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 = q->backing_dev_info->ra_pages <<
  92                                         (PAGE_SHIFT - 10);
  93
  94         return queue_var_show(ra_kb, (page));
  95 }
  96
  97 static ssize_t
  98 queue_ra_store(struct request_queue *q, const char *page, size_t count)
  99 {
 100         unsigned long ra_kb;
 101         ssize_t ret = queue_var_store(&ra_kb, page, count);
 102
 103         if (ret < 0)
 104                 return ret;
 105
 106         q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 107
 108         return ret;
 109 }
 110
 111 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 112 {
 113         int max_sectors_kb = queue_max_sectors(q) >> 1;
 114
 115         return queue_var_show(max_sectors_kb, (page));
 116 }
 117
 118 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 119 {
 120         return queue_var_show(queue_max_segments(q), (page));
 121 }
 122
 123 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 124                 char *page)
 125 {
 126         return queue_var_show(queue_max_discard_segments(q), (page));
 127 }
 128
 129 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 130 {
 131         return queue_var_show(q->limits.max_integrity_segments, (page));
 132 }
 133
 134 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 135 {
 136         return queue_var_show(queue_max_segment_size(q), (page));
 137 }
 138
 139 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 140 {
 141         return queue_var_show(queue_logical_block_size(q), page);
 142 }
 143
 144 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 145 {
 146         return queue_var_show(queue_physical_block_size(q), page);
 147 }
 148
 149 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 150 {
 151         return queue_var_show(q->limits.chunk_sectors, page);
 152 }
 153
 154 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 155 {
 156         return queue_var_show(queue_io_min(q), page);
 157 }
 158
 159 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 160 {
 161         return queue_var_show(queue_io_opt(q), page);
 162 }
 163
 164 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 165 {
 166         return queue_var_show(q->limits.discard_granularity, page);
 167 }
 168
 169 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 170 {
 171
 172         return sprintf(page, "%llu\n",
 173                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 174 }
 175
 176 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 177 {
 178         return sprintf(page, "%llu\n",
 179                        (unsigned long long)q->limits.max_discard_sectors << 9);
 180 }
 181
 182 static ssize_t queue_discard_max_store(struct request_queue *q,
 183                                        const char *page, size_t count)
 184 {
 185         unsigned long max_discard;
 186         ssize_t ret = queue_var_store(&max_discard, page, count);
 187
 188         if (ret < 0)
 189                 return ret;
 190
 191         if (max_discard & (q->limits.discard_granularity - 1))
 192                 return -EINVAL;
 193
 194         max_discard >>= 9;
 195         if (max_discard > UINT_MAX)
 196                 return -EINVAL;
 197
 198         if (max_discard > q->limits.max_hw_discard_sectors)
 199                 max_discard = q->limits.max_hw_discard_sectors;
 200
 201         q->limits.max_discard_sectors = max_discard;
 202         return ret;
 203 }
 204
 205 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 206 {
 207         return queue_var_show(0, page);
 208 }
 209
 210 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 211 {
 212         return sprintf(page, "%llu\n",
 213                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 214 }
 215
 216 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 217 {
 218         return sprintf(page, "%llu\n",
 219                 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 220 }
 221
 222 static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
 223 {
 224         unsigned long long max_sectors = q->limits.max_zone_append_sectors;
 225
 226         return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
 227 }
 228
 229 static ssize_t
 230 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 231 {
 232         unsigned long max_sectors_kb,
 233                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 234                         page_kb = 1 << (PAGE_SHIFT - 10);
 235         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 236
 237         if (ret < 0)
 238                 return ret;
 239
 240         max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
 241                                          q->limits.max_dev_sectors >> 1);
 242
 243         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 244                 return -EINVAL;
 245
 246         spin_lock_irq(&q->queue_lock);
 247         q->limits.max_sectors = max_sectors_kb << 1;
 248         q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 249         spin_unlock_irq(&q->queue_lock);
 250
 251         return ret;
 252 }
 253
 254 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 255 {
 256         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 257
 258         return queue_var_show(max_hw_sectors_kb, (page));
 259 }
 260
 261 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 262 static ssize_t                                                          \
 263 queue_show_##name(struct request_queue *q, char *page)                  \
 264 {                                                                       \
 265         int bit;                                                        \
 266         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 267         return queue_var_show(neg ? !bit : bit, page);                  \
 268 }                                                                       \
 269 static ssize_t                                                          \
 270 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
 271 {                                                                       \
 272         unsigned long val;                                              \
 273         ssize_t ret;                                                    \
 274         ret = queue_var_store(&val, page, count);                       \
 275         if (ret < 0)                                                    \
 276                  return ret;                                            \
 277         if (neg)                                                        \
 278                 val = !val;                                             \
 279                                                                         \
