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/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 (!q->request_fn && !q->mq_ops)
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 if (q->request_fn)
82 err = blk_update_nr_requests(q, nr);
83 else
84 err = blk_mq_update_nr_requests(q, nr);
85
86 if (err)
87 return err;
88
89 return ret;
90}
91
92static ssize_t queue_ra_show(struct request_queue *q, char *page)
93{
94 unsigned long ra_kb = q->backing_dev_info->ra_pages <<
95 (PAGE_SHIFT - 10);
96
97 return queue_var_show(ra_kb, (page));
98}
99
100static ssize_t
101queue_ra_store(struct request_queue *q, const char *page, size_t count)
102{
103 unsigned long ra_kb;
104 ssize_t ret = queue_var_store(&ra_kb, page, count);
105
106 if (ret < 0)
107 return ret;
108
109 q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
110
111 return ret;
112}
113
114static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
115{
116 int max_sectors_kb = queue_max_sectors(q) >> 1;
117
118 return queue_var_show(max_sectors_kb, (page));
119}
120
121static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
122{
123 return queue_var_show(queue_max_segments(q), (page));
124}
125
126static ssize_t queue_max_discard_segments_show(struct request_queue *q,
127 char *page)
128{
129 return queue_var_show(queue_max_discard_segments(q), (page));
130}
131
132static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
133{
134 return queue_var_show(q->limits.max_integrity_segments, (page));
135}
136
137static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
138{
139 if (blk_queue_cluster(q))
140 return queue_var_show(queue_max_segment_size(q), (page));
141
142 return queue_var_show(PAGE_SIZE, (page));
143}
144
145static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
146{
147 return queue_var_show(queue_logical_block_size(q), page);
148}
149
150static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
151{
152 return queue_var_show(queue_physical_block_size(q), page);
153}
154
155static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
156{
157 return queue_var_show(q->limits.chunk_sectors, page);
158}
159
160static ssize_t queue_io_min_show(struct request_queue *q, char *page)
161{
162 return queue_var_show(queue_io_min(q), page);
163}
164
165static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
166{
167 return queue_var_show(queue_io_opt(q), page);
168}
169
170static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
171{
172 return queue_var_show(q->limits.discard_granularity, page);
173}
174
175static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
176{
177
178 return sprintf(page, "%llu\n",
179 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
180}
181
182static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
183{
184 return sprintf(page, "%llu\n",
185 (unsigned long long)q->limits.max_discard_sectors << 9);
186}
187
188static ssize_t queue_discard_max_store(struct request_queue *q,
189 const char *page, size_t count)
190{
191 unsigned long max_discard;
192 ssize_t ret = queue_var_store(&max_discard, page, count);
193
194 if (ret < 0)
195 return ret;
196
197 if (max_discard & (q->limits.discard_granularity - 1))
198 return -EINVAL;
199
200 max_discard >>= 9;
201 if (max_discard > UINT_MAX)
202 return -EINVAL;
203
204 if (max_discard > q->limits.max_hw_discard_sectors)
205 max_discard = q->limits.max_hw_discard_sectors;
206
207 q->limits.max_discard_sectors = max_discard;
208 return ret;
209}
210
211static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
212{
213 return queue_var_show(0, page);
214}
215
216static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
217{
218 return sprintf(page, "%llu\n",
219 (unsigned long long)q->limits.max_write_same_sectors << 9);
220}
221
222static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
223{
224 return sprintf(page, "%llu\n",
225 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
226}
227
228static ssize_t
229queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
230{
231 unsigned long max_sectors_kb,
232 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
233 page_kb = 1 << (PAGE_SHIFT - 10);
234 ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
235
236 if (ret < 0)
237 return ret;
238
239 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
240 q->limits.max_dev_sectors >> 1);
