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