Loading...
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/debugfs.h>
14
15#include "blk.h"
16#include "blk-mq.h"
17#include "blk-mq-debugfs.h"
18#include "blk-mq-sched.h"
19#include "blk-wbt.h"
20#include "blk-cgroup.h"
21#include "blk-throttle.h"
22
23struct queue_sysfs_entry {
24 struct attribute attr;
25 ssize_t (*show)(struct request_queue *, char *);
26 ssize_t (*store)(struct request_queue *, const char *, size_t);
27};
28
29static ssize_t
30queue_var_show(unsigned long var, char *page)
31{
32 return sprintf(page, "%lu\n", var);
33}
34
35static ssize_t
36queue_var_store(unsigned long *var, const char *page, size_t count)
37{
38 int err;
39 unsigned long v;
40
41 err = kstrtoul(page, 10, &v);
42 if (err || v > UINT_MAX)
43 return -EINVAL;
44
45 *var = v;
46
47 return count;
48}
49
50static ssize_t queue_var_store64(s64 *var, const char *page)
51{
52 int err;
53 s64 v;
54
55 err = kstrtos64(page, 10, &v);
56 if (err < 0)
57 return err;
58
59 *var = v;
60 return 0;
61}
62
63static ssize_t queue_requests_show(struct request_queue *q, char *page)
64{
65 return queue_var_show(q->nr_requests, page);
66}
67
68static ssize_t
69queue_requests_store(struct request_queue *q, const char *page, size_t count)
70{
71 unsigned long nr;
72 int ret, err;
73
74 if (!queue_is_mq(q))
75 return -EINVAL;
76
77 ret = queue_var_store(&nr, page, count);
78 if (ret < 0)
79 return ret;
80
81 if (nr < BLKDEV_MIN_RQ)
82 nr = BLKDEV_MIN_RQ;
83
84 err = blk_mq_update_nr_requests(q, nr);
85 if (err)
86 return err;
87
88 return ret;
89}
90
91static ssize_t queue_ra_show(struct request_queue *q, char *page)
92{
93 unsigned long ra_kb;
94
95 if (!q->disk)
96 return -EINVAL;
97 ra_kb = q->disk->bdi->ra_pages << (PAGE_SHIFT - 10);
98 return queue_var_show(ra_kb, page);
99}
100
101static ssize_t
102queue_ra_store(struct request_queue *q, const char *page, size_t count)
103{
104 unsigned long ra_kb;
105 ssize_t ret;
106
107 if (!q->disk)
108 return -EINVAL;
109 ret = queue_var_store(&ra_kb, page, count);
110 if (ret < 0)
111 return ret;
112 q->disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
113 return ret;
114}
115
116static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
117{
118 int max_sectors_kb = queue_max_sectors(q) >> 1;
119
120 return queue_var_show(max_sectors_kb, page);
121}
122
123static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
124{
125 return queue_var_show(queue_max_segments(q), page);
126}
127
128static ssize_t queue_max_discard_segments_show(struct request_queue *q,
129 char *page)
130{
131 return queue_var_show(queue_max_discard_segments(q), page);
132}
133
134static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
135{
136 return queue_var_show(q->limits.max_integrity_segments, page);
137}
138
139static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
140{
141 return queue_var_show(queue_max_segment_size(q), page);
142}
143
144static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
145{
146 return queue_var_show(queue_logical_block_size(q), page);
147}
148
149static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
150{
151 return queue_var_show(queue_physical_block_size(q), page);
152}
153
154static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
155{
156 return queue_var_show(q->limits.chunk_sectors, page);
157}
158
159static ssize_t queue_io_min_show(struct request_queue *q, char *page)
160{
161 return queue_var_show(queue_io_min(q), page);
162}
163
164static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
165{
166 return queue_var_show(queue_io_opt(q), page);
167}
168
169static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
170{
171 return queue_var_show(q->limits.discard_granularity, page);
172}
173
174static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
175{
176
177 return sprintf(page, "%llu\n",
178 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
179}
180
181static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
182{
183 return sprintf(page, "%llu\n",
184 (unsigned long long)q->limits.max_discard_sectors << 9);
185}
186
187static ssize_t queue_discard_max_store(struct request_queue *q,
188 const char *page, size_t count)
189{
190 unsigned long max_discard;
191 ssize_t ret = queue_var_store(&max_discard, page, count);
192
193 if (ret < 0)
194 return ret;
195
196 if (max_discard & (q->limits.discard_granularity - 1))
197 return -EINVAL;
198
199 max_discard >>= 9;
200 if (max_discard > UINT_MAX)
201 return -EINVAL;
202
203 if (max_discard > q->limits.max_hw_discard_sectors)
204 max_discard = q->limits.max_hw_discard_sectors;
205
206 q->limits.max_discard_sectors = max_discard;
207 return ret;
208}
209
210static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
211{
212 return queue_var_show(0, page);
213}
214
215static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
216{
217 return queue_var_show(0, page);
218}
219
220static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
221{
222 return sprintf(page, "%llu\n",
223 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
224}
225
226static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
227 char *page)
228{
229 return queue_var_show(queue_zone_write_granularity(q), page);
230}
231
232static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
233{
234 unsigned long long max_sectors = q->limits.max_zone_append_sectors;
235
236 return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
237}
238
239static ssize_t
240queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
241{
242 unsigned long max_sectors_kb,
243 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
244 page_kb = 1 << (PAGE_SHIFT - 10);
245 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
246
247 if (ret < 0)
248 return ret;
249
250 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
251 q->limits.max_dev_sectors >> 1);
252
253 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
254 return -EINVAL;
255
256 spin_lock_irq(&q->queue_lock);
257 q->limits.max_sectors = max_sectors_kb << 1;
258 if (q->disk)
259 q->disk->bdi->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
260 spin_unlock_irq(&q->queue_lock);
261
262 return ret;
263}
264
265static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
266{
267 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
