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
  15 #include "blk.h"
  16 #include "blk-mq.h"
  17 #include "blk-mq-debugfs.h"
  18 #include "blk-wbt.h"
  19
  20 struct queue_sysfs_entry {
  21         struct attribute attr;
  22         ssize_t (*show)(struct request_queue *, char *);
  23         ssize_t (*store)(struct request_queue *, const char *, size_t);
  24 };
  25
  26 static ssize_t
  27 queue_var_show(unsigned long var, char *page)
  28 {
  29         return sprintf(page, "%lu\n", var);
  30 }
  31
  32 static ssize_t
  33 queue_var_store(unsigned long *var, const char *page, size_t count)
  34 {
  35         int err;
  36         unsigned long v;
  37
  38         err = kstrtoul(page, 10, &v);
  39         if (err || v > UINT_MAX)
  40                 return -EINVAL;
  41
  42         *var = v;
  43
  44         return count;
  45 }
  46
  47 static ssize_t queue_var_store64(s64 *var, const char *page)
  48 {
  49         int err;
  50         s64 v;
  51
  52         err = kstrtos64(page, 10, &v);
  53         if (err < 0)
  54                 return err;
  55
  56         *var = v;
  57         return 0;
  58 }
  59
  60 static ssize_t queue_requests_show(struct request_queue *q, char *page)
  61 {
  62         return queue_var_show(q->nr_requests, (page));
  63 }
  64
  65 static ssize_t
  66 queue_requests_store(struct request_queue *q, const char *page, size_t count)
  67 {
  68         unsigned long nr;
  69         int ret, err;
  70
  71         if (!queue_is_mq(q))
  72                 return -EINVAL;
  73
  74         ret = queue_var_store(&nr, page, count);
  75         if (ret < 0)
  76                 return ret;
  77
  78         if (nr < BLKDEV_MIN_RQ)
  79                 nr = BLKDEV_MIN_RQ;
  80
  81         err = blk_mq_update_nr_requests(q, nr);
  82         if (err)
  83                 return err;
  84
  85         return ret;
  86 }
  87
  88 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  89 {
  90         unsigned long ra_kb = q->backing_dev_info->ra_pages <<
  91                                         (PAGE_SHIFT - 10);
  92
  93         return queue_var_show(ra_kb, (page));
  94 }
  95
  96 static ssize_t
  97 queue_ra_store(struct request_queue *q, const char *page, size_t count)
  98 {
  99         unsigned long ra_kb;
 100         ssize_t ret = queue_var_store(&ra_kb, page, count);
 101
 102         if (ret < 0)
 103                 return ret;
 104
 105         q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 106
 107         return ret;
 108 }
 109
 110 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 111 {
 112         int max_sectors_kb = queue_max_sectors(q) >> 1;
 113
 114         return queue_var_show(max_sectors_kb, (page));
 115 }
 116
 117 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 118 {
 119         return queue_var_show(queue_max_segments(q), (page));
 120 }
 121
 122 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 123                 char *page)
 124 {
 125         return queue_var_show(queue_max_discard_segments(q), (page));
 126 }
 127
 128 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 129 {
 130         return queue_var_show(q->limits.max_integrity_segments, (page));
 131 }
 132
 133 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 134 {
 135         return queue_var_show(queue_max_segment_size(q), (page));
 136 }
 137
 138 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 139 {
 140         return queue_var_show(queue_logical_block_size(q), page);
 141 }
 142
 143 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 144 {
 145         return queue_var_show(queue_physical_block_size(q), page);
 146 }
 147
 148 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 149 {
 150         return queue_var_show(q->limits.chunk_sectors, page);
 151 }
 152
 153 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 154 {
 155         return queue_var_show(queue_io_min(q), page);
 156 }
 157
 158 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 159 {
 160         return queue_var_show(queue_io_opt(q), page);
 161 }
 162
 163 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 164 {
 165         return queue_var_show(q->limits.discard_granularity, page);
 166 }
 167
 168 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 169 {
 170
 171         return sprintf(page, "%llu\n",
 172                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 173 }
 174
 175 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 176 {
 177         return sprintf(page, "%llu\n",
 178                        (unsigned long long)q->limits.max_discard_sectors << 9);
 179 }
 180
 181 static ssize_t queue_discard_max_store(struct request_queue *q,
 182                                        const char *page, size_t count)
 183 {
 184         unsigned long max_discard;
 185         ssize_t ret = queue_var_store(&max_discard, page, count);
 186
 187         if (ret < 0)
 188                 return ret;
 189
 190         if (max_discard & (q->limits.discard_granularity - 1))
 191                 return -EINVAL;
 192
 193         max_discard >>= 9;
