block/blk-sysfs.c

   1 /*
   2  * Functions related to sysfs handling
   3  */
   4 #include <linux/kernel.h>
   5 #include <linux/slab.h>
   6 #include <linux/module.h>
   7 #include <linux/bio.h>
   8 #include <linux/blkdev.h>
   9 #include <linux/blktrace_api.h>
  10 #include <linux/blk-mq.h>
  11
  12 #include "blk.h"
  13 #include "blk-cgroup.h"
  14 #include "blk-mq.h"
  15
  16 struct queue_sysfs_entry {
  17         struct attribute attr;
  18         ssize_t (*show)(struct request_queue *, char *);
  19         ssize_t (*store)(struct request_queue *, const char *, size_t);
  20 };
  21
  22 static ssize_t
  23 queue_var_show(unsigned long var, char *page)
  24 {
  25         return sprintf(page, "%lu\n", var);
  26 }
  27
  28 static ssize_t
  29 queue_var_store(unsigned long *var, const char *page, size_t count)
  30 {
  31         int err;
  32         unsigned long v;
  33
  34         err = kstrtoul(page, 10, &v);
  35         if (err || v > UINT_MAX)
  36                 return -EINVAL;
  37
  38         *var = v;
  39
  40         return count;
  41 }
  42
  43 static ssize_t queue_requests_show(struct request_queue *q, char *page)
  44 {
  45         return queue_var_show(q->nr_requests, (page));
  46 }
  47
  48 static ssize_t
  49 queue_requests_store(struct request_queue *q, const char *page, size_t count)
  50 {
  51         unsigned long nr;
  52         int ret, err;
  53
  54         if (!q->request_fn && !q->mq_ops)
  55                 return -EINVAL;
  56
  57         ret = queue_var_store(&nr, page, count);
  58         if (ret < 0)
  59                 return ret;
  60
  61         if (nr < BLKDEV_MIN_RQ)
  62                 nr = BLKDEV_MIN_RQ;
  63
  64         if (q->request_fn)
  65                 err = blk_update_nr_requests(q, nr);
  66         else
  67                 err = blk_mq_update_nr_requests(q, nr);
  68
  69         if (err)
  70                 return err;
  71
  72         return ret;
  73 }
  74
  75 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  76 {
  77         unsigned long ra_kb = q->backing_dev_info.ra_pages <<
  78                                         (PAGE_CACHE_SHIFT - 10);
  79
  80         return queue_var_show(ra_kb, (page));
  81 }
  82
  83 static ssize_t
  84 queue_ra_store(struct request_queue *q, const char *page, size_t count)
  85 {
  86         unsigned long ra_kb;
  87         ssize_t ret = queue_var_store(&ra_kb, page, count);
  88
  89         if (ret < 0)
  90                 return ret;
  91
  92         q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10);
  93
  94         return ret;
  95 }
  96
  97 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
  98 {
  99         int max_sectors_kb = queue_max_sectors(q) >> 1;
 100
 101         return queue_var_show(max_sectors_kb, (page));
 102 }
 103
 104 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 105 {
 106         return queue_var_show(queue_max_segments(q), (page));
 107 }
 108
 109 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 110 {
 111         return queue_var_show(q->limits.max_integrity_segments, (page));
 112 }
 113
 114 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 115 {
 116         if (blk_queue_cluster(q))
 117                 return queue_var_show(queue_max_segment_size(q), (page));
 118
 119         return queue_var_show(PAGE_CACHE_SIZE, (page));
 120 }
 121
 122 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 123 {
 124         return queue_var_show(queue_logical_block_size(q), page);
 125 }
 126
 127 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 128 {
 129         return queue_var_show(queue_physical_block_size(q), page);
 130 }
 131
 132 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 133 {
 134         return queue_var_show(queue_io_min(q), page);
 135 }
 136
 137 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 138 {
 139         return queue_var_show(queue_io_opt(q), page);
 140 }
 141
 142 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 143 {
 144         return queue_var_show(q->limits.discard_granularity, page);
 145 }
 146
 147 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 148 {
 149         return sprintf(page, "%llu\n",
 150                        (unsigned long long)q->limits.max_discard_sectors << 9);
 151 }
 152
 153 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 154 {
 155         return queue_var_show(queue_discard_zeroes_data(q), page);
 156 }
 157
 158 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 159 {
 160         return sprintf(page, "%llu\n",
 161                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 162 }
 163
 164
 165 static ssize_t
 166 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 167 {
 168         unsigned long max_sectors_kb,
 169                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 170                         page_kb = 1 << (PAGE_CACHE_SHIFT - 10);
 171         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 172
 173         if (ret < 0)
 174                 return ret;
 175
 176         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 177                 return -EINVAL;
 178
 179         spin_lock_irq(q->queue_lock);
 180         q->limits.max_sectors = max_sectors_kb << 1;
 181         spin_unlock_irq(q->queue_lock);
 182
 183         return ret;
 184 }
 185
 186 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 187 {
 188         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 189
 190         return queue_var_show(max_hw_sectors_kb, (page));
 191 }
 192
 193 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 194 static ssize_t                                                          \
 195 queue_show_##name(struct request_queue *q, char *page)                  \
 196 {                                                                       \
 197         int bit;                                                        \
 198         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 199         return queue_var_show(neg ? !bit : bit, page);                  \
 200 }                                                                       \
 201 static ssize_t                                                          \
 202 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
 203 {                                                                       \
 204         unsigned long val;                                              \
 205         ssize_t ret;                                                    \
 206         ret = queue_var_store(&val, page, count);                       \
 207         if (ret < 0)                                                    \
 208                  return ret;                                            \
 209         if (neg)                                                        \
 210                 val = !val;                                             \
 211                                                                         \
