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
  21 struct queue_sysfs_entry {
  22         struct attribute attr;
  23         ssize_t (*show)(struct request_queue *, char *);
  24         ssize_t (*store)(struct request_queue *, const char *, size_t);
  25 };
  26
  27 static ssize_t
  28 queue_var_show(unsigned long var, char *page)
  29 {
  30         return sprintf(page, "%lu\n", var);
  31 }
  32
  33 static ssize_t
  34 queue_var_store(unsigned long *var, const char *page, size_t count)
  35 {
  36         int err;
  37         unsigned long v;
  38
  39         err = kstrtoul(page, 10, &v);
  40         if (err || v > UINT_MAX)
  41                 return -EINVAL;
  42
  43         *var = v;
  44
  45         return count;
  46 }
  47
  48 static ssize_t queue_var_store64(s64 *var, const char *page)
  49 {
  50         int err;
  51         s64 v;
  52
  53         err = kstrtos64(page, 10, &v);
  54         if (err < 0)
  55                 return err;
  56
  57         *var = v;
  58         return 0;
  59 }
  60
  61 static ssize_t queue_requests_show(struct request_queue *q, char *page)
  62 {
  63         return queue_var_show(q->nr_requests, page);
  64 }
  65
  66 static ssize_t
  67 queue_requests_store(struct request_queue *q, const char *page, size_t count)
  68 {
  69         unsigned long nr;
  70         int ret, err;
  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         if (nr < BLKDEV_MIN_RQ)
  80                 nr = BLKDEV_MIN_RQ;
  81
  82         err = blk_mq_update_nr_requests(q, nr);
  83         if (err)
  84                 return err;
  85
  86         return ret;
  87 }
  88
  89 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  90 {
  91         unsigned long ra_kb;
  92
  93         if (!q->disk)
  94                 return -EINVAL;
  95         ra_kb = q->disk->bdi->ra_pages << (PAGE_SHIFT - 10);
  96         return queue_var_show(ra_kb, page);
  97 }
  98
  99 static ssize_t
 100 queue_ra_store(struct request_queue *q, const char *page, size_t count)
 101 {
 102         unsigned long ra_kb;
 103         ssize_t ret;
 104
 105         if (!q->disk)
 106                 return -EINVAL;
 107         ret = queue_var_store(&ra_kb, page, count);
 108         if (ret < 0)
 109                 return ret;
 110         q->disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 111         return ret;
 112 }
 113
 114 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 115 {
 116         int max_sectors_kb = queue_max_sectors(q) >> 1;
 117
 118         return queue_var_show(max_sectors_kb, page);
 119 }
 120
 121 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 122 {
 123         return queue_var_show(queue_max_segments(q), page);
 124 }
 125
 126 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 127                 char *page)
 128 {
 129         return queue_var_show(queue_max_discard_segments(q), page);
 130 }
 131
 132 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 133 {
 134         return queue_var_show(q->limits.max_integrity_segments, page);
 135 }
 136
 137 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 138 {
 139         return queue_var_show(queue_max_segment_size(q), page);
 140 }
 141
 142 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 143 {
 144         return queue_var_show(queue_logical_block_size(q), page);
 145 }
 146
 147 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 148 {
 149         return queue_var_show(queue_physical_block_size(q), page);
 150 }
 151
 152 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 153 {
 154         return queue_var_show(q->limits.chunk_sectors, page);
 155 }
 156
 157 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 158 {
 159         return queue_var_show(queue_io_min(q), page);
 160 }
 161
 162 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 163 {
 164         return queue_var_show(queue_io_opt(q), page);
 165 }
 166
 167 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 168 {
 169         return queue_var_show(q->limits.discard_granularity, page);
 170 }
 171
 172 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 173 {
 174
 175         return sprintf(page, "%llu\n",
 176                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 177 }
 178
 179 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 180 {
 181         return sprintf(page, "%llu\n",
 182                        (unsigned long long)q->limits.max_discard_sectors << 9);
 183 }
 184
 185 static ssize_t queue_discard_max_store(struct request_queue *q,
 186                                        const char *page, size_t count)
 187 {
 188         unsigned long max_discard;
 189         ssize_t ret = queue_var_store(&max_discard, page, count);
 190
 191         if (ret < 0)
 192                 return ret;
 193
 194         if (max_discard & (q->limits.discard_granularity - 1))
 195                 return -EINVAL;
 196
 197         max_discard >>= 9;
 198         if (max_discard > UINT_MAX)
 199                 return -EINVAL;
 200
 201         if (max_discard > q->limits.max_hw_discard_sectors)
 202                 max_discard = q->limits.max_hw_discard_sectors;
 203
 204         q->limits.max_discard_sectors = max_discard;
 205         return ret;
 206 }
 207
 208 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 209 {
 210         return queue_var_show(0, page);
 211 }
 212
 213 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 214 {
 215         return sprintf(page, "%llu\n",
 216                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 217 }
 218
 219 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 220 {
 221         return sprintf(page, "%llu\n",
