xref: /linux/block/blk-sysfs.c (revision 3630400697a3d334a391c1dba1b601d852145f2c)
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 
23 struct queue_sysfs_entry {
24 	struct attribute attr;
25 	ssize_t (*show)(struct gendisk *disk, char *page);
26 	int (*load_module)(struct gendisk *disk, const char *page, size_t count);
27 	ssize_t (*store)(struct gendisk *disk, const char *page, size_t count);
28 };
29 
30 static ssize_t
31 queue_var_show(unsigned long var, char *page)
32 {
33 	return sprintf(page, "%lu\n", var);
34 }
35 
36 static ssize_t
37 queue_var_store(unsigned long *var, const char *page, size_t count)
38 {
39 	int err;
40 	unsigned long v;
41 
42 	err = kstrtoul(page, 10, &v);
43 	if (err || v > UINT_MAX)
44 		return -EINVAL;
45 
46 	*var = v;
47 
48 	return count;
49 }
50 
51 static ssize_t queue_requests_show(struct gendisk *disk, char *page)
52 {
53 	return queue_var_show(disk->queue->nr_requests, page);
54 }
55 
56 static ssize_t
57 queue_requests_store(struct gendisk *disk, const char *page, size_t count)
58 {
59 	unsigned long nr;
60 	int ret, err;
61 
62 	if (!queue_is_mq(disk->queue))
63 		return -EINVAL;
64 
65 	ret = queue_var_store(&nr, page, count);
66 	if (ret < 0)
67 		return ret;
68 
69 	if (nr < BLKDEV_MIN_RQ)
70 		nr = BLKDEV_MIN_RQ;
71 
72 	err = blk_mq_update_nr_requests(disk->queue, nr);
73 	if (err)
74 		return err;
75 
76 	return ret;
77 }
78 
79 static ssize_t queue_ra_show(struct gendisk *disk, char *page)
80 {
81 	return queue_var_show(disk->bdi->ra_pages << (PAGE_SHIFT - 10), page);
82 }
83 
84 static ssize_t
85 queue_ra_store(struct gendisk *disk, const char *page, size_t count)
86 {
87 	unsigned long ra_kb;
88 	ssize_t ret;
89 
90 	ret = queue_var_store(&ra_kb, page, count);
91 	if (ret < 0)
92 		return ret;
93 	disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
94 	return ret;
95 }
96 
97 #define QUEUE_SYSFS_LIMIT_SHOW(_field)					\
98 static ssize_t queue_##_field##_show(struct gendisk *disk, char *page)	\
99 {									\
100 	return queue_var_show(disk->queue->limits._field, page);	\
101 }
102 
103 QUEUE_SYSFS_LIMIT_SHOW(max_segments)
104 QUEUE_SYSFS_LIMIT_SHOW(max_discard_segments)
105 QUEUE_SYSFS_LIMIT_SHOW(max_integrity_segments)
106 QUEUE_SYSFS_LIMIT_SHOW(max_segment_size)
107 QUEUE_SYSFS_LIMIT_SHOW(logical_block_size)
108 QUEUE_SYSFS_LIMIT_SHOW(physical_block_size)
109 QUEUE_SYSFS_LIMIT_SHOW(chunk_sectors)
110 QUEUE_SYSFS_LIMIT_SHOW(io_min)
111 QUEUE_SYSFS_LIMIT_SHOW(io_opt)
112 QUEUE_SYSFS_LIMIT_SHOW(discard_granularity)
113 QUEUE_SYSFS_LIMIT_SHOW(zone_write_granularity)
114 QUEUE_SYSFS_LIMIT_SHOW(virt_boundary_mask)
115 QUEUE_SYSFS_LIMIT_SHOW(dma_alignment)
