Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
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/debugfs.h>
13
14#include "blk.h"
15#include "blk-mq.h"
16#include "blk-mq-debugfs.h"
17#include "blk-mq-sched.h"
18#include "blk-rq-qos.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 gendisk *disk, char *page);
26 ssize_t (*show_limit)(struct gendisk *disk, char *page);
27
28 ssize_t (*store)(struct gendisk *disk, const char *page, size_t count);
29 int (*store_limit)(struct gendisk *disk, const char *page,
30 size_t count, struct queue_limits *lim);
31};
32
33static ssize_t
34queue_var_show(unsigned long var, char *page)
35{
36 return sysfs_emit(page, "%lu\n", var);
37}
38
39static ssize_t
40queue_var_store(unsigned long *var, const char *page, size_t count)
41{
42 int err;
43 unsigned long v;
44
45 err = kstrtoul(page, 10, &v);
46 if (err || v > UINT_MAX)
47 return -EINVAL;
48
49 *var = v;
50
51 return count;
52}
53
54static ssize_t queue_requests_show(struct gendisk *disk, char *page)
55{
56 ssize_t ret;
57
58 mutex_lock(&disk->queue->elevator_lock);
59 ret = queue_var_show(disk->queue->nr_requests, page);
60 mutex_unlock(&disk->queue->elevator_lock);
61 return ret;
62}
63
64static ssize_t
65queue_requests_store(struct gendisk *disk, const char *page, size_t count)
66{
67 struct request_queue *q = disk->queue;
68 struct blk_mq_tag_set *set = q->tag_set;
69 struct elevator_tags *et = NULL;
70 unsigned int memflags;
71 unsigned long nr;
72 int ret;
73
74 ret = queue_var_store(&nr, page, count);
75 if (ret < 0)
76 return ret;
77
78 /*
79 * Serialize updating nr_requests with concurrent queue_requests_store()
80 * and switching elevator.
81 *
82 * Use trylock to avoid circular lock dependency with kernfs active
83 * reference during concurrent disk deletion:
84 * update_nr_hwq_lock -> kn->active (via del_gendisk -> kobject_del)
85 * kn->active -> update_nr_hwq_lock (via this sysfs write path)
86 */
87 if (!down_write_trylock(&set->update_nr_hwq_lock))
88 return -EBUSY;
89
90 if (nr == q->nr_requests)
91 goto unlock;
92
93 if (nr < BLKDEV_MIN_RQ)
94 nr = BLKDEV_MIN_RQ;
95
96 /*
97 * Switching elevator is protected by update_nr_hwq_lock:
98 * - read lock is held from elevator sysfs attribute;
99 * - write lock is held from updating nr_hw_queues;
100 * Hence it's safe to access q->elevator here with write lock held.
101 */
102 if (nr <= set->reserved_tags ||
103 (q->elevator && nr > MAX_SCHED_RQ) ||
104 (!q->elevator && nr > set->queue_depth)) {
105 ret = -EINVAL;
106 goto unlock;
107 }
108
109 if (!blk_mq_is_shared_tags(set->flags) && q->elevator &&
110 nr > q->elevator->et->nr_requests) {
111 /*
112 * Tags will grow, allocate memory before freezing queue to
113 * prevent deadlock.
114 */
115 et = blk_mq_alloc_sched_tags(set, q->nr_hw_queues, nr);
116 if (!et) {
117 ret = -ENOMEM;
118 goto unlock;
119 }
120 }
121
122 memflags = blk_mq_freeze_queue(q);
123 mutex_lock(&q->elevator_lock);
124 et = blk_mq_update_nr_requests(q, et, nr);
125 mutex_unlock(&q->elevator_lock);
126 blk_mq_unfreeze_queue(q, memflags);
127
128 if (et)
129 blk_mq_free_sched_tags(et, set);
130
131unlock:
132 up_write(&set->update_nr_hwq_lock);
133 return ret;
134}
135
136static ssize_t queue_async_depth_show(struct gendisk *disk, char *page)
137{
138 guard(mutex)(&disk->queue->elevator_lock);
139
140 return queue_var_show(disk->queue->async_depth, page);
141}
142
143static ssize_t
144queue_async_depth_store(struct gendisk *disk, const char *page, size_t count)
145{
146 struct request_queue *q = disk->queue;
147 unsigned int memflags;
148 unsigned long nr;
149 int ret;
150
151 if (!queue_is_mq(q))
152 return -EINVAL;
153
154 ret = queue_var_store(&nr, page, count);
155 if (ret < 0)
156 return ret;
157
158 if (nr == 0)
159 return -EINVAL;
160
161 memflags = blk_mq_freeze_queue(q);
162 scoped_guard(mutex, &q->elevator_lock) {
163 if (q->elevator) {
164 q->async_depth = min(q->nr_requests, nr);
165 if (q->elevator->type->ops.depth_updated)
166 q->elevator->type->ops.depth_updated(q);
167 } else {
168 ret = -EINVAL;
169 }
170 }
171 blk_mq_unfreeze_queue(q, memflags);
172
173 return ret;
174}
175
176static ssize_t queue_ra_show(struct gendisk *disk, char *page)
177{
178 ssize_t ret;
179
180 mutex_lock(&disk->queue->limits_lock);
181 ret = queue_var_show(disk->bdi->ra_pages << (PAGE_SHIFT - 10), page);
182 mutex_unlock(&disk->queue->limits_lock);
183
184 return ret;
185}
186
187static ssize_t
188queue_ra_store(struct gendisk *disk, const char *page, size_t count)
189{
190 unsigned long ra_kb;
191 ssize_t ret;
192 struct request_queue *q = disk->queue;
193
194 ret = queue_var_store(&ra_kb, page, count);
195 if (ret < 0)
196 return ret;
197 /*
198 * The ->ra_pages change below is protected by ->limits_lock because it
199 * is usually calculated from the queue limits by
200 * queue_limits_commit_update().
