Loading...
1#ifndef BLK_INTERNAL_H
2#define BLK_INTERNAL_H
3
4#include <linux/idr.h>
5
6/* Amount of time in which a process may batch requests */
7#define BLK_BATCH_TIME (HZ/50UL)
8
9/* Number of requests a "batching" process may submit */
10#define BLK_BATCH_REQ 32
11
12extern struct kmem_cache *blk_requestq_cachep;
13extern struct kmem_cache *request_cachep;
14extern struct kobj_type blk_queue_ktype;
15extern struct ida blk_queue_ida;
16
17static inline void __blk_get_queue(struct request_queue *q)
18{
19 kobject_get(&q->kobj);
20}
21
22int blk_init_rl(struct request_list *rl, struct request_queue *q,
23 gfp_t gfp_mask);
24void blk_exit_rl(struct request_list *rl);
25void init_request_from_bio(struct request *req, struct bio *bio);
26void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
27 struct bio *bio);
28int blk_rq_append_bio(struct request_queue *q, struct request *rq,
29 struct bio *bio);
30void blk_queue_bypass_start(struct request_queue *q);
31void blk_queue_bypass_end(struct request_queue *q);
32void blk_dequeue_request(struct request *rq);
33void __blk_queue_free_tags(struct request_queue *q);
34bool __blk_end_bidi_request(struct request *rq, int error,
35 unsigned int nr_bytes, unsigned int bidi_bytes);
36
37void blk_rq_timed_out_timer(unsigned long data);
38void blk_rq_check_expired(struct request *rq, unsigned long *next_timeout,
39 unsigned int *next_set);
40void __blk_add_timer(struct request *req, struct list_head *timeout_list);
41void blk_delete_timer(struct request *);
42void blk_add_timer(struct request *);
43
44
45bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
46 struct bio *bio);
47bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
48 struct bio *bio);
49bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
50 unsigned int *request_count);
51
52void blk_account_io_start(struct request *req, bool new_io);
53void blk_account_io_completion(struct request *req, unsigned int bytes);
54void blk_account_io_done(struct request *req);
55
56/*
57 * Internal atomic flags for request handling
58 */
59enum rq_atomic_flags {
60 REQ_ATOM_COMPLETE = 0,
61 REQ_ATOM_STARTED,
62};
63
64/*
65 * EH timer and IO completion will both attempt to 'grab' the request, make
66 * sure that only one of them succeeds
67 */
68static inline int blk_mark_rq_complete(struct request *rq)
69{
70 return test_and_set_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
71}
72
73static inline void blk_clear_rq_complete(struct request *rq)
74{
75 clear_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
76}
77
78/*
79 * Internal elevator interface
80 */
81#define ELV_ON_HASH(rq) ((rq)->cmd_flags & REQ_HASHED)
82
83void blk_insert_flush(struct request *rq);
84void blk_abort_flushes(struct request_queue *q);
85
86static inline struct request *__elv_next_request(struct request_queue *q)
87{
88 struct request *rq;
89
90 while (1) {
91 if (!list_empty(&q->queue_head)) {
92 rq = list_entry_rq(q->queue_head.next);
93 return rq;
94 }
95
96 /*
97 * Flush request is running and flush request isn't queueable
98 * in the drive, we can hold the queue till flush request is
99 * finished. Even we don't do this, driver can't dispatch next
100 * requests and will requeue them. And this can improve
101 * throughput too. For example, we have request flush1, write1,
102 * flush 2. flush1 is dispatched, then queue is hold, write1
103 * isn't inserted to queue. After flush1 is finished, flush2
104 * will be dispatched. Since disk cache is already clean,
105 * flush2 will be finished very soon, so looks like flush2 is
106 * folded to flush1.
107 * Since the queue is hold, a flag is set to indicate the queue
108 * should be restarted later. Please see flush_end_io() for
109 * details.