 194         if (max_discard > UINT_MAX)
 195                 return -EINVAL;
 196
 197         if (max_discard > q->limits.max_hw_discard_sectors)
 198                 max_discard = q->limits.max_hw_discard_sectors;
 199
 200         q->limits.max_discard_sectors = max_discard;
 201         return ret;
 202 }
 203
 204 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 205 {
 206         return queue_var_show(0, page);
 207 }
 208
 209 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 210 {
 211         return sprintf(page, "%llu\n",
 212                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 213 }
 214
 215 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 216 {
 217         return sprintf(page, "%llu\n",
 218                 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 219 }
 220
 221 static ssize_t
 222 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 223 {
 224         unsigned long max_sectors_kb,
 225                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 226                         page_kb = 1 << (PAGE_SHIFT - 10);
 227         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 228
 229         if (ret < 0)
 230                 return ret;
 231
 232         max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
 233                                          q->limits.max_dev_sectors >> 1);
 234
 235         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 236                 return -EINVAL;
 237
 238         spin_lock_irq(&q->queue_lock);
 239         q->limits.max_sectors = max_sectors_kb << 1;
 240         q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 241         spin_unlock_irq(&q->queue_lock);
 242
 243         return ret;
 244 }
 245
 246 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 247 {
 248         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 249
 250         return queue_var_show(max_hw_sectors_kb, (page));
 251 }
 252
 253 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 254 static ssize_t                                                          \
 255 queue_show_##name(struct request_queue *q, char *page)                  \
 256 {                                                                       \
 257         int bit;                                                        \
 258         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 259         return queue_var_show(neg ? !bit : bit, page);                  \
 260 }                                                                       \
 261 static ssize_t                                                          \
 262 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
 263 {                                                                       \
 264         unsigned long val;                                              \
 265         ssize_t ret;                                                    \
 266         ret = queue_var_store(&val, page, count);                       \
 267         if (ret < 0)                                                    \
 268                  return ret;                                            \
 269         if (neg)                                                        \
 270                 val = !val;                                             \
 271                                                                         \
 272         if (val)                                                        \
 273                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 274         else                                                            \
 275                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 276         return ret;                                                     \
 277 }
 278
 279 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 280 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 281 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 282 #undef QUEUE_SYSFS_BIT_FNS
 283
 284 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 285 {
 286         switch (blk_queue_zoned_model(q)) {
 287         case BLK_ZONED_HA:
 288                 return sprintf(page, "host-aware\n");
 289         case BLK_ZONED_HM:
 290                 return sprintf(page, "host-managed\n");
 291         default:
 292                 return sprintf(page, "none\n");
 293         }
 294 }
 295
 296 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
 297 {
 298         return queue_var_show(blk_queue_nr_zones(q), page);
 299 }
 300
 301 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 302 {
 303         return queue_var_show((blk_queue_nomerges(q) << 1) |
 304                                blk_queue_noxmerges(q), page);
 305 }
 306
 307 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 308                                     size_t count)
 309 {
 310         unsigned long nm;
 311         ssize_t ret = queue_var_store(&nm, page, count);
 312
 313         if (ret < 0)
 314                 return ret;
 315
 316         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 317         blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 318         if (nm == 2)
