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