201 *
202 * bdi->ra_pages reads are not serialized against bdi->ra_pages writes.
203 * Use WRITE_ONCE() to write bdi->ra_pages once.
204 */
205 mutex_lock(&q->limits_lock);
206 WRITE_ONCE(disk->bdi->ra_pages, ra_kb >> (PAGE_SHIFT - 10));
207 mutex_unlock(&q->limits_lock);
208
209 return ret;
210}
211
212#define QUEUE_SYSFS_LIMIT_SHOW(_field) \
213static ssize_t queue_##_field##_show(struct gendisk *disk, char *page) \
214{ \
215 return queue_var_show(disk->queue->limits._field, page); \
216}
217
218QUEUE_SYSFS_LIMIT_SHOW(max_segments)
219QUEUE_SYSFS_LIMIT_SHOW(max_discard_segments)
220QUEUE_SYSFS_LIMIT_SHOW(max_integrity_segments)
221QUEUE_SYSFS_LIMIT_SHOW(max_segment_size)
222QUEUE_SYSFS_LIMIT_SHOW(max_write_streams)
223QUEUE_SYSFS_LIMIT_SHOW(write_stream_granularity)
224QUEUE_SYSFS_LIMIT_SHOW(logical_block_size)
225QUEUE_SYSFS_LIMIT_SHOW(physical_block_size)
226QUEUE_SYSFS_LIMIT_SHOW(chunk_sectors)
227QUEUE_SYSFS_LIMIT_SHOW(io_min)
228QUEUE_SYSFS_LIMIT_SHOW(io_opt)
229QUEUE_SYSFS_LIMIT_SHOW(discard_granularity)
230QUEUE_SYSFS_LIMIT_SHOW(zone_write_granularity)
231QUEUE_SYSFS_LIMIT_SHOW(virt_boundary_mask)
232QUEUE_SYSFS_LIMIT_SHOW(dma_alignment)
233QUEUE_SYSFS_LIMIT_SHOW(max_open_zones)
234QUEUE_SYSFS_LIMIT_SHOW(max_active_zones)
235QUEUE_SYSFS_LIMIT_SHOW(atomic_write_unit_min)
236QUEUE_SYSFS_LIMIT_SHOW(atomic_write_unit_max)
237
238#define QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(_field) \
239static ssize_t queue_##_field##_show(struct gendisk *disk, char *page) \
240{ \
241 return sysfs_emit(page, "%llu\n", \
242 (unsigned long long)disk->queue->limits._field << \
243 SECTOR_SHIFT); \
244}
245
246QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_discard_sectors)
247QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_hw_discard_sectors)
248QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_write_zeroes_sectors)
249QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_hw_wzeroes_unmap_sectors)
250QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_wzeroes_unmap_sectors)
251QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(atomic_write_max_sectors)
252QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(atomic_write_boundary_sectors)
253QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_zone_append_sectors)
254
255#define QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(_field) \
256static ssize_t queue_##_field##_show(struct gendisk *disk, char *page) \
257{ \
258 return queue_var_show(disk->queue->limits._field >> 1, page); \
259}
260
261QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(max_sectors)
262QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(max_hw_sectors)
263
264#define QUEUE_SYSFS_SHOW_CONST(_name, _val) \
265static ssize_t queue_##_name##_show(struct gendisk *disk, char *page) \
266{ \
267 return sysfs_emit(page, "%d\n", _val); \
268}
269
270/* deprecated fields */
271QUEUE_SYSFS_SHOW_CONST(discard_zeroes_data, 0)
272QUEUE_SYSFS_SHOW_CONST(write_same_max, 0)
273QUEUE_SYSFS_SHOW_CONST(poll_delay, -1)
274
275static int queue_max_discard_sectors_store(struct gendisk *disk,
276 const char *page, size_t count, struct queue_limits *lim)
277{
278 unsigned long max_discard_bytes;
279 ssize_t ret;
280
281 ret = queue_var_store(&max_discard_bytes, page, count);
282 if (ret < 0)
283 return ret;
284
285 if (max_discard_bytes & (disk->queue->limits.discard_granularity - 1))
286 return -EINVAL;
287
288 if ((max_discard_bytes >> SECTOR_SHIFT) > UINT_MAX)
289 return -EINVAL;
290
291 lim->max_user_discard_sectors = max_discard_bytes >> SECTOR_SHIFT;
292 return 0;
293}
294
295static int queue_max_wzeroes_unmap_sectors_store(struct gendisk *disk,