116 QUEUE_SYSFS_LIMIT_SHOW(max_open_zones)
117 QUEUE_SYSFS_LIMIT_SHOW(max_active_zones)
118 QUEUE_SYSFS_LIMIT_SHOW(atomic_write_unit_min)
119 QUEUE_SYSFS_LIMIT_SHOW(atomic_write_unit_max)
120 
121 #define QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(_field)			\
122 static ssize_t queue_##_field##_show(struct gendisk *disk, char *page)	\
123 {									\
124 	return sprintf(page, "%llu\n",					\
125 		(unsigned long long)disk->queue->limits._field <<	\
126 			SECTOR_SHIFT);					\
127 }
128 
129 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_discard_sectors)
130 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_hw_discard_sectors)
131 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(max_write_zeroes_sectors)
132 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(atomic_write_max_sectors)
133 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES(atomic_write_boundary_sectors)
134 
135 #define QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(_field)			\
136 static ssize_t queue_##_field##_show(struct gendisk *disk, char *page)	\
137 {									\
138 	return queue_var_show(disk->queue->limits._field >> 1, page);	\
139 }
140 
141 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(max_sectors)
142 QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_KB(max_hw_sectors)
143 
144 #define QUEUE_SYSFS_SHOW_CONST(_name, _val)				\
145 static ssize_t queue_##_name##_show(struct gendisk *disk, char *page)	\
146 {									\
147 	return sprintf(page, "%d\n", _val);				\
148 }
149 
150 /* deprecated fields */
151 QUEUE_SYSFS_SHOW_CONST(discard_zeroes_data, 0)
152 QUEUE_SYSFS_SHOW_CONST(write_same_max, 0)
153 QUEUE_SYSFS_SHOW_CONST(poll_delay, -1)
154 
155 static ssize_t queue_max_discard_sectors_store(struct gendisk *disk,
156 		const char *page, size_t count)
157 {
158 	unsigned long max_discard_bytes;
159 	struct queue_limits lim;
160 	ssize_t ret;
161 	int err;
162 
163 	ret = queue_var_store(&max_discard_bytes, page, count);
164 	if (ret < 0)
165 		return ret;
166 
167 	if (max_discard_bytes & (disk->queue->limits.discard_granularity - 1))
168 		return -EINVAL;
169 
170 	if ((max_discard_bytes >> SECTOR_SHIFT) > UINT_MAX)
171 		return -EINVAL;
172 
173 	lim = queue_limits_start_update(disk->queue);
174 	lim.max_user_discard_sectors = max_discard_bytes >> SECTOR_SHIFT;
175 	err = queue_limits_commit_update(disk->queue, &lim);
176 	if (err)
177 		return err;
178 	return ret;
179 }
180 
181 /*
182  * For zone append queue_max_zone_append_sectors does not just return the
183  * underlying queue limits, but actually contains a calculation.  Because of
184  * that we can't simply use QUEUE_SYSFS_LIMIT_SHOW_SECTORS_TO_BYTES here.
