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 queue_zone_append_max_show(struct request_queue *q, char *page)
 222 {
 223         unsigned long long max_sectors = q->limits.max_zone_append_sectors;
 224
 225         return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
 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_nr_zones_show(struct request_queue *q, char *page)
 304 {
 305         return queue_var_show(blk_queue_nr_zones(q), page);
 306 }
 307
 308 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 309 {
 310         return queue_var_show((blk_queue_nomerges(q) << 1) |
 311                                blk_queue_noxmerges(q), page);
 312 }
 313
 314 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 315                                     size_t count)
 316 {
 317         unsigned long nm;
 318         ssize_t ret = queue_var_store(&nm, page, count);
 319
 320         if (ret < 0)
 321                 return ret;
 322
 323         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 324         blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 325         if (nm == 2)
 326                 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 327         else if (nm)
 328                 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 329
 330         return ret;
 331 }
 332
 333 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 334 {
 335         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 336         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 337
 338         return queue_var_show(set << force, page);
 339 }
 340
 341 static ssize_t
 342 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 343 {
 344         ssize_t ret = -EINVAL;
 345 #ifdef CONFIG_SMP
 346         unsigned long val;
 347
 348         ret = queue_var_store(&val, page, count);
 349         if (ret < 0)
 350                 return ret;
 351
 352         if (val == 2) {
 353                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 354                 blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 355         } else if (val == 1) {
 356                 blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 357                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 358         } else if (val == 0) {
 359                 blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 360                 blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 361         }
 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 == BLK_MQ_POLL_CLASSIC)
 371                 val = BLK_MQ_POLL_CLASSIC;
 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 == BLK_MQ_POLL_CLASSIC)
 391                 q->poll_nsec = BLK_MQ_POLL_CLASSIC;
 392         else if (val >= 0)
 393                 q->poll_nsec = val * 1000;
 394         else
 395                 return -EINVAL;
 396
 397         return count;
 398 }
 399
 400 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 401 {
 402         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 403 }
 404
 405 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 406                                 size_t count)
 407 {
 408         unsigned long poll_on;
 409         ssize_t ret;
 410
 411         if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
 412             !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
 413                 return -EINVAL;
 414
 415         ret = queue_var_store(&poll_on, page, count);
 416         if (ret < 0)
 417                 return ret;
 418
 419         if (poll_on)
 420                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 421         else
 422                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 423
 424         return ret;
 425 }
 426
 427 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
 428 {
 429         return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
 430 }
 431
 432 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
 433                                   size_t count)
 434 {
 435         unsigned int val;
 436         int err;
 437
 438         err = kstrtou32(page, 10, &val);
 439         if (err || val == 0)
 440                 return -EINVAL;
 441
 442         blk_queue_rq_timeout(q, msecs_to_jiffies(val));
 443
 444         return count;
 445 }
 446
 447 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 448 {
 449         if (!wbt_rq_qos(q))
 450                 return -EINVAL;
 451
 452         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 453 }
 454
 455 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 456                                   size_t count)
 457 {
 458         struct rq_qos *rqos;
 459         ssize_t ret;
 460         s64 val;
 461
 462         ret = queue_var_store64(&val, page);
 463         if (ret < 0)
 464                 return ret;
 465         if (val < -1)
 466                 return -EINVAL;
 467
 468         rqos = wbt_rq_qos(q);
 469         if (!rqos) {
 470                 ret = wbt_init(q);
 471                 if (ret)
 472                         return ret;
 473         }
 474
 475         if (val == -1)
 476                 val = wbt_default_latency_nsec(q);
 477         else if (val >= 0)
 478                 val *= 1000ULL;
 479
 480         if (wbt_get_min_lat(q) == val)
 481                 return count;
 482
 483         /*
 484          * Ensure that the queue is idled, in case the latency update
 485          * ends up either enabling or disabling wbt completely. We can't
 486          * have IO inflight if that happens.
