block/blk-sysfs.c

   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 #include <linux/debugfs.h>
  15
  16 #include "blk.h"
  17 #include "blk-mq.h"
  18 #include "blk-mq-debugfs.h"
  19 #include "blk-wbt.h"
  20 #include "blk-throttle.h"
  21
  22 struct queue_sysfs_entry {
  23         struct attribute attr;
  24         ssize_t (*show)(struct request_queue *, char *);
  25         ssize_t (*store)(struct request_queue *, const char *, size_t);
  26 };
  27
  28 static ssize_t
  29 queue_var_show(unsigned long var, char *page)
  30 {
  31         return sprintf(page, "%lu\n", var);
  32 }
  33
  34 static ssize_t
  35 queue_var_store(unsigned long *var, const char *page, size_t count)
  36 {
  37         int err;
  38         unsigned long v;
  39
  40         err = kstrtoul(page, 10, &v);
  41         if (err || v > UINT_MAX)
  42                 return -EINVAL;
  43
  44         *var = v;
  45
  46         return count;
  47 }
  48
  49 static ssize_t queue_var_store64(s64 *var, const char *page)
  50 {
  51         int err;
  52         s64 v;
  53
  54         err = kstrtos64(page, 10, &v);
  55         if (err < 0)
  56                 return err;
  57
  58         *var = v;
  59         return 0;
  60 }
  61
  62 static ssize_t queue_requests_show(struct request_queue *q, char *page)
  63 {
  64         return queue_var_show(q->nr_requests, page);
  65 }
  66
  67 static ssize_t
  68 queue_requests_store(struct request_queue *q, const char *page, size_t count)
  69 {
  70         unsigned long nr;
  71         int ret, err;
  72
  73         if (!queue_is_mq(q))
  74                 return -EINVAL;
  75
  76         ret = queue_var_store(&nr, page, count);
  77         if (ret < 0)
  78                 return ret;
  79
  80         if (nr < BLKDEV_MIN_RQ)
  81                 nr = BLKDEV_MIN_RQ;
  82
  83         err = blk_mq_update_nr_requests(q, nr);
  84         if (err)
  85                 return err;
  86
  87         return ret;
  88 }
  89
  90 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  91 {
  92         unsigned long ra_kb;
  93
  94         if (!q->disk)
  95                 return -EINVAL;
  96         ra_kb = q->disk->bdi->ra_pages << (PAGE_SHIFT - 10);
  97         return queue_var_show(ra_kb, page);
  98 }
  99
 100 static ssize_t
 101 queue_ra_store(struct request_queue *q, const char *page, size_t count)
 102 {
 103         unsigned long ra_kb;
 104         ssize_t ret;
 105
 106         if (!q->disk)
 107                 return -EINVAL;
 108         ret = queue_var_store(&ra_kb, page, count);
 109         if (ret < 0)
 110                 return ret;
 111         q->disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 112         return ret;
 113 }
 114
 115 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 116 {
 117         int max_sectors_kb = queue_max_sectors(q) >> 1;
 118
 119         return queue_var_show(max_sectors_kb, page);
 120 }
 121
 122 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 123 {
 124         return queue_var_show(queue_max_segments(q), page);
 125 }
 126
 127 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 128                 char *page)
 129 {
 130         return queue_var_show(queue_max_discard_segments(q), page);
 131 }
 132
 133 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 134 {
 135         return queue_var_show(q->limits.max_integrity_segments, page);
 136 }
 137
 138 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 139 {
 140         return queue_var_show(queue_max_segment_size(q), page);
 141 }
 142
 143 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 144 {
 145         return queue_var_show(queue_logical_block_size(q), page);
 146 }
 147
 148 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 149 {
 150         return queue_var_show(queue_physical_block_size(q), page);
 151 }
 152
 153 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 154 {
 155         return queue_var_show(q->limits.chunk_sectors, page);
 156 }
 157
 158 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 159 {
 160         return queue_var_show(queue_io_min(q), page);
 161 }
 162
 163 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 164 {
 165         return queue_var_show(queue_io_opt(q), page);
 166 }
 167
 168 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 169 {
 170         return queue_var_show(q->limits.discard_granularity, page);
 171 }
 172
 173 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 174 {
 175
 176         return sprintf(page, "%llu\n",
 177                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 178 }
 179
 180 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 181 {
 182         return sprintf(page, "%llu\n",
 183                        (unsigned long long)q->limits.max_discard_sectors << 9);
 184 }
 185
 186 static ssize_t queue_discard_max_store(struct request_queue *q,
 187                                        const char *page, size_t count)
 188 {
 189         unsigned long max_discard;
 190         ssize_t ret = queue_var_store(&max_discard, page, count);
 191
 192         if (ret < 0)
 193                 return ret;
 194
 195         if (max_discard & (q->limits.discard_granularity - 1))
 196                 return -EINVAL;
 197
 198         max_discard >>= 9;
 199         if (max_discard > UINT_MAX)
 200                 return -EINVAL;
 201
 202         if (max_discard > q->limits.max_hw_discard_sectors)
 203                 max_discard = q->limits.max_hw_discard_sectors;
 204
 205         q->limits.max_discard_sectors = max_discard;
 206         return ret;
 207 }
 208
 209 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 210 {
 211         return queue_var_show(0, page);
 212 }
 213
 214 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 215 {
 216         return sprintf(page, "%llu\n",
 217                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 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 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 278 static ssize_t                                                          \