185  */
186 static ssize_t queue_zone_append_max_show(struct gendisk *disk, char *page)
187 {
188 	return sprintf(page, "%llu\n",
189 		(u64)queue_max_zone_append_sectors(disk->queue) <<
190 			SECTOR_SHIFT);
191 }
192 
193 static ssize_t
194 queue_max_sectors_store(struct gendisk *disk, const char *page, size_t count)
195 {
196 	unsigned long max_sectors_kb;
197 	struct queue_limits lim;
198 	ssize_t ret;
199 	int err;
200 
201 	ret = queue_var_store(&max_sectors_kb, page, count);
202 	if (ret < 0)
203 		return ret;
204 
205 	lim = queue_limits_start_update(disk->queue);
206 	lim.max_user_sectors = max_sectors_kb << 1;
207 	err = queue_limits_commit_update(disk->queue, &lim);
208 	if (err)
209 		return err;
210 	return ret;
211 }
212 
213 static ssize_t queue_feature_store(struct gendisk *disk, const char *page,
214 		size_t count, blk_features_t feature)
215 {
216 	struct queue_limits lim;
217 	unsigned long val;
218 	ssize_t ret;
219 
220 	ret = queue_var_store(&val, page, count);
221 	if (ret < 0)
222 		return ret;
223 
224 	lim = queue_limits_start_update(disk->queue);
225 	if (val)
226 		lim.features |= feature;
227 	else
228 		lim.features &= ~feature;
229 	ret = queue_limits_commit_update(disk->queue, &lim);
230 	if (ret)
231 		return ret;
232 	return count;
233 }
234 
235 #define QUEUE_SYSFS_FEATURE(_name, _feature)				\
236 static ssize_t queue_##_name##_show(struct gendisk *disk, char *page)	\
237 {									\
238 	return sprintf(page, "%u\n",					\
239 		!!(disk->queue->limits.features & _feature));		\
240 }									\
241 static ssize_t queue_##_name##_store(struct gendisk *disk,		\
242 		const char *page, size_t count)				\
243 {									\
244 	return queue_feature_store(disk, page, count, _feature);	\
245 }
246 
247 QUEUE_SYSFS_FEATURE(rotational, BLK_FEAT_ROTATIONAL)
248 QUEUE_SYSFS_FEATURE(add_random, BLK_FEAT_ADD_RANDOM)
249 QUEUE_SYSFS_FEATURE(iostats, BLK_FEAT_IO_STAT)
250 QUEUE_SYSFS_FEATURE(stable_writes, BLK_FEAT_STABLE_WRITES);
251 
252 #define QUEUE_SYSFS_FEATURE_SHOW(_name, _feature)			\
253 static ssize_t queue_##_name##_show(struct gendisk *disk, char *page)	\
254 {									\
255 	return sprintf(page, "%u\n",					\
256 		!!(disk->queue->limits.features & _feature));		\
257 }
258 
259 QUEUE_SYSFS_FEATURE_SHOW(poll, BLK_FEAT_POLL);
260 QUEUE_SYSFS_FEATURE_SHOW(fua, BLK_FEAT_FUA);
261 QUEUE_SYSFS_FEATURE_SHOW(dax, BLK_FEAT_DAX);
262 
263 static ssize_t queue_zoned_show(struct gendisk *disk, char *page)
264 {
265 	if (blk_queue_is_zoned(disk->queue))
266 		return sprintf(page, "host-managed\n");
267 	return sprintf(page, "none\n");
268 }
269 
270 static ssize_t queue_nr_zones_show(struct gendisk *disk, char *page)
271 {
272 	return queue_var_show(disk_nr_zones(disk), page);
273 }
274 
275 static ssize_t queue_nomerges_show(struct gendisk *disk, char *page)
276 {
277 	return queue_var_show((blk_queue_nomerges(disk->queue) << 1) |
278 			       blk_queue_noxmerges(disk->queue), page);
279 }
280 
281 static ssize_t queue_nomerges_store(struct gendisk *disk, const char *page,
282 				    size_t count)
283 {
284 	unsigned long nm;
285 	ssize_t ret = queue_var_store(&nm, page, count);
286 
287 	if (ret < 0)
288 		return ret;
289 
290 	blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, disk->queue);
291 	blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, disk->queue);
292 	if (nm == 2)
293 		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, disk->queue);
294 	else if (nm)
295 		blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, disk->queue);
296 
297 	return ret;
298 }
299 
300 static ssize_t queue_rq_affinity_show(struct gendisk *disk, char *page)
301 {
302 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &disk->queue->queue_flags);
303 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &disk->queue->queue_flags);
304 
305 	return queue_var_show(set << force, page);