296 const char *page, size_t count, struct queue_limits *lim)
297{
298 unsigned long max_zeroes_bytes, max_hw_zeroes_bytes;
299 ssize_t ret;
300
301 ret = queue_var_store(&max_zeroes_bytes, page, count);
302 if (ret < 0)
303 return ret;
304
305 max_hw_zeroes_bytes = lim->max_hw_wzeroes_unmap_sectors << SECTOR_SHIFT;
306 if (max_zeroes_bytes != 0 && max_zeroes_bytes != max_hw_zeroes_bytes)
307 return -EINVAL;
308
309 lim->max_user_wzeroes_unmap_sectors = max_zeroes_bytes >> SECTOR_SHIFT;
310 return 0;
311}
312
313static int
314queue_max_sectors_store(struct gendisk *disk, const char *page, size_t count,
315 struct queue_limits *lim)
316{
317 unsigned long max_sectors_kb;
318 ssize_t ret;
319
320 ret = queue_var_store(&max_sectors_kb, page, count);
321 if (ret < 0)
322 return ret;
323
324 lim->max_user_sectors = max_sectors_kb << 1;
325 return 0;
326}
327
328static ssize_t queue_feature_store(struct gendisk *disk, const char *page,
329 size_t count, struct queue_limits *lim, blk_features_t feature)
330{
331 unsigned long val;
332 ssize_t ret;
333
334 ret = queue_var_store(&val, page, count);
335 if (ret < 0)
336 return ret;
337
338 if (val)
339 lim->features |= feature;
340 else
341 lim->features &= ~feature;
342 return 0;
343}
344
345#define QUEUE_SYSFS_FEATURE(_name, _feature) \
346static ssize_t queue_##_name##_show(struct gendisk *disk, char *page) \
347{ \
348 return sysfs_emit(page, "%u\n", \
349 !!(disk->queue->limits.features & _feature)); \
350} \
351static int queue_##_name##_store(struct gendisk *disk, \
352 const char *page, size_t count, struct queue_limits *lim) \
353{ \
354 return queue_feature_store(disk, page, count, lim, _feature); \
355}
356
357QUEUE_SYSFS_FEATURE(rotational, BLK_FEAT_ROTATIONAL)
358QUEUE_SYSFS_FEATURE(add_random, BLK_FEAT_ADD_RANDOM)
359QUEUE_SYSFS_FEATURE(iostats, BLK_FEAT_IO_STAT)
360QUEUE_SYSFS_FEATURE(stable_writes, BLK_FEAT_STABLE_WRITES);
361
362#define QUEUE_SYSFS_FEATURE_SHOW(_name, _feature) \
363static ssize_t queue_##_name##_show(struct gendisk *disk, char *page) \
364{ \
365 return sysfs_emit(page, "%u\n", \
366 !!(disk->queue->limits.features & _feature)); \
367}
368
369QUEUE_SYSFS_FEATURE_SHOW(fua, BLK_FEAT_FUA);
370QUEUE_SYSFS_FEATURE_SHOW(dax, BLK_FEAT_DAX);
371
372static ssize_t queue_poll_show(struct gendisk *disk, char *page)
373{
374 if (queue_is_mq(disk->queue))
375 return sysfs_emit(page, "%u\n", blk_mq_can_poll(disk->queue));
376
377 return sysfs_emit(page, "%u\n",
378 !!(disk->queue->limits.features & BLK_FEAT_POLL));
379}
380
381static ssize_t queue_zoned_show(struct gendisk *disk, char *page)
382{
383 if (blk_queue_is_zoned(disk->queue))
384 return sysfs_emit(page, "host-managed\n");
385 return sysfs_emit(page, "none\n");
386}
387
388static ssize_t queue_nr_zones_show(struct gendisk *disk, char *page)
389{
390 return queue_var_show(disk_nr_zones(disk), page);
391}
392
393static ssize_t queue_iostats_passthrough_show(struct gendisk *disk, char *page)
394{
395 return queue_var_show(!!blk_queue_passthrough_stat(disk->queue), page);
396}
397
398static int queue_iostats_passthrough_store(struct gendisk *disk,
399 const char *page, size_t count, struct queue_limits *lim)
400{
401 unsigned long ios;
402 ssize_t ret;
403
404 ret = queue_var_store(&ios, page, count);
405 if (ret < 0)
406 return ret;
407
408 if (ios)
409 lim->flags |= BLK_FLAG_IOSTATS_PASSTHROUGH;
410 else
411 lim->flags &= ~BLK_FLAG_IOSTATS_PASSTHROUGH;
412 return 0;
413}
414
415static ssize_t queue_nomerges_show(struct gendisk *disk, char *page)
416{
417 return queue_var_show((blk_queue_nomerges(disk->queue) << 1) |
418 blk_queue_noxmerges(disk->queue), page);
419}
420
