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// 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_append_max_show(struct request_queue *q, char *page)
223{
224 unsigned long long max_sectors = q->limits.max_zone_append_sectors;
225
226 return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
227}
228
229static ssize_t
230queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
231{
232 unsigned long max_sectors_kb,
233 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
234 page_kb = 1 << (PAGE_SHIFT - 10);
235 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
236
237 if (ret < 0)
238 return ret;
239
240 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
241 q->limits.max_dev_sectors >> 1);
242
243 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
244 return -EINVAL;
245
246 spin_lock_irq(&q->queue_lock);
247 q->limits.max_sectors = max_sectors_kb << 1;
248 q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
249 spin_unlock_irq(&q->queue_lock);
250
251 return ret;
252}
253
254static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
255{
256 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
257
258 return queue_var_show(max_hw_sectors_kb, (page));
259}
260
261#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
262static ssize_t \
263queue_show_##name(struct request_queue *q, char *page) \
264{ \
265 int bit; \
266 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
267 return queue_var_show(neg ? !bit : bit, page); \
268} \
269static ssize_t \
270queue_store_##name(struct request_queue *q, const char *page, size_t count) \
271{ \
272 unsigned long val; \
273 ssize_t ret; \
274 ret = queue_var_store(&val, page, count); \
275 if (ret < 0) \
276 return ret; \
277 if (neg) \
278 val = !val; \
279 \
280 if (val) \
281 blk_queue_flag_set(QUEUE_FLAG_##flag, q); \
282 else \
283 blk_queue_flag_clear(QUEUE_FLAG_##flag, q); \
284 return ret; \
285}
286
287QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
288QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
289QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
290#undef QUEUE_SYSFS_BIT_FNS
291
292static ssize_t queue_zoned_show(struct request_queue *q, char *page)
293{
294 switch (blk_queue_zoned_model(q)) {
295 case BLK_ZONED_HA:
296 return sprintf(page, "host-aware\n");
297 case BLK_ZONED_HM:
298 return sprintf(page, "host-managed\n");
299 default:
300 return sprintf(page, "none\n");
301 }
302}
303
304static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
305{
306 return queue_var_show(blk_queue_nr_zones(q), page);
307}
308
309static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
310{
311 return queue_var_show(queue_max_open_zones(q), page);
312}
313
314static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
315{
316 return queue_var_show(queue_max_active_zones(q), page);
317}
318
319static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
320{
321 return queue_var_show((blk_queue_nomerges(q) << 1) |
322 blk_queue_noxmerges(q), page);
323}
324
325static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
326 size_t count)
327{
328 unsigned long nm;
329 ssize_t ret = queue_var_store(&nm, page, count);
330
331 if (ret < 0)
332 return ret;
333
334 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
335 blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
336 if (nm == 2)
337 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
338 else if (nm)
339 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
340
341 return ret;
342}
343
344static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
345{
346 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
347 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
348
349 return queue_var_show(set << force, page);
350}
351
352static ssize_t
353queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
354{