306 }
307 
308 static ssize_t
309 queue_rq_affinity_store(struct gendisk *disk, const char *page, size_t count)
310 {
311 	ssize_t ret = -EINVAL;
312 #ifdef CONFIG_SMP
313 	struct request_queue *q = disk->queue;
314 	unsigned long val;
315 
316 	ret = queue_var_store(&val, page, count);
317 	if (ret < 0)
318 		return ret;
319 
320 	if (val == 2) {
321 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
322 		blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
323 	} else if (val == 1) {
324 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
325 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
326 	} else if (val == 0) {
327 		blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
328 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
329 	}
330 #endif
331 	return ret;
332 }
333 
334 static ssize_t queue_poll_delay_store(struct gendisk *disk, const char *page,
335 				size_t count)
336 {
337 	return count;
338 }
339 
340 static ssize_t queue_poll_store(struct gendisk *disk, const char *page,
341 				size_t count)
342 {
343 	if (!(disk->queue->limits.features & BLK_FEAT_POLL))
344 		return -EINVAL;
345 	pr_info_ratelimited("writes to the poll attribute are ignored.\n");
346 	pr_info_ratelimited("please use driver specific parameters instead.\n");
347 	return count;
348 }
349 
350 static ssize_t queue_io_timeout_show(struct gendisk *disk, char *page)
351 {
352 	return sprintf(page, "%u\n", jiffies_to_msecs(disk->queue->rq_timeout));
353 }
354 
355 static ssize_t queue_io_timeout_store(struct gendisk *disk, const char *page,
356 				  size_t count)
357 {
358 	unsigned int val;
359 	int err;
360 
361 	err = kstrtou32(page, 10, &val);
362 	if (err || val == 0)
363 		return -EINVAL;
364 
365 	blk_queue_rq_timeout(disk->queue, msecs_to_jiffies(val));
366 
367 	return count;
368 }
369 
370 static ssize_t queue_wc_show(struct gendisk *disk, char *page)
371 {
372 	if (blk_queue_write_cache(disk->queue))
373 		return sprintf(page, "write back\n");
374 	return sprintf(page, "write through\n");
375 }
376 
377 static ssize_t queue_wc_store(struct gendisk *disk, const char *page,
378 			      size_t count)
379 {
380 	struct queue_limits lim;
381 	bool disable;
382 	int err;
383 
384 	if (!strncmp(page, "write back", 10)) {
385 		disable = false;
386 	} else if (!strncmp(page, "write through", 13) ||
387 		   !strncmp(page, "none", 4)) {
388 		disable = true;
389 	} else {
390 		return -EINVAL;
391 	}
392 
393 	lim = queue_limits_start_update(disk->queue);
394 	if (disable)
395 		lim.flags |= BLK_FLAG_WRITE_CACHE_DISABLED;
396 	else
397 		lim.flags &= ~BLK_FLAG_WRITE_CACHE_DISABLED;
398 	err = queue_limits_commit_update(disk->queue, &lim);
399 	if (err)
400 		return err;
401 	return count;
402 }
403 
404 #define QUEUE_RO_ENTRY(_prefix, _name)			\
405 static struct queue_sysfs_entry _prefix##_entry = {	\
406 	.attr	= { .name = _name, .mode = 0444 },	\
407 	.show	= _prefix##_show,			\
408 };
409 
410 #define QUEUE_RW_ENTRY(_prefix, _name)			\
411 static struct queue_sysfs_entry _prefix##_entry = {	\
412 	.attr	= { .name = _name, .mode = 0644 },	\
413 	.show	= _prefix##_show,			\
414 	.store	= _prefix##_store,			\
415 };
416 
417 #define QUEUE_RW_LOAD_MODULE_ENTRY(_prefix, _name)		\
418 static struct queue_sysfs_entry _prefix##_entry = {		\
419 	.attr		= { .name = _name, .mode = 0644 },	\
420 	.show		= _prefix##_show,			\
421 	.load_module	= _prefix##_load_module,		\
422 	.store		= _prefix##_store,			\
423 }
424 
425 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
426 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
427 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
428 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
429 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
430 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