268
269 return queue_var_show(max_hw_sectors_kb, page);
270}
271
272static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
273{
274 return queue_var_show(q->limits.virt_boundary_mask, page);
275}
276
277static ssize_t queue_dma_alignment_show(struct request_queue *q, char *page)
278{
279 return queue_var_show(queue_dma_alignment(q), page);
280}
281
282#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
283static ssize_t \
284queue_##name##_show(struct request_queue *q, char *page) \
285{ \
286 int bit; \
287 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
288 return queue_var_show(neg ? !bit : bit, page); \
289} \
290static ssize_t \
291queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
292{ \
293 unsigned long val; \
294 ssize_t ret; \
295 ret = queue_var_store(&val, page, count); \
296 if (ret < 0) \
297 return ret; \
298 if (neg) \
299 val = !val; \
300 \
301 if (val) \
302 blk_queue_flag_set(QUEUE_FLAG_##flag, q); \
303 else \
304 blk_queue_flag_clear(QUEUE_FLAG_##flag, q); \
305 return ret; \
306}
307
308QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
309QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
310QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
311QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
312#undef QUEUE_SYSFS_BIT_FNS
313
314static ssize_t queue_zoned_show(struct request_queue *q, char *page)
315{
316 switch (blk_queue_zoned_model(q)) {
317 case BLK_ZONED_HA:
318 return sprintf(page, "host-aware\n");
319 case BLK_ZONED_HM:
320 return sprintf(page, "host-managed\n");
321 default:
322 return sprintf(page, "none\n");
323 }
324}
325
326static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
327{
328 return queue_var_show(disk_nr_zones(q->disk), page);
329}
330
331static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
332{
333 return queue_var_show(bdev_max_open_zones(q->disk->part0), page);
334}
335
336static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
337{
338 return queue_var_show(bdev_max_active_zones(q->disk->part0), page);
339}
340
341static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
342{
343 return queue_var_show((blk_queue_nomerges(q) << 1) |
344 blk_queue_noxmerges(q), page);
345}
346
347static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
348 size_t count)
349{
350 unsigned long nm;
351 ssize_t ret = queue_var_store(&nm, page, count);
352
353 if (ret < 0)
354 return ret;
355
356 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
357 blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
358 if (nm == 2)
359 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
360 else if (nm)
361 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
362
363 return ret;
364}
365
366static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
367{
368 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
369 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
370
371 return queue_var_show(set << force, page);
372}
373
374static ssize_t
375queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
376{
377 ssize_t ret = -EINVAL;
378#ifdef CONFIG_SMP
379 unsigned long val;
380
381 ret = queue_var_store(&val, page, count);
382 if (ret < 0)
383 return ret;
384
385 if (val == 2) {
386 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
387 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
388 } else if (val == 1) {
389 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
390 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
391 } else if (val == 0) {
392 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
393 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
394 }
395#endif
396 return ret;
397}
398
399static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
400{
401 int val;
402
403 if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
404 val = BLK_MQ_POLL_CLASSIC;
405 else
406 val = q->poll_nsec / 1000;
407
408 return sprintf(page, "%d\n", val);
409}
410
411static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
412 size_t count)
413{
414 int err, val;
415
416 if (!q->mq_ops || !q->mq_ops->poll)
417 return -EINVAL;
418
419 err = kstrtoint(page, 10, &val);
420 if (err < 0)
421 return err;
422
423 if (val == BLK_MQ_POLL_CLASSIC)
424 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
425 else if (val >= 0)
426 q->poll_nsec = val * 1000;
427 else
428 return -EINVAL;
429
430 return count;
431}
432
433static ssize_t queue_poll_show(struct request_queue *q, char *page)
434{
435 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
436}
437
438static ssize_t queue_poll_store(struct request_queue *q, const char *page,
439 size_t count)
440{
441 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
442 return -EINVAL;
443 pr_info_ratelimited("writes to the poll attribute are ignored.\n");
444 pr_info_ratelimited("please use driver specific parameters instead.\n");
445 return count;
446}
447
448static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
449{
450 return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
451}
452
453static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
454 size_t count)
455{
456 unsigned int val;
457 int err;
458
459 err = kstrtou32(page, 10, &val);
460 if (err || val == 0)
461 return -EINVAL;
462
463 blk_queue_rq_timeout(q, msecs_to_jiffies(val));
464
465 return count;
466}
467
468static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
469{
470 if (!wbt_rq_qos(q))
471 return -EINVAL;
472
473 if (wbt_disabled(q))
474 return sprintf(page, "0\n");
475
476 return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
477}
478
479static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
480 size_t count)
481{
482 struct rq_qos *rqos;
483 ssize_t ret;
484 s64 val;
485
486 ret = queue_var_store64(&val, page);
487 if (ret < 0)
488 return ret;
489 if (val < -1)
490 return -EINVAL;
491
492 rqos = wbt_rq_qos(q);
493 if (!rqos) {
494 ret = wbt_init(q);
495 if (ret)
496 return ret;
497 }
498
499 if (val == -1)
500 val = wbt_default_latency_nsec(q);
501 else if (val >= 0)
502 val *= 1000ULL;
503
504 if (wbt_get_min_lat(q) == val)
505 return count;
506
507 /*
508 * Ensure that the queue is idled, in case the latency update
509 * ends up either enabling or disabling wbt completely. We can't
510 * have IO inflight if that happens.