 222                 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 223 }
 224
 225 static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
 226                                                  char *page)
 227 {
 228         return queue_var_show(queue_zone_write_granularity(q), page);
 229 }
 230
 231 static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
 232 {
 233         unsigned long long max_sectors = q->limits.max_zone_append_sectors;
 234
 235         return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
 236 }
 237
 238 static ssize_t
 239 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 240 {
 241         unsigned long max_sectors_kb,
 242                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 243                         page_kb = 1 << (PAGE_SHIFT - 10);
 244         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 245
 246         if (ret < 0)
 247                 return ret;
 248
 249         max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
 250                                          q->limits.max_dev_sectors >> 1);
 251
 252         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 253                 return -EINVAL;
 254
 255         spin_lock_irq(&q->queue_lock);
 256         q->limits.max_sectors = max_sectors_kb << 1;
 257         if (q->disk)
 258                 q->disk->bdi->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 259         spin_unlock_irq(&q->queue_lock);
 260
 261         return ret;
 262 }
 263
 264 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 265 {
 266         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 267
 268         return queue_var_show(max_hw_sectors_kb, page);
 269 }
 270
 271 static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
 272 {
 273         return queue_var_show(q->limits.virt_boundary_mask, page);
 274 }
 275
 276 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 277 static ssize_t                                                          \
 278 queue_##name##_show(struct request_queue *q, char *page)                \
 279 {                                                                       \
 280         int bit;                                                        \
 281         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 282         return queue_var_show(neg ? !bit : bit, page);                  \
 283 }                                                                       \
 284 static ssize_t                                                          \
 285 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
 286 {                                                                       \
 287         unsigned long val;                                              \
 288         ssize_t ret;                                                    \
 289         ret = queue_var_store(&val, page, count);                       \
 290         if (ret < 0)                                                    \
 291                  return ret;                                            \
 292         if (neg)                                                        \
 293                 val = !val;                                             \
 294                                                                         \
 295         if (val)                                                        \
 296                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 297         else                                                            \
 298                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 299         return ret;                                                     \
 300 }
 301
 302 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 303 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 304 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 305 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
 306 #undef QUEUE_SYSFS_BIT_FNS
 307
 308 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 309 {
 310         switch (blk_queue_zoned_model(q)) {
 311         case BLK_ZONED_HA:
 312                 return sprintf(page, "host-aware\n");
 313         case BLK_ZONED_HM:
 314                 return sprintf(page, "host-managed\n");
 315         default:
 316                 return sprintf(page, "none\n");
 317         }
 318 }
 319
 320 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
 321 {
 322         return queue_var_show(blk_queue_nr_zones(q), page);
 323 }
 324
 325 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
 326 {
 327         return queue_var_show(queue_max_open_zones(q), page);
 328 }
 329
 330 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
 331 {
 332         return queue_var_show(queue_max_active_zones(q), page);
 333 }
 334
 335 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 336 {
 337         return queue_var_show((blk_queue_nomerges(q) << 1) |
 338                                blk_queue_noxmerges(q), page);
 339 }
 340
 341 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 342                                     size_t count)
 343 {
 344         unsigned long nm;
 345         ssize_t ret = queue_var_store(&nm, page, count);
 346
 347         if (ret < 0)
 348                 return ret;
 349
 350         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 351         blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 352         if (nm == 2)
 353                 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 354         else if (nm)
 355                 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 356
 357         return ret;
 358 }
 359