431 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
432 QUEUE_RW_LOAD_MODULE_ENTRY(elv_iosched, "scheduler");
433 
434 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
435 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
436 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
437 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
438 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
439 
440 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
441 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
442 QUEUE_RO_ENTRY(queue_max_hw_discard_sectors, "discard_max_hw_bytes");
443 QUEUE_RW_ENTRY(queue_max_discard_sectors, "discard_max_bytes");
444 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
445 
446 QUEUE_RO_ENTRY(queue_atomic_write_max_sectors, "atomic_write_max_bytes");
447 QUEUE_RO_ENTRY(queue_atomic_write_boundary_sectors,
448 		"atomic_write_boundary_bytes");
449 QUEUE_RO_ENTRY(queue_atomic_write_unit_max, "atomic_write_unit_max_bytes");
450 QUEUE_RO_ENTRY(queue_atomic_write_unit_min, "atomic_write_unit_min_bytes");
451 
452 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
453 QUEUE_RO_ENTRY(queue_max_write_zeroes_sectors, "write_zeroes_max_bytes");
454 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
455 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
456 
457 QUEUE_RO_ENTRY(queue_zoned, "zoned");
458 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
459 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
460 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
461 
462 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
463 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
464 QUEUE_RW_ENTRY(queue_poll, "io_poll");
465 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
466 QUEUE_RW_ENTRY(queue_wc, "write_cache");
467 QUEUE_RO_ENTRY(queue_fua, "fua");
468 QUEUE_RO_ENTRY(queue_dax, "dax");
469 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
470 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
471 QUEUE_RO_ENTRY(queue_dma_alignment, "dma_alignment");
472 
473 /* legacy alias for logical_block_size: */
474 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
475 	.attr = {.name = "hw_sector_size", .mode = 0444 },
476 	.show = queue_logical_block_size_show,
477 };
478 
479 QUEUE_RW_ENTRY(queue_rotational, "rotational");
480 QUEUE_RW_ENTRY(queue_iostats, "iostats");
481 QUEUE_RW_ENTRY(queue_add_random, "add_random");
482 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
483 
484 #ifdef CONFIG_BLK_WBT
485 static ssize_t queue_var_store64(s64 *var, const char *page)
486 {
487 	int err;
488 	s64 v;
489 
490 	err = kstrtos64(page, 10, &v);
491 	if (err < 0)
492 		return err;
493 
494 	*var = v;
495 	return 0;
496 }
497 
498 static ssize_t queue_wb_lat_show(struct gendisk *disk, char *page)
499 {
500 	if (!wbt_rq_qos(disk->queue))
501 		return -EINVAL;
502 
503 	if (wbt_disabled(disk->queue))
504 		return sprintf(page, "0\n");
505 
506 	return sprintf(page, "%llu\n",
507 		div_u64(wbt_get_min_lat(disk->queue), 1000));
508 }
509 
510 static ssize_t queue_wb_lat_store(struct gendisk *disk, const char *page,
511 				  size_t count)
512 {
513 	struct request_queue *q = disk->queue;
514 	struct rq_qos *rqos;
515 	ssize_t ret;
516 	s64 val;
517 
518 	ret = queue_var_store64(&val, page);
519 	if (ret < 0)
520 		return ret;
521 	if (val < -1)
522 		return -EINVAL;
523 
524 	rqos = wbt_rq_qos(q);
525 	if (!rqos) {
526 		ret = wbt_init(disk);
527 		if (ret)
528 			return ret;
529 	}
530 
531 	if (val == -1)
532 		val = wbt_default_latency_nsec(q);
533 	else if (val >= 0)
534 		val *= 1000ULL;
535 
536 	if (wbt_get_min_lat(q) == val)
537 		return count;
538 
539 	/*
540 	 * Ensure that the queue is idled, in case the latency update
541 	 * ends up either enabling or disabling wbt completely. We can't
542 	 * have IO inflight if that happens.