 279 queue_##name##_show(struct request_queue *q, char *page)                \
 280 {                                                                       \
 281         int bit;                                                        \
 282         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 283         return queue_var_show(neg ? !bit : bit, page);                  \
 284 }                                                                       \
 285 static ssize_t                                                          \
 286 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
 287 {                                                                       \
 288         unsigned long val;                                              \
 289         ssize_t ret;                                                    \
 290         ret = queue_var_store(&val, page, count);                       \
 291         if (ret < 0)                                                    \
 292                  return ret;                                            \
 293         if (neg)                                                        \
 294                 val = !val;                                             \
 295                                                                         \
 296         if (val)                                                        \
 297                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 298         else                                                            \
 299                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 300         return ret;                                                     \
 301 }
 302
 303 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 304 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 305 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 306 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
 307 #undef QUEUE_SYSFS_BIT_FNS
 308
 309 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 310 {
 311         switch (blk_queue_zoned_model(q)) {
 312         case BLK_ZONED_HA:
 313                 return sprintf(page, "host-aware\n");
 314         case BLK_ZONED_HM:
 315                 return sprintf(page, "host-managed\n");
 316         default:
 317                 return sprintf(page, "none\n");
 318         }
 319 }
 320
 321 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
 322 {
 323         return queue_var_show(blk_queue_nr_zones(q), page);
 324 }
 325
 326 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
 327 {
 328         return queue_var_show(queue_max_open_zones(q), page);
 329 }
 330
 331 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
 332 {
 333         return queue_var_show(queue_max_active_zones(q), page);
 334 }
 335
 336 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 337 {
 338         return queue_var_show((blk_queue_nomerges(q) << 1) |
 339                                blk_queue_noxmerges(q), page);
 340 }
 341
 342 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 343                                     size_t count)
 344 {
 345         unsigned long nm;
 346         ssize_t ret = queue_var_store(&nm, page, count);
 347
 348         if (ret < 0)
 349                 return ret;
 350
 351         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 352         blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 353         if (nm == 2)
 354                 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 355         else if (nm)
 356                 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 357
 358         return ret;
 359 }
 360
 361 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 362 {
 363         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 364         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 365
 366         return queue_var_show(set << force, page);
 367 }
 368
 369 static ssize_t
 370 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 371 {
 372         ssize_t ret = -EINVAL;
 373 #ifdef CONFIG_SMP
 374         unsigned long val;
 375
 376         ret = queue_var_store(&val, page, count);
 377         if (ret < 0)
 378                 return ret;
 379
 380         if (val == 2) {
 381                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 382                 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 383         } else if (val == 1) {
 384                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 385                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 386         } else if (val == 0) {
 387                 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 388                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 389         }
 390 #endif
 391         return ret;
 392 }
 393
 394 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 395 {
 396         int val;
 397
 398         if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
 399                 val = BLK_MQ_POLL_CLASSIC;
 400         else
 401                 val = q->poll_nsec / 1000;
 402
 403         return sprintf(page, "%d\n", val);
 404 }
 405
 406 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 407                                 size_t count)
 408 {
 409         int err, val;
 410
 411         if (!q->mq_ops || !q->mq_ops->poll)
 412                 return -EINVAL;
 413
 414         err = kstrtoint(page, 10, &val);
 415         if (err < 0)
 416                 return err;
 417
 418         if (val == BLK_MQ_POLL_CLASSIC)
 419                 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
 420         else if (val >= 0)
 421                 q->poll_nsec = val * 1000;
 422         else
 423                 return -EINVAL;
 424
 425         return count;
 426 }
 427
 428 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 429 {