421static ssize_t queue_nomerges_store(struct gendisk *disk, const char *page,
422 size_t count)
423{
424 unsigned long nm;
425 struct request_queue *q = disk->queue;
426 ssize_t ret = queue_var_store(&nm, page, count);
427
428 if (ret < 0)
429 return ret;
430
431 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
432 blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
433 if (nm == 2)
434 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
435 else if (nm)
436 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
437
438 return ret;
439}
440
441static ssize_t queue_rq_affinity_show(struct gendisk *disk, char *page)
442{
443 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &disk->queue->queue_flags);
444 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &disk->queue->queue_flags);
445
446 return queue_var_show(set << force, page);
447}
448
449static ssize_t
450queue_rq_affinity_store(struct gendisk *disk, const char *page, size_t count)
451{
452 ssize_t ret = -EINVAL;
453#ifdef CONFIG_SMP
454 struct request_queue *q = disk->queue;
455 unsigned long val;
456
457 ret = queue_var_store(&val, page, count);
458 if (ret < 0)
459 return ret;
460
461 /*
462 * Here we update two queue flags each using atomic bitops, although
463 * updating two flags isn't atomic it should be harmless as those flags
464 * are accessed individually using atomic test_bit operation. So we
465 * don't grab any lock while updating these flags.
466 */
467 if (val == 2) {
468 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
469 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
470 } else if (val == 1) {
471 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
472 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
473 } else if (val == 0) {
474 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
475 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
476 }
477#endif
478 return ret;
479}
480
481static ssize_t queue_poll_delay_store(struct gendisk *disk, const char *page,
482 size_t count)
483{
484 return count;
485}
486
487static ssize_t queue_poll_store(struct gendisk *disk, const char *page,
488 size_t count)
489{
490 ssize_t ret = count;
491 struct request_queue *q = disk->queue;
492
493 if (!(q->limits.features & BLK_FEAT_POLL)) {
494 ret = -EINVAL;
495 goto out;
496 }
497
498 pr_info_ratelimited("writes to the poll attribute are ignored.\n");
499 pr_info_ratelimited("please use driver specific parameters instead.\n");
500out:
501 return ret;
502}
503
504static ssize_t queue_io_timeout_show(struct gendisk *disk, char *page)
505{
506 return sysfs_emit(page, "%u\n",
507 jiffies_to_msecs(READ_ONCE(disk->queue->rq_timeout)));
508}
509
510static ssize_t queue_io_timeout_store(struct gendisk *disk, const char *page,
511 size_t count)
512{
513 unsigned int val;
514 int err;
515 struct request_queue *q = disk->queue;
516
517 err = kstrtou32(page, 10, &val);
518 if (err || val == 0)
519 return -EINVAL;
520
521 blk_queue_rq_timeout(q, msecs_to_jiffies(val));
522
523 return count;
524}
525
526static ssize_t queue_wc_show(struct gendisk *disk, char *page)
527{
528 if (blk_queue_write_cache(disk->queue))
529 return sysfs_emit(page, "write back\n");
530 return sysfs_emit(page, "write through\n");
531}
532
533static int queue_wc_store(struct gendisk *disk, const char *page,
534 size_t count, struct queue_limits *lim)
535{
536 bool disable;
537
538 if (!strncmp(page, "write back", 10)) {
539 disable = false;
540 } else if (!strncmp(page, "write through", 13) ||
541 !strncmp(page, "none", 4)) {
542 disable = true;
543 } else {
544 return -EINVAL;
545 }
546
547 if (disable)
548 lim->flags |= BLK_FLAG_WRITE_CACHE_DISABLED;
549 else
