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->request_fn && !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         if (q->request_fn)
  82                 err = blk_update_nr_requests(q, nr);
  83         else
  84                 err = blk_mq_update_nr_requests(q, nr);
  85
  86         if (err)
  87                 return err;
  88
  89         return ret;
  90 }
  91
  92 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  93 {
  94         unsigned long ra_kb = q->backing_dev_info->ra_pages <<
  95                                         (PAGE_SHIFT - 10);
  96
  97         return queue_var_show(ra_kb, (page));
  98 }
  99
 100 static ssize_t
 101 queue_ra_store(struct request_queue *q, const char *page, size_t count)
 102 {
 103         unsigned long ra_kb;
 104         ssize_t ret = queue_var_store(&ra_kb, page, count);
 105
 106         if (ret < 0)
 107                 return ret;
 108
 109         q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 110
 111         return ret;
 112 }
 113
 114 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 115 {
 116         int max_sectors_kb = queue_max_sectors(q) >> 1;
 117
 118         return queue_var_show(max_sectors_kb, (page));
 119 }
 120
 121 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 122 {
 123         return queue_var_show(queue_max_segments(q), (page));
 124 }
 125
 126 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 127                 char *page)
 128 {
 129         return queue_var_show(queue_max_discard_segments(q), (page));
 130 }
 131
 132 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 133 {
 134         return queue_var_show(q->limits.max_integrity_segments, (page));
 135 }
 136
 137 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 138 {
 139         if (blk_queue_cluster(q))
 140                 return queue_var_show(queue_max_segment_size(q), (page));
 141
 142         return queue_var_show(PAGE_SIZE, (page));
 143 }
 144
 145 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 146 {
 147         return queue_var_show(queue_logical_block_size(q), page);
 148 }
 149
 150 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 151 {
 152         return queue_var_show(queue_physical_block_size(q), page);
 153 }
 154
 155 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 156 {
 157         return queue_var_show(q->limits.chunk_sectors, page);
 158 }
 159
 160 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 161 {
 162         return queue_var_show(queue_io_min(q), page);
 163 }
 164
 165 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 166 {
 167         return queue_var_show(queue_io_opt(q), page);
 168 }
 169
 170 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 171 {
 172         return queue_var_show(q->limits.discard_granularity, page);
 173 }
 174
 175 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 176 {
 177
 178         return sprintf(page, "%llu\n",
 179                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 180 }
 181
 182 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 183 {
 184         return sprintf(page, "%llu\n",
 185                        (unsigned long long)q->limits.max_discard_sectors << 9);
 186 }
 187
 188 static ssize_t queue_discard_max_store(struct request_queue *q,
 189                                        const char *page, size_t count)
 190 {
 191         unsigned long max_discard;
 192         ssize_t ret = queue_var_store(&max_discard, page, count);
 193
 194         if (ret < 0)
 195                 return ret;
 196
 197         if (max_discard & (q->limits.discard_granularity - 1))
 198                 return -EINVAL;
 199
 200         max_discard >>= 9;
 201         if (max_discard > UINT_MAX)
 202                 return -EINVAL;
 203
 204         if (max_discard > q->limits.max_hw_discard_sectors)
 205                 max_discard = q->limits.max_hw_discard_sectors;
 206
 207         q->limits.max_discard_sectors = max_discard;
 208         return ret;
 209 }
 210
 211 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 212 {
 213         return queue_var_show(0, page);
 214 }
 215
 216 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 217 {
 218         return sprintf(page, "%llu\n",
 219                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 220 }
 221
 222 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 223 {
 224         return sprintf(page, "%llu\n",
 225                 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 226 }
 227
 228 static ssize_t
 229 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 230 {
 231         unsigned long max_sectors_kb,
 232                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 233                         page_kb = 1 << (PAGE_SHIFT - 10);
 234         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 235
 236         if (ret < 0)
 237                 return ret;
 238
 239         max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
 240                                          q->limits.max_dev_sectors >> 1);
 241
 242         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 243                 return -EINVAL;
 244
 245         spin_lock_irq(q->queue_lock);
 246         q->limits.max_sectors = max_sectors_kb << 1;
 247         q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 248         spin_unlock_irq(q->queue_lock);
 249
 250         return ret;
 251 }
 252
 253 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 254 {
 255         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 256
 257         return queue_var_show(max_hw_sectors_kb, (page));
 258 }
 259
 260 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 261 static ssize_t                                                          \
 262 queue_show_##name(struct request_queue *q, char *page)                  \
 263 {                                                                       \
 264         int bit;                                                        \
 265         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 266         return queue_var_show(neg ? !bit : bit, page);                  \
 267 }                                                                       \
 268 static ssize_t                                                          \
 269 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
 270 {                                                                       \
 271         unsigned long val;                                              \