355 ssize_t ret = -EINVAL;
356#ifdef CONFIG_SMP
357 unsigned long val;
358
359 ret = queue_var_store(&val, page, count);
360 if (ret < 0)
361 return ret;
362
363 if (val == 2) {
364 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
365 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
366 } else if (val == 1) {
367 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
368 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
369 } else if (val == 0) {
370 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
371 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
372 }
373#endif
374 return ret;
375}
376
377static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
378{
379 int val;
380
381 if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
382 val = BLK_MQ_POLL_CLASSIC;
383 else
384 val = q->poll_nsec / 1000;
385
386 return sprintf(page, "%d\n", val);
387}
388
389static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
390 size_t count)
391{
392 int err, val;
393
394 if (!q->mq_ops || !q->mq_ops->poll)
395 return -EINVAL;
396
397 err = kstrtoint(page, 10, &val);
398 if (err < 0)
399 return err;
400
401 if (val == BLK_MQ_POLL_CLASSIC)
402 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
403 else if (val >= 0)
404 q->poll_nsec = val * 1000;
405 else
406 return -EINVAL;
407
408 return count;
409}
410
411static ssize_t queue_poll_show(struct request_queue *q, char *page)
412{
413 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
414}
415
416static ssize_t queue_poll_store(struct request_queue *q, const char *page,
417 size_t count)
418{
419 unsigned long poll_on;
420 ssize_t ret;
421
422 if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
423 !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
424 return -EINVAL;
425
426 ret = queue_var_store(&poll_on, page, count);
427 if (ret < 0)
428 return ret;
429
430 if (poll_on)
431 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
432 else
433 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
434
435 return ret;
436}
437
438static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
439{
440 return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
441}
442
443static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
444 size_t count)
445{
446 unsigned int val;
447 int err;
448
449 err = kstrtou32(page, 10, &val);
450 if (err || val == 0)
451 return -EINVAL;
452
453 blk_queue_rq_timeout(q, msecs_to_jiffies(val));
454
455 return count;
456}
457
458static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
459{
460 if (!wbt_rq_qos(q))
461 return -EINVAL;
462
463 return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
464}
465
466static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
467 size_t count)
468{
469 struct rq_qos *rqos;
470 ssize_t ret;
471 s64 val;
472
473 ret = queue_var_store64(&val, page);
474 if (ret < 0)
475 return ret;
476 if (val < -1)
477 return -EINVAL;
478
479 rqos = wbt_rq_qos(q);
480 if (!rqos) {
481 ret = wbt_init(q);
482 if (ret)
483 return ret;
484 }
485
486 if (val == -1)
487 val = wbt_default_latency_nsec(q);
488 else if (val >= 0)
489 val *= 1000ULL;
490
491 if (wbt_get_min_lat(q) == val)
492 return count;
493
494 /*
495 * Ensure that the queue is idled, in case the latency update
496 * ends up either enabling or disabling wbt completely. We can't
497 * have IO inflight if that happens.
498 */
499 blk_mq_freeze_queue(q);
500 blk_mq_quiesce_queue(q);
501
502 wbt_set_min_lat(q, val);
503
504 blk_mq_unquiesce_queue(q);
505 blk_mq_unfreeze_queue(q);
506
507 return count;
508}
509
510static ssize_t queue_wc_show(struct request_queue *q, char *page)
511{
512 if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
513 return sprintf(page, "write back\n");
514
515 return sprintf(page, "write through\n");
516}
517
518static ssize_t queue_wc_store(struct request_queue *q, const char *page,