550 lim->flags &= ~BLK_FLAG_WRITE_CACHE_DISABLED;
551 return 0;
552}
553
554#define QUEUE_RO_ENTRY(_prefix, _name) \
555static struct queue_sysfs_entry _prefix##_entry = { \
556 .attr = { .name = _name, .mode = 0444 }, \
557 .show = _prefix##_show, \
558};
559
560#define QUEUE_RW_ENTRY(_prefix, _name) \
561static struct queue_sysfs_entry _prefix##_entry = { \
562 .attr = { .name = _name, .mode = 0644 }, \
563 .show = _prefix##_show, \
564 .store = _prefix##_store, \
565};
566
567#define QUEUE_LIM_RO_ENTRY(_prefix, _name) \
568static struct queue_sysfs_entry _prefix##_entry = { \
569 .attr = { .name = _name, .mode = 0444 }, \
570 .show_limit = _prefix##_show, \
571}
572
573#define QUEUE_LIM_RW_ENTRY(_prefix, _name) \
574static struct queue_sysfs_entry _prefix##_entry = { \
575 .attr = { .name = _name, .mode = 0644 }, \
576 .show_limit = _prefix##_show, \
577 .store_limit = _prefix##_store, \
578}
579
580QUEUE_RW_ENTRY(queue_requests, "nr_requests");
581QUEUE_RW_ENTRY(queue_async_depth, "async_depth");
582QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
583QUEUE_LIM_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
584QUEUE_LIM_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
585QUEUE_LIM_RO_ENTRY(queue_max_segments, "max_segments");
586QUEUE_LIM_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
587QUEUE_LIM_RO_ENTRY(queue_max_segment_size, "max_segment_size");
588QUEUE_LIM_RO_ENTRY(queue_max_write_streams, "max_write_streams");
589QUEUE_LIM_RO_ENTRY(queue_write_stream_granularity, "write_stream_granularity");
590QUEUE_RW_ENTRY(elv_iosched, "scheduler");
591
592QUEUE_LIM_RO_ENTRY(queue_logical_block_size, "logical_block_size");
593QUEUE_LIM_RO_ENTRY(queue_physical_block_size, "physical_block_size");
594QUEUE_LIM_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
595QUEUE_LIM_RO_ENTRY(queue_io_min, "minimum_io_size");
596QUEUE_LIM_RO_ENTRY(queue_io_opt, "optimal_io_size");
597
598QUEUE_LIM_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
599QUEUE_LIM_RO_ENTRY(queue_discard_granularity, "discard_granularity");
600QUEUE_LIM_RO_ENTRY(queue_max_hw_discard_sectors, "discard_max_hw_bytes");
601QUEUE_LIM_RW_ENTRY(queue_max_discard_sectors, "discard_max_bytes");
602QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
603
604QUEUE_LIM_RO_ENTRY(queue_atomic_write_max_sectors, "atomic_write_max_bytes");
605QUEUE_LIM_RO_ENTRY(queue_atomic_write_boundary_sectors,
606 "atomic_write_boundary_bytes");
607QUEUE_LIM_RO_ENTRY(queue_atomic_write_unit_max, "atomic_write_unit_max_bytes");
608QUEUE_LIM_RO_ENTRY(queue_atomic_write_unit_min, "atomic_write_unit_min_bytes");
609
610QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
611QUEUE_LIM_RO_ENTRY(queue_max_write_zeroes_sectors, "write_zeroes_max_bytes");
612QUEUE_LIM_RO_ENTRY(queue_max_hw_wzeroes_unmap_sectors,
613 "write_zeroes_unmap_max_hw_bytes");
614QUEUE_LIM_RW_ENTRY(queue_max_wzeroes_unmap_sectors,
615 "write_zeroes_unmap_max_bytes");
616QUEUE_LIM_RO_ENTRY(queue_max_zone_append_sectors, "zone_append_max_bytes");
617QUEUE_LIM_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
618
619QUEUE_LIM_RO_ENTRY(queue_zoned, "zoned");
620QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
621QUEUE_LIM_RO_ENTRY(queue_max_open_zones, "max_open_zones");
622QUEUE_LIM_RO_ENTRY(queue_max_active_zones, "max_active_zones");
623
624QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
625QUEUE_LIM_RW_ENTRY(queue_iostats_passthrough, "iostats_passthrough");
626QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
627QUEUE_RW_ENTRY(queue_poll, "io_poll");
628QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
629QUEUE_LIM_RW_ENTRY(queue_wc, "write_cache");
