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
15#include "blk.h"
16#include "blk-mq.h"
17#include "blk-mq-debugfs.h"
18#include "blk-wbt.h"
19
20struct queue_sysfs_entry {
21 struct attribute attr;
22 ssize_t (*show)(struct request_queue *, char *);
23 ssize_t (*store)(struct request_queue *, const char *, size_t);
24};
25
26static ssize_t
27queue_var_show(unsigned long var, char *page)
28{
29 return sprintf(page, "%lu\n", var);
30}
31
32static ssize_t
33queue_var_store(unsigned long *var, const char *page, size_t count)
34{
35 int err;
36 unsigned long v;
37
38 err = kstrtoul(page, 10, &v);
39 if (err || v > UINT_MAX)
40 return -EINVAL;
41
42 *var = v;
43
44 return count;
45}
46
47static ssize_t queue_var_store64(s64 *var, const char *page)
48{
49 int err;
50 s64 v;
51
52 err = kstrtos64(page, 10, &v);
53 if (err < 0)
54 return err;
55
56 *var = v;
57 return 0;
58}
59
60static ssize_t queue_requests_show(struct request_queue *q, char *page)
61{
62 return queue_var_show(q->nr_requests, (page));
63}
64
65static ssize_t
66queue_requests_store(struct request_queue *q, const char *page, size_t count)
67{
68 unsigned long nr;
69 int ret, err;
70
71 if (!queue_is_mq(q))
72 return -EINVAL;
73
74 ret = queue_var_store(&nr, page, count);
75 if (ret < 0)
76 return ret;
77
78 if (nr < BLKDEV_MIN_RQ)
79 nr = BLKDEV_MIN_RQ;
80
81 err = blk_mq_update_nr_requests(q, nr);
82 if (err)
83 return err;
84
85 return ret;
86}
87
88static ssize_t queue_ra_show(struct request_queue *q, char *page)
89{
90 unsigned long ra_kb = q->backing_dev_info->ra_pages <<
91 (PAGE_SHIFT - 10);
92
93 return queue_var_show(ra_kb, (page));
94}
95
96static ssize_t
97queue_ra_store(struct request_queue *q, const char *page, size_t count)
98{
99 unsigned long ra_kb;
100 ssize_t ret = queue_var_store(&ra_kb, page, count);
101
102 if (ret < 0)
103 return ret;
104
105 q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
106
107 return ret;
108}
109
110static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
111{
112 int max_sectors_kb = queue_max_sectors(q) >> 1;
113
114 return queue_var_show(max_sectors_kb, (page));
115}
116
117static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
118{
119 return queue_var_show(queue_max_segments(q), (page));
120}
121
122static ssize_t queue_max_discard_segments_show(struct request_queue *q,
123 char *page)
124{
125 return queue_var_show(queue_max_discard_segments(q), (page));
126}
127
128static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
129{
130 return queue_var_show(q->limits.max_integrity_segments, (page));
131}
132
133static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
134{
135 return queue_var_show(queue_max_segment_size(q), (page));
136}
137
138static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
139{
140 return queue_var_show(queue_logical_block_size(q), page);
141}
142
143static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
144{
145 return queue_var_show(queue_physical_block_size(q), page);
146}
147
148static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
149{
150 return queue_var_show(q->limits.chunk_sectors, page);
151}
152
153static ssize_t queue_io_min_show(struct request_queue *q, char *page)
154{
155 return queue_var_show(queue_io_min(q), page);
156}
157
158static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
159{
160 return queue_var_show(queue_io_opt(q), page);
161}
162
163static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
164{
165 return queue_var_show(q->limits.discard_granularity, page);
166}
167
168static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
169{
170
171 return sprintf(page, "%llu\n",
172 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
173}
174
175static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
176{
177 return sprintf(page, "%llu\n",
178 (unsigned long long)q->limits.max_discard_sectors << 9);
179}
180
181static ssize_t queue_discard_max_store(struct request_queue *q,
182 const char *page, size_t count)
183{
184 unsigned long max_discard;
185 ssize_t ret = queue_var_store(&max_discard, page, count);
186
187 if (ret < 0)
188 return ret;
189
190 if (max_discard & (q->limits.discard_granularity - 1))
191 return -EINVAL;
192
193 max_discard >>= 9;
194 if (max_discard > UINT_MAX)
195 return -EINVAL;
196
197 if (max_discard > q->limits.max_hw_discard_sectors)
198 max_discard = q->limits.max_hw_discard_sectors;
199
200 q->limits.max_discard_sectors = max_discard;
201 return ret;
202}
203
204static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