 430         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 431 }
 432
 433 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 434                                 size_t count)
 435 {
 436         unsigned long poll_on;
 437         ssize_t ret;
 438
 439         if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
 440             !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
 441                 return -EINVAL;
 442
 443         ret = queue_var_store(&poll_on, page, count);
 444         if (ret < 0)
 445                 return ret;
 446
 447         if (poll_on) {
 448                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 449         } else {
 450                 blk_mq_freeze_queue(q);
 451                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 452                 blk_mq_unfreeze_queue(q);
 453         }
 454
 455         return ret;
 456 }
 457
 458 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
 459 {
 460         return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
 461 }
 462
 463 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
 464                                   size_t count)
 465 {
 466         unsigned int val;
 467         int err;
 468
 469         err = kstrtou32(page, 10, &val);
 470         if (err || val == 0)
 471                 return -EINVAL;
 472
 473         blk_queue_rq_timeout(q, msecs_to_jiffies(val));
 474
 475         return count;
 476 }
 477
 478 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 479 {
 480         if (!wbt_rq_qos(q))
 481                 return -EINVAL;
 482
 483         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 484 }
 485
 486 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 487                                   size_t count)
 488 {
 489         struct rq_qos *rqos;
 490         ssize_t ret;
 491         s64 val;
 492
 493         ret = queue_var_store64(&val, page);
 494         if (ret < 0)
 495                 return ret;
 496         if (val < -1)
 497                 return -EINVAL;
 498
 499         rqos = wbt_rq_qos(q);
 500         if (!rqos) {
 501                 ret = wbt_init(q);
 502                 if (ret)
 503                         return ret;
 504         }
 505
 506         if (val == -1)
 507                 val = wbt_default_latency_nsec(q);
 508         else if (val >= 0)
 509                 val *= 1000ULL;
 510
 511         if (wbt_get_min_lat(q) == val)
 512                 return count;
 513
 514         /*
 515          * Ensure that the queue is idled, in case the latency update
 516          * ends up either enabling or disabling wbt completely. We can't
 517          * have IO inflight if that happens.
 518          */
 519         blk_mq_freeze_queue(q);
 520         blk_mq_quiesce_queue(q);
 521
 522         wbt_set_min_lat(q, val);
 523
 524         blk_mq_unquiesce_queue(q);
 525         blk_mq_unfreeze_queue(q);
 526
 527         return count;
 528 }
 529
 530 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 531 {
 532         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 533                 return sprintf(page, "write back\n");
 534
 535         return sprintf(page, "write through\n");
 536 }
 537
 538 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 539                               size_t count)
 540 {
 541         int set = -1;
 542
 543         if (!strncmp(page, "write back", 10))
 544                 set = 1;
 545         else if (!strncmp(page, "write through", 13) ||
 546                  !strncmp(page, "none", 4))
 547                 set = 0;
 548
 549         if (set == -1)
 550                 return -EINVAL;
 551
 552         if (set)
 553                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 554         else
 555                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 556
 557         return count;
 558 }
 559
 560 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 561 {
 562         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 563 }
 564
 565 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 566 {
 567         return queue_var_show(blk_queue_dax(q), page);
 568 }
 569
 570 #define QUEUE_RO_ENTRY(_prefix, _name)                  \
 571 static struct queue_sysfs_entry _prefix##_entry = {     \
 572         .attr   = { .name = _name, .mode = 0444 },      \
 573         .show   = _prefix##_show,                       \
 574 };
 575
 576 #define QUEUE_RW_ENTRY(_prefix, _name)                  \
 577 static struct queue_sysfs_entry _prefix##_entry = {     \
 578         .attr   = { .name = _name, .mode = 0644 },      \
 579         .show   = _prefix##_show,                       \
 580         .store  = _prefix##_store,                      \
 581 };
 582
 583 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
 584 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
 585 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
 586 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
 587 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
 588 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
 589 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
 590 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
 591
 592 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
 593 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
 594 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
 595 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
 596 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
 597
 598 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
 599 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
 600 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
 601 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
 602 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
 603
 604 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
 605 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