511 */
512 blk_mq_freeze_queue(q);
513 blk_mq_quiesce_queue(q);
514
515 wbt_set_min_lat(q, val);
516
517 blk_mq_unquiesce_queue(q);
518 blk_mq_unfreeze_queue(q);
519
520 return count;
521}
522
523static ssize_t queue_wc_show(struct request_queue *q, char *page)
524{
525 if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
526 return sprintf(page, "write back\n");
527
528 return sprintf(page, "write through\n");
529}
530
531static ssize_t queue_wc_store(struct request_queue *q, const char *page,
532 size_t count)
533{
534 int set = -1;
535
536 if (!strncmp(page, "write back", 10))
537 set = 1;
538 else if (!strncmp(page, "write through", 13) ||
539 !strncmp(page, "none", 4))
540 set = 0;
541
542 if (set == -1)
543 return -EINVAL;
544
545 if (set)
546 blk_queue_flag_set(QUEUE_FLAG_WC, q);
547 else
548 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
549
550 return count;
551}
552
553static ssize_t queue_fua_show(struct request_queue *q, char *page)
554{
555 return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
556}
557
558static ssize_t queue_dax_show(struct request_queue *q, char *page)
559{
560 return queue_var_show(blk_queue_dax(q), page);
561}
562
563#define QUEUE_RO_ENTRY(_prefix, _name) \
564static struct queue_sysfs_entry _prefix##_entry = { \
565 .attr = { .name = _name, .mode = 0444 }, \
566 .show = _prefix##_show, \
567};
568
569#define QUEUE_RW_ENTRY(_prefix, _name) \
570static struct queue_sysfs_entry _prefix##_entry = { \
571 .attr = { .name = _name, .mode = 0644 }, \
572 .show = _prefix##_show, \
573 .store = _prefix##_store, \
574};
575
576QUEUE_RW_ENTRY(queue_requests, "nr_requests");
577QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
578QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
579QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
580QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
581QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
582QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
583QUEUE_RW_ENTRY(elv_iosched, "scheduler");
584
585QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
586QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
587QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
588QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
589QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
590
591QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
592QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
593QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
594QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
595QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
596
597QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
598QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
599QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
600QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
601
602QUEUE_RO_ENTRY(queue_zoned, "zoned");
603QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
604QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
605QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
606
607QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
608QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
609QUEUE_RW_ENTRY(queue_poll, "io_poll");
610QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
611QUEUE_RW_ENTRY(queue_wc, "write_cache");
612QUEUE_RO_ENTRY(queue_fua, "fua");
613QUEUE_RO_ENTRY(queue_dax, "dax");
614QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
615QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
616QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
617QUEUE_RO_ENTRY(queue_dma_alignment, "dma_alignment");
618
619#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
620QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
621#endif
622
623/* legacy alias for logical_block_size: */
624static struct queue_sysfs_entry queue_hw_sector_size_entry = {
625 .attr = {.name = "hw_sector_size", .mode = 0444 },
626 .show = queue_logical_block_size_show,
627};
628
629QUEUE_RW_ENTRY(queue_nonrot, "rotational");
630QUEUE_RW_ENTRY(queue_iostats, "iostats");
631QUEUE_RW_ENTRY(queue_random, "add_random");
632QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
633
634static struct attribute *queue_attrs[] = {
635 &queue_requests_entry.attr,
636 &queue_ra_entry.attr,
637 &queue_max_hw_sectors_entry.attr,
638 &queue_max_sectors_entry.attr,
639 &queue_max_segments_entry.attr,
640 &queue_max_discard_segments_entry.attr,
641 &queue_max_integrity_segments_entry.attr,
642 &queue_max_segment_size_entry.attr,
643 &elv_iosched_entry.attr,
644 &queue_hw_sector_size_entry.attr,
645 &queue_logical_block_size_entry.attr,
646 &queue_physical_block_size_entry.attr,