 280         if (val)                                                        \
 281                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 282         else                                                            \
 283                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 284         return ret;                                                     \
 285 }
 286
 287 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 288 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 289 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 290 #undef QUEUE_SYSFS_BIT_FNS
 291
 292 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 293 {
 294         switch (blk_queue_zoned_model(q)) {
 295         case BLK_ZONED_HA:
 296                 return sprintf(page, "host-aware\n");
 297         case BLK_ZONED_HM:
 298                 return sprintf(page, "host-managed\n");
 299         default:
 300                 return sprintf(page, "none\n");
 301         }
 302 }
 303
 304 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
 305 {
 306         return queue_var_show(blk_queue_nr_zones(q), page);
 307 }
 308
 309 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 310 {
 311         return queue_var_show((blk_queue_nomerges(q) << 1) |
 312                                blk_queue_noxmerges(q), page);
 313 }
 314
 315 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 316                                     size_t count)
 317 {
 318         unsigned long nm;
 319         ssize_t ret = queue_var_store(&nm, page, count);
 320
 321         if (ret < 0)
 322                 return ret;
 323
 324         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 325         blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 326         if (nm == 2)
 327                 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 328         else if (nm)
 329                 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 330
 331         return ret;
 332 }
 333
 334 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 335 {
 336         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 337         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 338
 339         return queue_var_show(set << force, page);
 340 }
 341
 342 static ssize_t
 343 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 344 {
 345         ssize_t ret = -EINVAL;
 346 #ifdef CONFIG_SMP
 347         unsigned long val;
 348
 349         ret = queue_var_store(&val, page, count);
 350         if (ret < 0)
 351                 return ret;
 352
 353         if (val == 2) {
 354                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 355                 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 356         } else if (val == 1) {
 357                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 358                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 359         } else if (val == 0) {
 360                 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 361                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 362         }
 363 #endif
 364         return ret;
 365 }
 366
 367 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 368 {
 369         int val;
 370
 371         if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
 372                 val = BLK_MQ_POLL_CLASSIC;
 373         else
 374                 val = q->poll_nsec / 1000;
 375
 376         return sprintf(page, "%d\n", val);
 377 }
 378
 379 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 380                                 size_t count)
 381 {
 382         int err, val;
 383
 384         if (!q->mq_ops || !q->mq_ops->poll)
 385                 return -EINVAL;
 386
 387         err = kstrtoint(page, 10, &val);
 388         if (err < 0)
 389                 return err;
 390
 391         if (val == BLK_MQ_POLL_CLASSIC)
 392                 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
 393         else if (val >= 0)
 394                 q->poll_nsec = val * 1000;
 395         else
 396                 return -EINVAL;
 397
 398         return count;
 399 }
 400
 401 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 402 {
 403         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 404 }
 405
 406 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 407                                 size_t count)
 408 {
 409         unsigned long poll_on;
 410         ssize_t ret;
 411
 412         if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
 413             !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
 414                 return -EINVAL;
 415
 416         ret = queue_var_store(&poll_on, page, count);
 417         if (ret < 0)
 418                 return ret;
 419
 420         if (poll_on)
 421                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 422         else
 423                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 424
 425         return ret;
 426 }
 427
 428 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
 429 {
 430         return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
 431 }
 432
 433 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
 434                                   size_t count)
 435 {
 436         unsigned int val;
 437         int err;
 438
 439         err = kstrtou32(page, 10, &val);
 440         if (err || val == 0)
 441                 return -EINVAL;
 442
 443         blk_queue_rq_timeout(q, msecs_to_jiffies(val));
 444
 445         return count;
 446 }
 447
 448 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 449 {
 450         if (!wbt_rq_qos(q))
 451                 return -EINVAL;
 452
 453         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 454 }
 455
 456 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 457                                   size_t count)
 458 {
 459         struct rq_qos *rqos;
 460         ssize_t ret;
 461         s64 val;
 462
 463         ret = queue_var_store64(&val, page);
 464         if (ret < 0)
 465                 return ret;
 466         if (val < -1)
 467                 return -EINVAL;
 468
 469         rqos = wbt_rq_qos(q);
 470         if (!rqos) {
 471                 ret = wbt_init(q);
 472                 if (ret)
 473                         return ret;
 474         }
 475
 476         if (val == -1)
 477                 val = wbt_default_latency_nsec(q);
 478         else if (val >= 0)
 479                 val *= 1000ULL;
 480
 481         if (wbt_get_min_lat(q) == val)
 482                 return count;
 483
 484         /*
 485          * Ensure that the queue is idled, in case the latency update
 486          * ends up either enabling or disabling wbt completely. We can't
 487          * have IO inflight if that happens.
