xref: /linux/block/blk-sysfs.c (revision bf80eef2212a1e8451df13b52533f4bc31bb4f8e)
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 
23 struct 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 
29 static ssize_t
30 queue_var_show(unsigned long var, char *page)
31 {
32 	return sprintf(page, "%lu\n", var);
33 }
34 
35 static ssize_t
36 queue_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 
50 static 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 
63 static ssize_t queue_requests_show(struct request_queue *q, char *page)
64 {
65 	return queue_var_show(q->nr_requests, page);
66 }
67 
68 static ssize_t
69 queue_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 
91 static 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 
101 static ssize_t
102 queue_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 
116 static 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 
123 static 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 
128 static 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 
134 static 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 
139 static 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 
144 static 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 
149 static 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 
154 static 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 
159 static 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 
164 static 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 
169 static 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 
174 static 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 
181 static 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 
187 static 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 
210 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
211 {
212 	return queue_var_show(0, page);
213 }
214 
215 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
216 {
217 	return queue_var_show(0, page);
218 }
219 
220 static 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 
226 static 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 
232 static 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 
239 static ssize_t
240 queue_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 
265 static 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 
272 static 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 
277 static 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)				\
283 static ssize_t								\
284 queue_##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 }									\
290 static ssize_t								\
291 queue_##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 
308 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
309 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
310 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
311 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
312 #undef QUEUE_SYSFS_BIT_FNS
313 
314 static 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 
326 static 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 
331 static 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 
336 static 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 
341 static 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 
347 static 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 
366 static 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 
374 static ssize_t
375 queue_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 
399 static 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 
411 static 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 
433 static 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 
438 static 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 
448 static 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 
453 static 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 
468 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
469 {
470 	if (!wbt_rq_qos(q))
471 		return -EINVAL;
472 
473 	return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
474 }
475 
476 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
477 				  size_t count)
478 {
479 	struct rq_qos *rqos;
480 	ssize_t ret;
481 	s64 val;
482 
483 	ret = queue_var_store64(&val, page);
484 	if (ret < 0)
485 		return ret;
486 	if (val < -1)
487 		return -EINVAL;
488 
489 	rqos = wbt_rq_qos(q);
490 	if (!rqos) {
491 		ret = wbt_init(q);
492 		if (ret)
493 			return ret;
494 	}
495 
496 	if (val == -1)
497 		val = wbt_default_latency_nsec(q);
498 	else if (val >= 0)
499 		val *= 1000ULL;
500 
501 	if (wbt_get_min_lat(q) == val)
502 		return count;
503 
504 	/*
505 	 * Ensure that the queue is idled, in case the latency update
506 	 * ends up either enabling or disabling wbt completely. We can't
507 	 * have IO inflight if that happens.
508 	 */
