Loading...
1/*
2 * Functions related to sysfs handling
3 */
4#include <linux/kernel.h>
5#include <linux/slab.h>
6#include <linux/module.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/backing-dev.h>
10#include <linux/blktrace_api.h>
11#include <linux/blk-mq.h>
12#include <linux/blk-cgroup.h>
13
14#include "blk.h"
15#include "blk-mq.h"
16
17struct queue_sysfs_entry {
18 struct attribute attr;
19 ssize_t (*show)(struct request_queue *, char *);
20 ssize_t (*store)(struct request_queue *, const char *, size_t);
21};
22
23static ssize_t
24queue_var_show(unsigned long var, char *page)
25{
26 return sprintf(page, "%lu\n", var);
27}
28
29static ssize_t
30queue_var_store(unsigned long *var, const char *page, size_t count)
31{
32 int err;
33 unsigned long v;
34
35 err = kstrtoul(page, 10, &v);
36 if (err || v > UINT_MAX)
37 return -EINVAL;
38
39 *var = v;
40
41 return count;
42}
43
44static ssize_t queue_requests_show(struct request_queue *q, char *page)
45{
46 return queue_var_show(q->nr_requests, (page));
47}
48
49static ssize_t
50queue_requests_store(struct request_queue *q, const char *page, size_t count)
51{
52 unsigned long nr;
53 int ret, err;
54
55 if (!q->request_fn && !q->mq_ops)
56 return -EINVAL;
57
58 ret = queue_var_store(&nr, page, count);
59 if (ret < 0)
60 return ret;
61
62 if (nr < BLKDEV_MIN_RQ)
63 nr = BLKDEV_MIN_RQ;
64
65 if (q->request_fn)
66 err = blk_update_nr_requests(q, nr);
67 else
68 err = blk_mq_update_nr_requests(q, nr);
69
70 if (err)
71 return err;
72
73 return ret;
74}
75
76static ssize_t queue_ra_show(struct request_queue *q, char *page)
77{
78 unsigned long ra_kb = q->backing_dev_info.ra_pages <<
79 (PAGE_SHIFT - 10);
80
81 return queue_var_show(ra_kb, (page));
82}
83
84static ssize_t
85queue_ra_store(struct request_queue *q, const char *page, size_t count)
86{
87 unsigned long ra_kb;
88 ssize_t ret = queue_var_store(&ra_kb, page, count);
89
90 if (ret < 0)
91 return ret;
92
93 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_SHIFT - 10);
94
95 return ret;
96}
97
98static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
99{
100 int max_sectors_kb = queue_max_sectors(q) >> 1;
101
102 return queue_var_show(max_sectors_kb, (page));
103}
104
105static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
106{
107 return queue_var_show(queue_max_segments(q), (page));
108}
109
110static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
111{
112 return queue_var_show(q->limits.max_integrity_segments, (page));
113}
114
115static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
116{
117 if (blk_queue_cluster(q))
118 return queue_var_show(queue_max_segment_size(q), (page));
119
120 return queue_var_show(PAGE_SIZE, (page));
121}
122
123static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
124{
125 return queue_var_show(queue_logical_block_size(q), page);
126}
127
128static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
129{
130 return queue_var_show(queue_physical_block_size(q), page);
131}
132
133static ssize_t queue_io_min_show(struct request_queue *q, char *page)
134{
135 return queue_var_show(queue_io_min(q), page);
136}
137
138static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
139{
140 return queue_var_show(queue_io_opt(q), page);
141}
142
143static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
144{
145 return queue_var_show(q->limits.discard_granularity, page);
146}
147
148static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
149{
150
151 return sprintf(page, "%llu\n",
152 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
153}
154
155static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
156{
157 return sprintf(page, "%llu\n",
158 (unsigned long long)q->limits.max_discard_sectors << 9);
159}
160
161static ssize_t queue_discard_max_store(struct request_queue *q,
162 const char *page, size_t count)
163{
164 unsigned long max_discard;
165 ssize_t ret = queue_var_store(&max_discard, page, count);
166
167 if (ret < 0)
168 return ret;
169
170 if (max_discard & (q->limits.discard_granularity - 1))
171 return -EINVAL;
172
173 max_discard >>= 9;
174 if (max_discard > UINT_MAX)
175 return -EINVAL;
176
177 if (max_discard > q->limits.max_hw_discard_sectors)
178 max_discard = q->limits.max_hw_discard_sectors;
179
180 q->limits.max_discard_sectors = max_discard;
181 return ret;
182}
183
184static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
185{
186 return queue_var_show(queue_discard_zeroes_data(q), page);
187}
188
189static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
190{
191 return sprintf(page, "%llu\n",
192 (unsigned long long)q->limits.max_write_same_sectors << 9);
193}
194
195
196static ssize_t
197queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
198{
199 unsigned long max_sectors_kb,
200 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
201 page_kb = 1 << (PAGE_SHIFT - 10);
202 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
203
204 if (ret < 0)
205 return ret;
206
207 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
208 q->limits.max_dev_sectors >> 1);
209
210 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
211 return -EINVAL;
212
213 spin_lock_irq(q->queue_lock);
214 q->limits.max_sectors = max_sectors_kb << 1;
215 spin_unlock_irq(q->queue_lock);
216
217 return ret;
218}
219
220static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
221{
222 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
223
224 return queue_var_show(max_hw_sectors_kb, (page));
225}
226
227#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
228static ssize_t \
229queue_show_##name(struct request_queue *q, char *page) \
230{ \
231 int bit; \
232 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
233 return queue_var_show(neg ? !bit : bit, page); \
234} \
235static ssize_t \
236queue_store_##name(struct request_queue *q, const char *page, size_t count) \
237{ \
238 unsigned long val; \
239 ssize_t ret; \
240 ret = queue_var_store(&val, page, count); \