 360 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 361 {
 362         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 363         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 364
 365         return queue_var_show(set << force, page);
 366 }
 367
 368 static ssize_t
 369 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 370 {
 371         ssize_t ret = -EINVAL;
 372 #ifdef CONFIG_SMP
 373         unsigned long val;
 374
 375         ret = queue_var_store(&val, page, count);
 376         if (ret < 0)
 377                 return ret;
 378
 379         if (val == 2) {
 380                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 381                 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 382         } else if (val == 1) {
 383                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 384                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 385         } else if (val == 0) {
 386                 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 387                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 388         }
 389 #endif
 390         return ret;
 391 }
 392
 393 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 394 {
 395         int val;
 396
 397         if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
 398                 val = BLK_MQ_POLL_CLASSIC;
 399         else
 400                 val = q->poll_nsec / 1000;
 401
 402         return sprintf(page, "%d\n", val);
 403 }
 404
 405 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 406                                 size_t count)
 407 {
 408         int err, val;
 409
 410         if (!q->mq_ops || !q->mq_ops->poll)
 411                 return -EINVAL;
 412
 413         err = kstrtoint(page, 10, &val);
 414         if (err < 0)
 415                 return err;
 416
 417         if (val == BLK_MQ_POLL_CLASSIC)
 418                 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
 419         else if (val >= 0)
 420                 q->poll_nsec = val * 1000;
 421         else
 422                 return -EINVAL;
 423
 424         return count;
 425 }
 426
 427 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 428 {
 429         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 430 }
 431
 432 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 433                                 size_t count)
 434 {
 435         unsigned long poll_on;
 436         ssize_t ret;
 437
 438         if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
 439             !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
 440                 return -EINVAL;
 441
 442         ret = queue_var_store(&poll_on, page, count);
 443         if (ret < 0)
 444                 return ret;
 445
 446         if (poll_on) {
 447                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 448         } else {
 449                 blk_mq_freeze_queue(q);
 450                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 451                 blk_mq_unfreeze_queue(q);
 452         }
 453
 454         return ret;
 455 }
 456
 457 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
 458 {
 459         return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
 460 }
 461
 462 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
 463                                   size_t count)
 464 {
 465         unsigned int val;
 466         int err;
 467
 468         err = kstrtou32(page, 10, &val);
 469         if (err || val == 0)
 470                 return -EINVAL;
 471
 472         blk_queue_rq_timeout(q, msecs_to_jiffies(val));
 473
 474         return count;
 475 }
 476
 477 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 478 {
 479         if (!wbt_rq_qos(q))
 480                 return -EINVAL;
 481
 482         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 483 }
 484
 485 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 486                                   size_t count)
 487 {
 488         struct rq_qos *rqos;
 489         ssize_t ret;
 490         s64 val;
 491
 492         ret = queue_var_store64(&val, page);
 493         if (ret < 0)
 494                 return ret;
 495         if (val < -1)
 496                 return -EINVAL;
 497
 498         rqos = wbt_rq_qos(q);
 499         if (!rqos) {
 500                 ret = wbt_init(q);
 501                 if (ret)
 502                         return ret;
 503         }
 504
 505         if (val == -1)
 506                 val = wbt_default_latency_nsec(q);
 507         else if (val >= 0)
 508                 val *= 1000ULL;
 509
 510         if (wbt_get_min_lat(q) == val)
 511                 return count;
 512
 513         /*
 514          * Ensure that the queue is idled, in case the latency update
 515          * ends up either enabling or disabling wbt completely. We can't
 516          * have IO inflight if that happens.
 517          */
 518         blk_mq_freeze_queue(q);
 519         blk_mq_quiesce_queue(q);
 520
 521         wbt_set_min_lat(q, val);
 522
 523         blk_mq_unquiesce_queue(q);
 524         blk_mq_unfreeze_queue(q);
 525
 526         return count;
 527 }
 528
 529 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 530 {
 531         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 532                 return sprintf(page, "write back\n");
 533
 534         return sprintf(page, "write through\n");
 535 }
 536
 537 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 538                               size_t count)
 539 {
 540         int set = -1;
 541
 542         if (!strncmp(page, "write back", 10))
 543                 set = 1;