647 &queue_chunk_sectors_entry.attr,
648 &queue_io_min_entry.attr,
649 &queue_io_opt_entry.attr,
650 &queue_discard_granularity_entry.attr,
651 &queue_discard_max_entry.attr,
652 &queue_discard_max_hw_entry.attr,
653 &queue_discard_zeroes_data_entry.attr,
654 &queue_write_same_max_entry.attr,
655 &queue_write_zeroes_max_entry.attr,
656 &queue_zone_append_max_entry.attr,
657 &queue_zone_write_granularity_entry.attr,
658 &queue_nonrot_entry.attr,
659 &queue_zoned_entry.attr,
660 &queue_nr_zones_entry.attr,
661 &queue_max_open_zones_entry.attr,
662 &queue_max_active_zones_entry.attr,
663 &queue_nomerges_entry.attr,
664 &queue_rq_affinity_entry.attr,
665 &queue_iostats_entry.attr,
666 &queue_stable_writes_entry.attr,
667 &queue_random_entry.attr,
668 &queue_poll_entry.attr,
669 &queue_wc_entry.attr,
670 &queue_fua_entry.attr,
671 &queue_dax_entry.attr,
672 &queue_wb_lat_entry.attr,
673 &queue_poll_delay_entry.attr,
674 &queue_io_timeout_entry.attr,
675#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
676 &blk_throtl_sample_time_entry.attr,
677#endif
678 &queue_virt_boundary_mask_entry.attr,
679 &queue_dma_alignment_entry.attr,
680 NULL,
681};
682
683static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
684 int n)
685{
686 struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
687 struct request_queue *q = disk->queue;
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 gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
714 struct request_queue *q = disk->queue;
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 gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
731 struct request_queue *q = disk->queue;
732 ssize_t res;
733
734 if (!entry->store)
735 return -EIO;
736
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 const struct sysfs_ops queue_sysfs_ops = {
744 .show = queue_attr_show,
745 .store = queue_attr_store,
746};
747
748static const struct attribute_group *blk_queue_attr_groups[] = {
749 &queue_attr_group,
750 NULL
751};
752
753static void blk_queue_release(struct kobject *kobj)
754{
755 /* nothing to do here, all data is associated with the parent gendisk */
756}
757
758static struct kobj_type blk_queue_ktype = {
759 .default_groups = blk_queue_attr_groups,
760 .sysfs_ops = &queue_sysfs_ops,
761 .release = blk_queue_release,
762};
763
764static void blk_debugfs_remove(struct gendisk *disk)
765{
766 struct request_queue *q = disk->queue;
767
768 mutex_lock(&q->debugfs_mutex);
769 blk_trace_shutdown(q);
770 debugfs_remove_recursive(q->debugfs_dir);
771 q->debugfs_dir = NULL;
772 q->sched_debugfs_dir = NULL;
773 q->rqos_debugfs_dir = NULL;
774 mutex_unlock(&q->debugfs_mutex);
775}
776
777/**
778 * blk_register_queue - register a block layer queue with sysfs
779 * @disk: Disk of which the request queue should be registered with sysfs.
780 */
781int blk_register_queue(struct gendisk *disk)
782{
783 struct request_queue *q = disk->queue;
784 int ret;
785
786 mutex_lock(&q->sysfs_dir_lock);
787 kobject_init(&disk->queue_kobj, &blk_queue_ktype);
788 ret = kobject_add(&disk->queue_kobj, &disk_to_dev(disk)->kobj, "queue");
789 if (ret < 0)
790 goto out_put_queue_kobj;
791
792 if (queue_is_mq(q)) {
793 ret = blk_mq_sysfs_register(disk);
794 if (ret)
795 goto out_put_queue_kobj;
796 }
797 mutex_lock(&q->sysfs_lock);
798
799 mutex_lock(&q->debugfs_mutex);
800 q->debugfs_dir = debugfs_create_dir(disk->disk_name, blk_debugfs_root);
801 if (queue_is_mq(q))
802 blk_mq_debugfs_register(q);
803 mutex_unlock(&q->debugfs_mutex);
804
805 ret = disk_register_independent_access_ranges(disk);
806 if (ret)
807 goto out_debugfs_remove;
808
809 if (q->elevator) {
810 ret = elv_register_queue(q, false);
811 if (ret)
812 goto out_unregister_ia_ranges;
813 }
814
815 ret = blk_crypto_sysfs_register(disk);
816 if (ret)
817 goto out_elv_unregister;
818
819 blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
820 wbt_enable_default(q);
821 blk_throtl_register(disk);
822
823 /* Now everything is ready and send out KOBJ_ADD uevent */
824 kobject_uevent(&disk->queue_kobj, KOBJ_ADD);
825 if (q->elevator)
826 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
827 mutex_unlock(&q->sysfs_lock);
828 mutex_unlock(&q->sysfs_dir_lock);
829
830 /*
831 * SCSI probing may synchronously create and destroy a lot of
832 * request_queues for non-existent devices. Shutting down a fully
833 * functional queue takes measureable wallclock time as RCU grace
834 * periods are involved. To avoid excessive latency in these
835 * cases, a request_queue starts out in a degraded mode which is
836 * faster to shut down and is made fully functional here as
837 * request_queues for non-existent devices never get registered.