 272         ssize_t ret;                                                    \
 273         ret = queue_var_store(&val, page, count);                       \
 274         if (ret < 0)                                                    \
 275                  return ret;                                            \
 276         if (neg)                                                        \
 277                 val = !val;                                             \
 278                                                                         \
 279         if (val)                                                        \
 280                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 281         else                                                            \
 282                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 283         return ret;                                                     \
 284 }
 285
 286 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 287 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 288 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 289 #undef QUEUE_SYSFS_BIT_FNS
 290
 291 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 292 {
 293         switch (blk_queue_zoned_model(q)) {
 294         case BLK_ZONED_HA:
 295                 return sprintf(page, "host-aware\n");
 296         case BLK_ZONED_HM:
 297                 return sprintf(page, "host-managed\n");
 298         default:
 299                 return sprintf(page, "none\n");
 300         }
 301 }
 302
 303 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 304 {
 305         return queue_var_show((blk_queue_nomerges(q) << 1) |
 306                                blk_queue_noxmerges(q), page);
 307 }
 308
 309 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 310                                     size_t count)
 311 {
 312         unsigned long nm;
 313         ssize_t ret = queue_var_store(&nm, page, count);
 314
 315         if (ret < 0)
 316                 return ret;
 317
 318         spin_lock_irq(q->queue_lock);
 319         queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 320         queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 321         if (nm == 2)
 322                 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 323         else if (nm)
 324                 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 325         spin_unlock_irq(q->queue_lock);
 326
 327         return ret;
 328 }
 329
 330 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 331 {
 332         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 333         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 334
 335         return queue_var_show(set << force, page);
 336 }
 337
 338 static ssize_t
 339 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 340 {
 341         ssize_t ret = -EINVAL;
 342 #ifdef CONFIG_SMP
 343         unsigned long val;
 344
 345         ret = queue_var_store(&val, page, count);
 346         if (ret < 0)
 347                 return ret;
 348
 349         spin_lock_irq(q->queue_lock);
 350         if (val == 2) {
 351                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 352                 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 353         } else if (val == 1) {
 354                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 355                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 356         } else if (val == 0) {
 357                 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 358                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 359         }
 360         spin_unlock_irq(q->queue_lock);
 361 #endif
 362         return ret;
 363 }
 364
 365 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 366 {
 367         int val;
 368
 369         if (q->poll_nsec == -1)
 370                 val = -1;
 371         else
 372                 val = q->poll_nsec / 1000;
 373
 374         return sprintf(page, "%d\n", val);
 375 }
 376
 377 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 378                                 size_t count)
 379 {
 380         int err, val;
 381
 382         if (!q->mq_ops || !q->mq_ops->poll)
 383                 return -EINVAL;
 384
 385         err = kstrtoint(page, 10, &val);
 386         if (err < 0)
 387                 return err;
 388
 389         if (val == -1)
 390                 q->poll_nsec = -1;
 391         else
 392                 q->poll_nsec = val * 1000;
 393
 394         return count;
 395 }
 396
 397 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 398 {
 399         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 400 }
 401
 402 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 403                                 size_t count)
 404 {
 405         unsigned long poll_on;
 406         ssize_t ret;
 407
 408         if (!q->mq_ops || !q->mq_ops->poll)
 409                 return -EINVAL;
 410
 411         ret = queue_var_store(&poll_on, page, count);
 412         if (ret < 0)
 413                 return ret;
 414
 415         if (poll_on)
 416                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 417         else
 418                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 419
 420         return ret;
 421 }
 422
 423 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 424 {
 425         if (!wbt_rq_qos(q))
 426                 return -EINVAL;
 427
 428         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 429 }
 430
 431 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 432                                   size_t count)
 433 {
 434         struct rq_qos *rqos;
 435         ssize_t ret;
 436         s64 val;
 437
 438         ret = queue_var_store64(&val, page);
 439         if (ret < 0)
 440                 return ret;
 441         if (val < -1)
 442                 return -EINVAL;
 443
 444         rqos = wbt_rq_qos(q);
 445         if (!rqos) {
 446                 ret = wbt_init(q);
 447                 if (ret)
 448                         return ret;
 449         }
 450
 451         if (val == -1)
 452                 val = wbt_default_latency_nsec(q);
 453         else if (val >= 0)
 454                 val *= 1000ULL;
 455
 456         /*
 457          * Ensure that the queue is idled, in case the latency update
 458          * ends up either enabling or disabling wbt completely. We can't
 459          * have IO inflight if that happens.