 544         else if (!strncmp(page, "write through", 13) ||
 545                  !strncmp(page, "none", 4))
 546                 set = 0;
 547
 548         if (set == -1)
 549                 return -EINVAL;
 550
 551         if (set)
 552                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 553         else
 554                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 555
 556         return count;
 557 }
 558
 559 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 560 {
 561         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 562 }
 563
 564 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 565 {
 566         return queue_var_show(blk_queue_dax(q), page);
 567 }
 568
 569 #define QUEUE_RO_ENTRY(_prefix, _name)                  \
 570 static struct queue_sysfs_entry _prefix##_entry = {     \
 571         .attr   = { .name = _name, .mode = 0444 },      \
 572         .show   = _prefix##_show,                       \
 573 };
 574
 575 #define QUEUE_RW_ENTRY(_prefix, _name)                  \
 576 static struct queue_sysfs_entry _prefix##_entry = {     \
 577         .attr   = { .name = _name, .mode = 0644 },      \
 578         .show   = _prefix##_show,                       \
 579         .store  = _prefix##_store,                      \
 580 };
 581
 582 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
 583 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
 584 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
 585 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
 586 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
 587 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
 588 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
 589 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
 590
 591 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
 592 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
 593 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
 594 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
 595 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
 596
 597 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
 598 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
 599 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
 600 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
 601 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
 602
 603 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
 604 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
 605 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
 606 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
 607
 608 QUEUE_RO_ENTRY(queue_zoned, "zoned");
 609 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
 610 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
 611 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
 612
 613 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
 614 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
 615 QUEUE_RW_ENTRY(queue_poll, "io_poll");
 616 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
 617 QUEUE_RW_ENTRY(queue_wc, "write_cache");
 618 QUEUE_RO_ENTRY(queue_fua, "fua");
 619 QUEUE_RO_ENTRY(queue_dax, "dax");
 620 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
 621 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
 622 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
 623
 624 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 625 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
 626 #endif
 627
 628 /* legacy alias for logical_block_size: */
 629 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 630         .attr = {.name = "hw_sector_size", .mode = 0444 },
 631         .show = queue_logical_block_size_show,
 632 };
 633
 634 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
 635 QUEUE_RW_ENTRY(queue_iostats, "iostats");
 636 QUEUE_RW_ENTRY(queue_random, "add_random");
 637 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
 638
 639 static struct attribute *queue_attrs[] = {
 640         &queue_requests_entry.attr,
 641         &queue_ra_entry.attr,
 642         &queue_max_hw_sectors_entry.attr,
 643         &queue_max_sectors_entry.attr,
 644         &queue_max_segments_entry.attr,
 645         &queue_max_discard_segments_entry.attr,
 646         &queue_max_integrity_segments_entry.attr,
 647         &queue_max_segment_size_entry.attr,
 648         &elv_iosched_entry.attr,
 649         &queue_hw_sector_size_entry.attr,
 650         &queue_logical_block_size_entry.attr,
 651         &queue_physical_block_size_entry.attr,
 652         &queue_chunk_sectors_entry.attr,
 653         &queue_io_min_entry.attr,
 654         &queue_io_opt_entry.attr,
 655         &queue_discard_granularity_entry.attr,
 656         &queue_discard_max_entry.attr,
 657         &queue_discard_max_hw_entry.attr,
 658         &queue_discard_zeroes_data_entry.attr,
 659         &queue_write_same_max_entry.attr,
 660         &queue_write_zeroes_max_entry.attr,
 661         &queue_zone_append_max_entry.attr,
 662         &queue_zone_write_granularity_entry.attr,
 663         &queue_nonrot_entry.attr,
 664         &queue_zoned_entry.attr,
 665         &queue_nr_zones_entry.attr,
 666         &queue_max_open_zones_entry.attr,
 667         &queue_max_active_zones_entry.attr,
 668         &queue_nomerges_entry.attr,
 669         &queue_rq_affinity_entry.attr,
 670         &queue_iostats_entry.attr,
 671         &queue_stable_writes_entry.attr,
 672         &queue_random_entry.attr,
 673         &queue_poll_entry.attr,
 674         &queue_wc_entry.attr,
 675         &queue_fua_entry.attr,
 676         &queue_dax_entry.attr,
 677         &queue_wb_lat_entry.attr,