241
242 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
243 return -EINVAL;
244
245 spin_lock_irq(q->queue_lock);
246 q->limits.max_sectors = max_sectors_kb << 1;
247 q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
248 spin_unlock_irq(q->queue_lock);
249
250 return ret;
251}
252
253static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
254{
255 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
256
257 return queue_var_show(max_hw_sectors_kb, (page));
258}
259
260#define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \
261static ssize_t \
262queue_show_##name(struct request_queue *q, char *page) \
263{ \
264 int bit; \
265 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \
266 return queue_var_show(neg ? !bit : bit, page); \
267} \
268static ssize_t \
269queue_store_##name(struct request_queue *q, const char *page, size_t count) \
270{ \
271 unsigned long val; \
272 ssize_t ret; \
273 ret = queue_var_store(&val, page, count); \
274 if (ret < 0) \
275 return ret; \
276 if (neg) \
277 val = !val; \
278 \
279 if (val) \
280 blk_queue_flag_set(QUEUE_FLAG_##flag, q); \
281 else \
282 blk_queue_flag_clear(QUEUE_FLAG_##flag, q); \
283 return ret; \
284}
285
286QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
287QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
288QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
289#undef QUEUE_SYSFS_BIT_FNS
290
291static ssize_t queue_zoned_show(struct request_queue *q, char *page)
292{
293 switch (blk_queue_zoned_model(q)) {
294 case BLK_ZONED_HA:
295 return sprintf(page, "host-aware\n");
296 case BLK_ZONED_HM:
297 return sprintf(page, "host-managed\n");
298 default:
299 return sprintf(page, "none\n");
300 }
301}
302
303static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
304{
305 return queue_var_show((blk_queue_nomerges(q) << 1) |
306 blk_queue_noxmerges(q), page);
307}
308
309static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
310 size_t count)
311{
312 unsigned long nm;
313 ssize_t ret = queue_var_store(&nm, page, count);
314
315 if (ret < 0)
316 return ret;
317
318 spin_lock_irq(q->queue_lock);
319 queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
320 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
321 if (nm == 2)
322 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
323 else if (nm)
324 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
325 spin_unlock_irq(q->queue_lock);
326
327 return ret;
328}
329
330static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
331{
332 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
333 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
334
335 return queue_var_show(set << force, page);
336}
337
338static ssize_t
339queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
340{
341 ssize_t ret = -EINVAL;
342#ifdef CONFIG_SMP
343 unsigned long val;
344
345 ret = queue_var_store(&val, page, count);
346 if (ret < 0)
347 return ret;
348
349 spin_lock_irq(q->queue_lock);
350 if (val == 2) {
351 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
352 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
353 } else if (val == 1) {
354 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
355 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
356 } else if (val == 0) {
357 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
358 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
359 }
360 spin_unlock_irq(q->queue_lock);
361#endif
362 return ret;
363}
364
365static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
366{
367 int val;
368
369 if (q->poll_nsec == -1)
370 val = -1;
371 else
372 val = q->poll_nsec / 1000;
373
374 return sprintf(page, "%d\n", val);
375}
376
377static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
378 size_t count)
379{
380 int err, val;
381
382 if (!q->mq_ops || !q->mq_ops->poll)
383 return -EINVAL;
384
385 err = kstrtoint(page, 10, &val);
386 if (err < 0)
387 return err;
388
389 if (val == -1)
390 q->poll_nsec = -1;
391 else
392 q->poll_nsec = val * 1000;
393
394 return count;
395}
396
397static ssize_t queue_poll_show(struct request_queue *q, char *page)
398{
399 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
400}
401
402static ssize_t queue_poll_store(struct request_queue *q, const char *page,
403 size_t count)
404{
405 unsigned long poll_on;