 460          */
 461         if (q->mq_ops) {
 462                 blk_mq_freeze_queue(q);
 463                 blk_mq_quiesce_queue(q);
 464         } else
 465                 blk_queue_bypass_start(q);
 466
 467         wbt_set_min_lat(q, val);
 468         wbt_update_limits(q);
 469
 470         if (q->mq_ops) {
 471                 blk_mq_unquiesce_queue(q);
 472                 blk_mq_unfreeze_queue(q);
 473         } else
 474                 blk_queue_bypass_end(q);
 475
 476         return count;
 477 }
 478
 479 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 480 {
 481         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 482                 return sprintf(page, "write back\n");
 483
 484         return sprintf(page, "write through\n");
 485 }
 486
 487 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 488                               size_t count)
 489 {
 490         int set = -1;
 491
 492         if (!strncmp(page, "write back", 10))
 493                 set = 1;
 494         else if (!strncmp(page, "write through", 13) ||
 495                  !strncmp(page, "none", 4))
 496                 set = 0;
 497
 498         if (set == -1)
 499                 return -EINVAL;
 500
 501         if (set)
 502                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 503         else
 504                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 505
 506         return count;
 507 }
 508
 509 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 510 {
 511         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 512 }
 513
 514 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 515 {
 516         return queue_var_show(blk_queue_dax(q), page);
 517 }
 518
 519 static struct queue_sysfs_entry queue_requests_entry = {
 520         .attr = {.name = "nr_requests", .mode = 0644 },
 521         .show = queue_requests_show,
 522         .store = queue_requests_store,
 523 };
 524
 525 static struct queue_sysfs_entry queue_ra_entry = {
 526         .attr = {.name = "read_ahead_kb", .mode = 0644 },
 527         .show = queue_ra_show,
 528         .store = queue_ra_store,
 529 };
 530
 531 static struct queue_sysfs_entry queue_max_sectors_entry = {
 532         .attr = {.name = "max_sectors_kb", .mode = 0644 },
 533         .show = queue_max_sectors_show,
 534         .store = queue_max_sectors_store,
 535 };
 536
 537 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 538         .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
 539         .show = queue_max_hw_sectors_show,
 540 };
 541
 542 static struct queue_sysfs_entry queue_max_segments_entry = {
 543         .attr = {.name = "max_segments", .mode = 0444 },
 544         .show = queue_max_segments_show,
 545 };
 546
 547 static struct queue_sysfs_entry queue_max_discard_segments_entry = {
 548         .attr = {.name = "max_discard_segments", .mode = 0444 },
 549         .show = queue_max_discard_segments_show,
 550 };
 551
 552 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 553         .attr = {.name = "max_integrity_segments", .mode = 0444 },
 554         .show = queue_max_integrity_segments_show,
 555 };
 556
 557 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 558         .attr = {.name = "max_segment_size", .mode = 0444 },
 559         .show = queue_max_segment_size_show,
 560 };
 561
 562 static struct queue_sysfs_entry queue_iosched_entry = {
 563         .attr = {.name = "scheduler", .mode = 0644 },
 564         .show = elv_iosched_show,
 565         .store = elv_iosched_store,
 566 };
 567
 568 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 569         .attr = {.name = "hw_sector_size", .mode = 0444 },
 570         .show = queue_logical_block_size_show,
 571 };
 572
 573 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 574         .attr = {.name = "logical_block_size", .mode = 0444 },
 575         .show = queue_logical_block_size_show,
 576 };
 577
 578 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 579         .attr = {.name = "physical_block_size", .mode = 0444 },
 580         .show = queue_physical_block_size_show,
 581 };
 582
 583 static struct queue_sysfs_entry queue_chunk_sectors_entry = {
 584         .attr = {.name = "chunk_sectors", .mode = 0444 },
 585         .show = queue_chunk_sectors_show,
 586 };
 587
 588 static struct queue_sysfs_entry queue_io_min_entry = {
 589         .attr = {.name = "minimum_io_size", .mode = 0444 },
 590         .show = queue_io_min_show,
 591 };