110 */
111 if (q->flush_pending_idx != q->flush_running_idx &&
112 !queue_flush_queueable(q)) {
113 q->flush_queue_delayed = 1;
114 return NULL;
115 }
116 if (unlikely(blk_queue_bypass(q)) ||
117 !q->elevator->type->ops.elevator_dispatch_fn(q, 0))
118 return NULL;
119 }
120}
121
122static inline void elv_activate_rq(struct request_queue *q, struct request *rq)
123{
124 struct elevator_queue *e = q->elevator;
125
126 if (e->type->ops.elevator_activate_req_fn)
127 e->type->ops.elevator_activate_req_fn(q, rq);
128}
129
130static inline void elv_deactivate_rq(struct request_queue *q, struct request *rq)
131{
132 struct elevator_queue *e = q->elevator;
133
134 if (e->type->ops.elevator_deactivate_req_fn)
135 e->type->ops.elevator_deactivate_req_fn(q, rq);
136}
137
138#ifdef CONFIG_FAIL_IO_TIMEOUT
139int blk_should_fake_timeout(struct request_queue *);
140ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
141ssize_t part_timeout_store(struct device *, struct device_attribute *,
142 const char *, size_t);
143#else
144static inline int blk_should_fake_timeout(struct request_queue *q)
145{
146 return 0;
147}
148#endif
149
150int ll_back_merge_fn(struct request_queue *q, struct request *req,
151 struct bio *bio);
152int ll_front_merge_fn(struct request_queue *q, struct request *req,
153 struct bio *bio);
154int attempt_back_merge(struct request_queue *q, struct request *rq);
155int attempt_front_merge(struct request_queue *q, struct request *rq);
156int blk_attempt_req_merge(struct request_queue *q, struct request *rq,
157 struct request *next);
158void blk_recalc_rq_segments(struct request *rq);
159void blk_rq_set_mixed_merge(struct request *rq);
160bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
161int blk_try_merge(struct request *rq, struct bio *bio);
162
163void blk_queue_congestion_threshold(struct request_queue *q);
164
165void __blk_run_queue_uncond(struct request_queue *q);
166
167int blk_dev_init(void);
168
169
170/*
171 * Return the threshold (number of used requests) at which the queue is
172 * considered to be congested. It include a little hysteresis to keep the
173 * context switch rate down.
174 */
175static inline int queue_congestion_on_threshold(struct request_queue *q)
176{
177 return q->nr_congestion_on;
178}
179
180/*
181 * The threshold at which a queue is considered to be uncongested
182 */
183static inline int queue_congestion_off_threshold(struct request_queue *q)
184{
185 return q->nr_congestion_off;
186}
187
188/*
189 * Contribute to IO statistics IFF:
190 *
191 * a) it's attached to a gendisk, and
192 * b) the queue had IO stats enabled when this request was started, and
193 * c) it's a file system request
194 */
195static inline int blk_do_io_stat(struct request *rq)
196{
197 return rq->rq_disk &&
198 (rq->cmd_flags & REQ_IO_STAT) &&
199 (rq->cmd_type == REQ_TYPE_FS);
200}
201
202/*
203 * Internal io_context interface
204 */
205void get_io_context(struct io_context *ioc);
206struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
207struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
208 gfp_t gfp_mask);
209void ioc_clear_queue(struct request_queue *q);
210
211int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
212
213/**
214 * create_io_context - try to create task->io_context
215 * @gfp_mask: allocation mask
216 * @node: allocation node
217 *
218 * If %current->io_context is %NULL, allocate a new io_context and install
219 * it. Returns the current %current->io_context which may be %NULL if
220 * allocation failed.
221 *
222 * Note that this function can't be called with IRQ disabled because
223 * task_lock which protects %current->io_context is IRQ-unsafe.