205{
206 return queue_var_show(0, page);
207}
208
209static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
210{
211 return sprintf(page, "%llu\n",
212 (unsigned long long)q->limits.max_write_same_sectors << 9);
213}
214
215static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
216{
217 return sprintf(page, "%llu\n",
218 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
219}
220
221static ssize_t
222queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
223{
224 unsigned long max_sectors_kb,
225 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
226 page_kb = 1 << (PAGE_SHIFT - 10);
227 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
228
229 if (ret < 0)
230 return ret;
231
232 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
233 q->limits.max_dev_sectors >> 1);
234
235 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
236 return -EINVAL;
237
238 spin_lock_irq(&q->queue_lock);
239 q->limits.max_sectors = max_sectors_kb << 1;
240 q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
241 spin_unlock_irq(&q->queue_lock);
242
243 return ret;
244}
245
246static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
247{
248 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
249
250 return queue_var_show(max_hw_sectors_kb, (page));
251}
252
253#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
254static ssize_t \
255queue_show_##name(struct request_queue *q, char *page) \
256{ \
257 int bit; \
258 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
259 return queue_var_show(neg ? !bit : bit, page); \
260} \
261static ssize_t \
262queue_store_##name(struct request_queue *q, const char *page, size_t count) \
263{ \
264 unsigned long val; \
265 ssize_t ret; \
266 ret = queue_var_store(&val, page, count); \
267 if (ret < 0) \
268 return ret; \
269 if (neg) \
270 val = !val; \
271 \
272 if (val) \
273 blk_queue_flag_set(QUEUE_FLAG_##flag, q); \
274 else \
275 blk_queue_flag_clear(QUEUE_FLAG_##flag, q); \
276 return ret; \
277}
278
279QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
280QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
281QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
282#undef QUEUE_SYSFS_BIT_FNS
283
284static ssize_t queue_zoned_show(struct request_queue *q, char *page)
285{
286 switch (blk_queue_zoned_model(q)) {
287 case BLK_ZONED_HA:
288 return sprintf(page, "host-aware\n");
289 case BLK_ZONED_HM:
290 return sprintf(page, "host-managed\n");
291 default:
292 return sprintf(page, "none\n");
293 }
294}
295
296static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
297{
298 return queue_var_show(blk_queue_nr_zones(q), page);
299}
300
301static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
302{
303 return queue_var_show((blk_queue_nomerges(q) << 1) |
304 blk_queue_noxmerges(q), page);
305}
306
307static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
308 size_t count)
309{
310 unsigned long nm;
311 ssize_t ret = queue_var_store(&nm, page, count);
312
313 if (ret < 0)
314 return ret;
315
316 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
317 blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
318 if (nm == 2)
319 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
320 else if (nm)
321 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
322
323 return ret;
324}
325
326static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
327{
328 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
329 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
330
331 return queue_var_show(set << force, page);
332}
333
334static ssize_t
335queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
336{
337 ssize_t ret = -EINVAL;
338#ifdef CONFIG_SMP
339 unsigned long val;
340
341 ret = queue_var_store(&val, page, count);
342 if (ret < 0)
343 return ret;
344
345 if (val == 2) {
346 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
347 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
348 } else if (val == 1) {
349 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
350 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
351 } else if (val == 0) {
352 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
353 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
354 }
355#endif
356 return ret;
357}
358
359static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
360{
361 int val;
362
363 if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
364 val = BLK_MQ_POLL_CLASSIC;
365 else
366 val = q->poll_nsec / 1000;
367