406 ssize_t ret;
407
408 if (!q->mq_ops || !q->mq_ops->poll)
409 return -EINVAL;
410
411 ret = queue_var_store(&poll_on, page, count);
412 if (ret < 0)
413 return ret;
414
415 if (poll_on)
416 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
417 else
418 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
419
420 return ret;
421}
422
423static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
424{
425 if (!q->rq_wb)
426 return -EINVAL;
427
428 return sprintf(page, "%llu\n", div_u64(q->rq_wb->min_lat_nsec, 1000));
429}
430
431static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
432 size_t count)
433{
434 struct rq_wb *rwb;
435 ssize_t ret;
436 s64 val;
437
438 ret = queue_var_store64(&val, page);
439 if (ret < 0)
440 return ret;
441 if (val < -1)
442 return -EINVAL;
443
444 rwb = q->rq_wb;
445 if (!rwb) {
446 ret = wbt_init(q);
447 if (ret)
448 return ret;
449 }
450
451 rwb = q->rq_wb;
452 if (val == -1)
453 rwb->min_lat_nsec = wbt_default_latency_nsec(q);
454 else if (val >= 0)
455 rwb->min_lat_nsec = val * 1000ULL;
456
457 if (rwb->enable_state == WBT_STATE_ON_DEFAULT)
458 rwb->enable_state = WBT_STATE_ON_MANUAL;
459
460 wbt_update_limits(rwb);
461 return count;
462}
463
464static ssize_t queue_wc_show(struct request_queue *q, char *page)
465{
466 if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
467 return sprintf(page, "write back\n");
468
469 return sprintf(page, "write through\n");
470}
471
472static ssize_t queue_wc_store(struct request_queue *q, const char *page,
473 size_t count)
474{
475 int set = -1;
476
477 if (!strncmp(page, "write back", 10))
478 set = 1;
479 else if (!strncmp(page, "write through", 13) ||
480 !strncmp(page, "none", 4))
481 set = 0;
482
483 if (set == -1)
484 return -EINVAL;
485
486 if (set)
487 blk_queue_flag_set(QUEUE_FLAG_WC, q);
488 else
489 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
490
491 return count;
492}
493
494static ssize_t queue_dax_show(struct request_queue *q, char *page)
495{
496 return queue_var_show(blk_queue_dax(q), page);
497}
498
499static struct queue_sysfs_entry queue_requests_entry = {
500 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
501 .show = queue_requests_show,
502 .store = queue_requests_store,
503};
504
505static struct queue_sysfs_entry queue_ra_entry = {
506 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
507 .show = queue_ra_show,
508 .store = queue_ra_store,
509};
510
511static struct queue_sysfs_entry queue_max_sectors_entry = {
512 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
513 .show = queue_max_sectors_show,
514 .store = queue_max_sectors_store,
515};
516
517static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
518 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
519 .show = queue_max_hw_sectors_show,
520};
521
522static struct queue_sysfs_entry queue_max_segments_entry = {
523 .attr = {.name = "max_segments", .mode = S_IRUGO },
524 .show = queue_max_segments_show,
525};
526
527static struct queue_sysfs_entry queue_max_discard_segments_entry = {
528 .attr = {.name = "max_discard_segments", .mode = S_IRUGO },
529 .show = queue_max_discard_segments_show,
530};
531
532static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
533 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
534 .show = queue_max_integrity_segments_show,
535};
536
537static struct queue_sysfs_entry queue_max_segment_size_entry = {
538 .attr = {.name = "max_segment_size", .mode = S_IRUGO },
539 .show = queue_max_segment_size_show,
540};
541
542static struct queue_sysfs_entry queue_iosched_entry = {
543 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
544 .show = elv_iosched_show,
545 .store = elv_iosched_store,
546};
547
548static struct queue_sysfs_entry queue_hw_sector_size_entry = {
549 .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
550 .show = queue_logical_block_size_show,
551};
552
553static struct queue_sysfs_entry queue_logical_block_size_entry = {
554 .attr = {.name = "logical_block_size", .mode = S_IRUGO },
555 .show = queue_logical_block_size_show,
556};
557
558static struct queue_sysfs_entry queue_physical_block_size_entry = {
559 .attr = {.name = "physical_block_size", .mode = S_IRUGO },
560 .show = queue_physical_block_size_show,
561};
562
563static struct queue_sysfs_entry queue_chunk_sectors_entry = {