519 size_t count)
520{
521 int set = -1;
522
523 if (!strncmp(page, "write back", 10))
524 set = 1;
525 else if (!strncmp(page, "write through", 13) ||
526 !strncmp(page, "none", 4))
527 set = 0;
528
529 if (set == -1)
530 return -EINVAL;
531
532 if (set)
533 blk_queue_flag_set(QUEUE_FLAG_WC, q);
534 else
535 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
536
537 return count;
538}
539
540static ssize_t queue_fua_show(struct request_queue *q, char *page)
541{
542 return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
543}
544
545static ssize_t queue_dax_show(struct request_queue *q, char *page)
546{
547 return queue_var_show(blk_queue_dax(q), page);
548}
549
550static struct queue_sysfs_entry queue_requests_entry = {
551 .attr = {.name = "nr_requests", .mode = 0644 },
552 .show = queue_requests_show,
553 .store = queue_requests_store,
554};
555
556static struct queue_sysfs_entry queue_ra_entry = {
557 .attr = {.name = "read_ahead_kb", .mode = 0644 },
558 .show = queue_ra_show,
559 .store = queue_ra_store,
560};
561
562static struct queue_sysfs_entry queue_max_sectors_entry = {
563 .attr = {.name = "max_sectors_kb", .mode = 0644 },
564 .show = queue_max_sectors_show,
565 .store = queue_max_sectors_store,
566};
567
568static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
569 .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
570 .show = queue_max_hw_sectors_show,
571};
572
573static struct queue_sysfs_entry queue_max_segments_entry = {
574 .attr = {.name = "max_segments", .mode = 0444 },
575 .show = queue_max_segments_show,
576};
577
578static struct queue_sysfs_entry queue_max_discard_segments_entry = {
579 .attr = {.name = "max_discard_segments", .mode = 0444 },
580 .show = queue_max_discard_segments_show,
581};
582
583static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
584 .attr = {.name = "max_integrity_segments", .mode = 0444 },
585 .show = queue_max_integrity_segments_show,
586};
587
588static struct queue_sysfs_entry queue_max_segment_size_entry = {
589 .attr = {.name = "max_segment_size", .mode = 0444 },
590 .show = queue_max_segment_size_show,
591};
592
593static struct queue_sysfs_entry queue_iosched_entry = {
594 .attr = {.name = "scheduler", .mode = 0644 },
595 .show = elv_iosched_show,
596 .store = elv_iosched_store,
597};
598
599static struct queue_sysfs_entry queue_hw_sector_size_entry = {
600 .attr = {.name = "hw_sector_size", .mode = 0444 },
601 .show = queue_logical_block_size_show,
602};
603
604static struct queue_sysfs_entry queue_logical_block_size_entry = {
605 .attr = {.name = "logical_block_size", .mode = 0444 },
606 .show = queue_logical_block_size_show,
607};
608
609static struct queue_sysfs_entry queue_physical_block_size_entry = {
610 .attr = {.name = "physical_block_size", .mode = 0444 },
611 .show = queue_physical_block_size_show,
612};
613
614static struct queue_sysfs_entry queue_chunk_sectors_entry = {
615 .attr = {.name = "chunk_sectors", .mode = 0444 },
616 .show = queue_chunk_sectors_show,
617};
618
619static struct queue_sysfs_entry queue_io_min_entry = {
620 .attr = {.name = "minimum_io_size", .mode = 0444 },
621 .show = queue_io_min_show,
622};
623
624static struct queue_sysfs_entry queue_io_opt_entry = {
625 .attr = {.name = "optimal_io_size", .mode = 0444 },
626 .show = queue_io_opt_show,
627};
628
629static struct queue_sysfs_entry queue_discard_granularity_entry = {
630 .attr = {.name = "discard_granularity", .mode = 0444 },
631 .show = queue_discard_granularity_show,
632};
633
634static struct queue_sysfs_entry queue_discard_max_hw_entry = {
635 .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
636 .show = queue_discard_max_hw_show,
637};
638
639static struct queue_sysfs_entry queue_discard_max_entry = {
640 .attr = {.name = "discard_max_bytes", .mode = 0644 },