543 	 */
544 	blk_mq_quiesce_queue(q);
545 
546 	wbt_set_min_lat(q, val);
547 
548 	blk_mq_unquiesce_queue(q);
549 
550 	return count;
551 }
552 
553 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
554 #endif
555 
556 /* Common attributes for bio-based and request-based queues. */
557 static struct attribute *queue_attrs[] = {
558 	&queue_ra_entry.attr,
559 	&queue_max_hw_sectors_entry.attr,
560 	&queue_max_sectors_entry.attr,
561 	&queue_max_segments_entry.attr,
562 	&queue_max_discard_segments_entry.attr,
563 	&queue_max_integrity_segments_entry.attr,
564 	&queue_max_segment_size_entry.attr,
565 	&queue_hw_sector_size_entry.attr,
566 	&queue_logical_block_size_entry.attr,
567 	&queue_physical_block_size_entry.attr,
568 	&queue_chunk_sectors_entry.attr,
569 	&queue_io_min_entry.attr,
570 	&queue_io_opt_entry.attr,
571 	&queue_discard_granularity_entry.attr,
572 	&queue_max_discard_sectors_entry.attr,
573 	&queue_max_hw_discard_sectors_entry.attr,
574 	&queue_discard_zeroes_data_entry.attr,
575 	&queue_atomic_write_max_sectors_entry.attr,
576 	&queue_atomic_write_boundary_sectors_entry.attr,
577 	&queue_atomic_write_unit_min_entry.attr,
578 	&queue_atomic_write_unit_max_entry.attr,
579 	&queue_write_same_max_entry.attr,
580 	&queue_max_write_zeroes_sectors_entry.attr,
581 	&queue_zone_append_max_entry.attr,
582 	&queue_zone_write_granularity_entry.attr,
583 	&queue_rotational_entry.attr,
584 	&queue_zoned_entry.attr,
585 	&queue_nr_zones_entry.attr,
586 	&queue_max_open_zones_entry.attr,
587 	&queue_max_active_zones_entry.attr,
588 	&queue_nomerges_entry.attr,
589 	&queue_iostats_entry.attr,
590 	&queue_stable_writes_entry.attr,
591 	&queue_add_random_entry.attr,
592 	&queue_poll_entry.attr,
593 	&queue_wc_entry.attr,
594 	&queue_fua_entry.attr,
595 	&queue_dax_entry.attr,
596 	&queue_poll_delay_entry.attr,
597 	&queue_virt_boundary_mask_entry.attr,
598 	&queue_dma_alignment_entry.attr,
599 	NULL,
600 };
601 
602 /* Request-based queue attributes that are not relevant for bio-based queues. */
603 static struct attribute *blk_mq_queue_attrs[] = {
604 	&queue_requests_entry.attr,
605 	&elv_iosched_entry.attr,
606 	&queue_rq_affinity_entry.attr,
607 	&queue_io_timeout_entry.attr,
608 #ifdef CONFIG_BLK_WBT
609 	&queue_wb_lat_entry.attr,
610 #endif
611 	NULL,
612 };
613 
614 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
615 				int n)
616 {
617 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
618 	struct request_queue *q = disk->queue;
619 
620 	if ((attr == &queue_max_open_zones_entry.attr ||
621 	     attr == &queue_max_active_zones_entry.attr) &&
622 	    !blk_queue_is_zoned(q))
623 		return 0;
624 
625 	return attr->mode;
626 }
627 
628 static umode_t blk_mq_queue_attr_visible(struct kobject *kobj,
629 					 struct attribute *attr, int n)
630 {
631 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
632 	struct request_queue *q = disk->queue;
633 
634 	if (!queue_is_mq(q))
635 		return 0;
636 
637 	if (attr == &queue_io_timeout_entry.attr && !q->mq_ops->timeout)
638 		return 0;
639 
640 	return attr->mode;
641 }
642 
643 static struct attribute_group queue_attr_group = {
644 	.attrs = queue_attrs,
645 	.is_visible = queue_attr_visible,
646 };
647 
648 static struct attribute_group blk_mq_queue_attr_group = {
649 	.attrs = blk_mq_queue_attrs,
650 	.is_visible = blk_mq_queue_attr_visible,
651 };
652 
653 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
654 
655 static ssize_t
656 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
657 {
658 	struct queue_sysfs_entry *entry = to_queue(attr);
659 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
660 	ssize_t res;
661 
662 	if (!entry->show)
663 		return -EIO;
664 	mutex_lock(&disk->queue->sysfs_lock);
665 	res = entry->show(disk, page);
666 	mutex_unlock(&disk->queue->sysfs_lock);
667 	return res;
668 }
669 
670 static ssize_t
671 queue_attr_store(struct kobject *kobj, struct attribute *attr,
672 		    const char *page, size_t length)
673 {
674 	struct queue_sysfs_entry *entry = to_queue(attr);
675 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
676 	struct request_queue *q = disk->queue;
677 	ssize_t res;
678 
679 	if (!entry->store)
680 		return -EIO;
681 
682 	/*
683 	 * If the attribute needs to load a module, do it before freezing the
684 	 * queue to ensure that the module file can be read when the request
685 	 * queue is the one for the device storing the module file.