 487          */
 488         blk_mq_freeze_queue(q);
 489         blk_mq_quiesce_queue(q);
 490
 491         wbt_set_min_lat(q, val);
 492
 493         blk_mq_unquiesce_queue(q);
 494         blk_mq_unfreeze_queue(q);
 495
 496         return count;
 497 }
 498
 499 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 500 {
 501         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 502                 return sprintf(page, "write back\n");
 503
 504         return sprintf(page, "write through\n");
 505 }
 506
 507 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 508                               size_t count)
 509 {
 510         int set = -1;
 511
 512         if (!strncmp(page, "write back", 10))
 513                 set = 1;
 514         else if (!strncmp(page, "write through", 13) ||
 515                  !strncmp(page, "none", 4))
 516                 set = 0;
 517
 518         if (set == -1)
 519                 return -EINVAL;
 520
 521         if (set)
 522                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 523         else
 524                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 525
 526         return count;
 527 }
 528
 529 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 530 {
 531         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 532 }
 533
 534 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 535 {
 536         return queue_var_show(blk_queue_dax(q), page);
 537 }
 538
 539 static struct queue_sysfs_entry queue_requests_entry = {
 540         .attr = {.name = "nr_requests", .mode = 0644 },
 541         .show = queue_requests_show,
 542         .store = queue_requests_store,
 543 };
 544
 545 static struct queue_sysfs_entry queue_ra_entry = {
 546         .attr = {.name = "read_ahead_kb", .mode = 0644 },
 547         .show = queue_ra_show,
 548         .store = queue_ra_store,
 549 };
 550
 551 static struct queue_sysfs_entry queue_max_sectors_entry = {
 552         .attr = {.name = "max_sectors_kb", .mode = 0644 },
 553         .show = queue_max_sectors_show,
 554         .store = queue_max_sectors_store,
 555 };
 556
 557 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 558         .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
 559         .show = queue_max_hw_sectors_show,
 560 };
 561
 562 static struct queue_sysfs_entry queue_max_segments_entry = {
 563         .attr = {.name = "max_segments", .mode = 0444 },
 564         .show = queue_max_segments_show,
 565 };
 566
 567 static struct queue_sysfs_entry queue_max_discard_segments_entry = {
 568         .attr = {.name = "max_discard_segments", .mode = 0444 },
 569         .show = queue_max_discard_segments_show,
 570 };
 571
 572 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 573         .attr = {.name = "max_integrity_segments", .mode = 0444 },
 574         .show = queue_max_integrity_segments_show,
 575 };
 576
 577 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 578         .attr = {.name = "max_segment_size", .mode = 0444 },
 579         .show = queue_max_segment_size_show,
 580 };
 581
 582 static struct queue_sysfs_entry queue_iosched_entry = {
 583         .attr = {.name = "scheduler", .mode = 0644 },
 584         .show = elv_iosched_show,
 585         .store = elv_iosched_store,
 586 };
 587
 588 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 589         .attr = {.name = "hw_sector_size", .mode = 0444 },
 590         .show = queue_logical_block_size_show,
 591 };
 592
 593 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 594         .attr = {.name = "logical_block_size", .mode = 0444 },
 595         .show = queue_logical_block_size_show,
 596 };
 597
 598 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 599         .attr = {.name = "physical_block_size", .mode = 0444 },
 600         .show = queue_physical_block_size_show,
 601 };
 602
 603 static struct queue_sysfs_entry queue_chunk_sectors_entry = {
 604         .attr = {.name = "chunk_sectors", .mode = 0444 },
 605         .show = queue_chunk_sectors_show,
 606 };
 607
 608 static struct queue_sysfs_entry queue_io_min_entry = {
 609         .attr = {.name = "minimum_io_size", .mode = 0444 },
 610         .show = queue_io_min_show,
 611 };
 612
 613 static struct queue_sysfs_entry queue_io_opt_entry = {
 614         .attr = {.name = "optimal_io_size", .mode = 0444 },