838 */
839 if (!blk_queue_init_done(q)) {
840 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
841 percpu_ref_switch_to_percpu(&q->q_usage_counter);
842 }
843
844 return ret;
845
846out_elv_unregister:
847 elv_unregister_queue(q);
848out_unregister_ia_ranges:
849 disk_unregister_independent_access_ranges(disk);
850out_debugfs_remove:
851 blk_debugfs_remove(disk);
852 mutex_unlock(&q->sysfs_lock);
853out_put_queue_kobj:
854 kobject_put(&disk->queue_kobj);
855 mutex_unlock(&q->sysfs_dir_lock);
856 return ret;
857}
858
859/**
860 * blk_unregister_queue - counterpart of blk_register_queue()
861 * @disk: Disk of which the request queue should be unregistered from sysfs.
862 *
863 * Note: the caller is responsible for guaranteeing that this function is called
864 * after blk_register_queue() has finished.
865 */
866void blk_unregister_queue(struct gendisk *disk)
867{
868 struct request_queue *q = disk->queue;
869
870 if (WARN_ON(!q))
871 return;
872
873 /* Return early if disk->queue was never registered. */
874 if (!blk_queue_registered(q))
875 return;
876
877 /*
878 * Since sysfs_remove_dir() prevents adding new directory entries
879 * before removal of existing entries starts, protect against
880 * concurrent elv_iosched_store() calls.
881 */
882 mutex_lock(&q->sysfs_lock);
883 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
884 mutex_unlock(&q->sysfs_lock);
885
886 mutex_lock(&q->sysfs_dir_lock);
887 /*
888 * Remove the sysfs attributes before unregistering the queue data
889 * structures that can be modified through sysfs.
890 */
891 if (queue_is_mq(q))
892 blk_mq_sysfs_unregister(disk);
893 blk_crypto_sysfs_unregister(disk);
894
895 mutex_lock(&q->sysfs_lock);
896 elv_unregister_queue(q);
897 disk_unregister_independent_access_ranges(disk);
898 mutex_unlock(&q->sysfs_lock);
899
900 /* Now that we've deleted all child objects, we can delete the queue. */
901 kobject_uevent(&disk->queue_kobj, KOBJ_REMOVE);
902 kobject_del(&disk->queue_kobj);
903 mutex_unlock(&q->sysfs_dir_lock);
904
905 blk_debugfs_remove(disk);
906}
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/blktrace_api.h>
10#include <linux/blk-mq.h>
11
12#include "blk.h"
13#include "blk-cgroup.h"
14#include "blk-mq.h"
15
16struct queue_sysfs_entry {
17 struct attribute attr;
18 ssize_t (*show)(struct request_queue *, char *);
19 ssize_t (*store)(struct request_queue *, const char *, size_t);
20};
21
22static ssize_t
23queue_var_show(unsigned long var, char *page)
24{
25 return sprintf(page, "%lu\n", var);
26}
27
28static ssize_t
29queue_var_store(unsigned long *var, const char *page, size_t count)
30{
31 int err;
32 unsigned long v;
33
34 err = kstrtoul(page, 10, &v);
35 if (err || v > UINT_MAX)
36 return -EINVAL;
37
38 *var = v;
39
40 return count;
41}
42
43static ssize_t queue_requests_show(struct request_queue *q, char *page)
44{
45 return queue_var_show(q->nr_requests, (page));
46}
47
48static ssize_t
49queue_requests_store(struct request_queue *q, const char *page, size_t count)
50{
51 struct request_list *rl;
52 unsigned long nr;
53 int ret;
54
55 if (!q->request_fn)
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 spin_lock_irq(q->queue_lock);
66 q->nr_requests = nr;
67 blk_queue_congestion_threshold(q);
68
69 /* congestion isn't cgroup aware and follows root blkcg for now */
70 rl = &q->root_rl;
71
72 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
73 blk_set_queue_congested(q, BLK_RW_SYNC);
74 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
75 blk_clear_queue_congested(q, BLK_RW_SYNC);
76
77 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
78 blk_set_queue_congested(q, BLK_RW_ASYNC);
79 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
80 blk_clear_queue_congested(q, BLK_RW_ASYNC);
81
82 blk_queue_for_each_rl(rl, q) {
83 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
84 blk_set_rl_full(rl, BLK_RW_SYNC);
85 } else {
86 blk_clear_rl_full(rl, BLK_RW_SYNC);
87 wake_up(&rl->wait[BLK_RW_SYNC]);
88 }
89
90 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
91 blk_set_rl_full(rl, BLK_RW_ASYNC);
92 } else {
93 blk_clear_rl_full(rl, BLK_RW_ASYNC);
94 wake_up(&rl->wait[BLK_RW_ASYNC]);
95 }
96 }
97
98 spin_unlock_irq(q->queue_lock);
99 return ret;
100}
101
102static ssize_t queue_ra_show(struct request_queue *q, char *page)
103{
104 unsigned long ra_kb = q->backing_dev_info.ra_pages <<
105 (PAGE_CACHE_SHIFT - 10);
106
107 return queue_var_show(ra_kb, (page));
108}
109
110static ssize_t
111queue_ra_store(struct request_queue *q, const char *page, size_t count)
112{
113 unsigned long ra_kb;
114 ssize_t ret = queue_var_store(&ra_kb, page, count);
115
116 if (ret < 0)
117 return ret;
118
119 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
120
121 return ret;
122}
123
124static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
125{