 488          */
 489         blk_mq_freeze_queue(q);
 490         blk_mq_quiesce_queue(q);
 491
 492         wbt_set_min_lat(q, val);
 493
 494         blk_mq_unquiesce_queue(q);
 495         blk_mq_unfreeze_queue(q);
 496
 497         return count;
 498 }
 499
 500 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 501 {
 502         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 503                 return sprintf(page, "write back\n");
 504
 505         return sprintf(page, "write through\n");
 506 }
 507
 508 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 509                               size_t count)
 510 {
 511         int set = -1;
 512
 513         if (!strncmp(page, "write back", 10))
 514                 set = 1;
 515         else if (!strncmp(page, "write through", 13) ||
 516                  !strncmp(page, "none", 4))
 517                 set = 0;
 518
 519         if (set == -1)
 520                 return -EINVAL;
 521
 522         if (set)
 523                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 524         else
 525                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 526
 527         return count;
 528 }
 529
 530 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 531 {
 532         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 533 }
 534
 535 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 536 {
 537         return queue_var_show(blk_queue_dax(q), page);
 538 }
 539
 540 static struct queue_sysfs_entry queue_requests_entry = {
 541         .attr = {.name = "nr_requests", .mode = 0644 },
 542         .show = queue_requests_show,
 543         .store = queue_requests_store,
 544 };
 545
 546 static struct queue_sysfs_entry queue_ra_entry = {
 547         .attr = {.name = "read_ahead_kb", .mode = 0644 },
 548         .show = queue_ra_show,
 549         .store = queue_ra_store,
 550 };
 551
 552 static struct queue_sysfs_entry queue_max_sectors_entry = {
 553         .attr = {.name = "max_sectors_kb", .mode = 0644 },
 554         .show = queue_max_sectors_show,
 555         .store = queue_max_sectors_store,
 556 };
 557
 558 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 559         .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
 560         .show = queue_max_hw_sectors_show,
 561 };
 562
 563 static struct queue_sysfs_entry queue_max_segments_entry = {
 564         .attr = {.name = "max_segments", .mode = 0444 },
 565         .show = queue_max_segments_show,
 566 };
 567
 568 static struct queue_sysfs_entry queue_max_discard_segments_entry = {
 569         .attr = {.name = "max_discard_segments", .mode = 0444 },
 570         .show = queue_max_discard_segments_show,
 571 };
 572
 573 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 574         .attr = {.name = "max_integrity_segments", .mode = 0444 },
 575         .show = queue_max_integrity_segments_show,
 576 };
 577
 578 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 579         .attr = {.name = "max_segment_size", .mode = 0444 },
 580         .show = queue_max_segment_size_show,
 581 };
 582
 583 static struct queue_sysfs_entry queue_iosched_entry = {
 584         .attr = {.name = "scheduler", .mode = 0644 },
 585         .show = elv_iosched_show,
 586         .store = elv_iosched_store,
 587 };
 588
 589 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 590         .attr = {.name = "hw_sector_size", .mode = 0444 },
 591         .show = queue_logical_block_size_show,
 592 };
 593
 594 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 595         .attr = {.name = "logical_block_size", .mode = 0444 },
 596         .show = queue_logical_block_size_show,
 597 };
 598
 599 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 600         .attr = {.name = "physical_block_size", .mode = 0444 },
 601         .show = queue_physical_block_size_show,
 602 };
 603
 604 static struct queue_sysfs_entry queue_chunk_sectors_entry = {
 605         .attr = {.name = "chunk_sectors", .mode = 0444 },
 606         .show = queue_chunk_sectors_show,
 607 };
 608
 609 static struct queue_sysfs_entry queue_io_min_entry = {
 610         .attr = {.name = "minimum_io_size", .mode = 0444 },
 611         .show = queue_io_min_show,
 612 };
 613
 614 static struct queue_sysfs_entry queue_io_opt_entry = {
 615         .attr = {.name = "optimal_io_size", .mode = 0444 },