 319                 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 320         else if (nm)
 321                 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 322
 323         return ret;
 324 }
 325
 326 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 327 {
 328         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 329         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 330
 331         return queue_var_show(set << force, page);
 332 }
 333
 334 static ssize_t
 335 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 336 {
 337         ssize_t ret = -EINVAL;
 338 #ifdef CONFIG_SMP
 339         unsigned long val;
 340
 341         ret = queue_var_store(&val, page, count);
 342         if (ret < 0)
 343                 return ret;
 344
 345         if (val == 2) {
 346                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 347                 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 348         } else if (val == 1) {
 349                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 350                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 351         } else if (val == 0) {
 352                 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 353                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 354         }
 355 #endif
 356         return ret;
 357 }
 358
 359 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 360 {
 361         int val;
 362
 363         if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
 364                 val = BLK_MQ_POLL_CLASSIC;
 365         else
 366                 val = q->poll_nsec / 1000;
 367
 368         return sprintf(page, "%d\n", val);
 369 }
 370
 371 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 372                                 size_t count)
 373 {
 374         int err, val;
 375
 376         if (!q->mq_ops || !q->mq_ops->poll)
 377                 return -EINVAL;
 378
 379         err = kstrtoint(page, 10, &val);
 380         if (err < 0)
 381                 return err;
 382
 383         if (val == BLK_MQ_POLL_CLASSIC)
 384                 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
 385         else if (val >= 0)
 386                 q->poll_nsec = val * 1000;
 387         else
 388                 return -EINVAL;
 389
 390         return count;
 391 }
 392
 393 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 394 {
 395         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 396 }
 397
 398 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 399                                 size_t count)
 400 {
 401         unsigned long poll_on;
 402         ssize_t ret;
 403
 404         if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
 405             !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
 406                 return -EINVAL;
 407
 408         ret = queue_var_store(&poll_on, page, count);
 409         if (ret < 0)
 410                 return ret;
 411
 412         if (poll_on)
 413                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 414         else
 415                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 416
 417         return ret;
 418 }
 419
 420 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
 421 {
 422         return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
 423 }
 424
 425 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
 426                                   size_t count)
 427 {
 428         unsigned int val;
 429         int err;
 430
 431         err = kstrtou32(page, 10, &val);
 432         if (err || val == 0)
 433                 return -EINVAL;
 434
 435         blk_queue_rq_timeout(q, msecs_to_jiffies(val));
 436
 437         return count;
 438 }
 439
 440 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 441 {
 442         if (!wbt_rq_qos(q))
 443                 return -EINVAL;
 444
 445         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 446 }
 447
 448 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 449                                   size_t count)
 450 {
 451         struct rq_qos *rqos;
 452         ssize_t ret;
 453         s64 val;
 454
 455         ret = queue_var_store64(&val, page);
 456         if (ret < 0)
 457                 return ret;
 458         if (val < -1)
 459                 return -EINVAL;
 460
 461         rqos = wbt_rq_qos(q);
 462         if (!rqos) {
 463                 ret = wbt_init(q);
 464                 if (ret)
 465                         return ret;
 466         }
 467
 468         if (val == -1)
 469                 val = wbt_default_latency_nsec(q);
 470         else if (val >= 0)
 471                 val *= 1000ULL;
 472
 473         if (wbt_get_min_lat(q) == val)
 474                 return count;
 475
 476         /*
 477          * Ensure that the queue is idled, in case the latency update
 478          * ends up either enabling or disabling wbt completely. We can't
 479          * have IO inflight if that happens.
