr_segs);
301bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
302 unsigned int nr_segs);
303bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
304 struct bio *bio, unsigned int nr_segs);
305
306/*
307 * Plug flush limits
308 */
309#define BLK_MAX_REQUEST_COUNT 32
310#define BLK_PLUG_FLUSH_SIZE (128 * 1024)
311
312/*
313 * Internal elevator interface
314 */
315#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
316
317bool blk_insert_flush(struct request *rq);
318
319int elevator_switch(struct request_queue *q, struct elevator_type *new_e);
320void elevator_disable(struct request_queue *q);
321void elevator_exit(struct request_queue *q);
322int elv_register_queue(struct request_queue *q, bool uevent);
323void elv_unregister_queue(struct request_queue *q);
324
325ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
326 char *buf);
327ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
328 char *buf);
329ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
330 char *buf);
331ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
332 char *buf);
333ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
334 const char *buf, size_t count);
335ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
336ssize_t part_timeout_store(struct device *, struct device_attribute *,
337 const char *, size_t);
338
339struct bio *bio_split_discard(struct bio *bio, const struct queue_limits *lim,
340 unsigned *nsegs);
341struct bio *bio_split_write_zeroes(struct bio *bio,
342 const struct queue_limits *lim, unsigned *nsegs);
343struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim,
344 unsigned *nr_segs);
345struct bio *bio_split_zone_append(struct bio *bio,
346 const struct queue_limits *lim, unsigned *nr_segs);
347
348/*
349 * All drivers must accept single-segments bios that are smaller than PAGE_SIZE.
350 *
351 * This is a quick and dirty check that relies on the fact that bi_io_vec[0] is
352 * always valid if a bio has data. The check might lead to occasional false
353 * positives when bios are cloned, but compared to the performance impact of
354 * cloned bios themselves the loop below doesn't matter anyway.
355 */
356static inline bool bio_may_need_split(struct bio *bio,
357 const struct queue_limits *lim)
358{
359 return lim->chunk_sectors || bio->bi_vcnt != 1 ||
360 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
361}
362
363/**
364 * __bio_split_to_limits - split a bio to fit the queue limits
365 * @bio: bio to be split
366 * @lim: queue limits to split based on
367 * @nr_segs: returns the number of segments in the returned bio
368 *
369 * Check if @bio needs splitting based on the queue limits, and if so split off
370 * a bio fitting the limits from the beginning of @bio and return it. @bio is
371 * shortened to the remainder and re-submitted.
372 *
373 * The split bio is allocated from @q->bio_split, which is provided by the
374 * block layer.
375 */
376static inline struct bio *__bio_split_to_limits(struct bio *bio,
377 const struct queue_limits *lim, unsigned int *nr_segs)
378{
379 switch (bio_op(bio)) {
380 case REQ_OP_READ:
381 case REQ_OP_WRITE:
382 if (bio_may_need_split(bio, lim))
383 return bio_split_rw(bio, lim, nr_segs);
384 *nr_segs = 1;
385 return bio;
386 case REQ_OP_ZONE_APPEND:
387 return bio_split_zone_append(bio, lim, nr_segs);
388 case REQ_OP_DISCARD:
389 case REQ_OP_SECURE_ERASE:
390 return bio_split_discard(bio, lim, nr_segs);
391 case REQ_OP_WRITE_ZEROES:
392 return bio_split_write_zeroes(bio, lim, nr_segs);
393 default:
394 /* other operations can't be split */
395 *nr_segs = 0;
396 return bio;
397 }
398}
399
400int ll_back_merge_fn(struct request *req, struct bio *bio,
401 unsigned int nr_segs);
402bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
403 struct request *next);
404unsigned int blk_recalc_rq_segments(struct request *rq);
405bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
406enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
407
408int blk_set_default_limits(struct queue_limits *lim);
409void blk_apply_bdi_limits(struct backing_dev_info *bdi,
410 struct queue_limits *lim);
411int blk_dev_init(void);
412
413void update_io_ticks(struct block_device *part, unsigned long now, bool end);