 616         .show = queue_io_opt_show,
 617 };
 618
 619 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 620         .attr = {.name = "discard_granularity", .mode = 0444 },
 621         .show = queue_discard_granularity_show,
 622 };
 623
 624 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
 625         .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
 626         .show = queue_discard_max_hw_show,
 627 };
 628
 629 static struct queue_sysfs_entry queue_discard_max_entry = {
 630         .attr = {.name = "discard_max_bytes", .mode = 0644 },
 631         .show = queue_discard_max_show,
 632         .store = queue_discard_max_store,
 633 };
 634
 635 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 636         .attr = {.name = "discard_zeroes_data", .mode = 0444 },
 637         .show = queue_discard_zeroes_data_show,
 638 };
 639
 640 static struct queue_sysfs_entry queue_write_same_max_entry = {
 641         .attr = {.name = "write_same_max_bytes", .mode = 0444 },
 642         .show = queue_write_same_max_show,
 643 };
 644
 645 static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 646         .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
 647         .show = queue_write_zeroes_max_show,
 648 };
 649
 650 static struct queue_sysfs_entry queue_zone_append_max_entry = {
 651         .attr = {.name = "zone_append_max_bytes", .mode = 0444 },
 652         .show = queue_zone_append_max_show,
 653 };
 654
 655 static struct queue_sysfs_entry queue_nonrot_entry = {
 656         .attr = {.name = "rotational", .mode = 0644 },
 657         .show = queue_show_nonrot,
 658         .store = queue_store_nonrot,
 659 };
 660
 661 static struct queue_sysfs_entry queue_zoned_entry = {
 662         .attr = {.name = "zoned", .mode = 0444 },
 663         .show = queue_zoned_show,
 664 };
 665
 666 static struct queue_sysfs_entry queue_nr_zones_entry = {
 667         .attr = {.name = "nr_zones", .mode = 0444 },
 668         .show = queue_nr_zones_show,
 669 };
 670
 671 static struct queue_sysfs_entry queue_nomerges_entry = {
 672         .attr = {.name = "nomerges", .mode = 0644 },
 673         .show = queue_nomerges_show,
 674         .store = queue_nomerges_store,
 675 };
 676
 677 static struct queue_sysfs_entry queue_rq_affinity_entry = {
 678         .attr = {.name = "rq_affinity", .mode = 0644 },
 679         .show = queue_rq_affinity_show,
 680         .store = queue_rq_affinity_store,
 681 };
 682
 683 static struct queue_sysfs_entry queue_iostats_entry = {
 684         .attr = {.name = "iostats", .mode = 0644 },
 685         .show = queue_show_iostats,
 686         .store = queue_store_iostats,
 687 };
 688
 689 static struct queue_sysfs_entry queue_random_entry = {
 690         .attr = {.name = "add_random", .mode = 0644 },
 691         .show = queue_show_random,
 692         .store = queue_store_random,
 693 };
 694
 695 static struct queue_sysfs_entry queue_poll_entry = {
 696         .attr = {.name = "io_poll", .mode = 0644 },
 697         .show = queue_poll_show,
 698         .store = queue_poll_store,
 699 };
 700
 701 static struct queue_sysfs_entry queue_poll_delay_entry = {
 702         .attr = {.name = "io_poll_delay", .mode = 0644 },
 703         .show = queue_poll_delay_show,
 704         .store = queue_poll_delay_store,
 705 };
 706
 707 static struct queue_sysfs_entry queue_wc_entry = {
 708         .attr = {.name = "write_cache", .mode = 0644 },
 709         .show = queue_wc_show,
 710         .store = queue_wc_store,
 711 };
 712
 713 static struct queue_sysfs_entry queue_fua_entry = {
 714         .attr = {.name = "fua", .mode = 0444 },
 715         .show = queue_fua_show,
 716 };
 717
 718 static struct queue_sysfs_entry queue_dax_entry = {
 719         .attr = {.name = "dax", .mode = 0444 },
 720         .show = queue_dax_show,
 721 };
 722
 723 static struct queue_sysfs_entry queue_io_timeout_entry = {
 724         .attr = {.name = "io_timeout", .mode = 0644 },
 725         .show = queue_io_timeout_show,
 726         .store = queue_io_timeout_store,
 727 };
 728
 729 static struct queue_sysfs_entry queue_wb_lat_entry = {
 730         .attr = {.name = "wbt_lat_usec", .mode = 0644 },
 731         .show = queue_wb_lat_show,
 732         .store = queue_wb_lat_store,
 733 };
 734
 735 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 736 static struct queue_sysfs_entry throtl_sample_time_entry = {