686 	 */
687 	if (entry->load_module) {
688 		res = entry->load_module(disk, page, length);
689 		if (res)
690 			return res;
691 	}
692 
693 	blk_mq_freeze_queue(q);
694 	mutex_lock(&q->sysfs_lock);
695 	res = entry->store(disk, page, length);
696 	mutex_unlock(&q->sysfs_lock);
697 	blk_mq_unfreeze_queue(q);
698 	return res;
699 }
700 
701 static const struct sysfs_ops queue_sysfs_ops = {
702 	.show	= queue_attr_show,
703 	.store	= queue_attr_store,
704 };
705 
706 static const struct attribute_group *blk_queue_attr_groups[] = {
707 	&queue_attr_group,
708 	&blk_mq_queue_attr_group,
709 	NULL
710 };
711 
712 static void blk_queue_release(struct kobject *kobj)
713 {
714 	/* nothing to do here, all data is associated with the parent gendisk */
715 }
716 
717 static const struct kobj_type blk_queue_ktype = {
718 	.default_groups = blk_queue_attr_groups,
719 	.sysfs_ops	= &queue_sysfs_ops,
720 	.release	= blk_queue_release,
721 };
722 
723 static void blk_debugfs_remove(struct gendisk *disk)
724 {
725 	struct request_queue *q = disk->queue;
726 
727 	mutex_lock(&q->debugfs_mutex);
728 	blk_trace_shutdown(q);
729 	debugfs_remove_recursive(q->debugfs_dir);
730 	q->debugfs_dir = NULL;
731 	q->sched_debugfs_dir = NULL;
732 	q->rqos_debugfs_dir = NULL;
733 	mutex_unlock(&q->debugfs_mutex);
734 }
735 
736 /**
737  * blk_register_queue - register a block layer queue with sysfs
738  * @disk: Disk of which the request queue should be registered with sysfs.