 592
 593 static struct queue_sysfs_entry queue_io_opt_entry = {
 594         .attr = {.name = "optimal_io_size", .mode = 0444 },
 595         .show = queue_io_opt_show,
 596 };
 597
 598 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 599         .attr = {.name = "discard_granularity", .mode = 0444 },
 600         .show = queue_discard_granularity_show,
 601 };
 602
 603 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
 604         .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
 605         .show = queue_discard_max_hw_show,
 606 };
 607
 608 static struct queue_sysfs_entry queue_discard_max_entry = {
 609         .attr = {.name = "discard_max_bytes", .mode = 0644 },
 610         .show = queue_discard_max_show,
 611         .store = queue_discard_max_store,
 612 };
 613
 614 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 615         .attr = {.name = "discard_zeroes_data", .mode = 0444 },
 616         .show = queue_discard_zeroes_data_show,
 617 };
 618
 619 static struct queue_sysfs_entry queue_write_same_max_entry = {
 620         .attr = {.name = "write_same_max_bytes", .mode = 0444 },
 621         .show = queue_write_same_max_show,
 622 };
 623
 624 static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 625         .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
 626         .show = queue_write_zeroes_max_show,
 627 };
 628
 629 static struct queue_sysfs_entry queue_nonrot_entry = {
 630         .attr = {.name = "rotational", .mode = 0644 },
 631         .show = queue_show_nonrot,
 632         .store = queue_store_nonrot,
 633 };
 634
 635 static struct queue_sysfs_entry queue_zoned_entry = {
 636         .attr = {.name = "zoned", .mode = 0444 },
 637         .show = queue_zoned_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_nomerges_entry.attr,
 731         &queue_rq_affinity_entry.attr,
 732         &queue_iostats_entry.attr,
 733         &queue_random_entry.attr,
 734         &queue_poll_entry.attr,
 735         &queue_wc_entry.attr,
 736         &queue_fua_entry.attr,
 737         &queue_dax_entry.attr,
 738         &queue_wb_lat_entry.attr,
 739         &queue_poll_delay_entry.attr,
 740 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 741         &throtl_sample_time_entry.attr,
 742 #endif
 743         NULL,
 744 };
 745
 746 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 747
 748 static ssize_t
 749 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 750 {
 751         struct queue_sysfs_entry *entry = to_queue(attr);
 752         struct request_queue *q =
 753                 container_of(kobj, struct request_queue, kobj);
 754         ssize_t res;
 755
 756         if (!entry->show)
 757                 return -EIO;
 758         mutex_lock(&q->sysfs_lock);
 759         if (blk_queue_dying(q)) {
 760                 mutex_unlock(&q->sysfs_lock);
 761                 return -ENOENT;
 762         }
 763         res = entry->show(q, page);
 764         mutex_unlock(&q->sysfs_lock);
 765         return res;
 766 }
 767
 768 static ssize_t
 769 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 770                     const char *page, size_t length)
 771 {
 772         struct queue_sysfs_entry *entry = to_queue(attr);
 773         struct request_queue *q;
 774         ssize_t res;
 775
 776         if (!entry->store)
 777                 return -EIO;
 778
 779         q = container_of(kobj, struct request_queue, kobj);
 780         mutex_lock(&q->sysfs_lock);
 781         if (blk_queue_dying(q)) {
 782                 mutex_unlock(&q->sysfs_lock);
 783                 return -ENOENT;
 784         }
 785         res = entry->store(q, page, length);
 786         mutex_unlock(&q->sysfs_lock);
 787         return res;
 788 }
 789
 790 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 791 {
 792         struct request_queue *q = container_of(rcu_head, struct request_queue,
 793                                                rcu_head);
 794         kmem_cache_free(blk_requestq_cachep, q);
 795 }
 796
 797 /**
 798  * __blk_release_queue - release a request queue when it is no longer needed
 799  * @work: pointer to the release_work member of the request queue to be released
 800  *
 801  * Description:
 802  *     blk_release_queue is the counterpart of blk_init_queue(). It should be
 803  *     called when a request queue is being released; typically when a block
 804  *     device is being de-registered. Its primary task it to free the queue
 805  *     itself.