126 int max_sectors_kb = queue_max_sectors(q) >> 1;
127
128 return queue_var_show(max_sectors_kb, (page));
129}
130
131static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
132{
133 return queue_var_show(queue_max_segments(q), (page));
134}
135
136static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
137{
138 return queue_var_show(q->limits.max_integrity_segments, (page));
139}
140
141static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
142{
143 if (blk_queue_cluster(q))
144 return queue_var_show(queue_max_segment_size(q), (page));
145
146 return queue_var_show(PAGE_CACHE_SIZE, (page));
147}
148
149static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
150{
151 return queue_var_show(queue_logical_block_size(q), page);
152}
153
154static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
155{
156 return queue_var_show(queue_physical_block_size(q), page);
157}
158
159static ssize_t queue_io_min_show(struct request_queue *q, char *page)
160{
161 return queue_var_show(queue_io_min(q), page);
162}
163
164static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
165{
166 return queue_var_show(queue_io_opt(q), page);
167}
168
169static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
170{
171 return queue_var_show(q->limits.discard_granularity, page);
172}
173
174static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
175{
176 return sprintf(page, "%llu\n",
177 (unsigned long long)q->limits.max_discard_sectors << 9);
178}
179
180static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
181{
182 return queue_var_show(queue_discard_zeroes_data(q), page);
183}
184
185static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
186{
187 return sprintf(page, "%llu\n",
188 (unsigned long long)q->limits.max_write_same_sectors << 9);
189}
190
191
192static ssize_t
193queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
194{
195 unsigned long max_sectors_kb,
196 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
197 page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
198 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
199
200 if (ret < 0)
201 return ret;
202
203 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
204 return -EINVAL;
205
206 spin_lock_irq(q->queue_lock);
207 q->limits.max_sectors = max_sectors_kb << 1;
208 spin_unlock_irq(q->queue_lock);
209
210 return ret;
211}
212
213static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
214{
215 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
216
217 return queue_var_show(max_hw_sectors_kb, (page));
218}
219
220#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
221static ssize_t \
222queue_show_##name(struct request_queue *q, char *page) \
223{ \
224 int bit; \
225 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
226 return queue_var_show(neg ? !bit : bit, page); \
227} \
228static ssize_t \
229queue_store_##name(struct request_queue *q, const char *page, size_t count) \
230{ \
231 unsigned long val; \
232 ssize_t ret; \
233 ret = queue_var_store(&val, page, count); \
234 if (ret < 0) \
235 return ret; \
236 if (neg) \
237 val = !val; \
238 \
239 spin_lock_irq(q->queue_lock); \
240 if (val) \
241 queue_flag_set(QUEUE_FLAG_##flag, q); \
242 else \
243 queue_flag_clear(QUEUE_FLAG_##flag, q); \
244 spin_unlock_irq(q->queue_lock); \
245 return ret; \
246}
247
248QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
249QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
250QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
251#undef QUEUE_SYSFS_BIT_FNS
252
253static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
254{
255 return queue_var_show((blk_queue_nomerges(q) << 1) |
256 blk_queue_noxmerges(q), page);
257}
258
259static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
260 size_t count)
261{
262 unsigned long nm;
263 ssize_t ret = queue_var_store(&nm, page, count);
264
265 if (ret < 0)
266 return ret;
267
268 spin_lock_irq(q->queue_lock);
269 queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
270 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
271 if (nm == 2)
272 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
273 else if (nm)
274 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
275 spin_unlock_irq(q->queue_lock);
276
277 return ret;
278}
279
280static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
281{
282 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
283 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
284
285 return queue_var_show(set << force, page);
286}
287
288static ssize_t
289queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
290{
291 ssize_t ret = -EINVAL;
292#ifdef CONFIG_SMP
293 unsigned long val;
294
295 ret = queue_var_store(&val, page, count);
296 if (ret < 0)
297 return ret;
298
299 spin_lock_irq(q->queue_lock);
300 if (val == 2) {
301 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