224 */
225static inline struct io_context *create_io_context(gfp_t gfp_mask, int node)
226{
227 WARN_ON_ONCE(irqs_disabled());
228 if (unlikely(!current->io_context))
229 create_task_io_context(current, gfp_mask, node);
230 return current->io_context;
231}
232
233/*
234 * Internal throttling interface
235 */
236#ifdef CONFIG_BLK_DEV_THROTTLING
237extern bool blk_throtl_bio(struct request_queue *q, struct bio *bio);
238extern void blk_throtl_drain(struct request_queue *q);
239extern int blk_throtl_init(struct request_queue *q);
240extern void blk_throtl_exit(struct request_queue *q);
241#else /* CONFIG_BLK_DEV_THROTTLING */
242static inline bool blk_throtl_bio(struct request_queue *q, struct bio *bio)
243{
244 return false;
245}
246static inline void blk_throtl_drain(struct request_queue *q) { }
247static inline int blk_throtl_init(struct request_queue *q) { return 0; }
248static inline void blk_throtl_exit(struct request_queue *q) { }
249#endif /* CONFIG_BLK_DEV_THROTTLING */
250
251#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/idr.h>
6#include <linux/blk-mq.h>
7#include <linux/part_stat.h>
8#include <linux/blk-crypto.h>
9#include <linux/memblock.h> /* for max_pfn/max_low_pfn */
10#include <xen/xen.h>
11#include "blk-crypto-internal.h"
12#include "blk-mq.h"
13#include "blk-mq-sched.h"
14
15/* Max future timer expiry for timeouts */
16#define BLK_MAX_TIMEOUT (5 * HZ)
17
18extern struct dentry *blk_debugfs_root;
19
20struct blk_flush_queue {
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 spinlock_t mq_flush_lock;
30};
31
32extern struct kmem_cache *blk_requestq_cachep;
33extern struct kobj_type blk_queue_ktype;
34extern struct ida blk_queue_ida;
35
36static inline struct blk_flush_queue *
37blk_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
42static inline void __blk_get_queue(struct request_queue *q)
43{
44 kobject_get(&q->kobj);
45}
46
47bool is_flush_rq(struct request *req);
48
49struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
50 gfp_t flags);
51void blk_free_flush_queue(struct blk_flush_queue *q);
52
53void blk_freeze_queue(struct request_queue *q);
54
55#define BIO_INLINE_VECS 4
56struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
57 gfp_t gfp_mask);
58void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
59
60static inline bool biovec_phys_mergeable(struct request_queue *q,
61 struct bio_vec *vec1, struct bio_vec *vec2)
62{
63 unsigned long mask = queue_segment_boundary(q);
64 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
65 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
66
67 if (addr1 + vec1->bv_len != addr2)
68 return false;
69 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
70 return false;
71 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
72 return false;
73 return true;
74}
75
76static inline bool __bvec_gap_to_prev(struct request_queue *q,
77 struct bio_vec *bprv, unsigned int offset)
78{
79 return (offset & queue_virt_boundary(q)) ||
80 ((bprv->bv_offset + bprv->bv_len) & queue_virt_boundary(q));
81}
82
83/*
84 * Check if adding a bio_vec after bprv with offset would create a gap in
85 * the SG list. Most drivers don't care about this, but some do.
86 */
87static inline bool bvec_gap_to_prev(struct request_queue *q,
88 struct bio_vec *bprv, unsigned int offset)
89{
90 if (!queue_virt_boundary(q))
91 return false;
92 return __bvec_gap_to_prev(q, bprv, offset);
93}
94
95#ifdef CONFIG_BLK_DEV_INTEGRITY
96void blk_flush_integrity(void);
97bool __bio_integrity_endio(struct bio *);
98void bio_integrity_free(struct bio *bio);
99static inline bool bio_integrity_endio(struct bio *bio)
100{
101 if (bio_integrity(bio))
102 return __bio_integrity_endio(bio);
103 return true;
104}
105
106bool blk_integrity_merge_rq(struct request_queue *, struct request *,
107 struct request *);
108bool blk_integrity_merge_bio(struct request_queue *, struct request *,
109 struct bio *);
110
111static inline bool integrity_req_gap_back_merge(struct request *req,
112 struct bio *next)
113{
114 struct bio_integrity_payload *bip = bio_integrity(req->bio);
115 struct bio_integrity_payload *bip_next = bio_integrity(next);
116
117 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
118 bip_next->bip_vec[0].bv_offset);
119}
120
121static inline bool integrity_req_gap_front_merge(struct request *req,
122 struct bio *bio)
123{
124 struct bio_integrity_payload *bip = bio_integrity(bio);
125 struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
126