368 return sprintf(page, "%d\n", val);
369}
370
371static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
372 size_t count)
373{
374 int err, val;
375
376 if (!q->mq_ops || !q->mq_ops->poll)
377 return -EINVAL;
378
379 err = kstrtoint(page, 10, &val);
380 if (err < 0)
381 return err;
382
383 if (val == BLK_MQ_POLL_CLASSIC)
384 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
385 else if (val >= 0)
386 q->poll_nsec = val * 1000;
387 else
388 return -EINVAL;
389
390 return count;
391}
392
393static ssize_t queue_poll_show(struct request_queue *q, char *page)
394{
395 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
396}
397
398static ssize_t queue_poll_store(struct request_queue *q, const char *page,
399 size_t count)
400{
401 unsigned long poll_on;
402 ssize_t ret;
403
404 if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
405 !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
406 return -EINVAL;
407
408 ret = queue_var_store(&poll_on, page, count);
409 if (ret < 0)
410 return ret;
411
412 if (poll_on)
413 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
414 else
415 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
416
417 return ret;
418}
419
420static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
421{
422 return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
423}
424
425static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
426 size_t count)
427{
428 unsigned int val;
429 int err;
430
431 err = kstrtou32(page, 10, &val);
432 if (err || val == 0)
433 return -EINVAL;
434
435 blk_queue_rq_timeout(q, msecs_to_jiffies(val));
436
437 return count;
438}
439
440static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
441{
442 if (!wbt_rq_qos(q))
443 return -EINVAL;
444
445 return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
446}
447
448static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
449 size_t count)
450{
451 struct rq_qos *rqos;
452 ssize_t ret;
453 s64 val;
454
455 ret = queue_var_store64(&val, page);
456 if (ret < 0)
457 return ret;
458 if (val < -1)
459 return -EINVAL;
460
461 rqos = wbt_rq_qos(q);
462 if (!rqos) {
463 ret = wbt_init(q);
464 if (ret)
465 return ret;
466 }
467
468 if (val == -1)
469 val = wbt_default_latency_nsec(q);
470 else if (val >= 0)
471 val *= 1000ULL;
472
473 if (wbt_get_min_lat(q) == val)
474 return count;
475
476 /*
477 * Ensure that the queue is idled, in case the latency update
478 * ends up either enabling or disabling wbt completely. We can't
479 * have IO inflight if that happens.
480 */
481 blk_mq_freeze_queue(q);
482 blk_mq_quiesce_queue(q);
483
484 wbt_set_min_lat(q, val);
485
486 blk_mq_unquiesce_queue(q);
487 blk_mq_unfreeze_queue(q);
488
489 return count;
490}
491
492static ssize_t queue_wc_show(struct request_queue *q, char *page)
493{
494 if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
495 return sprintf(page, "write back\n");
496
497 return sprintf(page, "write through\n");
498}
499
500static ssize_t queue_wc_store(struct request_queue *q, const char *page,
501 size_t count)
502{
503 int set = -1;
504
505 if (!strncmp(page, "write back", 10))
506 set = 1;
507 else if (!strncmp(page, "write through", 13) ||
508 !strncmp(page, "none", 4))
509 set = 0;
510
511 if (set == -1)
512 return -EINVAL;
513
514 if (set)
515 blk_queue_flag_set(QUEUE_FLAG_WC, q);
516 else
517 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
518
519 return count;
520}
521
522static ssize_t queue_fua_show(struct request_queue *q, char *page)
523{
524 return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
525}
526
527static ssize_t queue_dax_show(struct request_queue *q, char *page)
528{
529 return queue_var_show(blk_queue_dax(q), page);
530}
531
532static struct queue_sysfs_entry queue_requests_entry = {
533 .attr = {.name = "nr_requests", .mode = 0644 },
534 .show = queue_requests_show,
535 .store = queue_requests_store,
536};
537
538static struct queue_sysfs_entry queue_ra_entry = {
539 .attr = {.name = "read_ahead_kb", .mode = 0644 },
540 .show = queue_ra_show,
541 .store = queue_ra_store,
542};
543
544static struct queue_sysfs_entry queue_max_sectors_entry = {
545 .attr = {.name = "max_sectors_kb", .mode = 0644 },
546 .show = queue_max_sectors_show,
547 .store = queue_max_sectors_store,
548};
549
550static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
551 .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
552 .show = queue_max_hw_sectors_show,
553};
554
555static struct queue_sysfs_entry queue_max_segments_entry = {
556 .attr = {.name = "max_segments", .mode = 0444 },