241 if (ret < 0) \
242 return ret; \
243 if (neg) \
244 val = !val; \
245 \
246 spin_lock_irq(q->queue_lock); \
247 if (val) \
248 queue_flag_set(QUEUE_FLAG_##flag, q); \
249 else \
250 queue_flag_clear(QUEUE_FLAG_##flag, q); \
251 spin_unlock_irq(q->queue_lock); \
252 return ret; \
253}
254
255QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
256QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
257QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
258#undef QUEUE_SYSFS_BIT_FNS
259
260static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
261{
262 return queue_var_show((blk_queue_nomerges(q) << 1) |
263 blk_queue_noxmerges(q), page);
264}
265
266static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
267 size_t count)
268{
269 unsigned long nm;
270 ssize_t ret = queue_var_store(&nm, page, count);
271
272 if (ret < 0)
273 return ret;
274
275 spin_lock_irq(q->queue_lock);
276 queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
277 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
278 if (nm == 2)
279 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
280 else if (nm)
281 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
282 spin_unlock_irq(q->queue_lock);
283
284 return ret;
285}
286
287static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
288{
289 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
290 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
291
292 return queue_var_show(set << force, page);
293}
294
295static ssize_t
296queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
297{
298 ssize_t ret = -EINVAL;
299#ifdef CONFIG_SMP
300 unsigned long val;
301
302 ret = queue_var_store(&val, page, count);
303 if (ret < 0)
304 return ret;
305
306 spin_lock_irq(q->queue_lock);
307 if (val == 2) {
308 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
309 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
310 } else if (val == 1) {
311 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
312 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
313 } else if (val == 0) {
314 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
315 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
316 }
317 spin_unlock_irq(q->queue_lock);
318#endif
319 return ret;
320}
321
322static ssize_t queue_poll_show(struct request_queue *q, char *page)
323{
324 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
325}
326
327static ssize_t queue_poll_store(struct request_queue *q, const char *page,
328 size_t count)
329{
330 unsigned long poll_on;
331 ssize_t ret;
332
333 if (!q->mq_ops || !q->mq_ops->poll)
334 return -EINVAL;
335
336 ret = queue_var_store(&poll_on, page, count);
337 if (ret < 0)
338 return ret;
339
340 spin_lock_irq(q->queue_lock);
341 if (poll_on)
342 queue_flag_set(QUEUE_FLAG_POLL, q);
343 else
344 queue_flag_clear(QUEUE_FLAG_POLL, q);
345 spin_unlock_irq(q->queue_lock);
346
347 return ret;
348}
349
350static struct queue_sysfs_entry queue_requests_entry = {
351 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
352 .show = queue_requests_show,
353 .store = queue_requests_store,
354};
355
356static struct queue_sysfs_entry queue_ra_entry = {
357 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
358 .show = queue_ra_show,
359 .store = queue_ra_store,
360};
361
362static struct queue_sysfs_entry queue_max_sectors_entry = {
363 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
364 .show = queue_max_sectors_show,
365 .store = queue_max_sectors_store,
366};
367
368static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
369 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
370 .show = queue_max_hw_sectors_show,
371};
372
373static struct queue_sysfs_entry queue_max_segments_entry = {
374 .attr = {.name = "max_segments", .mode = S_IRUGO },
375 .show = queue_max_segments_show,
376};
377
378static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
379 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
380 .show = queue_max_integrity_segments_show,
381};
382
383static struct queue_sysfs_entry queue_max_segment_size_entry = {
384 .attr = {.name = "max_segment_size", .mode = S_IRUGO },
385 .show = queue_max_segment_size_show,
386};
387
388static struct queue_sysfs_entry queue_iosched_entry = {
389 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
390 .show = elv_iosched_show,
391 .store = elv_iosched_store,
392};
393
394static struct queue_sysfs_entry queue_hw_sector_size_entry = {
395 .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
396 .show = queue_logical_block_size_show,
397};
398
399static struct queue_sysfs_entry queue_logical_block_size_entry = {
400 .attr = {.name = "logical_block_size", .mode = S_IRUGO },
401 .show = queue_logical_block_size_show,
402};
403
404static struct queue_sysfs_entry queue_physical_block_size_entry = {
405 .attr = {.name = "physical_block_size", .mode = S_IRUGO },
406 .show = queue_physical_block_size_show,
407};
408
409static struct queue_sysfs_entry queue_io_min_entry = {
410 .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
411 .show = queue_io_min_show,
412};
413
414static struct queue_sysfs_entry queue_io_opt_entry = {
415 .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
416 .show = queue_io_opt_show,
417};
418
419static struct queue_sysfs_entry queue_discard_granularity_entry = {
420 .attr = {.name = "discard_granularity", .mode = S_IRUGO },
421 .show = queue_discard_granularity_show,
422};
423
424static struct queue_sysfs_entry queue_discard_max_hw_entry = {
425 .attr = {.name = "discard_max_hw_bytes", .mode = S_IRUGO },
426 .show = queue_discard_max_hw_show,
427};
428
429static struct queue_sysfs_entry queue_discard_max_entry = {
430 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO | S_IWUSR },
431 .show = queue_discard_max_show,
432 .store = queue_discard_max_store,
433};
434
435static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
436 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