 606 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
 607 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
 608
 609 QUEUE_RO_ENTRY(queue_zoned, "zoned");
 610 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
 611 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
 612 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
 613
 614 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
 615 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
 616 QUEUE_RW_ENTRY(queue_poll, "io_poll");
 617 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
 618 QUEUE_RW_ENTRY(queue_wc, "write_cache");
 619 QUEUE_RO_ENTRY(queue_fua, "fua");
 620 QUEUE_RO_ENTRY(queue_dax, "dax");
 621 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
 622 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
 623 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
 624
 625 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 626 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
 627 #endif
 628
 629 /* legacy alias for logical_block_size: */
 630 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 631         .attr = {.name = "hw_sector_size", .mode = 0444 },
 632         .show = queue_logical_block_size_show,
 633 };
 634
 635 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
 636 QUEUE_RW_ENTRY(queue_iostats, "iostats");
 637 QUEUE_RW_ENTRY(queue_random, "add_random");
 638 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
 639
 640 static struct attribute *queue_attrs[] = {
 641         &queue_requests_entry.attr,
 642         &queue_ra_entry.attr,
 643         &queue_max_hw_sectors_entry.attr,
 644         &queue_max_sectors_entry.attr,
 645         &queue_max_segments_entry.attr,
 646         &queue_max_discard_segments_entry.attr,
 647         &queue_max_integrity_segments_entry.attr,
 648         &queue_max_segment_size_entry.attr,
 649         &elv_iosched_entry.attr,
 650         &queue_hw_sector_size_entry.attr,
 651         &queue_logical_block_size_entry.attr,
 652         &queue_physical_block_size_entry.attr,
 653         &queue_chunk_sectors_entry.attr,
 654         &queue_io_min_entry.attr,
 655         &queue_io_opt_entry.attr,
 656         &queue_discard_granularity_entry.attr,
 657         &queue_discard_max_entry.attr,
 658         &queue_discard_max_hw_entry.attr,
 659         &queue_discard_zeroes_data_entry.attr,
 660         &queue_write_same_max_entry.attr,
 661         &queue_write_zeroes_max_entry.attr,
 662         &queue_zone_append_max_entry.attr,
 663         &queue_zone_write_granularity_entry.attr,
 664         &queue_nonrot_entry.attr,
 665         &queue_zoned_entry.attr,
 666         &queue_nr_zones_entry.attr,
 667         &queue_max_open_zones_entry.attr,
 668         &queue_max_active_zones_entry.attr,
 669         &queue_nomerges_entry.attr,
 670         &queue_rq_affinity_entry.attr,
 671         &queue_iostats_entry.attr,
 672         &queue_stable_writes_entry.attr,
 673         &queue_random_entry.attr,
 674         &queue_poll_entry.attr,
 675         &queue_wc_entry.attr,
 676         &queue_fua_entry.attr,
 677         &queue_dax_entry.attr,
 678         &queue_wb_lat_entry.attr,
 679         &queue_poll_delay_entry.attr,
 680         &queue_io_timeout_entry.attr,
 681 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 682         &blk_throtl_sample_time_entry.attr,
 683 #endif
 684         &queue_virt_boundary_mask_entry.attr,
 685         NULL,
 686 };
 687
 688 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
 689                                 int n)
 690 {
 691         struct request_queue *q =
 692                 container_of(kobj, struct request_queue, kobj);
 693
 694         if (attr == &queue_io_timeout_entry.attr &&
 695                 (!q->mq_ops || !q->mq_ops->timeout))
 696                         return 0;
 697
 698         if ((attr == &queue_max_open_zones_entry.attr ||
 699              attr == &queue_max_active_zones_entry.attr) &&
 700             !blk_queue_is_zoned(q))
 701                 return 0;
 702
 703         return attr->mode;
 704 }
 705
 706 static struct attribute_group queue_attr_group = {
 707         .attrs = queue_attrs,
 708         .is_visible = queue_attr_visible,
 709 };
 710
 711
 712 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 713
 714 static ssize_t
 715 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 716 {
 717         struct queue_sysfs_entry *entry = to_queue(attr);
 718         struct request_queue *q =
 719                 container_of(kobj, struct request_queue, kobj);
 720         ssize_t res;
 721
 722         if (!entry->show)
 723                 return -EIO;
 724         mutex_lock(&q->sysfs_lock);
 725         res = entry->show(q, page);
 726         mutex_unlock(&q->sysfs_lock);
 727         return res;
 728 }
 729
 730 static ssize_t
 731 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 732                     const char *page, size_t length)
 733 {
 734         struct queue_sysfs_entry *entry = to_queue(attr);
 735         struct request_queue *q;
 736         ssize_t res;
 737
 738         if (!entry->store)
 739                 return -EIO;
 740
 741         q = container_of(kobj, struct request_queue, kobj);
 742         mutex_lock(&q->sysfs_lock);
 743         res = entry->store(q, page, length);
 744         mutex_unlock(&q->sysfs_lock);
 745         return res;
 746 }
 747
 748 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 749 {
 750         struct request_queue *q = container_of(rcu_head, struct request_queue,
 751                                                rcu_head);
 752         kmem_cache_free(blk_requestq_cachep, q);
 753 }
 754
 755 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 756 static void blk_exit_queue(struct request_queue *q)
 757 {
 758         /*
 759          * Since the I/O scheduler exit code may access cgroup information,
 760          * perform I/O scheduler exit before disassociating from the block
 761          * cgroup controller.