630QUEUE_LIM_RO_ENTRY(queue_fua, "fua");
631QUEUE_LIM_RO_ENTRY(queue_dax, "dax");
632QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
633QUEUE_LIM_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
634QUEUE_LIM_RO_ENTRY(queue_dma_alignment, "dma_alignment");
635
636/* legacy alias for logical_block_size: */
637static struct queue_sysfs_entry queue_hw_sector_size_entry = {
638 .attr = {.name = "hw_sector_size", .mode = 0444 },
639 .show_limit = queue_logical_block_size_show,
640};
641
642QUEUE_LIM_RW_ENTRY(queue_rotational, "rotational");
643QUEUE_LIM_RW_ENTRY(queue_iostats, "iostats");
644QUEUE_LIM_RW_ENTRY(queue_add_random, "add_random");
645QUEUE_LIM_RW_ENTRY(queue_stable_writes, "stable_writes");
646
647#ifdef CONFIG_BLK_WBT
648static ssize_t queue_var_store64(s64 *var, const char *page)
649{
650 int err;
651 s64 v;
652
653 err = kstrtos64(page, 10, &v);
654 if (err < 0)
655 return err;
656
657 *var = v;
658 return 0;
659}
660
661static ssize_t queue_wb_lat_show(struct gendisk *disk, char *page)
662{
663 ssize_t ret;
664 struct request_queue *q = disk->queue;
665
666 mutex_lock(&disk->rqos_state_mutex);
667 if (!wbt_rq_qos(q)) {
668 ret = -EINVAL;
669 goto out;
670 }
671
672 if (wbt_disabled(q)) {
673 ret = sysfs_emit(page, "0\n");
674 goto out;
675 }
676
677 ret = sysfs_emit(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
678out:
679 mutex_unlock(&disk->rqos_state_mutex);
680 return ret;
681}
682
683static ssize_t queue_wb_lat_store(struct gendisk *disk, const char *page,
684 size_t count)
685{
686 ssize_t ret;
687 s64 val;
688
689 ret = queue_var_store64(&val, page);
690 if (ret < 0)
691 return ret;
692 if (val < -1)
693 return -EINVAL;
694
695 ret = wbt_set_lat(disk, val);
696 return ret ? ret : count;
697}
698
699QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
700#endif
701
702/* Common attributes for bio-based and request-based queues. */
703static struct attribute *queue_attrs[] = {
704 /*
705 * Attributes which are protected with q->limits_lock.
706 */
707 &queue_max_hw_sectors_entry.attr,
708 &queue_max_sectors_entry.attr,
709 &queue_max_segments_entry.attr,
710 &queue_max_discard_segments_entry.attr,
711 &queue_max_integrity_segments_entry.attr,
712 &queue_max_segment_size_entry.attr,
713 &queue_max_write_streams_entry.attr,
714 &queue_write_stream_granularity_entry.attr,
715 &queue_hw_sector_size_entry.attr,
716 &queue_logical_block_size_entry.attr,
717 &queue_physical_block_size_entry.attr,
718 &queue_chunk_sectors_entry.attr,
719 &queue_io_min_entry.attr,
720 &queue_io_opt_entry.attr,
721 &queue_discard_granularity_entry.attr,
722 &queue_max_discard_sectors_entry.attr,
723 &queue_max_hw_discard_sectors_entry.attr,
724 &queue_atomic_write_max_sectors_entry.attr,
725 &queue_atomic_write_boundary_sectors_entry.attr,
726 &queue_atomic_write_unit_min_entry.attr,
727 &queue_atomic_write_unit_max_entry.attr,
728 &queue_max_write_zeroes_sectors_entry.attr,
729 &queue_max_hw_wzeroes_unmap_sectors_entry.attr,
730 &queue_max_wzeroes_unmap_sectors_entry.attr,
731 &queue_max_zone_append_sectors_entry.attr,
732 &queue_zone_write_granularity_entry.attr,
733 &queue_rotational_entry.attr,
734 &queue_zoned_entry.attr,
735 &queue_max_open_zones_entry.attr,
736 &queue_max_active_zones_entry.attr,
737 &queue_iostats_passthrough_entry.attr,
738 &queue_iostats_entry.attr,
739 &queue_stable_writes_entry.attr,
740 &queue_add_random_entry.attr,
741 &queue_wc_entry.attr,
742 &queue_fua_entry.attr,
743 &queue_dax_entry.attr,
744 &queue_virt_boundary_mask_entry.attr,
745 &queue_dma_alignment_entry.attr,
746 &queue_ra_entry.attr,
747
748 /*
749 * Attributes which don't require locking.