 480          */
 481         blk_mq_freeze_queue(q);
 482         blk_mq_quiesce_queue(q);
 483
 484         wbt_set_min_lat(q, val);
 485
 486         blk_mq_unquiesce_queue(q);
 487         blk_mq_unfreeze_queue(q);
 488
 489         return count;
 490 }
 491
 492 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 493 {
 494         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 495                 return sprintf(page, "write back\n");
 496
 497         return sprintf(page, "write through\n");
 498 }
 499
 500 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 501                               size_t count)
 502 {
 503         int set = -1;
 504
 505         if (!strncmp(page, "write back", 10))
 506                 set = 1;
 507         else if (!strncmp(page, "write through", 13) ||
 508                  !strncmp(page, "none", 4))
 509                 set = 0;
 510
 511         if (set == -1)
 512                 return -EINVAL;
 513
 514         if (set)
 515                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 516         else
 517                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 518
 519         return count;
 520 }
 521
 522 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 523 {
 524         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 525 }
 526
 527 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 528 {
 529         return queue_var_show(blk_queue_dax(q), page);
 530 }
 531
 532 static struct queue_sysfs_entry queue_requests_entry = {
 533         .attr = {.name = "nr_requests", .mode = 0644 },
 534         .show = queue_requests_show,
 535         .store = queue_requests_store,
 536 };
 537
 538 static struct queue_sysfs_entry queue_ra_entry = {
 539         .attr = {.name = "read_ahead_kb", .mode = 0644 },
 540         .show = queue_ra_show,
 541         .store = queue_ra_store,
 542 };
 543
 544 static struct queue_sysfs_entry queue_max_sectors_entry = {
 545         .attr = {.name = "max_sectors_kb", .mode = 0644 },
 546         .show = queue_max_sectors_show,
 547         .store = queue_max_sectors_store,
 548 };
 549
 550 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 551         .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
 552         .show = queue_max_hw_sectors_show,
 553 };
 554
 555 static struct queue_sysfs_entry queue_max_segments_entry = {
 556         .attr = {.name = "max_segments", .mode = 0444 },
 557         .show = queue_max_segments_show,
 558 };
 559
 560 static struct queue_sysfs_entry queue_max_discard_segments_entry = {
 561         .attr = {.name = "max_discard_segments", .mode = 0444 },
 562         .show = queue_max_discard_segments_show,
 563 };
 564
 565 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 566         .attr = {.name = "max_integrity_segments", .mode = 0444 },
 567         .show = queue_max_integrity_segments_show,
 568 };
 569
 570 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 571         .attr = {.name = "max_segment_size", .mode = 0444 },
 572         .show = queue_max_segment_size_show,
 573 };
 574
 575 static struct queue_sysfs_entry queue_iosched_entry = {
 576         .attr = {.name = "scheduler", .mode = 0644 },
 577         .show = elv_iosched_show,
 578         .store = elv_iosched_store,
 579 };
 580
 581 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 582         .attr = {.name = "hw_sector_size", .mode = 0444 },
 583         .show = queue_logical_block_size_show,
 584 };
 585
 586 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 587         .attr = {.name = "logical_block_size", .mode = 0444 },
 588         .show = queue_logical_block_size_show,
 589 };
 590
 591 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 592         .attr = {.name = "physical_block_size", .mode = 0444 },
 593         .show = queue_physical_block_size_show,
 594 };
 595
 596 static struct queue_sysfs_entry queue_chunk_sectors_entry = {
 597         .attr = {.name = "chunk_sectors", .mode = 0444 },
 598         .show = queue_chunk_sectors_show,
 599 };
 600
 601 static struct queue_sysfs_entry queue_io_min_entry = {
 602         .attr = {.name = "minimum_io_size", .mode = 0444 },
 603         .show = queue_io_min_show,
 604 };
 605
 606 static struct queue_sysfs_entry queue_io_opt_entry = {