 762          */
 763         if (q->elevator) {
 764                 ioc_clear_queue(q);
 765                 __elevator_exit(q, q->elevator);
 766         }
 767
 768         /*
 769          * Remove all references to @q from the block cgroup controller before
 770          * restoring @q->queue_lock to avoid that restoring this pointer causes
 771          * e.g. blkcg_print_blkgs() to crash.
 772          */
 773         blkcg_exit_queue(q);
 774 }
 775
 776 /**
 777  * blk_release_queue - releases all allocated resources of the request_queue
 778  * @kobj: pointer to a kobject, whose container is a request_queue
 779  *
 780  * This function releases all allocated resources of the request queue.
 781  *
 782  * The struct request_queue refcount is incremented with blk_get_queue() and
 783  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
 784  * is called.
 785  *
 786  * For drivers that have a request_queue on a gendisk and added with
 787  * __device_add_disk() the refcount to request_queue will reach 0 with
 788  * the last put_disk() called by the driver. For drivers which don't use
 789  * __device_add_disk() this happens with blk_cleanup_queue().
 790  *
 791  * Drivers exist which depend on the release of the request_queue to be
 792  * synchronous, it should not be deferred.
 793  *
 794  * Context: can sleep
 795  */
 796 static void blk_release_queue(struct kobject *kobj)
 797 {
 798         struct request_queue *q =
 799                 container_of(kobj, struct request_queue, kobj);
 800
 801         might_sleep();
 802
 803         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 804                 blk_stat_remove_callback(q, q->poll_cb);
 805         blk_stat_free_callback(q->poll_cb);
 806
 807         blk_free_queue_stats(q->stats);
 808
 809         if (queue_is_mq(q)) {
 810                 struct blk_mq_hw_ctx *hctx;
 811                 int i;
 812
 813                 cancel_delayed_work_sync(&q->requeue_work);
 814
 815                 queue_for_each_hw_ctx(q, hctx, i)
 816                         cancel_delayed_work_sync(&hctx->run_work);
 817         }
 818
 819         blk_exit_queue(q);
 820
 821         blk_queue_free_zone_bitmaps(q);
 822
 823         if (queue_is_mq(q))
 824                 blk_mq_release(q);
 825
 826         blk_trace_shutdown(q);
 827         mutex_lock(&q->debugfs_mutex);
 828         debugfs_remove_recursive(q->debugfs_dir);
 829         mutex_unlock(&q->debugfs_mutex);
 830
 831         if (queue_is_mq(q))
 832                 blk_mq_debugfs_unregister(q);
 833
 834         bioset_exit(&q->bio_split);
 835
 836         ida_simple_remove(&blk_queue_ida, q->id);
 837         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 838 }
 839
 840 static const struct sysfs_ops queue_sysfs_ops = {
 841         .show   = queue_attr_show,
 842         .store  = queue_attr_store,
 843 };
 844
 845 struct kobj_type blk_queue_ktype = {
 846         .sysfs_ops      = &queue_sysfs_ops,
 847         .release        = blk_release_queue,
 848 };
 849
 850 /**
 851  * blk_register_queue - register a block layer queue with sysfs
 852  * @disk: Disk of which the request queue should be registered with sysfs.