127 return bvec_gap_to_prev(req->q, &bip->bip_vec[bip->bip_vcnt - 1],
128 bip_next->bip_vec[0].bv_offset);
129}
130
131void blk_integrity_add(struct gendisk *);
132void blk_integrity_del(struct gendisk *);
133#else /* CONFIG_BLK_DEV_INTEGRITY */
134static inline bool blk_integrity_merge_rq(struct request_queue *rq,
135 struct request *r1, struct request *r2)
136{
137 return true;
138}
139static inline bool blk_integrity_merge_bio(struct request_queue *rq,
140 struct request *r, struct bio *b)
141{
142 return true;
143}
144static inline bool integrity_req_gap_back_merge(struct request *req,
145 struct bio *next)
146{
147 return false;
148}
149static inline bool integrity_req_gap_front_merge(struct request *req,
150 struct bio *bio)
151{
152 return false;
153}
154
155static inline void blk_flush_integrity(void)
156{
157}
158static inline bool bio_integrity_endio(struct bio *bio)
159{
160 return true;
161}
162static inline void bio_integrity_free(struct bio *bio)
163{
164}
165static inline void blk_integrity_add(struct gendisk *disk)
166{
167}
168static inline void blk_integrity_del(struct gendisk *disk)
169{
170}
171#endif /* CONFIG_BLK_DEV_INTEGRITY */
172
173unsigned long blk_rq_timeout(unsigned long timeout);
174void blk_add_timer(struct request *req);
175
176bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
177 unsigned int nr_segs, struct request **same_queue_rq);
178bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
179 struct bio *bio, unsigned int nr_segs);
180
181void blk_account_io_start(struct request *req);
182void blk_account_io_done(struct request *req, u64 now);
183
184/*
185 * Internal elevator interface
186 */
187#define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
188
189void blk_insert_flush(struct request *rq);
190
191int elevator_switch_mq(struct request_queue *q,
192 struct elevator_type *new_e);
193void __elevator_exit(struct request_queue *, struct elevator_queue *);
194int elv_register_queue(struct request_queue *q, bool uevent);
195void elv_unregister_queue(struct request_queue *q);
196
197static inline void elevator_exit(struct request_queue *q,
198 struct elevator_queue *e)
199{
200 lockdep_assert_held(&q->sysfs_lock);
201
202 blk_mq_sched_free_requests(q);
203 __elevator_exit(q, e);
204}
205
206ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
207 char *buf);
208ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
209 char *buf);
210ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
211 char *buf);
212ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
213 char *buf);
214ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
215 const char *buf, size_t count);
216ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
217ssize_t part_timeout_store(struct device *, struct device_attribute *,
218 const char *, size_t);
219
220void __blk_queue_split(struct bio **bio, unsigned int *nr_segs);
221int ll_back_merge_fn(struct request *req, struct bio *bio,
222 unsigned int nr_segs);
223bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
224 struct request *next);
225unsigned int blk_recalc_rq_segments(struct request *rq);
226void blk_rq_set_mixed_merge(struct request *rq);
227bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
228enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
229
230int blk_dev_init(void);
231
232/*
233 * Contribute to IO statistics IFF:
234 *
235 * a) it's attached to a gendisk, and
236 * b) the queue had IO stats enabled when this request was started
237 */
238static inline bool blk_do_io_stat(struct request *rq)
239{
240 return rq->rq_disk && (rq->rq_flags & RQF_IO_STAT);
241}
242
243static inline void req_set_nomerge(struct request_queue *q, struct request *req)
244{
245 req->cmd_flags |= REQ_NOMERGE;
246 if (req == q->last_merge)
247 q->last_merge = NULL;
248}
249
250/*
251 * The max size one bio can handle is UINT_MAX becasue bvec_iter.bi_size
252 * is defined as 'unsigned int', meantime it has to aligned to with logical
253 * block size which is the minimum accepted unit by hardware.
254 */
255static inline unsigned int bio_allowed_max_sectors(struct request_queue *q)
256{
257 return round_down(UINT_MAX, queue_logical_block_size(q)) >> 9;
258}
259
260/*
261 * The max bio size which is aligned to q->limits.discard_granularity. This
262 * is a hint to split large discard bio in generic block layer, then if device
263 * driver needs to split the discard bio into smaller ones, their bi_size can
264 * be very probably and easily aligned to discard_granularity of the device's
265 * queue.