641 .show = queue_discard_max_show,
642 .store = queue_discard_max_store,
643};
644
645static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
646 .attr = {.name = "discard_zeroes_data", .mode = 0444 },
647 .show = queue_discard_zeroes_data_show,
648};
649
650static struct queue_sysfs_entry queue_write_same_max_entry = {
651 .attr = {.name = "write_same_max_bytes", .mode = 0444 },
652 .show = queue_write_same_max_show,
653};
654
655static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
656 .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
657 .show = queue_write_zeroes_max_show,
658};
659
660static struct queue_sysfs_entry queue_zone_append_max_entry = {
661 .attr = {.name = "zone_append_max_bytes", .mode = 0444 },
662 .show = queue_zone_append_max_show,
663};
664
665static struct queue_sysfs_entry queue_nonrot_entry = {
666 .attr = {.name = "rotational", .mode = 0644 },
667 .show = queue_show_nonrot,
668 .store = queue_store_nonrot,
669};
670
671static struct queue_sysfs_entry queue_zoned_entry = {
672 .attr = {.name = "zoned", .mode = 0444 },
673 .show = queue_zoned_show,
674};
675
676static struct queue_sysfs_entry queue_nr_zones_entry = {
677 .attr = {.name = "nr_zones", .mode = 0444 },
678 .show = queue_nr_zones_show,
679};
680
681static struct queue_sysfs_entry queue_max_open_zones_entry = {
682 .attr = {.name = "max_open_zones", .mode = 0444 },
683 .show = queue_max_open_zones_show,
684};
685
686static struct queue_sysfs_entry queue_max_active_zones_entry = {
687 .attr = {.name = "max_active_zones", .mode = 0444 },
688 .show = queue_max_active_zones_show,
689};
690
691static struct queue_sysfs_entry queue_nomerges_entry = {
692 .attr = {.name = "nomerges", .mode = 0644 },
693 .show = queue_nomerges_show,
694 .store = queue_nomerges_store,
695};
696
697static struct queue_sysfs_entry queue_rq_affinity_entry = {
698 .attr = {.name = "rq_affinity", .mode = 0644 },
699 .show = queue_rq_affinity_show,
700 .store = queue_rq_affinity_store,
701};
702
703static struct queue_sysfs_entry queue_iostats_entry = {
704 .attr = {.name = "iostats", .mode = 0644 },
705 .show = queue_show_iostats,
706 .store = queue_store_iostats,
707};
708
709static struct queue_sysfs_entry queue_random_entry = {
710 .attr = {.name = "add_random", .mode = 0644 },
711 .show = queue_show_random,
712 .store = queue_store_random,
713};
714
715static struct queue_sysfs_entry queue_poll_entry = {
716 .attr = {.name = "io_poll", .mode = 0644 },
717 .show = queue_poll_show,
718 .store = queue_poll_store,
719};
720
721static struct queue_sysfs_entry queue_poll_delay_entry = {
722 .attr = {.name = "io_poll_delay", .mode = 0644 },
723 .show = queue_poll_delay_show,
724 .store = queue_poll_delay_store,
725};
726
727static struct queue_sysfs_entry queue_wc_entry = {
728 .attr = {.name = "write_cache", .mode = 0644 },
729 .show = queue_wc_show,
730 .store = queue_wc_store,
731};
732
733static struct queue_sysfs_entry queue_fua_entry = {
734 .attr = {.name = "fua", .mode = 0444 },
735 .show = queue_fua_show,
736};
737
738static struct queue_sysfs_entry queue_dax_entry = {
739 .attr = {.name = "dax", .mode = 0444 },
740 .show = queue_dax_show,
741};
742
743static struct queue_sysfs_entry queue_io_timeout_entry = {
744 .attr = {.name = "io_timeout", .mode = 0644 },
745 .show = queue_io_timeout_show,
746 .store = queue_io_timeout_store,
747};
748
749static struct queue_sysfs_entry queue_wb_lat_entry = {
750 .attr = {.name = "wbt_lat_usec", .mode = 0644 },
751 .show = queue_wb_lat_show,
752 .store = queue_wb_lat_store,
753};
754
755#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
756static struct queue_sysfs_entry throtl_sample_time_entry = {
757 .attr = {.name = "throttle_sample_time", .mode = 0644 },
758 .show = blk_throtl_sample_time_show,
759 .store = blk_throtl_sample_time_store,
760};
761#endif