414unsigned int part_in_flight(struct block_device *part);
415
416static inline void req_set_nomerge(struct request_queue *q, struct request *req)
417{
418 req->cmd_flags |= REQ_NOMERGE;
419 if (req == q->last_merge)
420 q->last_merge = NULL;
421}
422
423/*
424 * Internal io_context interface
425 */
426struct io_cq *ioc_find_get_icq(struct request_queue *q);
427struct io_cq *ioc_lookup_icq(struct request_queue *q);
428#ifdef CONFIG_BLK_ICQ
429void ioc_clear_queue(struct request_queue *q);
430#else
431static inline void ioc_clear_queue(struct request_queue *q)
432{
433}
434#endif /* CONFIG_BLK_ICQ */
435
436struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
437
438static inline bool blk_queue_may_bounce(struct request_queue *q)
439{
440 return IS_ENABLED(CONFIG_BOUNCE) &&
441 (q->limits.features & BLK_FEAT_BOUNCE_HIGH) &&
442 max_low_pfn >= max_pfn;
443}
444
445static inline struct bio *blk_queue_bounce(struct bio *bio,
446 struct request_queue *q)
447{
448 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
449 return __blk_queue_bounce(bio, q);
450 return bio;
451}
452
453#ifdef CONFIG_BLK_DEV_ZONED
454void disk_init_zone_resources(struct gendisk *disk);
455void disk_free_zone_resources(struct gendisk *disk);
456static inline bool bio_zone_write_plugging(struct bio *bio)
457{
458 return bio_flagged(bio, BIO_ZONE_WRITE_PLUGGING);
459}
460void blk_zone_write_plug_bio_merged(struct bio *bio);
461void blk_zone_write_plug_init_request(struct request *rq);
462static inline void blk_zone_update_request_bio(struct request *rq,
463 struct bio *bio)
464{
465 /*
466 * For zone append requests, the request sector indicates the location
467 * at which the BIO data was written. Return this value to the BIO
468 * issuer through the BIO iter sector.
469 * For plugged zone writes, which include emulated zone append, we need
470 * the original BIO sector so that blk_zone_write_plug_bio_endio() can
471 * lookup the zone write plug.
472 */
473 if (req_op(rq) == REQ_OP_ZONE_APPEND || bio_zone_write_plugging(bio))
474 bio->bi_iter.bi_sector = rq->__sector;
475}
476void blk_zone_write_plug_bio_endio(struct bio *bio);
477static inline void blk_zone_bio_endio(struct bio *bio)
478{
479 /*
480 * For write BIOs to zoned devices, signal the completion of the BIO so
481 * that the next write BIO can be submitted by zone write plugging.
482 */
483 if (bio_zone_write_plugging(bio))
484 blk_zone_write_plug_bio_endio(bio);
485}
486
487void blk_zone_write_plug_finish_request(struct request *rq);
488static inline void blk_zone_finish_request(struct request *rq)
489{
490 if (rq->rq_flags & RQF_ZONE_WRITE_PLUGGING)
491 blk_zone_write_plug_finish_request(rq);
492}
493int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd,
494 unsigned long arg);
495int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode,
496 unsigned int cmd, unsigned long arg);
497#else /* CONFIG_BLK_DEV_ZONED */
498static inline void disk_init_zone_resources(struct gendisk *disk)
499{
500}
501static inline void disk_free_zone_resources(struct gendisk *disk)
502{
503}
504static inline bool bio_zone_write_plugging(struct bio *bio)
505{
506 return false;
507}
508static inline void blk_zone_write_plug_bio_merged(struct bio *bio)
509{
510}
511static inline void blk_zone_write_plug_init_request(struct request *rq)
512{
513}
514static inline void blk_zone_update_request_bio(struct request *rq,
515 struct bio *bio)
516{
517}
518static inline void blk_zone_bio_endio(struct bio *bio)
519{
520}
521static inline void blk_zone_finish_request(struct request *rq)
522{
523}
524static inline int blkdev_report_zones_ioctl(struct block_device *bdev,
525 unsigned int cmd, unsigned long arg)
526{
527 return -ENOTTY;
528}
529static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev,
530 blk_mode_t mode, unsigned int cmd, unsigned long arg)
531{
532 return -ENOTTY;
533}
534#endif /* CONFIG_BLK_DEV_ZONED */
535
536struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
537void bdev_add(struct block_device *bdev, dev_t dev);
538void bdev_unhash(struct block_device *bdev);
539void bdev_drop(struct block_device *bdev);
540
541int blk_alloc_ext_minor(void);