557 .show = queue_max_segments_show,
558};
559
560static struct queue_sysfs_entry queue_max_discard_segments_entry = {
561 .attr = {.name = "max_discard_segments", .mode = 0444 },
562 .show = queue_max_discard_segments_show,
563};
564
565static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
566 .attr = {.name = "max_integrity_segments", .mode = 0444 },
567 .show = queue_max_integrity_segments_show,
568};
569
570static struct queue_sysfs_entry queue_max_segment_size_entry = {
571 .attr = {.name = "max_segment_size", .mode = 0444 },
572 .show = queue_max_segment_size_show,
573};
574
575static struct queue_sysfs_entry queue_iosched_entry = {
576 .attr = {.name = "scheduler", .mode = 0644 },
577 .show = elv_iosched_show,
578 .store = elv_iosched_store,
579};
580
581static struct queue_sysfs_entry queue_hw_sector_size_entry = {
582 .attr = {.name = "hw_sector_size", .mode = 0444 },
583 .show = queue_logical_block_size_show,
584};
585
586static struct queue_sysfs_entry queue_logical_block_size_entry = {
587 .attr = {.name = "logical_block_size", .mode = 0444 },
588 .show = queue_logical_block_size_show,
589};
590
591static struct queue_sysfs_entry queue_physical_block_size_entry = {
592 .attr = {.name = "physical_block_size", .mode = 0444 },
593 .show = queue_physical_block_size_show,
594};
595
596static struct queue_sysfs_entry queue_chunk_sectors_entry = {
597 .attr = {.name = "chunk_sectors", .mode = 0444 },
598 .show = queue_chunk_sectors_show,
599};
600
601static struct queue_sysfs_entry queue_io_min_entry = {
602 .attr = {.name = "minimum_io_size", .mode = 0444 },
603 .show = queue_io_min_show,
604};
605
606static struct queue_sysfs_entry queue_io_opt_entry = {
607 .attr = {.name = "optimal_io_size", .mode = 0444 },
608 .show = queue_io_opt_show,
609};
610
611static struct queue_sysfs_entry queue_discard_granularity_entry = {
612 .attr = {.name = "discard_granularity", .mode = 0444 },
613 .show = queue_discard_granularity_show,
614};
615
616static struct queue_sysfs_entry queue_discard_max_hw_entry = {
617 .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
618 .show = queue_discard_max_hw_show,
619};
620
621static struct queue_sysfs_entry queue_discard_max_entry = {
622 .attr = {.name = "discard_max_bytes", .mode = 0644 },
623 .show = queue_discard_max_show,
624 .store = queue_discard_max_store,
625};
626
627static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
628 .attr = {.name = "discard_zeroes_data", .mode = 0444 },
629 .show = queue_discard_zeroes_data_show,
630};
631
632static struct queue_sysfs_entry queue_write_same_max_entry = {
633 .attr = {.name = "write_same_max_bytes", .mode = 0444 },
634 .show = queue_write_same_max_show,
635};
636
637static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
638 .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
639 .show = queue_write_zeroes_max_show,
640};
641
642static struct queue_sysfs_entry queue_nonrot_entry = {
643 .attr = {.name = "rotational", .mode = 0644 },
644 .show = queue_show_nonrot,
645 .store = queue_store_nonrot,
646};
647
648static struct queue_sysfs_entry queue_zoned_entry = {
649 .attr = {.name = "zoned", .mode = 0444 },
650 .show = queue_zoned_show,
651};
652
653static struct queue_sysfs_entry queue_nr_zones_entry = {
654 .attr = {.name = "nr_zones", .mode = 0444 },
655 .show = queue_nr_zones_show,
656};
657
658static struct queue_sysfs_entry queue_nomerges_entry = {
659 .attr = {.name = "nomerges", .mode = 0644 },
660 .show = queue_nomerges_show,
661 .store = queue_nomerges_store,
662};
663
664static struct queue_sysfs_entry queue_rq_affinity_entry = {
665 .attr = {.name = "rq_affinity", .mode = 0644 },
666 .show = queue_rq_affinity_show,
667 .store = queue_rq_affinity_store,
668};
669
670static struct queue_sysfs_entry queue_iostats_entry = {
671 .attr = {.name = "iostats", .mode = 0644 },
672 .show = queue_show_iostats,
673 .store = queue_store_iostats,
674};
675
676static struct queue_sysfs_entry queue_random_entry = {
677 .attr = {.name = "add_random", .mode = 0644 },
678 .show = queue_show_random,
679 .store = queue_store_random,
680};
681
682static struct queue_sysfs_entry queue_poll_entry = {
683 .attr = {.name = "io_poll", .mode = 0644 },
684 .show = queue_poll_show,
685 .store = queue_poll_store,
686};
687
688static struct queue_sysfs_entry queue_poll_delay_entry = {
689 .attr = {.name = "io_poll_delay", .mode = 0644 },
690 .show = queue_poll_delay_show,
691 .store = queue_poll_delay_store,
692};
693