 806  *
 807  * Notes:
 808  *     The low level driver must have finished any outstanding requests first
 809  *     via blk_cleanup_queue().
 810  *
 811  *     Although blk_release_queue() may be called with preemption disabled,
 812  *     __blk_release_queue() may sleep.
 813  */
 814 static void __blk_release_queue(struct work_struct *work)
 815 {
 816         struct request_queue *q = container_of(work, typeof(*q), release_work);
 817
 818         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 819                 blk_stat_remove_callback(q, q->poll_cb);
 820         blk_stat_free_callback(q->poll_cb);
 821
 822         if (!blk_queue_dead(q)) {
 823                 /*
 824                  * Last reference was dropped without having called
 825                  * blk_cleanup_queue().
 826                  */
 827                 WARN_ONCE(blk_queue_init_done(q),
 828                           "request queue %p has been registered but blk_cleanup_queue() has not been called for that queue\n",
 829                           q);
 830                 blk_exit_queue(q);
 831         }
 832
 833         WARN(blk_queue_root_blkg(q),
 834              "request queue %p is being released but it has not yet been removed from the blkcg controller\n",
 835              q);
 836
 837         blk_free_queue_stats(q->stats);
 838
 839         blk_exit_rl(q, &q->root_rl);
 840
 841         if (q->queue_tags)
 842                 __blk_queue_free_tags(q);
 843
 844         if (!q->mq_ops) {
 845                 if (q->exit_rq_fn)
 846                         q->exit_rq_fn(q, q->fq->flush_rq);
 847                 blk_free_flush_queue(q->fq);
 848         } else {
 849                 blk_mq_release(q);
 850         }
 851
 852         blk_trace_shutdown(q);
 853
 854         if (q->mq_ops)
 855                 blk_mq_debugfs_unregister(q);
 856
 857         bioset_exit(&q->bio_split);
 858
 859         ida_simple_remove(&blk_queue_ida, q->id);
 860         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 861 }
 862
 863 static void blk_release_queue(struct kobject *kobj)
 864 {
 865         struct request_queue *q =
 866                 container_of(kobj, struct request_queue, kobj);
 867
 868         INIT_WORK(&q->release_work, __blk_release_queue);
 869         schedule_work(&q->release_work);
 870 }
 871
 872 static const struct sysfs_ops queue_sysfs_ops = {
 873         .show   = queue_attr_show,
 874         .store  = queue_attr_store,
 875 };
 876
 877 struct kobj_type blk_queue_ktype = {
 878         .sysfs_ops      = &queue_sysfs_ops,
 879         .default_attrs  = default_attrs,
 880         .release        = blk_release_queue,
 881 };
 882
 883 /**
 884  * blk_register_queue - register a block layer queue with sysfs
 885  * @disk: Disk of which the request queue should be registered with sysfs.