266 */
267static inline unsigned int bio_aligned_discard_max_sectors(
268 struct request_queue *q)
269{
270 return round_down(UINT_MAX, q->limits.discard_granularity) >>
271 SECTOR_SHIFT;
272}
273
274/*
275 * Internal io_context interface
276 */
277void get_io_context(struct io_context *ioc);
278struct io_cq *ioc_lookup_icq(struct io_context *ioc, struct request_queue *q);
279struct io_cq *ioc_create_icq(struct io_context *ioc, struct request_queue *q,
280 gfp_t gfp_mask);
281void ioc_clear_queue(struct request_queue *q);
282
283int create_task_io_context(struct task_struct *task, gfp_t gfp_mask, int node);
284
285/*
286 * Internal throttling interface
287 */
288#ifdef CONFIG_BLK_DEV_THROTTLING
289extern int blk_throtl_init(struct request_queue *q);
290extern void blk_throtl_exit(struct request_queue *q);
291extern void blk_throtl_register_queue(struct request_queue *q);
292extern void blk_throtl_charge_bio_split(struct bio *bio);
293bool blk_throtl_bio(struct bio *bio);
294#else /* CONFIG_BLK_DEV_THROTTLING */
295static inline int blk_throtl_init(struct request_queue *q) { return 0; }
296static inline void blk_throtl_exit(struct request_queue *q) { }
297static inline void blk_throtl_register_queue(struct request_queue *q) { }
298static inline void blk_throtl_charge_bio_split(struct bio *bio) { }
299static inline bool blk_throtl_bio(struct bio *bio) { return false; }
300#endif /* CONFIG_BLK_DEV_THROTTLING */
301#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
302extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
303extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
304 const char *page, size_t count);
305extern void blk_throtl_bio_endio(struct bio *bio);
306extern void blk_throtl_stat_add(struct request *rq, u64 time);
307#else
308static inline void blk_throtl_bio_endio(struct bio *bio) { }
309static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
310#endif
311
312void __blk_queue_bounce(struct request_queue *q, struct bio **bio);
313
314static inline bool blk_queue_may_bounce(struct request_queue *q)
315{
316 return IS_ENABLED(CONFIG_BOUNCE) &&
317 q->limits.bounce == BLK_BOUNCE_HIGH &&
318 max_low_pfn >= max_pfn;
319}
320
321static inline void blk_queue_bounce(struct request_queue *q, struct bio **bio)
322{
323 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(*bio)))
324 __blk_queue_bounce(q, bio);
325}
326
327#ifdef CONFIG_BLK_CGROUP_IOLATENCY
328extern int blk_iolatency_init(struct request_queue *q);
329#else
330static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
331#endif
332
333struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp);
334
335#ifdef CONFIG_BLK_DEV_ZONED
336void blk_queue_free_zone_bitmaps(struct request_queue *q);
337void blk_queue_clear_zone_settings(struct request_queue *q);
338#else
339static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
340static inline void blk_queue_clear_zone_settings(struct request_queue *q) {}
341#endif
342
343int blk_alloc_ext_minor(void);
344void blk_free_ext_minor(unsigned int minor);
345char *disk_name(struct gendisk *hd, int partno, char *buf);
346#define ADDPART_FLAG_NONE 0
347#define ADDPART_FLAG_RAID 1
348#define ADDPART_FLAG_WHOLEDISK 2
349int bdev_add_partition(struct block_device *bdev, int partno,
350 sector_t start, sector_t length);
351int bdev_del_partition(struct block_device *bdev, int partno);
352int bdev_resize_partition(struct block_device *bdev, int partno,
353 sector_t start, sector_t length);
354
355int bio_add_hw_page(struct request_queue *q, struct bio *bio,
356 struct page *page, unsigned int len, unsigned int offset,
357 unsigned int max_sectors, bool *same_page);
358
359struct request_queue *blk_alloc_queue(int node_id);
360
361void disk_alloc_events(struct gendisk *disk);
362void disk_add_events(struct gendisk *disk);
363void disk_del_events(struct gendisk *disk);
364void disk_release_events(struct gendisk *disk);
365extern struct device_attribute dev_attr_events;
366extern struct device_attribute dev_attr_events_async;
367extern struct device_attribute dev_attr_events_poll_msecs;
368
369#endif /* BLK_INTERNAL_H */