 678         &queue_poll_delay_entry.attr,
 679         &queue_io_timeout_entry.attr,
 680 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 681         &blk_throtl_sample_time_entry.attr,
 682 #endif
 683         &queue_virt_boundary_mask_entry.attr,
 684         NULL,
 685 };
 686
 687 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
 688                                 int n)
 689 {
 690         struct request_queue *q =
 691                 container_of(kobj, struct request_queue, kobj);
 692
 693         if (attr == &queue_io_timeout_entry.attr &&
 694                 (!q->mq_ops || !q->mq_ops->timeout))
 695                         return 0;
 696
 697         if ((attr == &queue_max_open_zones_entry.attr ||
 698              attr == &queue_max_active_zones_entry.attr) &&
 699             !blk_queue_is_zoned(q))
 700                 return 0;
 701
 702         return attr->mode;
 703 }
 704
 705 static struct attribute_group queue_attr_group = {
 706         .attrs = queue_attrs,
 707         .is_visible = queue_attr_visible,
 708 };
 709
 710
 711 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 712
 713 static ssize_t
 714 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 715 {
 716         struct queue_sysfs_entry *entry = to_queue(attr);
 717         struct request_queue *q =
 718                 container_of(kobj, struct request_queue, kobj);
 719         ssize_t res;
 720
 721         if (!entry->show)
 722                 return -EIO;
 723         mutex_lock(&q->sysfs_lock);
 724         res = entry->show(q, page);
 725         mutex_unlock(&q->sysfs_lock);
 726         return res;
 727 }
 728
 729 static ssize_t
 730 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 731                     const char *page, size_t length)
 732 {
 733         struct queue_sysfs_entry *entry = to_queue(attr);
 734         struct request_queue *q;
 735         ssize_t res;
 736
 737         if (!entry->store)
 738                 return -EIO;
 739
 740         q = container_of(kobj, struct request_queue, kobj);
 741         mutex_lock(&q->sysfs_lock);
 742         res = entry->store(q, page, length);
 743         mutex_unlock(&q->sysfs_lock);
 744         return res;
 745 }
 746
 747 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 748 {
 749         struct request_queue *q = container_of(rcu_head, struct request_queue,
 750                                                rcu_head);
 751         kmem_cache_free(blk_requestq_cachep, q);
 752 }
 753
 754 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 755 static void blk_exit_queue(struct request_queue *q)
 756 {
 757         /*
 758          * Since the I/O scheduler exit code may access cgroup information,
 759          * perform I/O scheduler exit before disassociating from the block
 760          * cgroup controller.
 761          */
 762         if (q->elevator) {
 763                 ioc_clear_queue(q);
 764                 __elevator_exit(q, q->elevator);
 765         }
 766
 767         /*
 768          * Remove all references to @q from the block cgroup controller before
 769          * restoring @q->queue_lock to avoid that restoring this pointer causes
 770          * e.g. blkcg_print_blkgs() to crash.
 771          */
 772         blkcg_exit_queue(q);
 773 }
 774
 775 /**
 776  * blk_release_queue - releases all allocated resources of the request_queue
 777  * @kobj: pointer to a kobject, whose container is a request_queue
 778  *
 779  * This function releases all allocated resources of the request queue.
 780  *
 781  * The struct request_queue refcount is incremented with blk_get_queue() and
 782  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
 783  * is called.
 784  *
 785  * For drivers that have a request_queue on a gendisk and added with
 786  * __device_add_disk() the refcount to request_queue will reach 0 with
 787  * the last put_disk() called by the driver. For drivers which don't use
 788  * __device_add_disk() this happens with blk_cleanup_queue().
 789  *
 790  * Drivers exist which depend on the release of the request_queue to be
 791  * synchronous, it should not be deferred.
 792  *
 793  * Context: can sleep
 794  */
 795 static void blk_release_queue(struct kobject *kobj)
 796 {
 797         struct request_queue *q =
 798                 container_of(kobj, struct request_queue, kobj);
 799
 800         might_sleep();
 801
 802         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 803                 blk_stat_remove_callback(q, q->poll_cb);
 804         blk_stat_free_callback(q->poll_cb);
 805
 806         blk_free_queue_stats(q->stats);
 807
 808         blk_exit_queue(q);
 809
 810         blk_queue_free_zone_bitmaps(q);
 811
 812         if (queue_is_mq(q))
 813                 blk_mq_release(q);
 814
 815         blk_trace_shutdown(q);
 816         mutex_lock(&q->debugfs_mutex);
 817         debugfs_remove_recursive(q->debugfs_dir);
 818         mutex_unlock(&q->debugfs_mutex);
 819
 820         if (queue_is_mq(q))
 821                 blk_mq_debugfs_unregister(q);
 822
 823         bioset_exit(&q->bio_split);
 824
 825         ida_simple_remove(&blk_queue_ida, q->id);
 826         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 827 }
 828
 829 static const struct sysfs_ops queue_sysfs_ops = {
 830         .show   = queue_attr_show,
 831         .store  = queue_attr_store,
 832 };
 833
 834 struct kobj_type blk_queue_ktype = {
 835         .sysfs_ops      = &queue_sysfs_ops,
 836         .release        = blk_release_queue,
 837 };
 838
 839 /**
 840  * blk_register_queue - register a block layer queue with sysfs
 841  * @disk: Disk of which the request queue should be registered with sysfs.