762
763static struct attribute *queue_attrs[] = {
764 &queue_requests_entry.attr,
765 &queue_ra_entry.attr,
766 &queue_max_hw_sectors_entry.attr,
767 &queue_max_sectors_entry.attr,
768 &queue_max_segments_entry.attr,
769 &queue_max_discard_segments_entry.attr,
770 &queue_max_integrity_segments_entry.attr,
771 &queue_max_segment_size_entry.attr,
772 &queue_iosched_entry.attr,
773 &queue_hw_sector_size_entry.attr,
774 &queue_logical_block_size_entry.attr,
775 &queue_physical_block_size_entry.attr,
776 &queue_chunk_sectors_entry.attr,
777 &queue_io_min_entry.attr,
778 &queue_io_opt_entry.attr,
779 &queue_discard_granularity_entry.attr,
780 &queue_discard_max_entry.attr,
781 &queue_discard_max_hw_entry.attr,
782 &queue_discard_zeroes_data_entry.attr,
783 &queue_write_same_max_entry.attr,
784 &queue_write_zeroes_max_entry.attr,
785 &queue_zone_append_max_entry.attr,
786 &queue_nonrot_entry.attr,
787 &queue_zoned_entry.attr,
788 &queue_nr_zones_entry.attr,
789 &queue_max_open_zones_entry.attr,
790 &queue_max_active_zones_entry.attr,
791 &queue_nomerges_entry.attr,
792 &queue_rq_affinity_entry.attr,
793 &queue_iostats_entry.attr,
794 &queue_random_entry.attr,
795 &queue_poll_entry.attr,
796 &queue_wc_entry.attr,
797 &queue_fua_entry.attr,
798 &queue_dax_entry.attr,
799 &queue_wb_lat_entry.attr,
800 &queue_poll_delay_entry.attr,
801 &queue_io_timeout_entry.attr,
802#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
803 &throtl_sample_time_entry.attr,
804#endif
805 NULL,
806};
807
808static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
809 int n)
810{
811 struct request_queue *q =
812 container_of(kobj, struct request_queue, kobj);
813
814 if (attr == &queue_io_timeout_entry.attr &&
815 (!q->mq_ops || !q->mq_ops->timeout))
816 return 0;
817
818 if ((attr == &queue_max_open_zones_entry.attr ||
819 attr == &queue_max_active_zones_entry.attr) &&
820 !blk_queue_is_zoned(q))
821 return 0;
822
823 return attr->mode;
824}
825
826static struct attribute_group queue_attr_group = {
827 .attrs = queue_attrs,
828 .is_visible = queue_attr_visible,
829};
830
831
832#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
833
834static ssize_t
835queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
836{
837 struct queue_sysfs_entry *entry = to_queue(attr);
838 struct request_queue *q =
839 container_of(kobj, struct request_queue, kobj);
840 ssize_t res;
841
842 if (!entry->show)
843 return -EIO;
844 mutex_lock(&q->sysfs_lock);
845 res = entry->show(q, page);
846 mutex_unlock(&q->sysfs_lock);
847 return res;
848}
849
850static ssize_t
851queue_attr_store(struct kobject *kobj, struct attribute *attr,
852 const char *page, size_t length)
853{
854 struct queue_sysfs_entry *entry = to_queue(attr);
855 struct request_queue *q;
856 ssize_t res;
857
858 if (!entry->store)
859 return -EIO;
860
861 q = container_of(kobj, struct request_queue, kobj);
862 mutex_lock(&q->sysfs_lock);
863 res = entry->store(q, page, length);
864 mutex_unlock(&q->sysfs_lock);
865 return res;
866}
867
868static void blk_free_queue_rcu(struct rcu_head *rcu_head)
869{
870 struct request_queue *q = container_of(rcu_head, struct request_queue,
871 rcu_head);
872 kmem_cache_free(blk_requestq_cachep, q);
873}
874
875/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
876static void blk_exit_queue(struct request_queue *q)
877{
878 /*
879 * Since the I/O scheduler exit code may access cgroup information,
880 * perform I/O scheduler exit before disassociating from the block
881 * cgroup controller.
882 */
883 if (q->elevator) {
884 ioc_clear_queue(q);
885 __elevator_exit(q, q->elevator);
886 q->elevator = NULL;
887 }
888
889 /*
890 * Remove all references to @q from the block cgroup controller before
891 * restoring @q->queue_lock to avoid that restoring this pointer causes
892 * e.g. blkcg_print_blkgs() to crash.