437 .show = queue_discard_zeroes_data_show,
438};
439
440static struct queue_sysfs_entry queue_write_same_max_entry = {
441 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
442 .show = queue_write_same_max_show,
443};
444
445static struct queue_sysfs_entry queue_nonrot_entry = {
446 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
447 .show = queue_show_nonrot,
448 .store = queue_store_nonrot,
449};
450
451static struct queue_sysfs_entry queue_nomerges_entry = {
452 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
453 .show = queue_nomerges_show,
454 .store = queue_nomerges_store,
455};
456
457static struct queue_sysfs_entry queue_rq_affinity_entry = {
458 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
459 .show = queue_rq_affinity_show,
460 .store = queue_rq_affinity_store,
461};
462
463static struct queue_sysfs_entry queue_iostats_entry = {
464 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
465 .show = queue_show_iostats,
466 .store = queue_store_iostats,
467};
468
469static struct queue_sysfs_entry queue_random_entry = {
470 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
471 .show = queue_show_random,
472 .store = queue_store_random,
473};
474
475static struct queue_sysfs_entry queue_poll_entry = {
476 .attr = {.name = "io_poll", .mode = S_IRUGO | S_IWUSR },
477 .show = queue_poll_show,
478 .store = queue_poll_store,
479};
480
481static struct attribute *default_attrs[] = {
482 &queue_requests_entry.attr,
483 &queue_ra_entry.attr,
484 &queue_max_hw_sectors_entry.attr,
485 &queue_max_sectors_entry.attr,
486 &queue_max_segments_entry.attr,
487 &queue_max_integrity_segments_entry.attr,
488 &queue_max_segment_size_entry.attr,
489 &queue_iosched_entry.attr,
490 &queue_hw_sector_size_entry.attr,
491 &queue_logical_block_size_entry.attr,
492 &queue_physical_block_size_entry.attr,
493 &queue_io_min_entry.attr,
494 &queue_io_opt_entry.attr,
495 &queue_discard_granularity_entry.attr,
496 &queue_discard_max_entry.attr,
497 &queue_discard_max_hw_entry.attr,
498 &queue_discard_zeroes_data_entry.attr,
499 &queue_write_same_max_entry.attr,
500 &queue_nonrot_entry.attr,
501 &queue_nomerges_entry.attr,
502 &queue_rq_affinity_entry.attr,
503 &queue_iostats_entry.attr,
504 &queue_random_entry.attr,
505 &queue_poll_entry.attr,
506 NULL,
507};
508
509#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
510
511static ssize_t
512queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
513{
514 struct queue_sysfs_entry *entry = to_queue(attr);
515 struct request_queue *q =
516 container_of(kobj, struct request_queue, kobj);
517 ssize_t res;
518
519 if (!entry->show)
520 return -EIO;
521 mutex_lock(&q->sysfs_lock);
522 if (blk_queue_dying(q)) {
523 mutex_unlock(&q->sysfs_lock);
524 return -ENOENT;
525 }
526 res = entry->show(q, page);
527 mutex_unlock(&q->sysfs_lock);
528 return res;
529}
530
531static ssize_t
532queue_attr_store(struct kobject *kobj, struct attribute *attr,
533 const char *page, size_t length)
534{
535 struct queue_sysfs_entry *entry = to_queue(attr);
536 struct request_queue *q;
537 ssize_t res;
538
539 if (!entry->store)
540 return -EIO;
541
542 q = container_of(kobj, struct request_queue, kobj);
543 mutex_lock(&q->sysfs_lock);
544 if (blk_queue_dying(q)) {
545 mutex_unlock(&q->sysfs_lock);
546 return -ENOENT;
547 }
548 res = entry->store(q, page, length);
549 mutex_unlock(&q->sysfs_lock);
550 return res;
551}
552
553static void blk_free_queue_rcu(struct rcu_head *rcu_head)
554{
555 struct request_queue *q = container_of(rcu_head, struct request_queue,
556 rcu_head);
557 kmem_cache_free(blk_requestq_cachep, q);
558}
559
560/**
561 * blk_release_queue: - release a &struct request_queue when it is no longer needed
562 * @kobj: the kobj belonging to the request queue to be released
563 *
564 * Description:
565 * blk_release_queue is the pair to blk_init_queue() or
566 * blk_queue_make_request(). It should be called when a request queue is
567 * being released; typically when a block device is being de-registered.
568 * Currently, its primary task it to free all the &struct request
569 * structures that were allocated to the queue and the queue itself.
570 *
571 * Note:
572 * The low level driver must have finished any outstanding requests first
573 * via blk_cleanup_queue().
574 **/
575static void blk_release_queue(struct kobject *kobj)
576{
577 struct request_queue *q =
578 container_of(kobj, struct request_queue, kobj);
579
580 bdi_exit(&q->backing_dev_info);
581 blkcg_exit_queue(q);
582
583 if (q->elevator) {
584 spin_lock_irq(q->queue_lock);
585 ioc_clear_queue(q);
586 spin_unlock_irq(q->queue_lock);
587 elevator_exit(q->elevator);
588 }
589
590 blk_exit_rl(&q->root_rl);
591
592 if (q->queue_tags)
593 __blk_queue_free_tags(q);
594
595 if (!q->mq_ops)
596 blk_free_flush_queue(q->fq);
597 else
598 blk_mq_release(q);
599
600 blk_trace_shutdown(q);
601
602 if (q->bio_split)
603 bioset_free(q->bio_split);
604
605 ida_simple_remove(&blk_queue_ida, q->id);
606 call_rcu(&q->rcu_head, blk_free_queue_rcu);
607}
608
609static const struct sysfs_ops queue_sysfs_ops = {
610 .show = queue_attr_show,
611 .store = queue_attr_store,
612};
613
614struct kobj_type blk_queue_ktype = {
615 .sysfs_ops = &queue_sysfs_ops,
616 .default_attrs = default_attrs,
617 .release = blk_release_queue,
618};
619
620int blk_register_queue(struct gendisk *disk)
621{
622 int ret;
623 struct device *dev = disk_to_dev(disk);
624 struct request_queue *q = disk->queue;
625
626 if (WARN_ON(!q))
627 return -ENXIO;
628
629 /*
630 * SCSI probing may synchronously create and destroy a lot of
631 * request_queues for non-existent devices. Shutting down a fully
632 * functional queue takes measureable wallclock time as RCU grace
633 * periods are involved. To avoid excessive latency in these
634 * cases, a request_queue starts out in a degraded mode which is
635 * faster to shut down and is made fully functional here as
636 * request_queues for non-existent devices never get registered.