 842  */
 843 int blk_register_queue(struct gendisk *disk)
 844 {
 845         int ret;
 846         struct device *dev = disk_to_dev(disk);
 847         struct request_queue *q = disk->queue;
 848
 849         ret = blk_trace_init_sysfs(dev);
 850         if (ret)
 851                 return ret;
 852
 853         mutex_lock(&q->sysfs_dir_lock);
 854
 855         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 856         if (ret < 0) {
 857                 blk_trace_remove_sysfs(dev);
 858                 goto unlock;
 859         }
 860
 861         ret = sysfs_create_group(&q->kobj, &queue_attr_group);
 862         if (ret) {
 863                 blk_trace_remove_sysfs(dev);
 864                 kobject_del(&q->kobj);
 865                 kobject_put(&dev->kobj);
 866                 goto unlock;
 867         }
 868
 869         mutex_lock(&q->debugfs_mutex);
 870         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 871                                             blk_debugfs_root);
 872         mutex_unlock(&q->debugfs_mutex);
 873
 874         if (queue_is_mq(q)) {
 875                 __blk_mq_register_dev(dev, q);
 876                 blk_mq_debugfs_register(q);
 877         }
 878
 879         mutex_lock(&q->sysfs_lock);
 880         if (q->elevator) {
 881                 ret = elv_register_queue(q, false);
 882                 if (ret) {
 883                         mutex_unlock(&q->sysfs_lock);
 884                         mutex_unlock(&q->sysfs_dir_lock);
 885                         kobject_del(&q->kobj);
 886                         blk_trace_remove_sysfs(dev);
 887                         kobject_put(&dev->kobj);
 888                         return ret;
 889                 }
 890         }
 891
 892         blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
 893         wbt_enable_default(q);
 894         blk_throtl_register_queue(q);
 895
 896         /* Now everything is ready and send out KOBJ_ADD uevent */
 897         kobject_uevent(&q->kobj, KOBJ_ADD);
 898         if (q->elevator)
 899                 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
 900         mutex_unlock(&q->sysfs_lock);
 901
 902         ret = 0;
 903 unlock:
 904         mutex_unlock(&q->sysfs_dir_lock);
 905
 906         /*
 907          * SCSI probing may synchronously create and destroy a lot of
 908          * request_queues for non-existent devices.  Shutting down a fully
 909          * functional queue takes measureable wallclock time as RCU grace
 910          * periods are involved.  To avoid excessive latency in these
 911          * cases, a request_queue starts out in a degraded mode which is
 912          * faster to shut down and is made fully functional here as
 913          * request_queues for non-existent devices never get registered.
 914          */
 915         if (!blk_queue_init_done(q)) {
 916                 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
 917                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 918         }
 919
 920         return ret;
 921 }
 922
 923 /**
 924  * blk_unregister_queue - counterpart of blk_register_queue()
 925  * @disk: Disk of which the request queue should be unregistered from sysfs.
 926  *
 927  * Note: the caller is responsible for guaranteeing that this function is called
 928  * after blk_register_queue() has finished.
 929  */
 930 void blk_unregister_queue(struct gendisk *disk)
 931 {
 932         struct request_queue *q = disk->queue;
 933
 934         if (WARN_ON(!q))
 935                 return;
 936
 937         /* Return early if disk->queue was never registered. */
 938         if (!blk_queue_registered(q))
 939                 return;
 940
 941         /*
 942          * Since sysfs_remove_dir() prevents adding new directory entries
 943          * before removal of existing entries starts, protect against
 944          * concurrent elv_iosched_store() calls.
 945          */
 946         mutex_lock(&q->sysfs_lock);
 947         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
 948         mutex_unlock(&q->sysfs_lock);
 949
 950         mutex_lock(&q->sysfs_dir_lock);
 951         /*
 952          * Remove the sysfs attributes before unregistering the queue data
 953          * structures that can be modified through sysfs.
 954          */
 955         if (queue_is_mq(q))
 956                 blk_mq_unregister_dev(disk_to_dev(disk), q);
 957         blk_trace_remove_sysfs(disk_to_dev(disk));
 958
 959         mutex_lock(&q->sysfs_lock);
 960         if (q->elevator)
 961                 elv_unregister_queue(q);
 962         mutex_unlock(&q->sysfs_lock);
 963
 964         /* Now that we've deleted all child objects, we can delete the queue. */
 965         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 966         kobject_del(&q->kobj);
 967
 968         mutex_unlock(&q->sysfs_dir_lock);
 969
 970         kobject_put(&disk_to_dev(disk)->kobj);
 971 }