564 .attr = {.name = "chunk_sectors", .mode = S_IRUGO },
565 .show = queue_chunk_sectors_show,
566};
567
568static struct queue_sysfs_entry queue_io_min_entry = {
569 .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
570 .show = queue_io_min_show,
571};
572
573static struct queue_sysfs_entry queue_io_opt_entry = {
574 .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
575 .show = queue_io_opt_show,
576};
577
578static struct queue_sysfs_entry queue_discard_granularity_entry = {
579 .attr = {.name = "discard_granularity", .mode = S_IRUGO },
580 .show = queue_discard_granularity_show,
581};
582
583static struct queue_sysfs_entry queue_discard_max_hw_entry = {
584 .attr = {.name = "discard_max_hw_bytes", .mode = S_IRUGO },
585 .show = queue_discard_max_hw_show,
586};
587
588static struct queue_sysfs_entry queue_discard_max_entry = {
589 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO | S_IWUSR },
590 .show = queue_discard_max_show,
591 .store = queue_discard_max_store,
592};
593
594static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
595 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
596 .show = queue_discard_zeroes_data_show,
597};
598
599static struct queue_sysfs_entry queue_write_same_max_entry = {
600 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
601 .show = queue_write_same_max_show,
602};
603
604static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
605 .attr = {.name = "write_zeroes_max_bytes", .mode = S_IRUGO },
606 .show = queue_write_zeroes_max_show,
607};
608
609static struct queue_sysfs_entry queue_nonrot_entry = {
610 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
611 .show = queue_show_nonrot,
612 .store = queue_store_nonrot,
613};
614
615static struct queue_sysfs_entry queue_zoned_entry = {
616 .attr = {.name = "zoned", .mode = S_IRUGO },
617 .show = queue_zoned_show,
618};
619
620static struct queue_sysfs_entry queue_nomerges_entry = {
621 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
622 .show = queue_nomerges_show,
623 .store = queue_nomerges_store,
624};
625
626static struct queue_sysfs_entry queue_rq_affinity_entry = {
627 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
628 .show = queue_rq_affinity_show,
629 .store = queue_rq_affinity_store,
630};
631
632static struct queue_sysfs_entry queue_iostats_entry = {
633 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
634 .show = queue_show_iostats,
635 .store = queue_store_iostats,
636};
637
638static struct queue_sysfs_entry queue_random_entry = {
639 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
640 .show = queue_show_random,
641 .store = queue_store_random,
642};
643
644static struct queue_sysfs_entry queue_poll_entry = {
645 .attr = {.name = "io_poll", .mode = S_IRUGO | S_IWUSR },
646 .show = queue_poll_show,
647 .store = queue_poll_store,
648};
649
650static struct queue_sysfs_entry queue_poll_delay_entry = {
651 .attr = {.name = "io_poll_delay", .mode = S_IRUGO | S_IWUSR },
652 .show = queue_poll_delay_show,
653 .store = queue_poll_delay_store,
654};
655
656static struct queue_sysfs_entry queue_wc_entry = {
657 .attr = {.name = "write_cache", .mode = S_IRUGO | S_IWUSR },
658 .show = queue_wc_show,
659 .store = queue_wc_store,
660};
661
662static struct queue_sysfs_entry queue_dax_entry = {
663 .attr = {.name = "dax", .mode = S_IRUGO },
664 .show = queue_dax_show,
665};
666
667static struct queue_sysfs_entry queue_wb_lat_entry = {
668 .attr = {.name = "wbt_lat_usec", .mode = S_IRUGO | S_IWUSR },
669 .show = queue_wb_lat_show,
670 .store = queue_wb_lat_store,
671};
672
673#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
674static struct queue_sysfs_entry throtl_sample_time_entry = {
675 .attr = {.name = "throttle_sample_time", .mode = S_IRUGO | S_IWUSR },
676 .show = blk_throtl_sample_time_show,
677 .store = blk_throtl_sample_time_store,
678};
679#endif
680
681static struct attribute *default_attrs[] = {
682 &queue_requests_entry.attr,
683 &queue_ra_entry.attr,
684 &queue_max_hw_sectors_entry.attr,
685 &queue_max_sectors_entry.attr,
686 &queue_max_segments_entry.attr,
687 &queue_max_discard_segments_entry.attr,
688 &queue_max_integrity_segments_entry.attr,
689 &queue_max_segment_size_entry.attr,
690 &queue_iosched_entry.attr,
691 &queue_hw_sector_size_entry.attr,
692 &queue_logical_block_size_entry.attr,