 607         .attr = {.name = "optimal_io_size", .mode = 0444 },
 608         .show = queue_io_opt_show,
 609 };
 610
 611 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 612         .attr = {.name = "discard_granularity", .mode = 0444 },
 613         .show = queue_discard_granularity_show,
 614 };
 615
 616 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
 617         .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
 618         .show = queue_discard_max_hw_show,
 619 };
 620
 621 static struct queue_sysfs_entry queue_discard_max_entry = {
 622         .attr = {.name = "discard_max_bytes", .mode = 0644 },
 623         .show = queue_discard_max_show,
 624         .store = queue_discard_max_store,
 625 };
 626
 627 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 628         .attr = {.name = "discard_zeroes_data", .mode = 0444 },
 629         .show = queue_discard_zeroes_data_show,
 630 };
 631
 632 static struct queue_sysfs_entry queue_write_same_max_entry = {
 633         .attr = {.name = "write_same_max_bytes", .mode = 0444 },
 634         .show = queue_write_same_max_show,
 635 };
 636
 637 static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 638         .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
 639         .show = queue_write_zeroes_max_show,
 640 };
 641
 642 static struct queue_sysfs_entry queue_nonrot_entry = {
 643         .attr = {.name = "rotational", .mode = 0644 },
 644         .show = queue_show_nonrot,
 645         .store = queue_store_nonrot,
 646 };
 647
 648 static struct queue_sysfs_entry queue_zoned_entry = {
 649         .attr = {.name = "zoned", .mode = 0444 },
 650         .show = queue_zoned_show,
 651 };
 652
 653 static struct queue_sysfs_entry queue_nr_zones_entry = {
 654         .attr = {.name = "nr_zones", .mode = 0444 },
 655         .show = queue_nr_zones_show,
 656 };
 657
 658 static struct queue_sysfs_entry queue_nomerges_entry = {
 659         .attr = {.name = "nomerges", .mode = 0644 },
 660         .show = queue_nomerges_show,
 661         .store = queue_nomerges_store,
 662 };
 663
 664 static struct queue_sysfs_entry queue_rq_affinity_entry = {
 665         .attr = {.name = "rq_affinity", .mode = 0644 },
 666         .show = queue_rq_affinity_show,
 667         .store = queue_rq_affinity_store,
 668 };
 669
 670 static struct queue_sysfs_entry queue_iostats_entry = {
 671         .attr = {.name = "iostats", .mode = 0644 },
 672         .show = queue_show_iostats,
 673         .store = queue_store_iostats,
 674 };
 675
 676 static struct queue_sysfs_entry queue_random_entry = {
 677         .attr = {.name = "add_random", .mode = 0644 },
 678         .show = queue_show_random,
 679         .store = queue_store_random,
 680 };
 681
 682 static struct queue_sysfs_entry queue_poll_entry = {
 683         .attr = {.name = "io_poll", .mode = 0644 },
 684         .show = queue_poll_show,
 685         .store = queue_poll_store,
 686 };
 687
 688 static struct queue_sysfs_entry queue_poll_delay_entry = {
 689         .attr = {.name = "io_poll_delay", .mode = 0644 },
 690         .show = queue_poll_delay_show,
 691         .store = queue_poll_delay_store,
 692 };
 693
 694 static struct queue_sysfs_entry queue_wc_entry = {
 695         .attr = {.name = "write_cache", .mode = 0644 },
 696         .show = queue_wc_show,
 697         .store = queue_wc_store,
 698 };
 699
 700 static struct queue_sysfs_entry queue_fua_entry = {
 701         .attr = {.name = "fua", .mode = 0444 },
 702         .show = queue_fua_show,
 703 };
 704
 705 static struct queue_sysfs_entry queue_dax_entry = {
 706         .attr = {.name = "dax", .mode = 0444 },
 707         .show = queue_dax_show,
 708 };
 709
 710 static struct queue_sysfs_entry queue_io_timeout_entry = {
 711         .attr = {.name = "io_timeout", .mode = 0644 },
 712         .show = queue_io_timeout_show,
 713         .store = queue_io_timeout_store,
 714 };
 715
 716 static struct queue_sysfs_entry queue_wb_lat_entry = {
 717         .attr = {.name = "wbt_lat_usec", .mode = 0644 },
 718         .show = queue_wb_lat_show,
 719         .store = queue_wb_lat_store,
 720 };
 721
 722 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 723 static struct queue_sysfs_entry throtl_sample_time_entry = {