750 */
751 &queue_discard_zeroes_data_entry.attr,
752 &queue_write_same_max_entry.attr,
753 &queue_nr_zones_entry.attr,
754 &queue_nomerges_entry.attr,
755 &queue_poll_entry.attr,
756 &queue_poll_delay_entry.attr,
757
758 NULL,
759};
760
761/* Request-based queue attributes that are not relevant for bio-based queues. */
762static struct attribute *blk_mq_queue_attrs[] = {
763 /*
764 * Attributes which require some form of locking other than
765 * q->sysfs_lock.
766 */
767 &elv_iosched_entry.attr,
768 &queue_requests_entry.attr,
769 &queue_async_depth_entry.attr,
770#ifdef CONFIG_BLK_WBT
771 &queue_wb_lat_entry.attr,
772#endif
773 /*
774 * Attributes which don't require locking.
775 */
776 &queue_rq_affinity_entry.attr,
777 &queue_io_timeout_entry.attr,
778
779 NULL,
780};
781
782static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
783 int n)
784{
785 struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
786 struct request_queue *q = disk->queue;
787
788 if ((attr == &queue_max_open_zones_entry.attr ||
789 attr == &queue_max_active_zones_entry.attr) &&
790 !blk_queue_is_zoned(q))
791 return 0;
792
793 return attr->mode;
794}
795
796static umode_t blk_mq_queue_attr_visible(struct kobject *kobj,
797 struct attribute *attr, int n)
798{
799 struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
800 struct request_queue *q = disk->queue;
801
802 if (!queue_is_mq(q))
803 return 0;
804
805 if (attr == &queue_io_timeout_entry.attr && !q->mq_ops->timeout)
806 return 0;
807
808 return attr->mode;
809}
810
811static struct attribute_group queue_attr_group = {
812 .attrs = queue_attrs,
813 .is_visible = queue_attr_visible,
814};
815
816static struct attribute_group blk_mq_queue_attr_group = {
817 .attrs = blk_mq_queue_attrs,
818 .is_visible = blk_mq_queue_attr_visible,
819};
820
821#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
822
823static ssize_t
824queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
825{
826 struct queue_sysfs_entry *entry = to_queue(attr);
827 struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
828
829 if (!entry->show && !entry->show_limit)
830 return -EIO;
831
832 if (entry->show_limit) {
833 ssize_t res;
834
835 mutex_lock(&disk->queue->limits_lock);
836 res = entry->show_limit(disk, page);
837 mutex_unlock(&disk->queue->limits_lock);
838 return res;
839 }
840
841 return entry->show(disk, page);
842}
843
844static ssize_t
845queue_attr_store(struct kobject *kobj, struct attribute *attr,
846 const char *page, size_t length)
847{
848 struct queue_sysfs_entry *entry = to_queue(attr);
849 struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
850 struct request_queue *q = disk->queue;
851
852 if (!entry->store_limit && !entry->store)
853 return -EIO;
854
855 if (entry->store_limit) {
856 ssize_t res;
857
858 struct queue_limits lim = queue_limits_start_update(q);
859
860 res = entry->store_limit(disk, page, length, &lim);
861 if (res < 0) {
862 queue_limits_cancel_update(q);
863 return res;
864 }
865
866 res = queue_limits_commit_update_frozen(q, &lim);
867 if (res)
868 return res;
869 return length;
870 }
871
872 return entry->store(disk, page, length);
873}
874
875static const struct sysfs_ops queue_sysfs_ops = {
876 .show = queue_attr_show,
877 .store = queue_attr_store,
878};
879
880static const struct attribute_group *blk_queue_attr_groups[] = {
881 &queue_attr_group,
882 &blk_mq_queue_attr_group,
883 NULL
884};
885
886static void blk_queue_release(struct kobject *kobj)
887{
888 /* nothing to do here, all data is associated with the parent gendisk */
889}
890
891const struct kobj_type blk_queue_ktype = {
892 .default_groups = blk_queue_attr_groups,
893 .sysfs_ops = &queue_sysfs_ops,
894 .release = blk_queue_release,
895};
896
897static void blk_debugfs_remove(struct gendisk *disk)
898{
899 struct request_queue *q = disk->queue;
900
901 blk_debugfs_lock_nomemsave(q);
902 blk_trace_shutdown(q);
903 debugfs_remove_recursive(q->debugfs_dir);
904 q->debugfs_dir = NULL;
905 q->sched_debugfs_dir = NULL;
906 q->rqos_debugfs_dir = NULL;
907 blk_debugfs_unlock_nomemrestore(q);
908}
909
910/**
911 * blk_register_queue - register a block layer queue with sysfs
912 * @disk: Disk of which the request queue should be registered with sysfs.