694static struct queue_sysfs_entry queue_wc_entry = {
695 .attr = {.name = "write_cache", .mode = 0644 },
696 .show = queue_wc_show,
697 .store = queue_wc_store,
698};
699
700static struct queue_sysfs_entry queue_fua_entry = {
701 .attr = {.name = "fua", .mode = 0444 },
702 .show = queue_fua_show,
703};
704
705static struct queue_sysfs_entry queue_dax_entry = {
706 .attr = {.name = "dax", .mode = 0444 },
707 .show = queue_dax_show,
708};
709
710static struct queue_sysfs_entry queue_io_timeout_entry = {
711 .attr = {.name = "io_timeout", .mode = 0644 },
712 .show = queue_io_timeout_show,
713 .store = queue_io_timeout_store,
714};
715
716static struct queue_sysfs_entry queue_wb_lat_entry = {
717 .attr = {.name = "wbt_lat_usec", .mode = 0644 },
718 .show = queue_wb_lat_show,
719 .store = queue_wb_lat_store,
720};
721
722#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
723static struct queue_sysfs_entry throtl_sample_time_entry = {
724 .attr = {.name = "throttle_sample_time", .mode = 0644 },
725 .show = blk_throtl_sample_time_show,
726 .store = blk_throtl_sample_time_store,
727};
728#endif
729
730static struct attribute *queue_attrs[] = {
731 &queue_requests_entry.attr,
732 &queue_ra_entry.attr,
733 &queue_max_hw_sectors_entry.attr,
734 &queue_max_sectors_entry.attr,
735 &queue_max_segments_entry.attr,
736 &queue_max_discard_segments_entry.attr,
737 &queue_max_integrity_segments_entry.attr,
738 &queue_max_segment_size_entry.attr,
739 &queue_iosched_entry.attr,
740 &queue_hw_sector_size_entry.attr,
741 &queue_logical_block_size_entry.attr,
742 &queue_physical_block_size_entry.attr,
743 &queue_chunk_sectors_entry.attr,
744 &queue_io_min_entry.attr,
745 &queue_io_opt_entry.attr,
746 &queue_discard_granularity_entry.attr,
747 &queue_discard_max_entry.attr,
748 &queue_discard_max_hw_entry.attr,
749 &queue_discard_zeroes_data_entry.attr,
750 &queue_write_same_max_entry.attr,
751 &queue_write_zeroes_max_entry.attr,
752 &queue_nonrot_entry.attr,
753 &queue_zoned_entry.attr,
754 &queue_nr_zones_entry.attr,
755 &queue_nomerges_entry.attr,
756 &queue_rq_affinity_entry.attr,
757 &queue_iostats_entry.attr,
758 &queue_random_entry.attr,
759 &queue_poll_entry.attr,
760 &queue_wc_entry.attr,
761 &queue_fua_entry.attr,
762 &queue_dax_entry.attr,
763 &queue_wb_lat_entry.attr,
764 &queue_poll_delay_entry.attr,
765 &queue_io_timeout_entry.attr,
766#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
767 &throtl_sample_time_entry.attr,
768#endif
769 NULL,
770};
771
772static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
773 int n)
774{
775 struct request_queue *q =
776 container_of(kobj, struct request_queue, kobj);
777
778 if (attr == &queue_io_timeout_entry.attr &&
779 (!q->mq_ops || !q->mq_ops->timeout))
780 return 0;
781
782 return attr->mode;
783}
784
785static struct attribute_group queue_attr_group = {
786 .attrs = queue_attrs,
787 .is_visible = queue_attr_visible,
788};
789
790
791#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
792
793static ssize_t
794queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
795{
796 struct queue_sysfs_entry *entry = to_queue(attr);
797 struct request_queue *q =
798 container_of(kobj, struct request_queue, kobj);
799 ssize_t res;
800
801 if (!entry->show)
802 return -EIO;
803 mutex_lock(&q->sysfs_lock);
804 if (blk_queue_dying(q)) {
805 mutex_unlock(&q->sysfs_lock);
806 return -ENOENT;
807 }
808 res = entry->show(q, page);
809 mutex_unlock(&q->sysfs_lock);
810 return res;
811}
812
813static ssize_t
814queue_attr_store(struct kobject *kobj, struct attribute *attr,
815 const char *page, size_t length)
816{
817 struct queue_sysfs_entry *entry = to_queue(attr);
818 struct request_queue *q;
819 ssize_t res;
820
821 if (!entry->store)
822 return -EIO;
823
824 q = container_of(kobj, struct request_queue, kobj);
825 mutex_lock(&q->sysfs_lock);
826 if (blk_queue_dying(q)) {
827 mutex_unlock(&q->sysfs_lock);
828 return -ENOENT;
829 }
830 res = entry->store(q, page, length);
831 mutex_unlock(&q->sysfs_lock);
832 return res;
833}
834
835static void blk_free_queue_rcu(struct rcu_head *rcu_head)
836{
837 struct request_queue *q = container_of(rcu_head, struct request_queue,
838 rcu_head);
839 kmem_cache_free(blk_requestq_cachep, q);
840}
841
842/* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
843static void blk_exit_queue(struct request_queue *q)
844{
845 /*
846 * Since the I/O scheduler exit code may access cgroup information,
847 * perform I/O scheduler exit before disassociating from the block
848 * cgroup controller.