 724         .attr = {.name = "throttle_sample_time", .mode = 0644 },
 725         .show = blk_throtl_sample_time_show,
 726         .store = blk_throtl_sample_time_store,
 727 };
 728 #endif
 729
 730 static struct attribute *queue_attrs[] = {
 731         &queue_requests_entry.attr,
 732         &queue_ra_entry.attr,
 733         &queue_max_hw_sectors_entry.attr,
 734         &queue_max_sectors_entry.attr,
 735         &queue_max_segments_entry.attr,
 736         &queue_max_discard_segments_entry.attr,
 737         &queue_max_integrity_segments_entry.attr,
 738         &queue_max_segment_size_entry.attr,
 739         &queue_iosched_entry.attr,
 740         &queue_hw_sector_size_entry.attr,
 741         &queue_logical_block_size_entry.attr,
 742         &queue_physical_block_size_entry.attr,
 743         &queue_chunk_sectors_entry.attr,
 744         &queue_io_min_entry.attr,
 745         &queue_io_opt_entry.attr,
 746         &queue_discard_granularity_entry.attr,
 747         &queue_discard_max_entry.attr,
 748         &queue_discard_max_hw_entry.attr,
 749         &queue_discard_zeroes_data_entry.attr,
 750         &queue_write_same_max_entry.attr,
 751         &queue_write_zeroes_max_entry.attr,
 752         &queue_nonrot_entry.attr,
 753         &queue_zoned_entry.attr,
 754         &queue_nr_zones_entry.attr,
 755         &queue_nomerges_entry.attr,
 756         &queue_rq_affinity_entry.attr,
 757         &queue_iostats_entry.attr,
 758         &queue_random_entry.attr,
 759         &queue_poll_entry.attr,
 760         &queue_wc_entry.attr,
 761         &queue_fua_entry.attr,
 762         &queue_dax_entry.attr,
 763         &queue_wb_lat_entry.attr,
 764         &queue_poll_delay_entry.attr,
 765         &queue_io_timeout_entry.attr,
 766 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 767         &throtl_sample_time_entry.attr,
 768 #endif
 769         NULL,
 770 };
 771
 772 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
 773                                 int n)
 774 {
 775         struct request_queue *q =
 776                 container_of(kobj, struct request_queue, kobj);
 777
 778         if (attr == &queue_io_timeout_entry.attr &&
 779                 (!q->mq_ops || !q->mq_ops->timeout))
 780                         return 0;
 781
 782         return attr->mode;
 783 }
 784
 785 static struct attribute_group queue_attr_group = {
 786         .attrs = queue_attrs,
 787         .is_visible = queue_attr_visible,
 788 };
 789
 790
 791 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 792
 793 static ssize_t
 794 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 795 {
 796         struct queue_sysfs_entry *entry = to_queue(attr);
 797         struct request_queue *q =
 798                 container_of(kobj, struct request_queue, kobj);
 799         ssize_t res;
 800
 801         if (!entry->show)
 802                 return -EIO;
 803         mutex_lock(&q->sysfs_lock);
 804         res = entry->show(q, page);
 805         mutex_unlock(&q->sysfs_lock);
 806         return res;
 807 }
 808
 809 static ssize_t
 810 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 811                     const char *page, size_t length)
 812 {
 813         struct queue_sysfs_entry *entry = to_queue(attr);
 814         struct request_queue *q;
 815         ssize_t res;
 816
 817         if (!entry->store)
 818                 return -EIO;
 819
 820         q = container_of(kobj, struct request_queue, kobj);
 821         mutex_lock(&q->sysfs_lock);
 822         res = entry->store(q, page, length);
 823         mutex_unlock(&q->sysfs_lock);
 824         return res;
 825 }
 826
 827 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 828 {
 829         struct request_queue *q = container_of(rcu_head, struct request_queue,
 830                                                rcu_head);
 831         kmem_cache_free(blk_requestq_cachep, q);
 832 }
 833
 834 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 835 static void blk_exit_queue(struct request_queue *q)
 836 {
 837         /*
 838          * Since the I/O scheduler exit code may access cgroup information,
 839          * perform I/O scheduler exit before disassociating from the block
 840          * cgroup controller.