509 	blk_mq_freeze_queue(q);
510 	blk_mq_quiesce_queue(q);
511 
512 	wbt_set_min_lat(q, val);
513 
514 	blk_mq_unquiesce_queue(q);
515 	blk_mq_unfreeze_queue(q);
516 
517 	return count;
518 }
519 
520 static ssize_t queue_wc_show(struct request_queue *q, char *page)
521 {
522 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
523 		return sprintf(page, "write back\n");
524 
525 	return sprintf(page, "write through\n");
526 }
527 
528 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
529 			      size_t count)
530 {
531 	int set = -1;
532 
533 	if (!strncmp(page, "write back", 10))
534 		set = 1;
535 	else if (!strncmp(page, "write through", 13) ||
536 		 !strncmp(page, "none", 4))
537 		set = 0;
538 
539 	if (set == -1)
540 		return -EINVAL;
541 
542 	if (set)
543 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
544 	else
545 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
546 
547 	return count;
548 }
549 
550 static ssize_t queue_fua_show(struct request_queue *q, char *page)
551 {
552 	return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
553 }
554 
555 static ssize_t queue_dax_show(struct request_queue *q, char *page)
556 {
557 	return queue_var_show(blk_queue_dax(q), page);
558 }
559 
560 #define QUEUE_RO_ENTRY(_prefix, _name)			\
561 static struct queue_sysfs_entry _prefix##_entry = {	\
562 	.attr	= { .name = _name, .mode = 0444 },	\
563 	.show	= _prefix##_show,			\
564 };
565 
566 #define QUEUE_RW_ENTRY(_prefix, _name)			\
567 static struct queue_sysfs_entry _prefix##_entry = {	\
568 	.attr	= { .name = _name, .mode = 0644 },	\
569 	.show	= _prefix##_show,			\
570 	.store	= _prefix##_store,			\
571 };
572 
573 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
574 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
575 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
576 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
577 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
578 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
579 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
580 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
581 
582 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
583 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
584 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
585 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
586 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
587 
588 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
589 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
590 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
591 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
592 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
593 
594 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
595 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
596 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
597 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
598 
599 QUEUE_RO_ENTRY(queue_zoned, "zoned");
600 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
601 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
602 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
603 
604 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
605 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
606 QUEUE_RW_ENTRY(queue_poll, "io_poll");
607 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
608 QUEUE_RW_ENTRY(queue_wc, "write_cache");
609 QUEUE_RO_ENTRY(queue_fua, "fua");
610 QUEUE_RO_ENTRY(queue_dax, "dax");
611 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
612 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
613 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
614 QUEUE_RO_ENTRY(queue_dma_alignment, "dma_alignment");
615 
616 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
617 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
618 #endif
619 
620 /* legacy alias for logical_block_size: */
621 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
622 	.attr = {.name = "hw_sector_size", .mode = 0444 },
623 	.show = queue_logical_block_size_show,
624 };
625 
626 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
627 QUEUE_RW_ENTRY(queue_iostats, "iostats");
628 QUEUE_RW_ENTRY(queue_random, "add_random");
629 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
630 
631 static struct attribute *queue_attrs[] = {
632 	&queue_requests_entry.attr,
633 	&queue_ra_entry.attr,
634 	&queue_max_hw_sectors_entry.attr,
635 	&queue_max_sectors_entry.attr,
636 	&queue_max_segments_entry.attr,
637 	&queue_max_discard_segments_entry.attr,
638 	&queue_max_integrity_segments_entry.attr,
639 	&queue_max_segment_size_entry.attr,
640 	&elv_iosched_entry.attr,
641 	&queue_hw_sector_size_entry.attr,
642 	&queue_logical_block_size_entry.attr,
643 	&queue_physical_block_size_entry.attr,