 615         .show = queue_io_opt_show,
 616 };
 617
 618 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 619         .attr = {.name = "discard_granularity", .mode = 0444 },
 620         .show = queue_discard_granularity_show,
 621 };
 622
 623 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
 624         .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
 625         .show = queue_discard_max_hw_show,
 626 };
 627
 628 static struct queue_sysfs_entry queue_discard_max_entry = {
 629         .attr = {.name = "discard_max_bytes", .mode = 0644 },
 630         .show = queue_discard_max_show,
 631         .store = queue_discard_max_store,
 632 };
 633
 634 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 635         .attr = {.name = "discard_zeroes_data", .mode = 0444 },
 636         .show = queue_discard_zeroes_data_show,
 637 };
 638
 639 static struct queue_sysfs_entry queue_write_same_max_entry = {
 640         .attr = {.name = "write_same_max_bytes", .mode = 0444 },
 641         .show = queue_write_same_max_show,
 642 };
 643
 644 static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 645         .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
 646         .show = queue_write_zeroes_max_show,
 647 };
 648
 649 static struct queue_sysfs_entry queue_zone_append_max_entry = {
 650         .attr = {.name = "zone_append_max_bytes", .mode = 0444 },
 651         .show = queue_zone_append_max_show,
 652 };
 653
 654 static struct queue_sysfs_entry queue_nonrot_entry = {
 655         .attr = {.name = "rotational", .mode = 0644 },
 656         .show = queue_show_nonrot,
 657         .store = queue_store_nonrot,
 658 };
 659
 660 static struct queue_sysfs_entry queue_zoned_entry = {
 661         .attr = {.name = "zoned", .mode = 0444 },
 662         .show = queue_zoned_show,
 663 };
 664
 665 static struct queue_sysfs_entry queue_nr_zones_entry = {
 666         .attr = {.name = "nr_zones", .mode = 0444 },
 667         .show = queue_nr_zones_show,
 668 };
 669
 670 static struct queue_sysfs_entry queue_nomerges_entry = {
 671         .attr = {.name = "nomerges", .mode = 0644 },
 672         .show = queue_nomerges_show,
 673         .store = queue_nomerges_store,
 674 };
 675
 676 static struct queue_sysfs_entry queue_rq_affinity_entry = {
 677         .attr = {.name = "rq_affinity", .mode = 0644 },
 678         .show = queue_rq_affinity_show,
 679         .store = queue_rq_affinity_store,
 680 };
 681
 682 static struct queue_sysfs_entry queue_iostats_entry = {
 683         .attr = {.name = "iostats", .mode = 0644 },
 684         .show = queue_show_iostats,
 685         .store = queue_store_iostats,
 686 };
 687
 688 static struct queue_sysfs_entry queue_random_entry = {
 689         .attr = {.name = "add_random", .mode = 0644 },
 690         .show = queue_show_random,
 691         .store = queue_store_random,
 692 };
 693
 694 static struct queue_sysfs_entry queue_poll_entry = {
 695         .attr = {.name = "io_poll", .mode = 0644 },
 696         .show = queue_poll_show,
 697         .store = queue_poll_store,
 698 };
 699
 700 static struct queue_sysfs_entry queue_poll_delay_entry = {
 701         .attr = {.name = "io_poll_delay", .mode = 0644 },
 702         .show = queue_poll_delay_show,
 703         .store = queue_poll_delay_store,
 704 };
 705
 706 static struct queue_sysfs_entry queue_wc_entry = {
 707         .attr = {.name = "write_cache", .mode = 0644 },
 708         .show = queue_wc_show,
 709         .store = queue_wc_store,
 710 };
 711
 712 static struct queue_sysfs_entry queue_fua_entry = {
 713         .attr = {.name = "fua", .mode = 0444 },
 714         .show = queue_fua_show,
 715 };
 716
 717 static struct queue_sysfs_entry queue_dax_entry = {
 718         .attr = {.name = "dax", .mode = 0444 },
 719         .show = queue_dax_show,
 720 };
 721
 722 static struct queue_sysfs_entry queue_io_timeout_entry = {
 723         .attr = {.name = "io_timeout", .mode = 0644 },
 724         .show = queue_io_timeout_show,
 725         .store = queue_io_timeout_store,
 726 };
 727
 728 static struct queue_sysfs_entry queue_wb_lat_entry = {
 729         .attr = {.name = "wbt_lat_usec", .mode = 0644 },
 730         .show = queue_wb_lat_show,
 731         .store = queue_wb_lat_store,
 732 };
 733
 734 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 735 static struct queue_sysfs_entry throtl_sample_time_entry = {