302 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
303 } else if (val == 1) {
304 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
305 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
306 } else if (val == 0) {
307 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
308 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
309 }
310 spin_unlock_irq(q->queue_lock);
311#endif
312 return ret;
313}
314
315static struct queue_sysfs_entry queue_requests_entry = {
316 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
317 .show = queue_requests_show,
318 .store = queue_requests_store,
319};
320
321static struct queue_sysfs_entry queue_ra_entry = {
322 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
323 .show = queue_ra_show,
324 .store = queue_ra_store,
325};
326
327static struct queue_sysfs_entry queue_max_sectors_entry = {
328 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
329 .show = queue_max_sectors_show,
330 .store = queue_max_sectors_store,
331};
332
333static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
334 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
335 .show = queue_max_hw_sectors_show,
336};
337
338static struct queue_sysfs_entry queue_max_segments_entry = {
339 .attr = {.name = "max_segments", .mode = S_IRUGO },
340 .show = queue_max_segments_show,
341};
342
343static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
344 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
345 .show = queue_max_integrity_segments_show,
346};
347
348static struct queue_sysfs_entry queue_max_segment_size_entry = {
349 .attr = {.name = "max_segment_size", .mode = S_IRUGO },
350 .show = queue_max_segment_size_show,
351};
352
353static struct queue_sysfs_entry queue_iosched_entry = {
354 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
355 .show = elv_iosched_show,
356 .store = elv_iosched_store,
357};
358
359static struct queue_sysfs_entry queue_hw_sector_size_entry = {
360 .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
361 .show = queue_logical_block_size_show,
362};
363
364static struct queue_sysfs_entry queue_logical_block_size_entry = {
365 .attr = {.name = "logical_block_size", .mode = S_IRUGO },
366 .show = queue_logical_block_size_show,
367};
368
369static struct queue_sysfs_entry queue_physical_block_size_entry = {
370 .attr = {.name = "physical_block_size", .mode = S_IRUGO },
371 .show = queue_physical_block_size_show,
372};
373
374static struct queue_sysfs_entry queue_io_min_entry = {
375 .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
376 .show = queue_io_min_show,
377};
378
379static struct queue_sysfs_entry queue_io_opt_entry = {
380 .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
381 .show = queue_io_opt_show,
382};
383
384static struct queue_sysfs_entry queue_discard_granularity_entry = {
385 .attr = {.name = "discard_granularity", .mode = S_IRUGO },
386 .show = queue_discard_granularity_show,
387};
388
389static struct queue_sysfs_entry queue_discard_max_entry = {
390 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO },
391 .show = queue_discard_max_show,
392};
393
394static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
395 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
396 .show = queue_discard_zeroes_data_show,
397};
398
399static struct queue_sysfs_entry queue_write_same_max_entry = {
400 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
401 .show = queue_write_same_max_show,
402};
403
404static struct queue_sysfs_entry queue_nonrot_entry = {
405 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
406 .show = queue_show_nonrot,
407 .store = queue_store_nonrot,
408};
409
410static struct queue_sysfs_entry queue_nomerges_entry = {
411 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
412 .show = queue_nomerges_show,
413 .store = queue_nomerges_store,
414};
415
416static struct queue_sysfs_entry queue_rq_affinity_entry = {
417 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
418 .show = queue_rq_affinity_show,
419 .store = queue_rq_affinity_store,
420};
421
422static struct queue_sysfs_entry queue_iostats_entry = {
423 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
424 .show = queue_show_iostats,
425 .store = queue_store_iostats,
426};
427
428static struct queue_sysfs_entry queue_random_entry = {
429 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
430 .show = queue_show_random,
431 .store = queue_store_random,
432};
433
434static struct attribute *default_attrs[] = {
435 &queue_requests_entry.attr,
436 &queue_ra_entry.attr,
437 &queue_max_hw_sectors_entry.attr,
438 &queue_max_sectors_entry.attr,
439 &queue_max_segments_entry.attr,
440 &queue_max_integrity_segments_entry.attr,
441 &queue_max_segment_size_entry.attr,
442 &queue_iosched_entry.attr,
443 &queue_hw_sector_size_entry.attr,
444 &queue_logical_block_size_entry.attr,