644 	&queue_chunk_sectors_entry.attr,
645 	&queue_io_min_entry.attr,
646 	&queue_io_opt_entry.attr,
647 	&queue_discard_granularity_entry.attr,
648 	&queue_discard_max_entry.attr,
649 	&queue_discard_max_hw_entry.attr,
650 	&queue_discard_zeroes_data_entry.attr,
651 	&queue_write_same_max_entry.attr,
652 	&queue_write_zeroes_max_entry.attr,
653 	&queue_zone_append_max_entry.attr,
654 	&queue_zone_write_granularity_entry.attr,
655 	&queue_nonrot_entry.attr,
656 	&queue_zoned_entry.attr,
657 	&queue_nr_zones_entry.attr,
658 	&queue_max_open_zones_entry.attr,
659 	&queue_max_active_zones_entry.attr,
660 	&queue_nomerges_entry.attr,
661 	&queue_rq_affinity_entry.attr,
662 	&queue_iostats_entry.attr,
663 	&queue_stable_writes_entry.attr,
664 	&queue_random_entry.attr,
665 	&queue_poll_entry.attr,
666 	&queue_wc_entry.attr,
667 	&queue_fua_entry.attr,
668 	&queue_dax_entry.attr,
669 	&queue_wb_lat_entry.attr,
670 	&queue_poll_delay_entry.attr,
671 	&queue_io_timeout_entry.attr,
672 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
673 	&blk_throtl_sample_time_entry.attr,
674 #endif
675 	&queue_virt_boundary_mask_entry.attr,
676 	&queue_dma_alignment_entry.attr,
677 	NULL,
678 };
679 
680 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
681 				int n)
682 {
683 	struct request_queue *q =
684 		container_of(kobj, struct request_queue, kobj);
685 
686 	if (attr == &queue_io_timeout_entry.attr &&
687 		(!q->mq_ops || !q->mq_ops->timeout))
688 			return 0;
689 
690 	if ((attr == &queue_max_open_zones_entry.attr ||
691 	     attr == &queue_max_active_zones_entry.attr) &&
692 	    !blk_queue_is_zoned(q))
693 		return 0;
694 
695 	return attr->mode;
696 }
697 
698 static struct attribute_group queue_attr_group = {
699 	.attrs = queue_attrs,
700 	.is_visible = queue_attr_visible,
701 };
702 
703 
704 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
705 
706 static ssize_t
707 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
708 {
709 	struct queue_sysfs_entry *entry = to_queue(attr);
710 	struct request_queue *q =
711 		container_of(kobj, struct request_queue, kobj);
712 	ssize_t res;
713 
714 	if (!entry->show)
715 		return -EIO;
716 	mutex_lock(&q->sysfs_lock);
717 	res = entry->show(q, page);
718 	mutex_unlock(&q->sysfs_lock);
719 	return res;
720 }
721 
722 static ssize_t
723 queue_attr_store(struct kobject *kobj, struct attribute *attr,
724 		    const char *page, size_t length)
725 {
726 	struct queue_sysfs_entry *entry = to_queue(attr);
727 	struct request_queue *q;
728 	ssize_t res;
729 
730 	if (!entry->store)
731 		return -EIO;
732 
733 	q = container_of(kobj, struct request_queue, kobj);
734 	mutex_lock(&q->sysfs_lock);
735 	res = entry->store(q, page, length);
736 	mutex_unlock(&q->sysfs_lock);
737 	return res;
738 }
739 
740 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
741 {
742 	struct request_queue *q = container_of(rcu_head, struct request_queue,
743 					       rcu_head);
744 
745 	kmem_cache_free(blk_get_queue_kmem_cache(blk_queue_has_srcu(q)), q);
746 }
747 
748 /**
749  * blk_release_queue - releases all allocated resources of the request_queue
750  * @kobj: pointer to a kobject, whose container is a request_queue
751  *
752  * This function releases all allocated resources of the request queue.
753  *
754  * The struct request_queue refcount is incremented with blk_get_queue() and
755  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
756  * is called.
757  *
758  * Drivers exist which depend on the release of the request_queue to be
759  * synchronous, it should not be deferred.
760  *
761  * Context: can sleep
762  */
763 static void blk_release_queue(struct kobject *kobj)
764 {
765 	struct request_queue *q =
766 		container_of(kobj, struct request_queue, kobj);
767 
768 	might_sleep();
769 
770 	percpu_ref_exit(&q->q_usage_counter);
771 
772 	if (q->poll_stat)
773 		blk_stat_remove_callback(q, q->poll_cb);
774 	blk_stat_free_callback(q->poll_cb);
775 
776 	blk_free_queue_stats(q->stats);
777 	kfree(q->poll_stat);
778 
779 	if (queue_is_mq(q))
780 		blk_mq_release(q);
781 
782 	if (blk_queue_has_srcu(q))
783 		cleanup_srcu_struct(q->srcu);
784 
785 	ida_free(&blk_queue_ida, q->id);
786 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
787 }
788 
789 static const struct sysfs_ops queue_sysfs_ops = {
790 	.show	= queue_attr_show,
791 	.store	= queue_attr_store,
792 };
793 
794 static const struct attribute_group *blk_queue_attr_groups[] = {
795 	&queue_attr_group,
796 	NULL
797 };
798 
799 struct kobj_type blk_queue_ktype = {
800 	.default_groups = blk_queue_attr_groups,
801 	.sysfs_ops	= &queue_sysfs_ops,
802 	.release	= blk_release_queue,
803 };
804 
805 /**
806  * blk_register_queue - register a block layer queue with sysfs
807  * @disk: Disk of which the request queue should be registered with sysfs.