 736         .attr = {.name = "throttle_sample_time", .mode = 0644 },
 737         .show = blk_throtl_sample_time_show,
 738         .store = blk_throtl_sample_time_store,
 739 };
 740 #endif
 741
 742 static struct attribute *queue_attrs[] = {
 743         &queue_requests_entry.attr,
 744         &queue_ra_entry.attr,
 745         &queue_max_hw_sectors_entry.attr,
 746         &queue_max_sectors_entry.attr,
 747         &queue_max_segments_entry.attr,
 748         &queue_max_discard_segments_entry.attr,
 749         &queue_max_integrity_segments_entry.attr,
 750         &queue_max_segment_size_entry.attr,
 751         &queue_iosched_entry.attr,
 752         &queue_hw_sector_size_entry.attr,
 753         &queue_logical_block_size_entry.attr,
 754         &queue_physical_block_size_entry.attr,
 755         &queue_chunk_sectors_entry.attr,
 756         &queue_io_min_entry.attr,
 757         &queue_io_opt_entry.attr,
 758         &queue_discard_granularity_entry.attr,
 759         &queue_discard_max_entry.attr,
 760         &queue_discard_max_hw_entry.attr,
 761         &queue_discard_zeroes_data_entry.attr,
 762         &queue_write_same_max_entry.attr,
 763         &queue_write_zeroes_max_entry.attr,
 764         &queue_zone_append_max_entry.attr,
 765         &queue_nonrot_entry.attr,
 766         &queue_zoned_entry.attr,
 767         &queue_nr_zones_entry.attr,
 768         &queue_nomerges_entry.attr,
 769         &queue_rq_affinity_entry.attr,
 770         &queue_iostats_entry.attr,
 771         &queue_random_entry.attr,
 772         &queue_poll_entry.attr,
 773         &queue_wc_entry.attr,
 774         &queue_fua_entry.attr,
 775         &queue_dax_entry.attr,
 776         &queue_wb_lat_entry.attr,
 777         &queue_poll_delay_entry.attr,
 778         &queue_io_timeout_entry.attr,
 779 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 780         &throtl_sample_time_entry.attr,
 781 #endif
 782         NULL,
 783 };
 784
 785 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
 786                                 int n)
 787 {
 788         struct request_queue *q =
 789                 container_of(kobj, struct request_queue, kobj);
 790
 791         if (attr == &queue_io_timeout_entry.attr &&
 792                 (!q->mq_ops || !q->mq_ops->timeout))
 793                         return 0;
 794
 795         return attr->mode;
 796 }
 797
 798 static struct attribute_group queue_attr_group = {
 799         .attrs = queue_attrs,
 800         .is_visible = queue_attr_visible,
 801 };
 802
 803
 804 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 805
 806 static ssize_t
 807 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 808 {
 809         struct queue_sysfs_entry *entry = to_queue(attr);
 810         struct request_queue *q =
 811                 container_of(kobj, struct request_queue, kobj);
 812         ssize_t res;
 813
 814         if (!entry->show)
 815                 return -EIO;
 816         mutex_lock(&q->sysfs_lock);
 817         res = entry->show(q, page);
 818         mutex_unlock(&q->sysfs_lock);
 819         return res;
 820 }
 821
 822 static ssize_t
 823 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 824                     const char *page, size_t length)
 825 {
 826         struct queue_sysfs_entry *entry = to_queue(attr);
 827         struct request_queue *q;
 828         ssize_t res;
 829
 830         if (!entry->store)
 831                 return -EIO;
 832
 833         q = container_of(kobj, struct request_queue, kobj);
 834         mutex_lock(&q->sysfs_lock);
 835         res = entry->store(q, page, length);
 836         mutex_unlock(&q->sysfs_lock);
 837         return res;
 838 }
 839
 840 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 841 {
 842         struct request_queue *q = container_of(rcu_head, struct request_queue,
 843                                                rcu_head);
 844         kmem_cache_free(blk_requestq_cachep, q);
 845 }
 846
 847 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
 848 static void blk_exit_queue(struct request_queue *q)
 849 {
 850         /*
 851          * Since the I/O scheduler exit code may access cgroup information,
 852          * perform I/O scheduler exit before disassociating from the block
 853          * cgroup controller.