 841          */
 842         if (q->elevator) {
 843                 ioc_clear_queue(q);
 844                 __elevator_exit(q, q->elevator);
 845                 q->elevator = NULL;
 846         }
 847
 848         /*
 849          * Remove all references to @q from the block cgroup controller before
 850          * restoring @q->queue_lock to avoid that restoring this pointer causes
 851          * e.g. blkcg_print_blkgs() to crash.
 852          */
 853         blkcg_exit_queue(q);
 854
 855         /*
 856          * Since the cgroup code may dereference the @q->backing_dev_info
 857          * pointer, only decrease its reference count after having removed the
 858          * association with the block cgroup controller.
 859          */
 860         bdi_put(q->backing_dev_info);
 861 }
 862
 863
 864 /**
 865  * __blk_release_queue - release a request queue
 866  * @work: pointer to the release_work member of the request queue to be released
 867  *
 868  * Description:
 869  *     This function is called when a block device is being unregistered. The
 870  *     process of releasing a request queue starts with blk_cleanup_queue, which
 871  *     set the appropriate flags and then calls blk_put_queue, that decrements
 872  *     the reference counter of the request queue. Once the reference counter
 873  *     of the request queue reaches zero, blk_release_queue is called to release
 874  *     all allocated resources of the request queue.
 875  */
 876 static void __blk_release_queue(struct work_struct *work)
 877 {
 878         struct request_queue *q = container_of(work, typeof(*q), release_work);
 879
 880         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 881                 blk_stat_remove_callback(q, q->poll_cb);
 882         blk_stat_free_callback(q->poll_cb);
 883
 884         blk_free_queue_stats(q->stats);
 885
 886         if (queue_is_mq(q))
 887                 cancel_delayed_work_sync(&q->requeue_work);
 888
 889         blk_exit_queue(q);
 890
 891         blk_queue_free_zone_bitmaps(q);
 892
 893         if (queue_is_mq(q))
 894                 blk_mq_release(q);
 895
 896         blk_trace_shutdown(q);
 897
 898         if (queue_is_mq(q))
 899                 blk_mq_debugfs_unregister(q);
 900
 901         bioset_exit(&q->bio_split);
 902
 903         ida_simple_remove(&blk_queue_ida, q->id);
 904         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 905 }
 906
 907 static void blk_release_queue(struct kobject *kobj)
 908 {
 909         struct request_queue *q =
 910                 container_of(kobj, struct request_queue, kobj);
 911
 912         INIT_WORK(&q->release_work, __blk_release_queue);
 913         schedule_work(&q->release_work);
 914 }
 915
 916 static const struct sysfs_ops queue_sysfs_ops = {
 917         .show   = queue_attr_show,
 918         .store  = queue_attr_store,
 919 };
 920
 921 struct kobj_type blk_queue_ktype = {
 922         .sysfs_ops      = &queue_sysfs_ops,
 923         .release        = blk_release_queue,
 924 };
 925
 926 /**
 927  * blk_register_queue - register a block layer queue with sysfs
 928  * @disk: Disk of which the request queue should be registered with sysfs.
 929  */
 930 int blk_register_queue(struct gendisk *disk)
 931 {
 932         int ret;
 933         struct device *dev = disk_to_dev(disk);
 934         struct request_queue *q = disk->queue;
 935         bool has_elevator = false;
 936
 937         if (WARN_ON(!q))
 938                 return -ENXIO;
 939
 940         WARN_ONCE(blk_queue_registered(q),
 941                   "%s is registering an already registered queue\n",
 942                   kobject_name(&dev->kobj));
 943
 944         /*
 945          * SCSI probing may synchronously create and destroy a lot of
 946          * request_queues for non-existent devices.  Shutting down a fully
 947          * functional queue takes measureable wallclock time as RCU grace
 948          * periods are involved.  To avoid excessive latency in these
 949          * cases, a request_queue starts out in a degraded mode which is
 950          * faster to shut down and is made fully functional here as
 951          * request_queues for non-existent devices never get registered.