913 */
914int blk_register_queue(struct gendisk *disk)
915{
916 struct request_queue *q = disk->queue;
917 unsigned int memflags;
918 int ret;
919
920 ret = kobject_add(&disk->queue_kobj, &disk_to_dev(disk)->kobj, "queue");
921 if (ret < 0)
922 return ret;
923
924 if (queue_is_mq(q)) {
925 ret = blk_mq_sysfs_register(disk);
926 if (ret)
927 goto out_del_queue_kobj;
928 }
929 mutex_lock(&q->sysfs_lock);
930
931 memflags = blk_debugfs_lock(q);
932 q->debugfs_dir = debugfs_create_dir(disk->disk_name, blk_debugfs_root);
933 if (queue_is_mq(q))
934 blk_mq_debugfs_register(q);
935 blk_debugfs_unlock(q, memflags);
936
937 ret = disk_register_independent_access_ranges(disk);
938 if (ret)
939 goto out_debugfs_remove;
940
941 ret = blk_crypto_sysfs_register(disk);
942 if (ret)
943 goto out_unregister_ia_ranges;
944
945 if (queue_is_mq(q))
946 elevator_set_default(q);
947
948 blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
949 wbt_init_enable_default(disk);
950
951 /* Now everything is ready and send out KOBJ_ADD uevent */
952 kobject_uevent(&disk->queue_kobj, KOBJ_ADD);
953 if (q->elevator)
954 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
955 mutex_unlock(&q->sysfs_lock);
956
957 /*
958 * SCSI probing may synchronously create and destroy a lot of
959 * request_queues for non-existent devices. Shutting down a fully
960 * functional queue takes measureable wallclock time as RCU grace
961 * periods are involved. To avoid excessive latency in these
962 * cases, a request_queue starts out in a degraded mode which is
963 * faster to shut down and is made fully functional here as
964 * request_queues for non-existent devices never get registered.
965 */
966 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
967 percpu_ref_switch_to_percpu(&q->q_usage_counter);
968
969 return ret;
970
971out_unregister_ia_ranges:
972 disk_unregister_independent_access_ranges(disk);
973out_debugfs_remove:
974 blk_debugfs_remove(disk);
975 mutex_unlock(&q->sysfs_lock);
976 if (queue_is_mq(q))
977 blk_mq_sysfs_unregister(disk);
978out_del_queue_kobj:
979 kobject_del(&disk->queue_kobj);
980 return ret;
981}
982
983/**
984 * blk_unregister_queue - counterpart of blk_register_queue()
985 * @disk: Disk of which the request queue should be unregistered from sysfs.
986 *
987 * Note: the caller is responsible for guaranteeing that this function is called
988 * after blk_register_queue() has finished.
989 */
990void blk_unregister_queue(struct gendisk *disk)
991{
992 struct request_queue *q = disk->queue;
993
994 if (WARN_ON(!q))
995 return;
996
997 /* Return early if disk->queue was never registered. */
998 if (!blk_queue_registered(q))
999 return;
1000
1001 /*
1002 * Since sysfs_remove_dir() prevents adding new directory entries
1003 * before removal of existing entries starts, protect against
1004 * concurrent elv_iosched_store() calls.
1005 */
1006 mutex_lock(&q->sysfs_lock);
1007 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1008 mutex_unlock(&q->sysfs_lock);
1009
1010 /*
1011 * Remove the sysfs attributes before unregistering the queue data
1012 * structures that can be modified through sysfs.
1013 */
1014 if (queue_is_mq(q))
1015 blk_mq_sysfs_unregister(disk);
1016 blk_crypto_sysfs_unregister(disk);
1017
1018 mutex_lock(&q->sysfs_lock);
1019 disk_unregister_independent_access_ranges(disk);
1020 mutex_unlock(&q->sysfs_lock);
1021
1022 /* Now that we've deleted all child objects, we can delete the queue. */
1023 kobject_uevent(&disk->queue_kobj, KOBJ_REMOVE);
1024 kobject_del(&disk->queue_kobj);
1025
1026 if (queue_is_mq(q))
1027 elevator_set_none(q);
1028
1029 blk_debugfs_remove(disk);
1030}