739  */
740 int blk_register_queue(struct gendisk *disk)
741 {
742 	struct request_queue *q = disk->queue;
743 	int ret;
744 
745 	mutex_lock(&q->sysfs_dir_lock);
746 	kobject_init(&disk->queue_kobj, &blk_queue_ktype);
747 	ret = kobject_add(&disk->queue_kobj, &disk_to_dev(disk)->kobj, "queue");
748 	if (ret < 0)
749 		goto out_put_queue_kobj;
750 
751 	if (queue_is_mq(q)) {
752 		ret = blk_mq_sysfs_register(disk);
753 		if (ret)
754 			goto out_put_queue_kobj;
755 	}
756 	mutex_lock(&q->sysfs_lock);
757 
758 	mutex_lock(&q->debugfs_mutex);
759 	q->debugfs_dir = debugfs_create_dir(disk->disk_name, blk_debugfs_root);
760 	if (queue_is_mq(q))
761 		blk_mq_debugfs_register(q);
762 	mutex_unlock(&q->debugfs_mutex);
763 
764 	ret = disk_register_independent_access_ranges(disk);
765 	if (ret)
766 		goto out_debugfs_remove;
767 
768 	if (q->elevator) {
769 		ret = elv_register_queue(q, false);
770 		if (ret)
771 			goto out_unregister_ia_ranges;
772 	}
773 
774 	ret = blk_crypto_sysfs_register(disk);
775 	if (ret)
776 		goto out_elv_unregister;
777 
778 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
779 	wbt_enable_default(disk);
780 
781 	/* Now everything is ready and send out KOBJ_ADD uevent */
782 	kobject_uevent(&disk->queue_kobj, KOBJ_ADD);
783 	if (q->elevator)
784 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
785 	mutex_unlock(&q->sysfs_lock);
786 	mutex_unlock(&q->sysfs_dir_lock);
787 
788 	/*
789 	 * SCSI probing may synchronously create and destroy a lot of
790 	 * request_queues for non-existent devices.  Shutting down a fully
791 	 * functional queue takes measureable wallclock time as RCU grace
792 	 * periods are involved.  To avoid excessive latency in these
793 	 * cases, a request_queue starts out in a degraded mode which is
794 	 * faster to shut down and is made fully functional here as
795 	 * request_queues for non-existent devices never get registered.
796 	 */
797 	if (!blk_queue_init_done(q)) {
798 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
799 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
800 	}
801 
802 	return ret;
803 
804 out_elv_unregister:
805 	elv_unregister_queue(q);
806 out_unregister_ia_ranges:
807 	disk_unregister_independent_access_ranges(disk);
808 out_debugfs_remove:
809 	blk_debugfs_remove(disk);
810 	mutex_unlock(&q->sysfs_lock);
811 out_put_queue_kobj:
812 	kobject_put(&disk->queue_kobj);
813 	mutex_unlock(&q->sysfs_dir_lock);
814 	return ret;
815 }
816 
817 /**
818  * blk_unregister_queue - counterpart of blk_register_queue()
819  * @disk: Disk of which the request queue should be unregistered from sysfs.
820  *
821  * Note: the caller is responsible for guaranteeing that this function is called
822  * after blk_register_queue() has finished.
823  */
824 void blk_unregister_queue(struct gendisk *disk)
825 {
826 	struct request_queue *q = disk->queue;
827 
828 	if (WARN_ON(!q))
829 		return;
830 
831 	/* Return early if disk->queue was never registered. */
832 	if (!blk_queue_registered(q))
833 		return;
834 
835 	/*
836 	 * Since sysfs_remove_dir() prevents adding new directory entries
837 	 * before removal of existing entries starts, protect against
838 	 * concurrent elv_iosched_store() calls.
839 	 */
840 	mutex_lock(&q->sysfs_lock);
841 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
842 	mutex_unlock(&q->sysfs_lock);
843 
844 	mutex_lock(&q->sysfs_dir_lock);
845 	/*
846 	 * Remove the sysfs attributes before unregistering the queue data
847 	 * structures that can be modified through sysfs.
848 	 */
849 	if (queue_is_mq(q))
850 		blk_mq_sysfs_unregister(disk);
851 	blk_crypto_sysfs_unregister(disk);
852 
853 	mutex_lock(&q->sysfs_lock);
854 	elv_unregister_queue(q);
855 	disk_unregister_independent_access_ranges(disk);
856 	mutex_unlock(&q->sysfs_lock);
857 
858 	/* Now that we've deleted all child objects, we can delete the queue. */
859 	kobject_uevent(&disk->queue_kobj, KOBJ_REMOVE);
860 	kobject_del(&disk->queue_kobj);
861 	mutex_unlock(&q->sysfs_dir_lock);
862 
863 	blk_debugfs_remove(disk);
864 }
865