445 &queue_physical_block_size_entry.attr,
446 &queue_io_min_entry.attr,
447 &queue_io_opt_entry.attr,
448 &queue_discard_granularity_entry.attr,
449 &queue_discard_max_entry.attr,
450 &queue_discard_zeroes_data_entry.attr,
451 &queue_write_same_max_entry.attr,
452 &queue_nonrot_entry.attr,
453 &queue_nomerges_entry.attr,
454 &queue_rq_affinity_entry.attr,
455 &queue_iostats_entry.attr,
456 &queue_random_entry.attr,
457 NULL,
458};
459
460#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
461
462static ssize_t
463queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
464{
465 struct queue_sysfs_entry *entry = to_queue(attr);
466 struct request_queue *q =
467 container_of(kobj, struct request_queue, kobj);
468 ssize_t res;
469
470 if (!entry->show)
471 return -EIO;
472 mutex_lock(&q->sysfs_lock);
473 if (blk_queue_dying(q)) {
474 mutex_unlock(&q->sysfs_lock);
475 return -ENOENT;
476 }
477 res = entry->show(q, page);
478 mutex_unlock(&q->sysfs_lock);
479 return res;
480}
481
482static ssize_t
483queue_attr_store(struct kobject *kobj, struct attribute *attr,
484 const char *page, size_t length)
485{
486 struct queue_sysfs_entry *entry = to_queue(attr);
487 struct request_queue *q;
488 ssize_t res;
489
490 if (!entry->store)
491 return -EIO;
492
493 q = container_of(kobj, struct request_queue, kobj);
494 mutex_lock(&q->sysfs_lock);
495 if (blk_queue_dying(q)) {
496 mutex_unlock(&q->sysfs_lock);
497 return -ENOENT;
498 }
499 res = entry->store(q, page, length);
500 mutex_unlock(&q->sysfs_lock);
501 return res;
502}
503
504static void blk_free_queue_rcu(struct rcu_head *rcu_head)
505{
506 struct request_queue *q = container_of(rcu_head, struct request_queue,
507 rcu_head);
508 kmem_cache_free(blk_requestq_cachep, q);
509}
510
511/**
512 * blk_release_queue: - release a &struct request_queue when it is no longer needed
513 * @kobj: the kobj belonging to the request queue to be released
514 *
515 * Description:
516 * blk_release_queue is the pair to blk_init_queue() or
517 * blk_queue_make_request(). It should be called when a request queue is
518 * being released; typically when a block device is being de-registered.
519 * Currently, its primary task it to free all the &struct request
520 * structures that were allocated to the queue and the queue itself.
521 *
522 * Caveat:
523 * Hopefully the low level driver will have finished any
524 * outstanding requests first...
525 **/
526static void blk_release_queue(struct kobject *kobj)
527{
528 struct request_queue *q =
529 container_of(kobj, struct request_queue, kobj);
530
531 blk_sync_queue(q);
532
533 blkcg_exit_queue(q);
534
535 if (q->elevator) {
536 spin_lock_irq(q->queue_lock);
537 ioc_clear_queue(q);
538 spin_unlock_irq(q->queue_lock);
539 elevator_exit(q->elevator);
540 }
541
542 blk_exit_rl(&q->root_rl);
543
544 if (q->queue_tags)
545 __blk_queue_free_tags(q);
546
547 percpu_counter_destroy(&q->mq_usage_counter);
548
549 if (q->mq_ops)
550 blk_mq_free_queue(q);
551
552 kfree(q->flush_rq);
553
554 blk_trace_shutdown(q);
555
556 bdi_destroy(&q->backing_dev_info);
557
558 ida_simple_remove(&blk_queue_ida, q->id);
559 call_rcu(&q->rcu_head, blk_free_queue_rcu);
560}
561
562static const struct sysfs_ops queue_sysfs_ops = {
563 .show = queue_attr_show,
564 .store = queue_attr_store,
565};
566
567struct kobj_type blk_queue_ktype = {
568 .sysfs_ops = &queue_sysfs_ops,
569 .default_attrs = default_attrs,
570 .release = blk_release_queue,
571};
572
573int blk_register_queue(struct gendisk *disk)
574{
575 int ret;
576 struct device *dev = disk_to_dev(disk);
577 struct request_queue *q = disk->queue;
578
579 if (WARN_ON(!q))
580 return -ENXIO;
581
582 /*
583 * Initialization must be complete by now. Finish the initial
584 * bypass from queue allocation.
585 */
586 blk_queue_bypass_end(q);
587 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
588
589 ret = blk_trace_init_sysfs(dev);
590 if (ret)
591 return ret;
592
593 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
594 if (ret < 0) {
595 blk_trace_remove_sysfs(dev);
596 return ret;
597 }
598
599 kobject_uevent(&q->kobj, KOBJ_ADD);
600
601 if (q->mq_ops)
602 blk_mq_register_disk(disk);
603
604 if (!q->request_fn)
605 return 0;
606
607 ret = elv_register_queue(q);
608 if (ret) {
609 kobject_uevent(&q->kobj, KOBJ_REMOVE);
610 kobject_del(&q->kobj);
611 blk_trace_remove_sysfs(dev);
612 kobject_put(&dev->kobj);
613 return ret;
614 }
615
616 return 0;
617}
618
619void blk_unregister_queue(struct gendisk *disk)
620{
621 struct request_queue *q = disk->queue;
622
623 if (WARN_ON(!q))
624 return;
625
626 if (q->mq_ops)
627 blk_mq_unregister_disk(disk);
628
629 if (q->request_fn)
630 elv_unregister_queue(q);
631
632 kobject_uevent(&q->kobj, KOBJ_REMOVE);
633 kobject_del(&q->kobj);
634 blk_trace_remove_sysfs(disk_to_dev(disk));
635 kobject_put(&disk_to_dev(disk)->kobj);
636}