 854          */
 855         if (q->elevator) {
 856                 ioc_clear_queue(q);
 857                 __elevator_exit(q, q->elevator);
 858                 q->elevator = NULL;
 859         }
 860
 861         /*
 862          * Remove all references to @q from the block cgroup controller before
 863          * restoring @q->queue_lock to avoid that restoring this pointer causes
 864          * e.g. blkcg_print_blkgs() to crash.
 865          */
 866         blkcg_exit_queue(q);
 867
 868         /*
 869          * Since the cgroup code may dereference the @q->backing_dev_info
 870          * pointer, only decrease its reference count after having removed the
 871          * association with the block cgroup controller.
 872          */
 873         bdi_put(q->backing_dev_info);
 874 }
 875
 876
 877 /**
 878  * __blk_release_queue - release a request queue
 879  * @work: pointer to the release_work member of the request queue to be released
 880  *
 881  * Description:
 882  *     This function is called when a block device is being unregistered. The
 883  *     process of releasing a request queue starts with blk_cleanup_queue, which
 884  *     set the appropriate flags and then calls blk_put_queue, that decrements
 885  *     the reference counter of the request queue. Once the reference counter
 886  *     of the request queue reaches zero, blk_release_queue is called to release
 887  *     all allocated resources of the request queue.
 888  */
 889 static void __blk_release_queue(struct work_struct *work)
 890 {
 891         struct request_queue *q = container_of(work, typeof(*q), release_work);
 892
 893         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 894                 blk_stat_remove_callback(q, q->poll_cb);
 895         blk_stat_free_callback(q->poll_cb);
 896
 897         blk_free_queue_stats(q->stats);
 898
 899         if (queue_is_mq(q))
 900                 cancel_delayed_work_sync(&q->requeue_work);
 901
 902         blk_exit_queue(q);
 903
 904         blk_queue_free_zone_bitmaps(q);
 905
 906         if (queue_is_mq(q))
 907                 blk_mq_release(q);
 908
 909         blk_trace_shutdown(q);
 910
 911         if (queue_is_mq(q))
 912                 blk_mq_debugfs_unregister(q);
 913
 914         bioset_exit(&q->bio_split);
 915
 916         ida_simple_remove(&blk_queue_ida, q->id);
 917         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 918 }
 919
 920 static void blk_release_queue(struct kobject *kobj)
 921 {
 922         struct request_queue *q =
 923                 container_of(kobj, struct request_queue, kobj);
 924
 925         INIT_WORK(&q->release_work, __blk_release_queue);
 926         schedule_work(&q->release_work);
 927 }
 928
 929 static const struct sysfs_ops queue_sysfs_ops = {
 930         .show   = queue_attr_show,
 931         .store  = queue_attr_store,
 932 };
 933
 934 struct kobj_type blk_queue_ktype = {
 935         .sysfs_ops      = &queue_sysfs_ops,
 936         .release        = blk_release_queue,
 937 };
 938
 939 /**
 940  * blk_register_queue - register a block layer queue with sysfs
 941  * @disk: Disk of which the request queue should be registered with sysfs.
 942  */
 943 int blk_register_queue(struct gendisk *disk)
 944 {
 945         int ret;
 946         struct device *dev = disk_to_dev(disk);
 947         struct request_queue *q = disk->queue;
 948         bool has_elevator = false;
 949
 950         if (WARN_ON(!q))
 951                 return -ENXIO;
 952
 953         WARN_ONCE(blk_queue_registered(q),
 954                   "%s is registering an already registered queue\n",
 955                   kobject_name(&dev->kobj));
 956
 957         /*
 958          * SCSI probing may synchronously create and destroy a lot of
 959          * request_queues for non-existent devices.  Shutting down a fully
 960          * functional queue takes measureable wallclock time as RCU grace
 961          * periods are involved.  To avoid excessive latency in these
 962          * cases, a request_queue starts out in a degraded mode which is
 963          * faster to shut down and is made fully functional here as
 964          * request_queues for non-existent devices never get registered.