 737         .attr = {.name = "throttle_sample_time", .mode = 0644 },
 738         .show = blk_throtl_sample_time_show,
 739         .store = blk_throtl_sample_time_store,
 740 };
 741 #endif
 742
 743 static struct attribute *queue_attrs[] = {
 744         &queue_requests_entry.attr,
 745         &queue_ra_entry.attr,
 746         &queue_max_hw_sectors_entry.attr,
 747         &queue_max_sectors_entry.attr,
 748         &queue_max_segments_entry.attr,
 749         &queue_max_discard_segments_entry.attr,
 750         &queue_max_integrity_segments_entry.attr,
 751         &queue_max_segment_size_entry.attr,
 752         &queue_iosched_entry.attr,
 753         &queue_hw_sector_size_entry.attr,
 754         &queue_logical_block_size_entry.attr,
 755         &queue_physical_block_size_entry.attr,
 756         &queue_chunk_sectors_entry.attr,
 757         &queue_io_min_entry.attr,
 758         &queue_io_opt_entry.attr,
 759         &queue_discard_granularity_entry.attr,
 760         &queue_discard_max_entry.attr,
 761         &queue_discard_max_hw_entry.attr,
 762         &queue_discard_zeroes_data_entry.attr,
 763         &queue_write_same_max_entry.attr,
 764         &queue_write_zeroes_max_entry.attr,
 765         &queue_zone_append_max_entry.attr,
 766         &queue_nonrot_entry.attr,
 767         &queue_zoned_entry.attr,
 768         &queue_nr_zones_entry.attr,
 769         &queue_nomerges_entry.attr,
 770         &queue_rq_affinity_entry.attr,
 771         &queue_iostats_entry.attr,
 772         &queue_random_entry.attr,
 773         &queue_poll_entry.attr,
 774         &queue_wc_entry.attr,
 775         &queue_fua_entry.attr,
 776         &queue_dax_entry.attr,
 777         &queue_wb_lat_entry.attr,
 778         &queue_poll_delay_entry.attr,
 779         &queue_io_timeout_entry.attr,
 780 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 781         &throtl_sample_time_entry.attr,
 782 #endif
 783         NULL,
 784 };
 785
 786 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
 787                                 int n)
 788 {
 789         struct request_queue *q =
 790                 container_of(kobj, struct request_queue, kobj);
 791
 792         if (attr == &queue_io_timeout_entry.attr &&
 793                 (!q->mq_ops || !q->mq_ops->timeout))
 794                         return 0;
 795
 796         return attr->mode;
 797 }
 798
 799 static struct attribute_group queue_attr_group = {
 800         .attrs = queue_attrs,
 801         .is_visible = queue_attr_visible,
 802 };
 803
 804
 805 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 806
 807 static ssize_t
 808 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 809 {
 810         struct queue_sysfs_entry *entry = to_queue(attr);
 811         struct request_queue *q =
 812                 container_of(kobj, struct request_queue, kobj);
 813         ssize_t res;
 814
 815         if (!entry->show)
 816                 return -EIO;
 817         mutex_lock(&q->sysfs_lock);
 818         res = entry->show(q, page);
 819         mutex_unlock(&q->sysfs_lock);
 820         return res;
 821 }
 822
 823 static ssize_t
 824 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 825                     const char *page, size_t length)
 826 {
 827         struct queue_sysfs_entry *entry = to_queue(attr);
 828         struct request_queue *q;
 829         ssize_t res;
 830
 831         if (!entry->store)
 832                 return -EIO;
 833
 834         q = container_of(kobj, struct request_queue, kobj);
 835         mutex_lock(&q->sysfs_lock);
 836         res = entry->store(q, page, length);
 837         mutex_unlock(&q->sysfs_lock);
 838         return res;
 839 }
 840
 841 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 842 {
 843         struct request_queue *q = container_of(rcu_head, struct request_queue,
 844                                                rcu_head);
 845         kmem_cache_free(blk_requestq_cachep, q);
 846 }
 847
 848 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 849 static void blk_exit_queue(struct request_queue *q)
 850 {
 851         /*
 852          * Since the I/O scheduler exit code may access cgroup information,
 853          * perform I/O scheduler exit before disassociating from the block
 854          * cgroup controller.