808  */
809 int blk_register_queue(struct gendisk *disk)
810 {
811 	struct request_queue *q = disk->queue;
812 	int ret;
813 
814 	mutex_lock(&q->sysfs_dir_lock);
815 
816 	ret = kobject_add(&q->kobj, &disk_to_dev(disk)->kobj, "queue");
817 	if (ret < 0)
818 		goto unlock;
819 
820 	if (queue_is_mq(q))
821 		blk_mq_sysfs_register(disk);
822 	mutex_lock(&q->sysfs_lock);
823 
824 	mutex_lock(&q->debugfs_mutex);
825 	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
826 					    blk_debugfs_root);
827 	if (queue_is_mq(q))
828 		blk_mq_debugfs_register(q);
829 	mutex_unlock(&q->debugfs_mutex);
830 
831 	ret = disk_register_independent_access_ranges(disk);
832 	if (ret)
833 		goto put_dev;
834 
835 	if (q->elevator) {
836 		ret = elv_register_queue(q, false);
837 		if (ret)
838 			goto put_dev;
839 	}
840 
841 	ret = blk_crypto_sysfs_register(q);
842 	if (ret)
843 		goto put_dev;
844 
845 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
846 	wbt_enable_default(q);
847 	blk_throtl_register(disk);
848 
849 	/* Now everything is ready and send out KOBJ_ADD uevent */
850 	kobject_uevent(&q->kobj, KOBJ_ADD);
851 	if (q->elevator)
852 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
853 	mutex_unlock(&q->sysfs_lock);
854 
855 unlock:
856 	mutex_unlock(&q->sysfs_dir_lock);
857 
858 	/*
859 	 * SCSI probing may synchronously create and destroy a lot of
860 	 * request_queues for non-existent devices.  Shutting down a fully
861 	 * functional queue takes measureable wallclock time as RCU grace
862 	 * periods are involved.  To avoid excessive latency in these
863 	 * cases, a request_queue starts out in a degraded mode which is
864 	 * faster to shut down and is made fully functional here as
865 	 * request_queues for non-existent devices never get registered.
866 	 */
867 	if (!blk_queue_init_done(q)) {
868 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
869 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
870 	}
871 
872 	return ret;
873 
874 put_dev:
875 	elv_unregister_queue(q);
876 	disk_unregister_independent_access_ranges(disk);
877 	mutex_unlock(&q->sysfs_lock);
878 	mutex_unlock(&q->sysfs_dir_lock);
879 	kobject_del(&q->kobj);
880 
881 	return ret;
882 }
883 
884 /**
885  * blk_unregister_queue - counterpart of blk_register_queue()
886  * @disk: Disk of which the request queue should be unregistered from sysfs.
887  *
888  * Note: the caller is responsible for guaranteeing that this function is called
889  * after blk_register_queue() has finished.
890  */
891 void blk_unregister_queue(struct gendisk *disk)
892 {
893 	struct request_queue *q = disk->queue;
894 
895 	if (WARN_ON(!q))
896 		return;
897 
898 	/* Return early if disk->queue was never registered. */
899 	if (!blk_queue_registered(q))
900 		return;
901 
902 	/*
903 	 * Since sysfs_remove_dir() prevents adding new directory entries
904 	 * before removal of existing entries starts, protect against
905 	 * concurrent elv_iosched_store() calls.
906 	 */
907 	mutex_lock(&q->sysfs_lock);
908 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
909 	mutex_unlock(&q->sysfs_lock);
910 
911 	mutex_lock(&q->sysfs_dir_lock);
912 	/*
913 	 * Remove the sysfs attributes before unregistering the queue data
914 	 * structures that can be modified through sysfs.
915 	 */
916 	if (queue_is_mq(q))
917 		blk_mq_sysfs_unregister(disk);
918 	blk_crypto_sysfs_unregister(q);
919 
920 	mutex_lock(&q->sysfs_lock);
921 	elv_unregister_queue(q);
922 	disk_unregister_independent_access_ranges(disk);
923 	mutex_unlock(&q->sysfs_lock);
924 
925 	/* Now that we've deleted all child objects, we can delete the queue. */
926 	kobject_uevent(&q->kobj, KOBJ_REMOVE);
927 	kobject_del(&q->kobj);
928 	mutex_unlock(&q->sysfs_dir_lock);
929 
930 	mutex_lock(&q->debugfs_mutex);
931 	blk_trace_shutdown(q);
932 	debugfs_remove_recursive(q->debugfs_dir);
933 	q->debugfs_dir = NULL;
934 	q->sched_debugfs_dir = NULL;
935 	q->rqos_debugfs_dir = NULL;
936 	mutex_unlock(&q->debugfs_mutex);
937 }
938