637 */
638 if (!blk_queue_init_done(q)) {
639 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
640 percpu_ref_switch_to_percpu(&q->q_usage_counter);
641 blk_queue_bypass_end(q);
642 }
643
644 ret = blk_trace_init_sysfs(dev);
645 if (ret)
646 return ret;
647
648 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
649 if (ret < 0) {
650 blk_trace_remove_sysfs(dev);
651 return ret;
652 }
653
654 kobject_uevent(&q->kobj, KOBJ_ADD);
655
656 if (q->mq_ops)
657 blk_mq_register_disk(disk);
658
659 if (!q->request_fn)
660 return 0;
661
662 ret = elv_register_queue(q);
663 if (ret) {
664 kobject_uevent(&q->kobj, KOBJ_REMOVE);
665 kobject_del(&q->kobj);
666 blk_trace_remove_sysfs(dev);
667 kobject_put(&dev->kobj);
668 return ret;
669 }
670
671 return 0;
672}
673
674void blk_unregister_queue(struct gendisk *disk)
675{
676 struct request_queue *q = disk->queue;
677
678 if (WARN_ON(!q))
679 return;
680
681 if (q->mq_ops)
682 blk_mq_unregister_disk(disk);
683
684 if (q->request_fn)
685 elv_unregister_queue(q);
686
687 kobject_uevent(&q->kobj, KOBJ_REMOVE);
688 kobject_del(&q->kobj);
689 blk_trace_remove_sysfs(disk_to_dev(disk));
690 kobject_put(&disk_to_dev(disk)->kobj);
691}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Functions related to sysfs handling
4 */
5#include <linux/kernel.h>
6#include <linux/slab.h>
7#include <linux/module.h>
8#include <linux/bio.h>
9#include <linux/blkdev.h>
10#include <linux/backing-dev.h>
11#include <linux/blktrace_api.h>
12#include <linux/blk-mq.h>
13#include <linux/blk-cgroup.h>
14#include <linux/debugfs.h>
15
16#include "blk.h"
17#include "blk-mq.h"
18#include "blk-mq-debugfs.h"
19#include "blk-wbt.h"
20
21struct queue_sysfs_entry {
22 struct attribute attr;
23 ssize_t (*show)(struct request_queue *, char *);
24 ssize_t (*store)(struct request_queue *, const char *, size_t);
25};
26
27static ssize_t
28queue_var_show(unsigned long var, char *page)
29{
30 return sprintf(page, "%lu\n", var);
31}
32
33static ssize_t
34queue_var_store(unsigned long *var, const char *page, size_t count)
35{
36 int err;
37 unsigned long v;
38
39 err = kstrtoul(page, 10, &v);
40 if (err || v > UINT_MAX)
41 return -EINVAL;
42
43 *var = v;
44
45 return count;
46}
47
48static ssize_t queue_var_store64(s64 *var, const char *page)
49{
50 int err;
51 s64 v;
52
53 err = kstrtos64(page, 10, &v);
54 if (err < 0)
55 return err;
56
57 *var = v;
58 return 0;
59}
60
61static ssize_t queue_requests_show(struct request_queue *q, char *page)
62{
63 return queue_var_show(q->nr_requests, page);
64}
65
66static ssize_t
67queue_requests_store(struct request_queue *q, const char *page, size_t count)
68{
69 unsigned long nr;
70 int ret, err;
71
72 if (!queue_is_mq(q))
73 return -EINVAL;
74
75 ret = queue_var_store(&nr, page, count);
76 if (ret < 0)
77 return ret;
78
79 if (nr < BLKDEV_MIN_RQ)
80 nr = BLKDEV_MIN_RQ;
81
82 err = blk_mq_update_nr_requests(q, nr);
83 if (err)
84 return err;
85
86 return ret;
87}
88
89static ssize_t queue_ra_show(struct request_queue *q, char *page)
90{
91 unsigned long ra_kb = q->backing_dev_info->ra_pages <<
92 (PAGE_SHIFT - 10);
93
94 return queue_var_show(ra_kb, page);
95}
96
97static ssize_t
98queue_ra_store(struct request_queue *q, const char *page, size_t count)
99{
100 unsigned long ra_kb;
101 ssize_t ret = queue_var_store(&ra_kb, page, count);
102
103 if (ret < 0)
104 return ret;
105
106 q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
107
108 return ret;
109}
110
111static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
112{
113 int max_sectors_kb = queue_max_sectors(q) >> 1;
114
115 return queue_var_show(max_sectors_kb, page);
116}
117
118static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
119{
120 return queue_var_show(queue_max_segments(q), page);
121}
122
123static ssize_t queue_max_discard_segments_show(struct request_queue *q,
124 char *page)
125{
126 return queue_var_show(queue_max_discard_segments(q), page);
127}
128
129static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
130{
131 return queue_var_show(q->limits.max_integrity_segments, page);
132}
133
134static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
135{
136 return queue_var_show(queue_max_segment_size(q), page);
137}
138
139static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
140{
141 return queue_var_show(queue_logical_block_size(q), page);
142}
143
144static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
145{
146 return queue_var_show(queue_physical_block_size(q), page);
147}
148
149static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
150{
151 return queue_var_show(q->limits.chunk_sectors, page);
152}
153
154static ssize_t queue_io_min_show(struct request_queue *q, char *page)
155{
156 return queue_var_show(queue_io_min(q), page);
157}
158
159static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
160{
161 return queue_var_show(queue_io_opt(q), page);
162}
163
164static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
165{
166 return queue_var_show(q->limits.discard_granularity, page);
167}
168
169static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
170{
171
172 return sprintf(page, "%llu\n",
173 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
174}
175
176static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
177{
178 return sprintf(page, "%llu\n",
179 (unsigned long long)q->limits.max_discard_sectors << 9);
180}
181
182static ssize_t queue_discard_max_store(struct request_queue *q,
183 const char *page, size_t count)
184{
185 unsigned long max_discard;
186 ssize_t ret = queue_var_store(&max_discard, page, count);
187
188 if (ret < 0)
189 return ret;
190
191 if (max_discard & (q->limits.discard_granularity - 1))
192 return -EINVAL;
193
194 max_discard >>= 9;
195 if (max_discard > UINT_MAX)
196 return -EINVAL;