 855          */
 856         if (q->elevator) {
 857                 ioc_clear_queue(q);
 858                 __elevator_exit(q, q->elevator);
 859                 q->elevator = NULL;
 860         }
 861
 862         /*
 863          * Remove all references to @q from the block cgroup controller before
 864          * restoring @q->queue_lock to avoid that restoring this pointer causes
 865          * e.g. blkcg_print_blkgs() to crash.
 866          */
 867         blkcg_exit_queue(q);
 868
 869         /*
 870          * Since the cgroup code may dereference the @q->backing_dev_info
 871          * pointer, only decrease its reference count after having removed the
 872          * association with the block cgroup controller.
 873          */
 874         bdi_put(q->backing_dev_info);
 875 }
 876
 877 /**
 878  * blk_release_queue - releases all allocated resources of the request_queue
 879  * @kobj: pointer to a kobject, whose container is a request_queue
 880  *
 881  * This function releases all allocated resources of the request queue.
 882  *
 883  * The struct request_queue refcount is incremented with blk_get_queue() and
 884  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
 885  * is called.
 886  *
 887  * For drivers that have a request_queue on a gendisk and added with
 888  * __device_add_disk() the refcount to request_queue will reach 0 with
 889  * the last put_disk() called by the driver. For drivers which don't use
 890  * __device_add_disk() this happens with blk_cleanup_queue().
 891  *
 892  * Drivers exist which depend on the release of the request_queue to be
 893  * synchronous, it should not be deferred.
 894  *
 895  * Context: can sleep
 896  */
 897 static void blk_release_queue(struct kobject *kobj)
 898 {
 899         struct request_queue *q =
 900                 container_of(kobj, struct request_queue, kobj);
 901
 902         might_sleep();
 903
 904         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 905                 blk_stat_remove_callback(q, q->poll_cb);
 906         blk_stat_free_callback(q->poll_cb);
 907
 908         blk_free_queue_stats(q->stats);
 909
 910         if (queue_is_mq(q))
 911                 cancel_delayed_work_sync(&q->requeue_work);
 912
 913         blk_exit_queue(q);
 914
 915         blk_queue_free_zone_bitmaps(q);
 916
 917         if (queue_is_mq(q))
 918                 blk_mq_release(q);
 919
 920         blk_trace_shutdown(q);
 921         mutex_lock(&q->debugfs_mutex);
 922         debugfs_remove_recursive(q->debugfs_dir);
 923         mutex_unlock(&q->debugfs_mutex);
 924
 925         if (queue_is_mq(q))
 926                 blk_mq_debugfs_unregister(q);
 927
 928         bioset_exit(&q->bio_split);
 929
 930         ida_simple_remove(&blk_queue_ida, q->id);
 931         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 932 }
 933
 934 static const struct sysfs_ops queue_sysfs_ops = {
 935         .show   = queue_attr_show,
 936         .store  = queue_attr_store,
 937 };
 938
 939 struct kobj_type blk_queue_ktype = {
 940         .sysfs_ops      = &queue_sysfs_ops,
 941         .release        = blk_release_queue,
 942 };
 943
 944 /**
 945  * blk_register_queue - register a block layer queue with sysfs
 946  * @disk: Disk of which the request queue should be registered with sysfs.
 947  */
 948 int blk_register_queue(struct gendisk *disk)
 949 {
 950         int ret;
 951         struct device *dev = disk_to_dev(disk);
 952         struct request_queue *q = disk->queue;
 953         bool has_elevator = false;
 954
 955         if (WARN_ON(!q))
 956                 return -ENXIO;
 957
 958         WARN_ONCE(blk_queue_registered(q),
 959                   "%s is registering an already registered queue\n",
 960                   kobject_name(&dev->kobj));
 961
 962         /*
 963          * SCSI probing may synchronously create and destroy a lot of
 964          * request_queues for non-existent devices.  Shutting down a fully
 965          * functional queue takes measureable wallclock time as RCU grace
 966          * periods are involved.  To avoid excessive latency in these
 967          * cases, a request_queue starts out in a degraded mode which is
 968          * faster to shut down and is made fully functional here as
 969          * request_queues for non-existent devices never get registered.