542void blk_free_ext_minor(unsigned int minor);
543#define ADDPART_FLAG_NONE 0
544#define ADDPART_FLAG_RAID 1
545#define ADDPART_FLAG_WHOLEDISK 2
546#define ADDPART_FLAG_READONLY 4
547int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
548 sector_t length);
549int bdev_del_partition(struct gendisk *disk, int partno);
550int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
551 sector_t length);
552void drop_partition(struct block_device *part);
553
554void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors);
555
556struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
557 struct lock_class_key *lkclass);
558
559int bio_add_hw_page(struct request_queue *q, struct bio *bio,
560 struct page *page, unsigned int len, unsigned int offset,
561 unsigned int max_sectors, bool *same_page);
562
563int bio_add_hw_folio(struct request_queue *q, struct bio *bio,
564 struct folio *folio, size_t len, size_t offset,
565 unsigned int max_sectors, bool *same_page);
566
567/*
568 * Clean up a page appropriately, where the page may be pinned, may have a
569 * ref taken on it or neither.
570 */
571static inline void bio_release_page(struct bio *bio, struct page *page)
572{
573 if (bio_flagged(bio, BIO_PAGE_PINNED))
574 unpin_user_page(page);
575}
576
577struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id);
578
579int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode);
580
581int disk_alloc_events(struct gendisk *disk);
582void disk_add_events(struct gendisk *disk);
583void disk_del_events(struct gendisk *disk);
584void disk_release_events(struct gendisk *disk);
585void disk_block_events(struct gendisk *disk);
586void disk_unblock_events(struct gendisk *disk);
587void disk_flush_events(struct gendisk *disk, unsigned int mask);
588extern struct device_attribute dev_attr_events;
589extern struct device_attribute dev_attr_events_async;
590extern struct device_attribute dev_attr_events_poll_msecs;
591
592extern struct attribute_group blk_trace_attr_group;
593
594blk_mode_t file_to_blk_mode(struct file *file);
595int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode,
596 loff_t lstart, loff_t lend);
597long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
598int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags);
599long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
600
601extern const struct address_space_operations def_blk_aops;
602
603int disk_register_independent_access_ranges(struct gendisk *disk);
604void disk_unregister_independent_access_ranges(struct gendisk *disk);
605
606#ifdef CONFIG_FAIL_MAKE_REQUEST
607bool should_fail_request(struct block_device *part, unsigned int bytes);
608#else /* CONFIG_FAIL_MAKE_REQUEST */
609static inline bool should_fail_request(struct block_device *part,
610 unsigned int bytes)
611{
612 return false;
613}
614#endif /* CONFIG_FAIL_MAKE_REQUEST */
615
616/*
617 * Optimized request reference counting. Ideally we'd make timeouts be more
618 * clever, as that's the only reason we need references at all... But until
619 * this happens, this is faster than using refcount_t. Also see:
620 *
621 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
622 */
623#define req_ref_zero_or_close_to_overflow(req) \
624 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
625
626static inline bool req_ref_inc_not_zero(struct request *req)
627{
628 return atomic_inc_not_zero(&req->ref);
629}
630
631static inline bool req_ref_put_and_test(struct request *req)
632{
633 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
634 return atomic_dec_and_test(&req->ref);
635}
636
637static inline void req_ref_set(struct request *req, int value)
638{
639 atomic_set(&req->ref, value);
640}
641
642static inline int req_ref_read(struct request *req)
643{
644 return atomic_read(&req->ref);
645}
646
647static inline u64 blk_time_get_ns(void)
648{
649 struct blk_plug *plug = current->plug;
650
651 if (!plug || !in_task())
652 return ktime_get_ns();
653
654 /*
655 * 0 could very well be a valid time, but rather than flag "this is
656 * a valid timestamp" separately, just accept that we'll do an extra
657 * ktime_get_ns() if we just happen to get 0 as the current time.