693 &queue_physical_block_size_entry.attr,
694 &queue_chunk_sectors_entry.attr,
695 &queue_io_min_entry.attr,
696 &queue_io_opt_entry.attr,
697 &queue_discard_granularity_entry.attr,
698 &queue_discard_max_entry.attr,
699 &queue_discard_max_hw_entry.attr,
700 &queue_discard_zeroes_data_entry.attr,
701 &queue_write_same_max_entry.attr,
702 &queue_write_zeroes_max_entry.attr,
703 &queue_nonrot_entry.attr,
704 &queue_zoned_entry.attr,
705 &queue_nomerges_entry.attr,
706 &queue_rq_affinity_entry.attr,
707 &queue_iostats_entry.attr,
708 &queue_random_entry.attr,
709 &queue_poll_entry.attr,
710 &queue_wc_entry.attr,
711 &queue_dax_entry.attr,
712 &queue_wb_lat_entry.attr,
713 &queue_poll_delay_entry.attr,
714#ifdef CONFIG_BLK_DEV_THROTTLING_LOW
715 &throtl_sample_time_entry.attr,
716#endif
717 NULL,
718};
719
720#define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
721
722static ssize_t
723queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
724{
725 struct queue_sysfs_entry *entry = to_queue(attr);
726 struct request_queue *q =
727 container_of(kobj, struct request_queue, kobj);
728 ssize_t res;
729
730 if (!entry->show)
731 return -EIO;
732 mutex_lock(&q->sysfs_lock);
733 if (blk_queue_dying(q)) {
734 mutex_unlock(&q->sysfs_lock);
735 return -ENOENT;
736 }
737 res = entry->show(q, page);
738 mutex_unlock(&q->sysfs_lock);
739 return res;
740}
741
742static ssize_t
743queue_attr_store(struct kobject *kobj, struct attribute *attr,
744 const char *page, size_t length)
745{
746 struct queue_sysfs_entry *entry = to_queue(attr);
747 struct request_queue *q;
748 ssize_t res;
749
750 if (!entry->store)
751 return -EIO;
752
753 q = container_of(kobj, struct request_queue, kobj);
754 mutex_lock(&q->sysfs_lock);
755 if (blk_queue_dying(q)) {
756 mutex_unlock(&q->sysfs_lock);
757 return -ENOENT;
758 }
759 res = entry->store(q, page, length);
760 mutex_unlock(&q->sysfs_lock);
761 return res;
762}
763
764static void blk_free_queue_rcu(struct rcu_head *rcu_head)
765{
766 struct request_queue *q = container_of(rcu_head, struct request_queue,
767 rcu_head);
768 kmem_cache_free(blk_requestq_cachep, q);
769}
770
771/**
772 * __blk_release_queue - release a request queue when it is no longer needed
773 * @work: pointer to the release_work member of the request queue to be released
774 *
775 * Description:
776 * blk_release_queue is the counterpart of blk_init_queue(). It should be
777 * called when a request queue is being released; typically when a block
778 * device is being de-registered. Its primary task it to free the queue
779 * itself.
780 *
781 * Notes:
782 * The low level driver must have finished any outstanding requests first
783 * via blk_cleanup_queue().
784 *
785 * Although blk_release_queue() may be called with preemption disabled,
786 * __blk_release_queue() may sleep.
787 */
788static void __blk_release_queue(struct work_struct *work)
789{
790 struct request_queue *q = container_of(work, typeof(*q), release_work);
791
792 if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
793 blk_stat_remove_callback(q, q->poll_cb);
794 blk_stat_free_callback(q->poll_cb);
795
796 blk_free_queue_stats(q->stats);
797
798 blk_exit_rl(q, &q->root_rl);
799
800 if (q->queue_tags)
801 __blk_queue_free_tags(q);
802
803 if (!q->mq_ops) {
804 if (q->exit_rq_fn)
805 q->exit_rq_fn(q, q->fq->flush_rq);
806 blk_free_flush_queue(q->fq);
807 } else {
808 blk_mq_release(q);
809 }
810
811 blk_trace_shutdown(q);
812
813 if (q->mq_ops)
814 blk_mq_debugfs_unregister(q);
815
816 if (q->bio_split)
817 bioset_free(q->bio_split);
818
819 ida_simple_remove(&blk_queue_ida, q->id);
820 call_rcu(&q->rcu_head, blk_free_queue_rcu);
821}
822
823static void blk_release_queue(struct kobject *kobj)
824{
825 struct request_queue *q =
826 container_of(kobj, struct request_queue, kobj);
827
828 INIT_WORK(&q->release_work, __blk_release_queue);
829 schedule_work(&q->release_work);
830}
831
832static const struct sysfs_ops queue_sysfs_ops = {
833 .show = queue_attr_show,
834 .store = queue_attr_store,
835};
836
837struct kobj_type blk_queue_ktype = {
838 .sysfs_ops = &queue_sysfs_ops,
839 .default_attrs = default_attrs,
840 .release = blk_release_queue,
841};
842
843/**
844 * blk_register_queue - register a block layer queue with sysfs
845 * @disk: Disk of which the request queue should be registered with sysfs.