 853  */
 854 int blk_register_queue(struct gendisk *disk)
 855 {
 856         int ret;
 857         struct device *dev = disk_to_dev(disk);
 858         struct request_queue *q = disk->queue;
 859
 860         ret = blk_trace_init_sysfs(dev);
 861         if (ret)
 862                 return ret;
 863
 864         mutex_lock(&q->sysfs_dir_lock);
 865
 866         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 867         if (ret < 0) {
 868                 blk_trace_remove_sysfs(dev);
 869                 goto unlock;
 870         }
 871
 872         ret = sysfs_create_group(&q->kobj, &queue_attr_group);
 873         if (ret) {
 874                 blk_trace_remove_sysfs(dev);
 875                 kobject_del(&q->kobj);
 876                 kobject_put(&dev->kobj);
 877                 goto unlock;
 878         }
 879
 880         mutex_lock(&q->debugfs_mutex);
 881         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 882                                             blk_debugfs_root);
 883         mutex_unlock(&q->debugfs_mutex);
 884
 885         if (queue_is_mq(q)) {
 886                 __blk_mq_register_dev(dev, q);
 887                 blk_mq_debugfs_register(q);
 888         }
 889
 890         mutex_lock(&q->sysfs_lock);
 891         if (q->elevator) {
 892                 ret = elv_register_queue(q, false);
 893                 if (ret) {
 894                         mutex_unlock(&q->sysfs_lock);
 895                         mutex_unlock(&q->sysfs_dir_lock);
 896                         kobject_del(&q->kobj);
 897                         blk_trace_remove_sysfs(dev);
 898                         kobject_put(&dev->kobj);
 899                         return ret;
 900                 }
 901         }
 902
 903         blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
 904         wbt_enable_default(q);
 905         blk_throtl_register_queue(q);
 906
 907         /* Now everything is ready and send out KOBJ_ADD uevent */
 908         kobject_uevent(&q->kobj, KOBJ_ADD);
 909         if (q->elevator)
 910                 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
 911         mutex_unlock(&q->sysfs_lock);
 912
 913         ret = 0;
 914 unlock:
 915         mutex_unlock(&q->sysfs_dir_lock);
 916
 917         /*
 918          * SCSI probing may synchronously create and destroy a lot of
 919          * request_queues for non-existent devices.  Shutting down a fully
 920          * functional queue takes measureable wallclock time as RCU grace
 921          * periods are involved.  To avoid excessive latency in these
 922          * cases, a request_queue starts out in a degraded mode which is
 923          * faster to shut down and is made fully functional here as
 924          * request_queues for non-existent devices never get registered.
 925          */
 926         if (!blk_queue_init_done(q)) {
 927                 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
 928                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 929         }
 930
 931         return ret;
 932 }
 933
 934 /**
 935  * blk_unregister_queue - counterpart of blk_register_queue()
 936  * @disk: Disk of which the request queue should be unregistered from sysfs.
 937  *
 938  * Note: the caller is responsible for guaranteeing that this function is called
 939  * after blk_register_queue() has finished.
 940  */
 941 void blk_unregister_queue(struct gendisk *disk)
 942 {
 943         struct request_queue *q = disk->queue;
 944
 945         if (WARN_ON(!q))
 946                 return;
 947
 948         /* Return early if disk->queue was never registered. */
 949         if (!blk_queue_registered(q))
 950                 return;
 951
 952         /*
 953          * Since sysfs_remove_dir() prevents adding new directory entries
 954          * before removal of existing entries starts, protect against
 955          * concurrent elv_iosched_store() calls.
 956          */
 957         mutex_lock(&q->sysfs_lock);
 958         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
 959         mutex_unlock(&q->sysfs_lock);
 960
 961         mutex_lock(&q->sysfs_dir_lock);
 962         /*
 963          * Remove the sysfs attributes before unregistering the queue data
 964          * structures that can be modified through sysfs.
 965          */
 966         if (queue_is_mq(q))
 967                 blk_mq_unregister_dev(disk_to_dev(disk), q);
 968
 969         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 970         kobject_del(&q->kobj);
 971         blk_trace_remove_sysfs(disk_to_dev(disk));
 972
 973         mutex_lock(&q->sysfs_lock);
 974         if (q->elevator)
 975                 elv_unregister_queue(q);
 976         mutex_unlock(&q->sysfs_lock);
 977         mutex_unlock(&q->sysfs_dir_lock);
 978
 979         kobject_put(&disk_to_dev(disk)->kobj);
 980 }