893 */
894 blkcg_exit_queue(q);
895
896 /*
897 * Since the cgroup code may dereference the @q->backing_dev_info
898 * pointer, only decrease its reference count after having removed the
899 * association with the block cgroup controller.
900 */
901 bdi_put(q->backing_dev_info);
902}
903
904/**
905 * blk_release_queue - releases all allocated resources of the request_queue
906 * @kobj: pointer to a kobject, whose container is a request_queue
907 *
908 * This function releases all allocated resources of the request queue.
909 *
910 * The struct request_queue refcount is incremented with blk_get_queue() and
911 * decremented with blk_put_queue(). Once the refcount reaches 0 this function
912 * is called.
913 *
914 * For drivers that have a request_queue on a gendisk and added with
915 * __device_add_disk() the refcount to request_queue will reach 0 with
916 * the last put_disk() called by the driver. For drivers which don't use
917 * __device_add_disk() this happens with blk_cleanup_queue().
918 *
919 * Drivers exist which depend on the release of the request_queue to be
920 * synchronous, it should not be deferred.
921 *
922 * Context: can sleep
923 */
924static void blk_release_queue(struct kobject *kobj)
925{
926 struct request_queue *q =
927 container_of(kobj, struct request_queue, kobj);
928
929 might_sleep();
930
931 if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
932 blk_stat_remove_callback(q, q->poll_cb);
933 blk_stat_free_callback(q->poll_cb);
934
935 blk_free_queue_stats(q->stats);
936
937 if (queue_is_mq(q))
938 cancel_delayed_work_sync(&q->requeue_work);
939
940 blk_exit_queue(q);
941
942 blk_queue_free_zone_bitmaps(q);
943
944 if (queue_is_mq(q))
945 blk_mq_release(q);
946
947 blk_trace_shutdown(q);
948 mutex_lock(&q->debugfs_mutex);
949 debugfs_remove_recursive(q->debugfs_dir);
950 mutex_unlock(&q->debugfs_mutex);
951
952 if (queue_is_mq(q))
953 blk_mq_debugfs_unregister(q);
954
955 bioset_exit(&q->bio_split);
956
957 ida_simple_remove(&blk_queue_ida, q->id);
958 call_rcu(&q->rcu_head, blk_free_queue_rcu);
959}
960
961static const struct sysfs_ops queue_sysfs_ops = {
962 .show = queue_attr_show,
963 .store = queue_attr_store,
964};
965
966struct kobj_type blk_queue_ktype = {
967 .sysfs_ops = &queue_sysfs_ops,
968 .release = blk_release_queue,
969};
970
971/**
972 * blk_register_queue - register a block layer queue with sysfs
973 * @disk: Disk of which the request queue should be registered with sysfs.
974 */
975int blk_register_queue(struct gendisk *disk)
976{
977 int ret;
978 struct device *dev = disk_to_dev(disk);
979 struct request_queue *q = disk->queue;
980 bool has_elevator = false;
981
982 if (WARN_ON(!q))
983 return -ENXIO;
984
985 WARN_ONCE(blk_queue_registered(q),
986 "%s is registering an already registered queue\n",
987 kobject_name(&dev->kobj));
988
989 /*
990 * SCSI probing may synchronously create and destroy a lot of
991 * request_queues for non-existent devices. Shutting down a fully
992 * functional queue takes measureable wallclock time as RCU grace
993 * periods are involved. To avoid excessive latency in these
994 * cases, a request_queue starts out in a degraded mode which is
995 * faster to shut down and is made fully functional here as
996 * request_queues for non-existent devices never get registered.