849 */
850 if (q->elevator) {
851 ioc_clear_queue(q);
852 __elevator_exit(q, q->elevator);
853 q->elevator = NULL;
854 }
855
856 /*
857 * Remove all references to @q from the block cgroup controller before
858 * restoring @q->queue_lock to avoid that restoring this pointer causes
859 * e.g. blkcg_print_blkgs() to crash.
860 */
861 blkcg_exit_queue(q);
862
863 /*
864 * Since the cgroup code may dereference the @q->backing_dev_info
865 * pointer, only decrease its reference count after having removed the
866 * association with the block cgroup controller.
867 */
868 bdi_put(q->backing_dev_info);
869}
870
871
872/**
873 * __blk_release_queue - release a request queue
874 * @work: pointer to the release_work member of the request queue to be released
875 *
876 * Description:
877 * This function is called when a block device is being unregistered. The
878 * process of releasing a request queue starts with blk_cleanup_queue, which
879 * set the appropriate flags and then calls blk_put_queue, that decrements
880 * the reference counter of the request queue. Once the reference counter
881 * of the request queue reaches zero, blk_release_queue is called to release
882 * all allocated resources of the request queue.
883 */
884static void __blk_release_queue(struct work_struct *work)
885{
886 struct request_queue *q = container_of(work, typeof(*q), release_work);
887
888 if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
889 blk_stat_remove_callback(q, q->poll_cb);
890 blk_stat_free_callback(q->poll_cb);
891
892 blk_free_queue_stats(q->stats);
893
894 if (queue_is_mq(q))
895 cancel_delayed_work_sync(&q->requeue_work);
896
897 blk_exit_queue(q);
898
899 blk_queue_free_zone_bitmaps(q);
900
901 if (queue_is_mq(q))
902 blk_mq_release(q);
903
904 blk_trace_shutdown(q);
905
906 if (queue_is_mq(q))
907 blk_mq_debugfs_unregister(q);
908
909 bioset_exit(&q->bio_split);
910
911 ida_simple_remove(&blk_queue_ida, q->id);
912 call_rcu(&q->rcu_head, blk_free_queue_rcu);
913}
914
915static void blk_release_queue(struct kobject *kobj)
916{
917 struct request_queue *q =
918 container_of(kobj, struct request_queue, kobj);
919
920 INIT_WORK(&q->release_work, __blk_release_queue);
921 schedule_work(&q->release_work);
922}
923
924static const struct sysfs_ops queue_sysfs_ops = {
925 .show = queue_attr_show,
926 .store = queue_attr_store,
927};
928
929struct kobj_type blk_queue_ktype = {
930 .sysfs_ops = &queue_sysfs_ops,
931 .release = blk_release_queue,
932};
933
934/**
935 * blk_register_queue - register a block layer queue with sysfs
936 * @disk: Disk of which the request queue should be registered with sysfs.