 965          */
 966         if (!blk_queue_init_done(q)) {
 967                 blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
 968                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 969         }
 970
 971         ret = blk_trace_init_sysfs(dev);
 972         if (ret)
 973                 return ret;
 974
 975         mutex_lock(&q->sysfs_dir_lock);
 976
 977         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 978         if (ret < 0) {
 979                 blk_trace_remove_sysfs(dev);
 980                 goto unlock;
 981         }
 982
 983         ret = sysfs_create_group(&q->kobj, &queue_attr_group);
 984         if (ret) {
 985                 blk_trace_remove_sysfs(dev);
 986                 kobject_del(&q->kobj);
 987                 kobject_put(&dev->kobj);
 988                 goto unlock;
 989         }
 990
 991         if (queue_is_mq(q)) {
 992                 __blk_mq_register_dev(dev, q);
 993                 blk_mq_debugfs_register(q);
 994         }
 995
 996         mutex_lock(&q->sysfs_lock);
 997         if (q->elevator) {
 998                 ret = elv_register_queue(q, false);
 999                 if (ret) {
1000                         mutex_unlock(&q->sysfs_lock);
1001                         mutex_unlock(&q->sysfs_dir_lock);
1002                         kobject_del(&q->kobj);
1003                         blk_trace_remove_sysfs(dev);
1004                         kobject_put(&dev->kobj);
1005                         return ret;
1006                 }
1007                 has_elevator = true;
1008         }
1009
1010         blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
1011         wbt_enable_default(q);
1012         blk_throtl_register_queue(q);
1013
1014         /* Now everything is ready and send out KOBJ_ADD uevent */
1015         kobject_uevent(&q->kobj, KOBJ_ADD);
1016         if (has_elevator)
1017                 kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
1018         mutex_unlock(&q->sysfs_lock);
1019
1020         ret = 0;
1021 unlock:
1022         mutex_unlock(&q->sysfs_dir_lock);
1023         return ret;
1024 }
1025 EXPORT_SYMBOL_GPL(blk_register_queue);
1026
1027 /**
1028  * blk_unregister_queue - counterpart of blk_register_queue()
1029  * @disk: Disk of which the request queue should be unregistered from sysfs.
1030  *
1031  * Note: the caller is responsible for guaranteeing that this function is called
1032  * after blk_register_queue() has finished.
1033  */
1034 void blk_unregister_queue(struct gendisk *disk)
1035 {
1036         struct request_queue *q = disk->queue;
1037
1038         if (WARN_ON(!q))
1039                 return;
1040
1041         /* Return early if disk->queue was never registered. */
1042         if (!blk_queue_registered(q))
1043                 return;
1044
1045         /*
1046          * Since sysfs_remove_dir() prevents adding new directory entries
1047          * before removal of existing entries starts, protect against
1048          * concurrent elv_iosched_store() calls.
1049          */
1050         mutex_lock(&q->sysfs_lock);
1051         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
1052         mutex_unlock(&q->sysfs_lock);
1053
1054         mutex_lock(&q->sysfs_dir_lock);
1055         /*
1056          * Remove the sysfs attributes before unregistering the queue data
1057          * structures that can be modified through sysfs.
1058          */
1059         if (queue_is_mq(q))
1060                 blk_mq_unregister_dev(disk_to_dev(disk), q);
1061
1062         kobject_uevent(&q->kobj, KOBJ_REMOVE);
1063         kobject_del(&q->kobj);
1064         blk_trace_remove_sysfs(disk_to_dev(disk));
1065
1066         mutex_lock(&q->sysfs_lock);
1067         if (q->elevator)
1068                 elv_unregister_queue(q);
1069         mutex_unlock(&q->sysfs_lock);
1070         mutex_unlock(&q->sysfs_dir_lock);
1071
1072         kobject_put(&disk_to_dev(disk)->kobj);
1073 }