 970          */
 971         if (!blk_queue_init_done(q)) {
 972                 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
 973                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 974         }
 975
 976         ret = blk_trace_init_sysfs(dev);
 977         if (ret)
 978                 return ret;
 979
 980         mutex_lock(&q->sysfs_dir_lock);
 981
 982         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 983         if (ret < 0) {
 984                 blk_trace_remove_sysfs(dev);
 985                 goto unlock;
 986         }
 987
 988         ret = sysfs_create_group(&q->kobj, &queue_attr_group);
 989         if (ret) {
 990                 blk_trace_remove_sysfs(dev);
 991                 kobject_del(&q->kobj);
 992                 kobject_put(&dev->kobj);
 993                 goto unlock;
 994         }
 995
 996         mutex_lock(&q->debugfs_mutex);
 997         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 998                                             blk_debugfs_root);
 999         mutex_unlock(&q->debugfs_mutex);
1000
1001         if (queue_is_mq(q)) {
1002                 __blk_mq_register_dev(dev, q);
1003                 blk_mq_debugfs_register(q);
1004         }
1005
1006         mutex_lock(&q->sysfs_lock);
1007         if (q->elevator) {
1008                 ret = elv_register_queue(q, false);
1009                 if (ret) {
1010                         mutex_unlock(&q->sysfs_lock);
1011                         mutex_unlock(&q->sysfs_dir_lock);
1012                         kobject_del(&q->kobj);
1013                         blk_trace_remove_sysfs(dev);
1014                         kobject_put(&dev->kobj);
1015                         return ret;
1016                 }
1017                 has_elevator = true;
1018         }
1019
1020         blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
1021         wbt_enable_default(q);
1022         blk_throtl_register_queue(q);
1023
1024         /* Now everything is ready and send out KOBJ_ADD uevent */
1025         kobject_uevent(&q->kobj, KOBJ_ADD);
1026         if (has_elevator)
1027                 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
1028         mutex_unlock(&q->sysfs_lock);
1029
1030         ret = 0;
1031 unlock:
1032         mutex_unlock(&q->sysfs_dir_lock);
1033         return ret;
1034 }
1035 EXPORT_SYMBOL_GPL(blk_register_queue);
1036
1037 /**
1038  * blk_unregister_queue - counterpart of blk_register_queue()
1039  * @disk: Disk of which the request queue should be unregistered from sysfs.
1040  *
1041  * Note: the caller is responsible for guaranteeing that this function is called
1042  * after blk_register_queue() has finished.
1043  */
1044 void blk_unregister_queue(struct gendisk *disk)
1045 {
1046         struct request_queue *q = disk->queue;
1047
1048         if (WARN_ON(!q))
1049                 return;
1050
1051         /* Return early if disk->queue was never registered. */
1052         if (!blk_queue_registered(q))
1053                 return;
1054
1055         /*
1056          * Since sysfs_remove_dir() prevents adding new directory entries
1057          * before removal of existing entries starts, protect against
1058          * concurrent elv_iosched_store() calls.
1059          */
1060         mutex_lock(&q->sysfs_lock);
1061         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1062         mutex_unlock(&q->sysfs_lock);
1063
1064         mutex_lock(&q->sysfs_dir_lock);
1065         /*
1066          * Remove the sysfs attributes before unregistering the queue data
1067          * structures that can be modified through sysfs.
1068          */
1069         if (queue_is_mq(q))
1070                 blk_mq_unregister_dev(disk_to_dev(disk), q);
1071
1072         kobject_uevent(&q->kobj, KOBJ_REMOVE);
1073         kobject_del(&q->kobj);
1074         blk_trace_remove_sysfs(disk_to_dev(disk));
1075
1076         mutex_lock(&q->sysfs_lock);
1077         if (q->elevator)
1078                 elv_unregister_queue(q);
1079         mutex_unlock(&q->sysfs_lock);
1080         mutex_unlock(&q->sysfs_dir_lock);
1081
1082         kobject_put(&disk_to_dev(disk)->kobj);
1083 }