 952          */
 953         if (!blk_queue_init_done(q)) {
 954                 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
 955                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 956         }
 957
 958         ret = blk_trace_init_sysfs(dev);
 959         if (ret)
 960                 return ret;
 961
 962         mutex_lock(&q->sysfs_dir_lock);
 963
 964         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 965         if (ret < 0) {
 966                 blk_trace_remove_sysfs(dev);
 967                 goto unlock;
 968         }
 969
 970         ret = sysfs_create_group(&q->kobj, &queue_attr_group);
 971         if (ret) {
 972                 blk_trace_remove_sysfs(dev);
 973                 kobject_del(&q->kobj);
 974                 kobject_put(&dev->kobj);
 975                 goto unlock;
 976         }
 977
 978         if (queue_is_mq(q)) {
 979                 __blk_mq_register_dev(dev, q);
 980                 blk_mq_debugfs_register(q);
 981         }
 982
 983         mutex_lock(&q->sysfs_lock);
 984         if (q->elevator) {
 985                 ret = elv_register_queue(q, false);
 986                 if (ret) {
 987                         mutex_unlock(&q->sysfs_lock);
 988                         mutex_unlock(&q->sysfs_dir_lock);
 989                         kobject_del(&q->kobj);
 990                         blk_trace_remove_sysfs(dev);
 991                         kobject_put(&dev->kobj);
 992                         return ret;
 993                 }
 994                 has_elevator = true;
 995         }
 996
 997         blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
 998         wbt_enable_default(q);
 999         blk_throtl_register_queue(q);
1000
1001         /* Now everything is ready and send out KOBJ_ADD uevent */
1002         kobject_uevent(&q->kobj, KOBJ_ADD);
1003         if (has_elevator)
1004                 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
1005         mutex_unlock(&q->sysfs_lock);
1006
1007         ret = 0;
1008 unlock:
1009         mutex_unlock(&q->sysfs_dir_lock);
1010         return ret;
1011 }
1012 EXPORT_SYMBOL_GPL(blk_register_queue);
1013
1014 /**
1015  * blk_unregister_queue - counterpart of blk_register_queue()
1016  * @disk: Disk of which the request queue should be unregistered from sysfs.
1017  *
1018  * Note: the caller is responsible for guaranteeing that this function is called
1019  * after blk_register_queue() has finished.
1020  */
1021 void blk_unregister_queue(struct gendisk *disk)
1022 {
1023         struct request_queue *q = disk->queue;
1024
1025         if (WARN_ON(!q))
1026                 return;
1027
1028         /* Return early if disk->queue was never registered. */
1029         if (!blk_queue_registered(q))
1030                 return;
1031
1032         /*
1033          * Since sysfs_remove_dir() prevents adding new directory entries
1034          * before removal of existing entries starts, protect against
1035          * concurrent elv_iosched_store() calls.
1036          */
1037         mutex_lock(&q->sysfs_lock);
1038         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1039         mutex_unlock(&q->sysfs_lock);
1040
1041         mutex_lock(&q->sysfs_dir_lock);
1042         /*
1043          * Remove the sysfs attributes before unregistering the queue data
1044          * structures that can be modified through sysfs.
1045          */
1046         if (queue_is_mq(q))
1047                 blk_mq_unregister_dev(disk_to_dev(disk), q);
1048
1049         kobject_uevent(&q->kobj, KOBJ_REMOVE);
1050         kobject_del(&q->kobj);
1051         blk_trace_remove_sysfs(disk_to_dev(disk));
1052
1053         mutex_lock(&q->sysfs_lock);
1054         if (q->elevator)
1055                 elv_unregister_queue(q);
1056         mutex_unlock(&q->sysfs_lock);
1057         mutex_unlock(&q->sysfs_dir_lock);
1058
1059         kobject_put(&disk_to_dev(disk)->kobj);
1060 }