 886  */
 887 int blk_register_queue(struct gendisk *disk)
 888 {
 889         int ret;
 890         struct device *dev = disk_to_dev(disk);
 891         struct request_queue *q = disk->queue;
 892
 893         if (WARN_ON(!q))
 894                 return -ENXIO;
 895
 896         WARN_ONCE(test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags),
 897                   "%s is registering an already registered queue\n",
 898                   kobject_name(&dev->kobj));
 899         queue_flag_set_unlocked(QUEUE_FLAG_REGISTERED, q);
 900
 901         /*
 902          * SCSI probing may synchronously create and destroy a lot of
 903          * request_queues for non-existent devices.  Shutting down a fully
 904          * functional queue takes measureable wallclock time as RCU grace
 905          * periods are involved.  To avoid excessive latency in these
 906          * cases, a request_queue starts out in a degraded mode which is
 907          * faster to shut down and is made fully functional here as
 908          * request_queues for non-existent devices never get registered.
 909          */
 910         if (!blk_queue_init_done(q)) {
 911                 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
 912                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 913                 blk_queue_bypass_end(q);
 914         }
 915
 916         ret = blk_trace_init_sysfs(dev);
 917         if (ret)
 918                 return ret;
 919
 920         /* Prevent changes through sysfs until registration is completed. */
 921         mutex_lock(&q->sysfs_lock);
 922
 923         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 924         if (ret < 0) {
 925                 blk_trace_remove_sysfs(dev);
 926                 goto unlock;
 927         }
 928
 929         if (q->mq_ops) {
 930                 __blk_mq_register_dev(dev, q);
 931                 blk_mq_debugfs_register(q);
 932         }
 933
 934         kobject_uevent(&q->kobj, KOBJ_ADD);
 935
 936         wbt_enable_default(q);
 937
 938         blk_throtl_register_queue(q);
 939
 940         if (q->request_fn || (q->mq_ops && q->elevator)) {
 941                 ret = elv_register_queue(q);
 942                 if (ret) {
 943                         mutex_unlock(&q->sysfs_lock);
 944                         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 945                         kobject_del(&q->kobj);
 946                         blk_trace_remove_sysfs(dev);
 947                         kobject_put(&dev->kobj);
 948                         return ret;
 949                 }
 950         }
 951         ret = 0;
 952 unlock:
 953         mutex_unlock(&q->sysfs_lock);
 954         return ret;
 955 }
 956 EXPORT_SYMBOL_GPL(blk_register_queue);
 957
 958 /**
 959  * blk_unregister_queue - counterpart of blk_register_queue()
 960  * @disk: Disk of which the request queue should be unregistered from sysfs.
 961  *
 962  * Note: the caller is responsible for guaranteeing that this function is called
 963  * after blk_register_queue() has finished.
 964  */
 965 void blk_unregister_queue(struct gendisk *disk)
 966 {
 967         struct request_queue *q = disk->queue;
 968
 969         if (WARN_ON(!q))
 970                 return;
 971
 972         /* Return early if disk->queue was never registered. */
 973         if (!test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags))
 974                 return;
 975
 976         /*
 977          * Since sysfs_remove_dir() prevents adding new directory entries
 978          * before removal of existing entries starts, protect against
 979          * concurrent elv_iosched_store() calls.
 980          */
 981         mutex_lock(&q->sysfs_lock);
 982
 983         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
 984
 985         /*
 986          * Remove the sysfs attributes before unregistering the queue data
 987          * structures that can be modified through sysfs.
 988          */
 989         if (q->mq_ops)
 990                 blk_mq_unregister_dev(disk_to_dev(disk), q);
 991         mutex_unlock(&q->sysfs_lock);
 992
 993         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 994         kobject_del(&q->kobj);
 995         blk_trace_remove_sysfs(disk_to_dev(disk));
 996
 997         rq_qos_exit(q);
 998
 999         mutex_lock(&q->sysfs_lock);
1000         if (q->request_fn || (q->mq_ops && q->elevator))
1001                 elv_unregister_queue(q);
1002         mutex_unlock(&q->sysfs_lock);
1003
1004         kobject_put(&disk_to_dev(disk)->kobj);
1005 }