997 */
998 if (!blk_queue_init_done(q)) {
999 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
1000 percpu_ref_switch_to_percpu(&q->q_usage_counter);
1001 }
1002
1003 ret = blk_trace_init_sysfs(dev);
1004 if (ret)
1005 return ret;
1006
1007 mutex_lock(&q->sysfs_dir_lock);
1008
1009 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
1010 if (ret < 0) {
1011 blk_trace_remove_sysfs(dev);
1012 goto unlock;
1013 }
1014
1015 ret = sysfs_create_group(&q->kobj, &queue_attr_group);
1016 if (ret) {
1017 blk_trace_remove_sysfs(dev);
1018 kobject_del(&q->kobj);
1019 kobject_put(&dev->kobj);
1020 goto unlock;
1021 }
1022
1023 mutex_lock(&q->debugfs_mutex);
1024 q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
1025 blk_debugfs_root);
1026 mutex_unlock(&q->debugfs_mutex);
1027
1028 if (queue_is_mq(q)) {
1029 __blk_mq_register_dev(dev, q);
1030 blk_mq_debugfs_register(q);
1031 }
1032
1033 mutex_lock(&q->sysfs_lock);
1034 if (q->elevator) {
1035 ret = elv_register_queue(q, false);
1036 if (ret) {
1037 mutex_unlock(&q->sysfs_lock);
1038 mutex_unlock(&q->sysfs_dir_lock);
1039 kobject_del(&q->kobj);
1040 blk_trace_remove_sysfs(dev);
1041 kobject_put(&dev->kobj);
1042 return ret;
1043 }
1044 has_elevator = true;
1045 }
1046
1047 blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
1048 wbt_enable_default(q);
1049 blk_throtl_register_queue(q);
1050
1051 /* Now everything is ready and send out KOBJ_ADD uevent */
1052 kobject_uevent(&q->kobj, KOBJ_ADD);
1053 if (has_elevator)
1054 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
1055 mutex_unlock(&q->sysfs_lock);
1056
1057 ret = 0;
1058unlock:
1059 mutex_unlock(&q->sysfs_dir_lock);
1060 return ret;
1061}
1062EXPORT_SYMBOL_GPL(blk_register_queue);
1063
1064/**
1065 * blk_unregister_queue - counterpart of blk_register_queue()
1066 * @disk: Disk of which the request queue should be unregistered from sysfs.
1067 *
1068 * Note: the caller is responsible for guaranteeing that this function is called
1069 * after blk_register_queue() has finished.
1070 */
1071void blk_unregister_queue(struct gendisk *disk)
1072{
1073 struct request_queue *q = disk->queue;
1074
1075 if (WARN_ON(!q))
1076 return;
1077
1078 /* Return early if disk->queue was never registered. */
1079 if (!blk_queue_registered(q))
1080 return;
1081
1082 /*
1083 * Since sysfs_remove_dir() prevents adding new directory entries
1084 * before removal of existing entries starts, protect against
1085 * concurrent elv_iosched_store() calls.
1086 */
1087 mutex_lock(&q->sysfs_lock);
1088 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1089 mutex_unlock(&q->sysfs_lock);
1090
1091 mutex_lock(&q->sysfs_dir_lock);
1092 /*
1093 * Remove the sysfs attributes before unregistering the queue data
1094 * structures that can be modified through sysfs.
1095 */
1096 if (queue_is_mq(q))
1097 blk_mq_unregister_dev(disk_to_dev(disk), q);
1098
1099 kobject_uevent(&q->kobj, KOBJ_REMOVE);
1100 kobject_del(&q->kobj);
1101 blk_trace_remove_sysfs(disk_to_dev(disk));
1102
1103 mutex_lock(&q->sysfs_lock);
1104 if (q->elevator)
1105 elv_unregister_queue(q);
1106 mutex_unlock(&q->sysfs_lock);
1107 mutex_unlock(&q->sysfs_dir_lock);
1108
1109 kobject_put(&disk_to_dev(disk)->kobj);
1110}