937 */
938int blk_register_queue(struct gendisk *disk)
939{
940 int ret;
941 struct device *dev = disk_to_dev(disk);
942 struct request_queue *q = disk->queue;
943 bool has_elevator = false;
944
945 if (WARN_ON(!q))
946 return -ENXIO;
947
948 WARN_ONCE(blk_queue_registered(q),
949 "%s is registering an already registered queue\n",
950 kobject_name(&dev->kobj));
951
952 /*
953 * SCSI probing may synchronously create and destroy a lot of
954 * request_queues for non-existent devices. Shutting down a fully
955 * functional queue takes measureable wallclock time as RCU grace
956 * periods are involved. To avoid excessive latency in these
957 * cases, a request_queue starts out in a degraded mode which is
958 * faster to shut down and is made fully functional here as
959 * request_queues for non-existent devices never get registered.
960 */
961 if (!blk_queue_init_done(q)) {
962 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
963 percpu_ref_switch_to_percpu(&q->q_usage_counter);
964 }
965
966 ret = blk_trace_init_sysfs(dev);
967 if (ret)
968 return ret;
969
970 mutex_lock(&q->sysfs_dir_lock);
971
972 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
973 if (ret < 0) {
974 blk_trace_remove_sysfs(dev);
975 goto unlock;
976 }
977
978 ret = sysfs_create_group(&q->kobj, &queue_attr_group);
979 if (ret) {
980 blk_trace_remove_sysfs(dev);
981 kobject_del(&q->kobj);
982 kobject_put(&dev->kobj);
983 goto unlock;
984 }
985
986 if (queue_is_mq(q)) {
987 __blk_mq_register_dev(dev, q);
988 blk_mq_debugfs_register(q);
989 }
990
991 mutex_lock(&q->sysfs_lock);
992 if (q->elevator) {
993 ret = elv_register_queue(q, false);
994 if (ret) {
995 mutex_unlock(&q->sysfs_lock);
996 mutex_unlock(&q->sysfs_dir_lock);
997 kobject_del(&q->kobj);
998 blk_trace_remove_sysfs(dev);
999 kobject_put(&dev->kobj);
1000 return ret;
1001 }
1002 has_elevator = true;
1003 }
1004
1005 blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
1006 wbt_enable_default(q);
1007 blk_throtl_register_queue(q);
1008
1009 /* Now everything is ready and send out KOBJ_ADD uevent */
1010 kobject_uevent(&q->kobj, KOBJ_ADD);
1011 if (has_elevator)
1012 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
1013 mutex_unlock(&q->sysfs_lock);
1014
1015 ret = 0;
1016unlock:
1017 mutex_unlock(&q->sysfs_dir_lock);
1018 return ret;
1019}
1020EXPORT_SYMBOL_GPL(blk_register_queue);
1021
1022/**
1023 * blk_unregister_queue - counterpart of blk_register_queue()
1024 * @disk: Disk of which the request queue should be unregistered from sysfs.
1025 *
1026 * Note: the caller is responsible for guaranteeing that this function is called
1027 * after blk_register_queue() has finished.
1028 */
1029void blk_unregister_queue(struct gendisk *disk)
1030{
1031 struct request_queue *q = disk->queue;
1032
1033 if (WARN_ON(!q))
1034 return;
1035
1036 /* Return early if disk->queue was never registered. */
1037 if (!blk_queue_registered(q))
1038 return;
1039
1040 /*
1041 * Since sysfs_remove_dir() prevents adding new directory entries
1042 * before removal of existing entries starts, protect against
1043 * concurrent elv_iosched_store() calls.
1044 */
1045 mutex_lock(&q->sysfs_lock);
1046 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1047 mutex_unlock(&q->sysfs_lock);
1048
1049 mutex_lock(&q->sysfs_dir_lock);
1050 /*
1051 * Remove the sysfs attributes before unregistering the queue data
1052 * structures that can be modified through sysfs.
1053 */
1054 if (queue_is_mq(q))
1055 blk_mq_unregister_dev(disk_to_dev(disk), q);
1056
1057 kobject_uevent(&q->kobj, KOBJ_REMOVE);
1058 kobject_del(&q->kobj);
1059 blk_trace_remove_sysfs(disk_to_dev(disk));
1060
1061 mutex_lock(&q->sysfs_lock);
1062 if (q->elevator)
1063 elv_unregister_queue(q);
1064 mutex_unlock(&q->sysfs_lock);
1065 mutex_unlock(&q->sysfs_dir_lock);
1066
1067 kobject_put(&disk_to_dev(disk)->kobj);
1068}