846 */
847int blk_register_queue(struct gendisk *disk)
848{
849 int ret;
850 struct device *dev = disk_to_dev(disk);
851 struct request_queue *q = disk->queue;
852
853 if (WARN_ON(!q))
854 return -ENXIO;
855
856 WARN_ONCE(test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags),
857 "%s is registering an already registered queue\n",
858 kobject_name(&dev->kobj));
859 queue_flag_set_unlocked(QUEUE_FLAG_REGISTERED, q);
860
861 /*
862 * SCSI probing may synchronously create and destroy a lot of
863 * request_queues for non-existent devices. Shutting down a fully
864 * functional queue takes measureable wallclock time as RCU grace
865 * periods are involved. To avoid excessive latency in these
866 * cases, a request_queue starts out in a degraded mode which is
867 * faster to shut down and is made fully functional here as
868 * request_queues for non-existent devices never get registered.
869 */
870 if (!blk_queue_init_done(q)) {
871 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
872 percpu_ref_switch_to_percpu(&q->q_usage_counter);
873 blk_queue_bypass_end(q);
874 }
875
876 ret = blk_trace_init_sysfs(dev);
877 if (ret)
878 return ret;
879
880 /* Prevent changes through sysfs until registration is completed. */
881 mutex_lock(&q->sysfs_lock);
882
883 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
884 if (ret < 0) {
885 blk_trace_remove_sysfs(dev);
886 goto unlock;
887 }
888
889 if (q->mq_ops) {
890 __blk_mq_register_dev(dev, q);
891 blk_mq_debugfs_register(q);
892 }
893
894 kobject_uevent(&q->kobj, KOBJ_ADD);
895
896 wbt_enable_default(q);
897
898 blk_throtl_register_queue(q);
899
900 if (q->request_fn || (q->mq_ops && q->elevator)) {
901 ret = elv_register_queue(q);
902 if (ret) {
903 mutex_unlock(&q->sysfs_lock);
904 kobject_uevent(&q->kobj, KOBJ_REMOVE);
905 kobject_del(&q->kobj);
906 blk_trace_remove_sysfs(dev);
907 kobject_put(&dev->kobj);
908 return ret;
909 }
910 }
911 ret = 0;
912unlock:
913 mutex_unlock(&q->sysfs_lock);
914 return ret;
915}
916EXPORT_SYMBOL_GPL(blk_register_queue);
917
918/**
919 * blk_unregister_queue - counterpart of blk_register_queue()
920 * @disk: Disk of which the request queue should be unregistered from sysfs.
921 *
922 * Note: the caller is responsible for guaranteeing that this function is called
923 * after blk_register_queue() has finished.
924 */
925void blk_unregister_queue(struct gendisk *disk)
926{
927 struct request_queue *q = disk->queue;
928
929 if (WARN_ON(!q))
930 return;
931
932 /* Return early if disk->queue was never registered. */
933 if (!test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags))
934 return;
935
936 /*
937 * Since sysfs_remove_dir() prevents adding new directory entries
938 * before removal of existing entries starts, protect against
939 * concurrent elv_iosched_store() calls.
940 */
941 mutex_lock(&q->sysfs_lock);
942
943 blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
944
945 /*
946 * Remove the sysfs attributes before unregistering the queue data
947 * structures that can be modified through sysfs.
948 */
949 if (q->mq_ops)
950 blk_mq_unregister_dev(disk_to_dev(disk), q);
951 mutex_unlock(&q->sysfs_lock);
952
953 kobject_uevent(&q->kobj, KOBJ_REMOVE);
954 kobject_del(&q->kobj);
955 blk_trace_remove_sysfs(disk_to_dev(disk));
956
957 wbt_exit(q);
958
959 mutex_lock(&q->sysfs_lock);
960 if (q->request_fn || (q->mq_ops && q->elevator))
961 elv_unregister_queue(q);
962 mutex_unlock(&q->sysfs_lock);
963
964 kobject_put(&disk_to_dev(disk)->kobj);
965}
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}