197
198 if (max_discard > q->limits.max_hw_discard_sectors)
199 max_discard = q->limits.max_hw_discard_sectors;
200
201 q->limits.max_discard_sectors = max_discard;
202 return ret;
203}
204
205static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
206{
207 return queue_var_show(0, page);
208}
209
210static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
211{
212 return sprintf(page, "%llu\n",
213 (unsigned long long)q->limits.max_write_same_sectors << 9);
214}
215
216static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
217{
218 return sprintf(page, "%llu\n",
219 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
220}
221
222static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
223 char *page)
224{
225 return queue_var_show(queue_zone_write_granularity(q), page);
226}
227
228static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
229{
230 unsigned long long max_sectors = q->limits.max_zone_append_sectors;
231
232 return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
233}
234
235static ssize_t
236queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
237{
238 unsigned long max_sectors_kb,
239 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
240 page_kb = 1 << (PAGE_SHIFT - 10);
241 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
242
243 if (ret < 0)
244 return ret;
245
246 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
247 q->limits.max_dev_sectors >> 1);
248
249 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
250 return -EINVAL;
251
252 spin_lock_irq(&q->queue_lock);
253 q->limits.max_sectors = max_sectors_kb << 1;
254 q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
255 spin_unlock_irq(&q->queue_lock);
256
257 return ret;
258}
259
260static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
261{
262 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
263
264 return queue_var_show(max_hw_sectors_kb, page);
265}
266
267static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
268{
269 return queue_var_show(q->limits.virt_boundary_mask, page);
270}
271
272#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
273static ssize_t \
274queue_##name##_show(struct request_queue *q, char *page) \
275{ \
276 int bit; \
277 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
278 return queue_var_show(neg ? !bit : bit, page); \
279} \
280static ssize_t \
281queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
282{ \
283 unsigned long val; \
284 ssize_t ret; \
285 ret = queue_var_store(&val, page, count); \
286 if (ret < 0) \
287 return ret; \
288 if (neg) \
289 val = !val; \
290 \
291 if (val) \
292 blk_queue_flag_set(QUEUE_FLAG_##flag, q); \
293 else \
294 blk_queue_flag_clear(QUEUE_FLAG_##flag, q); \
295 return ret; \
296}
297
298QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
299QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
300QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
301QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
302#undef QUEUE_SYSFS_BIT_FNS
303
304static ssize_t queue_zoned_show(struct request_queue *q, char *page)
305{
306 switch (blk_queue_zoned_model(q)) {
307 case BLK_ZONED_HA:
308 return sprintf(page, "host-aware\n");
309 case BLK_ZONED_HM:
310 return sprintf(page, "host-managed\n");
311 default:
312 return sprintf(page, "none\n");
313 }
314}
315
316static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
317{
318 return queue_var_show(blk_queue_nr_zones(q), page);
319}
320
321static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
322{
323 return queue_var_show(queue_max_open_zones(q), page);
324}
325
326static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
327{
328 return queue_var_show(queue_max_active_zones(q), page);
329}
330
331static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
332{
333 return queue_var_show((blk_queue_nomerges(q) << 1) |
334 blk_queue_noxmerges(q), page);
335}
336
337static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
338 size_t count)
339{
340 unsigned long nm;
341 ssize_t ret = queue_var_store(&nm, page, count);
342
343 if (ret < 0)
344 return ret;
345
346 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
347 blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
348 if (nm == 2)
349 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
350 else if (nm)
351 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
352
353 return ret;
354}
355
356static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
357{
358 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
359 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
360
361 return queue_var_show(set << force, page);
362}
363
364static ssize_t
365queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
366{
367 ssize_t ret = -EINVAL;
368#ifdef CONFIG_SMP
369 unsigned long val;
370
371 ret = queue_var_store(&val, page, count);
372 if (ret < 0)
373 return ret;
374
375 if (val == 2) {
376 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
377 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
378 } else if (val == 1) {
379 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
380 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
381 } else if (val == 0) {
382 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
383 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
384 }
385#endif
386 return ret;
387}
388
389static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
390{
391 int val;
392
393 if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
394 val = BLK_MQ_POLL_CLASSIC;
395 else
396 val = q->poll_nsec / 1000;
397
398 return sprintf(page, "%d\n", val);
399}
400
401static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
402 size_t count)
403{
404 int err, val;
405
406 if (!q->mq_ops || !q->mq_ops->poll)
407 return -EINVAL;
408