658 */
659 if (!plug->cur_ktime) {
660 plug->cur_ktime = ktime_get_ns();
661 current->flags |= PF_BLOCK_TS;
662 }
663 return plug->cur_ktime;
664}
665
666static inline ktime_t blk_time_get(void)
667{
668 return ns_to_ktime(blk_time_get_ns());
669}
670
671/*
672 * From most significant bit:
673 * 1 bit: reserved for other usage, see below
674 * 12 bits: original size of bio
675 * 51 bits: issue time of bio
676 */
677#define BIO_ISSUE_RES_BITS 1
678#define BIO_ISSUE_SIZE_BITS 12
679#define BIO_ISSUE_RES_SHIFT (64 - BIO_ISSUE_RES_BITS)
680#define BIO_ISSUE_SIZE_SHIFT (BIO_ISSUE_RES_SHIFT - BIO_ISSUE_SIZE_BITS)
681#define BIO_ISSUE_TIME_MASK ((1ULL << BIO_ISSUE_SIZE_SHIFT) - 1)
682#define BIO_ISSUE_SIZE_MASK \
683 (((1ULL << BIO_ISSUE_SIZE_BITS) - 1) << BIO_ISSUE_SIZE_SHIFT)
684#define BIO_ISSUE_RES_MASK (~((1ULL << BIO_ISSUE_RES_SHIFT) - 1))
685
686/* Reserved bit for blk-throtl */
687#define BIO_ISSUE_THROTL_SKIP_LATENCY (1ULL << 63)
688
689static inline u64 __bio_issue_time(u64 time)
690{
691 return time & BIO_ISSUE_TIME_MASK;
692}
693
694static inline u64 bio_issue_time(struct bio_issue *issue)
695{
696 return __bio_issue_time(issue->value);
697}
698
699static inline sector_t bio_issue_size(struct bio_issue *issue)
700{
701 return ((issue->value & BIO_ISSUE_SIZE_MASK) >> BIO_ISSUE_SIZE_SHIFT);
702}
703
704static inline void bio_issue_init(struct bio_issue *issue,
705 sector_t size)
706{
707 size &= (1ULL << BIO_ISSUE_SIZE_BITS) - 1;
708 issue->value = ((issue->value & BIO_ISSUE_RES_MASK) |
709 (blk_time_get_ns() & BIO_ISSUE_TIME_MASK) |
710 ((u64)size << BIO_ISSUE_SIZE_SHIFT));
711}
712
713void bdev_release(struct file *bdev_file);
714int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder,
715 const struct blk_holder_ops *hops, struct file *bdev_file);
716int bdev_permission(dev_t dev, blk_mode_t mode, void *holder);
717
718void blk_integrity_generate(struct bio *bio);
719void blk_integrity_verify(struct bio *bio);
720void blk_integrity_prepare(struct request *rq);
721void blk_integrity_complete(struct request *rq, unsigned int nr_bytes);
722
723static inline void blk_freeze_acquire_lock(struct request_queue *q, bool
724 disk_dead, bool queue_dying)
725{
726 if (!disk_dead)
727 rwsem_acquire(&q->io_lockdep_map, 0, 1, _RET_IP_);
728 if (!queue_dying)
729 rwsem_acquire(&q->q_lockdep_map, 0, 1, _RET_IP_);
730}
731
732static inline void blk_unfreeze_release_lock(struct request_queue *q, bool
733 disk_dead, bool queue_dying)
734{
735 if (!queue_dying)
736 rwsem_release(&q->q_lockdep_map, _RET_IP_);
737 if (!disk_dead)
738 rwsem_release(&q->io_lockdep_map, _RET_IP_);
739}
740
741#endif /* BLK_INTERNAL_H */