409 err = kstrtoint(page, 10, &val);
410 if (err < 0)
411 return err;
412
413 if (val == BLK_MQ_POLL_CLASSIC)
414 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
415 else if (val >= 0)
416 q->poll_nsec = val * 1000;
417 else
418 return -EINVAL;
419
420 return count;
421}
422
423static ssize_t queue_poll_show(struct request_queue *q, char *page)
424{
425 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
426}
427
428static ssize_t queue_poll_store(struct request_queue *q, const char *page,
429 size_t count)
430{
431 unsigned long poll_on;
432 ssize_t ret;
433
434 if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
435 !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
436 return -EINVAL;
437
438 ret = queue_var_store(&poll_on, page, count);
439 if (ret < 0)
440 return ret;
441
442 if (poll_on) {
443 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
444 } else {
445 blk_mq_freeze_queue(q);
446 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
447 blk_mq_unfreeze_queue(q);
448 }
449
450 return ret;
451}
452
453static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
454{
455 return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
456}
457
458static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
459 size_t count)
460{
461 unsigned int val;
462 int err;
463
464 err = kstrtou32(page, 10, &val);
465 if (err || val == 0)
466 return -EINVAL;
467
468 blk_queue_rq_timeout(q, msecs_to_jiffies(val));
469
470 return count;
471}
472
473static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
474{
475 if (!wbt_rq_qos(q))
476 return -EINVAL;
477
478 return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
479}
480
481static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
482 size_t count)
483{
484 struct rq_qos *rqos;
485 ssize_t ret;
486 s64 val;
487
488 ret = queue_var_store64(&val, page);
489 if (ret < 0)
490 return ret;
491 if (val < -1)
492 return -EINVAL;
493
494 rqos = wbt_rq_qos(q);
495 if (!rqos) {
496 ret = wbt_init(q);
497 if (ret)
498 return ret;
499 }
500
501 if (val == -1)
502 val = wbt_default_latency_nsec(q);
503 else if (val >= 0)
504 val *= 1000ULL;
505
506 if (wbt_get_min_lat(q) == val)
507 return count;
508
509 /*
510 * Ensure that the queue is idled, in case the latency update
511 * ends up either enabling or disabling wbt completely. We can't
512 * have IO inflight if that happens.
513 */
514 blk_mq_freeze_queue(q);
515 blk_mq_quiesce_queue(q);
516
517 wbt_set_min_lat(q, val);
518
519 blk_mq_unquiesce_queue(q);
520 blk_mq_unfreeze_queue(q);
521
522 return count;
523}
524
525static ssize_t queue_wc_show(struct request_queue *q, char *page)
526{
527 if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
528 return sprintf(page, "write back\n");
529
530 return sprintf(page, "write through\n");
531}
532
533static ssize_t queue_wc_store(struct request_queue *q, const char *page,
534 size_t count)
535{
536 int set = -1;
537
538 if (!strncmp(page, "write back", 10))
539 set = 1;
540 else if (!strncmp(page, "write through", 13) ||
541 !strncmp(page, "none", 4))
542 set = 0;
543
544 if (set == -1)
545 return -EINVAL;
546
547 if (set)
548 blk_queue_flag_set(QUEUE_FLAG_WC, q);
549 else
550 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
551
552 return count;
553}
554
555static ssize_t queue_fua_show(struct request_queue *q, char *page)
556{
557 return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
558}
559
560static ssize_t queue_dax_show(struct request_queue *q, char *page)
561{
562 return queue_var_show(blk_queue_dax(q), page);
563}
564
565#define QUEUE_RO_ENTRY(_prefix, _name) \
566static struct queue_sysfs_entry _prefix##_entry = { \
567 .attr = { .name = _name, .mode = 0444 }, \
568 .show = _prefix##_show, \
569};
570
571#define QUEUE_RW_ENTRY(_prefix, _name) \
572static struct queue_sysfs_entry _prefix##_entry = { \
573 .attr = { .name = _name, .mode = 0644 }, \
574 .show = _prefix##_show, \
575 .store = _prefix##_store, \
576};
577
578QUEUE_RW_ENTRY(queue_requests, "nr_requests");
579QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
580QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
581QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
582QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
583QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
584QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
585QUEUE_RW_ENTRY(elv_iosched, "scheduler");
586
587QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
588QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
589QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
590QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
591QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
592
593QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
594QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
595QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
596QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
597QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
598
599QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
600QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
601QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
602QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
603
604QUEUE_RO_ENTRY(queue_zoned, "zoned");
605QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
606QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
607QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
608
609QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
610QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
611QUEUE_RW_ENTRY(queue_poll, "io_poll");
612QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
613QUEUE_RW_ENTRY(queue_wc, "write_cache");
614QUEUE_RO_ENTRY(queue_fua, "fua");
615QUEUE_RO_ENTRY(queue_dax, "dax");
616QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
617QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
618QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
619
620#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
621QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
622#endif
623
624/* legacy alias for logical_block_size: */
625static struct queue_sysfs_entry queue_hw_sector_size_entry = {
626 .attr = {.name = "hw_sector_size", .mode = 0444 },
627 .show = queue_logical_block_size_show,
628};
629
630QUEUE_RW_ENTRY(queue_nonrot, "rotational");
631QUEUE_RW_ENTRY(queue_iostats, "iostats");
632QUEUE_RW_ENTRY(queue_random, "add_random");
633QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
634
635static struct attribute *queue_attrs[] = {
636 &queue_requests_entry.attr,
637 &queue_ra_entry.attr,
638 &queue_max_hw_sectors_entry.attr,
639 &queue_max_sectors_entry.attr,
640 &queue_max_segments_entry.attr,
641 &queue_max_discard_segments_entry.attr,
642 &queue_max_integrity_segments_entry.attr,
643 &queue_max_segment_size_entry.attr,
644 &elv_iosched_entry.attr,
645 &queue_hw_sector_size_entry.attr,
646 &queue_logical_block_size_entry.attr,
647 &queue_physical_block_size_entry.attr,
648 &queue_chunk_sectors_entry.attr,
649 &queue_io_min_entry.attr,
650 &queue_io_opt_entry.attr,
651 &queue_discard_granularity_entry.attr,
652 &queue_discard_max_entry.attr,
653 &queue_discard_max_hw_entry.attr,
654 &queue_discard_zeroes_data_entry.attr,
655 &queue_write_same_max_entry.attr,
656 &queue_write_zeroes_max_entry.attr,
657 &queue_zone_append_max_entry.attr,
658 &queue_zone_write_granularity_entry.attr,
659 &queue_nonrot_entry.attr,
660 &queue_zoned_entry.attr,
661 &queue_nr_zones_entry.attr,
662 &queue_max_open_zones_entry.attr,
663 &queue_max_active_zones_entry.attr,
664 &queue_nomerges_entry.attr,
665 &queue_rq_affinity_entry.attr,
666 &queue_iostats_entry.attr,
667 &queue_stable_writes_entry.attr,
668 &queue_random_entry.attr,
669 &queue_poll_entry.attr,
670 &queue_wc_entry.attr,
671 &queue_fua_entry.attr,
672 &queue_dax_entry.attr,
673 &queue_wb_lat_entry.attr,
674 &queue_poll_delay_entry.attr,
675 &queue_io_timeout_entry.attr,
676#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
677 &blk_throtl_sample_time_entry.attr,
678#endif
679 &queue_virt_boundary_mask_entry.attr,
680 NULL,
681};
682
683static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
684 int n)
685{
686 struct request_queue *q =
687 container_of(kobj, struct request_queue, kobj);
688
689 if (attr == &queue_io_timeout_entry.attr &&
690 (!q->mq_ops || !q->mq_ops->timeout))
691 return 0;
692
693 if ((attr == &queue_max_open_zones_entry.attr ||
694 attr == &queue_max_active_zones_entry.attr) &&
695 !blk_queue_is_zoned(q))
696 return 0;
697
698 return attr->mode;
699}
700
701static struct attribute_group queue_attr_group = {
702 .attrs = queue_attrs,
703 .is_visible = queue_attr_visible,
704};
705
706
707#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
708
709static ssize_t
710queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
711{
712 struct queue_sysfs_entry *entry = to_queue(attr);
713 struct request_queue *q =
714 container_of(kobj, struct request_queue, kobj);
715 ssize_t res;
716
717 if (!entry->show)
718 return -EIO;
719 mutex_lock(&q->sysfs_lock);
720 res = entry->show(q, page);
721 mutex_unlock(&q->sysfs_lock);
722 return res;
723}
724
725static ssize_t
726queue_attr_store(struct kobject *kobj, struct attribute *attr,
727 const char *page, size_t length)
728{
729 struct queue_sysfs_entry *entry = to_queue(attr);
730 struct request_queue *q;
731 ssize_t res;
732
733 if (!entry->store)
734 return -EIO;
735
736 q = container_of(kobj, struct request_queue, kobj);
737 mutex_lock(&q->sysfs_lock);
738 res = entry->store(q, page, length);
739 mutex_unlock(&q->sysfs_lock);
740 return res;
741}
742
743static void blk_free_queue_rcu(struct rcu_head *rcu_head)
744{
745 struct request_queue *q = container_of(rcu_head, struct request_queue,
746 rcu_head);
747 kmem_cache_free(blk_requestq_cachep, q);
748}
749
750/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
751static void blk_exit_queue(struct request_queue *q)
752{
753 /*
754 * Since the I/O scheduler exit code may access cgroup information,
755 * perform I/O scheduler exit before disassociating from the block
756 * cgroup controller.
757 */
758 if (q->elevator) {
759 ioc_clear_queue(q);
760 __elevator_exit(q, q->elevator);
761 }
762
763 /*
764 * Remove all references to @q from the block cgroup controller before
765 * restoring @q->queue_lock to avoid that restoring this pointer causes
766 * e.g. blkcg_print_blkgs() to crash.
767 */
768 blkcg_exit_queue(q);
769
770 /*
771 * Since the cgroup code may dereference the @q->backing_dev_info
772 * pointer, only decrease its reference count after having removed the
773 * association with the block cgroup controller.
774 */
775 bdi_put(q->backing_dev_info);
776}
777
778/**
779 * blk_release_queue - releases all allocated resources of the request_queue
780 * @kobj: pointer to a kobject, whose container is a request_queue
781 *
782 * This function releases all allocated resources of the request queue.
783 *
784 * The struct request_queue refcount is incremented with blk_get_queue() and
785 * decremented with blk_put_queue(). Once the refcount reaches 0 this function
786 * is called.
787 *
788 * For drivers that have a request_queue on a gendisk and added with
789 * __device_add_disk() the refcount to request_queue will reach 0 with
790 * the last put_disk() called by the driver. For drivers which don't use
791 * __device_add_disk() this happens with blk_cleanup_queue().
792 *
793 * Drivers exist which depend on the release of the request_queue to be
794 * synchronous, it should not be deferred.
795 *
796 * Context: can sleep
797 */
798static void blk_release_queue(struct kobject *kobj)
799{
800 struct request_queue *q =
801 container_of(kobj, struct request_queue, kobj);
802
803 might_sleep();
804
805 if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
806 blk_stat_remove_callback(q, q->poll_cb);
807 blk_stat_free_callback(q->poll_cb);
808
809 blk_free_queue_stats(q->stats);
810
811 if (queue_is_mq(q)) {
812 struct blk_mq_hw_ctx *hctx;
813 int i;
814
815 cancel_delayed_work_sync(&q->requeue_work);
816
817 queue_for_each_hw_ctx(q, hctx, i)
818 cancel_delayed_work_sync(&hctx->run_work);
819 }
820
821 blk_exit_queue(q);
822
823 blk_queue_free_zone_bitmaps(q);
824
825 if (queue_is_mq(q))
826 blk_mq_release(q);
827
828 blk_trace_shutdown(q);
829 mutex_lock(&q->debugfs_mutex);
830 debugfs_remove_recursive(q->debugfs_dir);
831 mutex_unlock(&q->debugfs_mutex);
832
833 if (queue_is_mq(q))
834 blk_mq_debugfs_unregister(q);
835
836 bioset_exit(&q->bio_split);
837
838 ida_simple_remove(&blk_queue_ida, q->id);
839 call_rcu(&q->rcu_head, blk_free_queue_rcu);
840}
841
842static const struct sysfs_ops queue_sysfs_ops = {
843 .show = queue_attr_show,
844 .store = queue_attr_store,
845};
846
847struct kobj_type blk_queue_ktype = {
848 .sysfs_ops = &queue_sysfs_ops,
849 .release = blk_release_queue,
850};
851
852/**
853 * blk_register_queue - register a block layer queue with sysfs
854 * @disk: Disk of which the request queue should be registered with sysfs.
855 */
856int blk_register_queue(struct gendisk *disk)
857{
858 int ret;
859 struct device *dev = disk_to_dev(disk);
860 struct request_queue *q = disk->queue;
861
862 if (WARN_ON(!q))
863 return -ENXIO;
864
865 WARN_ONCE(blk_queue_registered(q),
866 "%s is registering an already registered queue\n",
867 kobject_name(&dev->kobj));
868
869 blk_queue_update_readahead(q);
870
871 ret = blk_trace_init_sysfs(dev);
872 if (ret)
873 return ret;
874
875 mutex_lock(&q->sysfs_dir_lock);
876
877 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
878 if (ret < 0) {
879 blk_trace_remove_sysfs(dev);
880 goto unlock;
881 }
882
883 ret = sysfs_create_group(&q->kobj, &queue_attr_group);
884 if (ret) {
885 blk_trace_remove_sysfs(dev);
886 kobject_del(&q->kobj);
887 kobject_put(&dev->kobj);
888 goto unlock;
889 }
890
891 mutex_lock(&q->debugfs_mutex);
892 q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
893 blk_debugfs_root);
894 mutex_unlock(&q->debugfs_mutex);
895
896 if (queue_is_mq(q)) {
897 __blk_mq_register_dev(dev, q);
898 blk_mq_debugfs_register(q);
899 }
900
901 mutex_lock(&q->sysfs_lock);
902 if (q->elevator) {
903 ret = elv_register_queue(q, false);
904 if (ret) {
905 mutex_unlock(&q->sysfs_lock);
906 mutex_unlock(&q->sysfs_dir_lock);
907 kobject_del(&q->kobj);
908 blk_trace_remove_sysfs(dev);
909 kobject_put(&dev->kobj);
910 return ret;
911 }
912 }
913
914 blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
915 wbt_enable_default(q);
916 blk_throtl_register_queue(q);
917
918 /* Now everything is ready and send out KOBJ_ADD uevent */
919 kobject_uevent(&q->kobj, KOBJ_ADD);
920 if (q->elevator)
921 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
922 mutex_unlock(&q->sysfs_lock);
923
924 ret = 0;
925unlock:
926 mutex_unlock(&q->sysfs_dir_lock);
927
928 /*
929 * SCSI probing may synchronously create and destroy a lot of
930 * request_queues for non-existent devices. Shutting down a fully
931 * functional queue takes measureable wallclock time as RCU grace
932 * periods are involved. To avoid excessive latency in these
933 * cases, a request_queue starts out in a degraded mode which is
934 * faster to shut down and is made fully functional here as
935 * request_queues for non-existent devices never get registered.
936 */
937 if (!blk_queue_init_done(q)) {
938 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
939 percpu_ref_switch_to_percpu(&q->q_usage_counter);
940 }
941
942 return ret;
943}
944EXPORT_SYMBOL_GPL(blk_register_queue);
945
946/**
947 * blk_unregister_queue - counterpart of blk_register_queue()
948 * @disk: Disk of which the request queue should be unregistered from sysfs.
949 *
950 * Note: the caller is responsible for guaranteeing that this function is called
951 * after blk_register_queue() has finished.
952 */
953void blk_unregister_queue(struct gendisk *disk)
954{
955 struct request_queue *q = disk->queue;
956
957 if (WARN_ON(!q))
958 return;
959
960 /* Return early if disk->queue was never registered. */
961 if (!blk_queue_registered(q))
962 return;
963
964 /*
965 * Since sysfs_remove_dir() prevents adding new directory entries
966 * before removal of existing entries starts, protect against
967 * concurrent elv_iosched_store() calls.
968 */
969 mutex_lock(&q->sysfs_lock);
970 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
971 mutex_unlock(&q->sysfs_lock);
972
973 mutex_lock(&q->sysfs_dir_lock);
974 /*
975 * Remove the sysfs attributes before unregistering the queue data
976 * structures that can be modified through sysfs.
977 */
978 if (queue_is_mq(q))
979 blk_mq_unregister_dev(disk_to_dev(disk), q);
980
981 kobject_uevent(&q->kobj, KOBJ_REMOVE);
982 kobject_del(&q->kobj);
983 blk_trace_remove_sysfs(disk_to_dev(disk));
984
985 mutex_lock(&q->sysfs_lock);
986 if (q->elevator)
987 elv_unregister_queue(q);
988 mutex_unlock(&q->sysfs_lock);
989 mutex_unlock(&q->sysfs_dir_lock);
990
991 kobject_put(&disk_to_dev(disk)->kobj);
992}