 212         spin_lock_irq(q->queue_lock);                                   \
 213         if (val)                                                        \
 214                 queue_flag_set(QUEUE_FLAG_##flag, q);                   \
 215         else                                                            \
 216                 queue_flag_clear(QUEUE_FLAG_##flag, q);                 \
 217         spin_unlock_irq(q->queue_lock);                                 \
 218         return ret;                                                     \
 219 }
 220
 221 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 222 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 223 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 224 #undef QUEUE_SYSFS_BIT_FNS
 225
 226 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 227 {
 228         return queue_var_show((blk_queue_nomerges(q) << 1) |
 229                                blk_queue_noxmerges(q), page);
 230 }
 231
 232 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 233                                     size_t count)
 234 {
 235         unsigned long nm;
 236         ssize_t ret = queue_var_store(&nm, page, count);
 237
 238         if (ret < 0)
 239                 return ret;
 240
 241         spin_lock_irq(q->queue_lock);
 242         queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 243         queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 244         if (nm == 2)
 245                 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 246         else if (nm)
 247                 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 248         spin_unlock_irq(q->queue_lock);
 249
 250         return ret;
 251 }
 252
 253 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 254 {
 255         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 256         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 257
 258         return queue_var_show(set << force, page);
 259 }
 260
 261 static ssize_t
 262 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 263 {
 264         ssize_t ret = -EINVAL;
 265 #ifdef CONFIG_SMP
 266         unsigned long val;
 267
 268         ret = queue_var_store(&val, page, count);
 269         if (ret < 0)
 270                 return ret;
 271
 272         spin_lock_irq(q->queue_lock);
 273         if (val == 2) {
 274                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 275                 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 276         } else if (val == 1) {
 277                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 278                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 279         } else if (val == 0) {
 280                 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 281                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 282         }
 283         spin_unlock_irq(q->queue_lock);
 284 #endif
 285         return ret;
 286 }
 287
 288 static struct queue_sysfs_entry queue_requests_entry = {
 289         .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR },
 290         .show = queue_requests_show,
 291         .store = queue_requests_store,
 292 };
 293
 294 static struct queue_sysfs_entry queue_ra_entry = {
 295         .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR },
 296         .show = queue_ra_show,
 297         .store = queue_ra_store,
 298 };
 299
 300 static struct queue_sysfs_entry queue_max_sectors_entry = {
 301         .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR },
 302         .show = queue_max_sectors_show,
 303         .store = queue_max_sectors_store,
 304 };
 305
 306 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 307         .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO },
 308         .show = queue_max_hw_sectors_show,
 309 };
 310
 311 static struct queue_sysfs_entry queue_max_segments_entry = {
 312         .attr = {.name = "max_segments", .mode = S_IRUGO },
 313         .show = queue_max_segments_show,
 314 };
 315
 316 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 317         .attr = {.name = "max_integrity_segments", .mode = S_IRUGO },
 318         .show = queue_max_integrity_segments_show,
 319 };
 320
 321 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 322         .attr = {.name = "max_segment_size", .mode = S_IRUGO },
 323         .show = queue_max_segment_size_show,
 324 };
 325
 326 static struct queue_sysfs_entry queue_iosched_entry = {
 327         .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR },
 328         .show = elv_iosched_show,
 329         .store = elv_iosched_store,
 330 };
 331
 332 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 333         .attr = {.name = "hw_sector_size", .mode = S_IRUGO },
 334         .show = queue_logical_block_size_show,
 335 };
 336
 337 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 338         .attr = {.name = "logical_block_size", .mode = S_IRUGO },
 339         .show = queue_logical_block_size_show,
 340 };
 341
 342 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 343         .attr = {.name = "physical_block_size", .mode = S_IRUGO },
 344         .show = queue_physical_block_size_show,
 345 };
 346
 347 static struct queue_sysfs_entry queue_io_min_entry = {
 348         .attr = {.name = "minimum_io_size", .mode = S_IRUGO },
 349         .show = queue_io_min_show,
 350 };
 351
 352 static struct queue_sysfs_entry queue_io_opt_entry = {
 353         .attr = {.name = "optimal_io_size", .mode = S_IRUGO },
 354         .show = queue_io_opt_show,
 355 };
 356
 357 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 358         .attr = {.name = "discard_granularity", .mode = S_IRUGO },
 359         .show = queue_discard_granularity_show,
 360 };
 361
 362 static struct queue_sysfs_entry queue_discard_max_entry = {
 363         .attr = {.name = "discard_max_bytes", .mode = S_IRUGO },
 364         .show = queue_discard_max_show,
 365 };
 366
 367 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 368         .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO },
 369         .show = queue_discard_zeroes_data_show,
 370 };
 371
 372 static struct queue_sysfs_entry queue_write_same_max_entry = {
 373         .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO },
 374         .show = queue_write_same_max_show,
 375 };
 376
 377 static struct queue_sysfs_entry queue_nonrot_entry = {
 378         .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR },
 379         .show = queue_show_nonrot,
 380         .store = queue_store_nonrot,
 381 };
 382
 383 static struct queue_sysfs_entry queue_nomerges_entry = {
 384         .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR },
 385         .show = queue_nomerges_show,
 386         .store = queue_nomerges_store,
 387 };
 388
 389 static struct queue_sysfs_entry queue_rq_affinity_entry = {
 390         .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR },
 391         .show = queue_rq_affinity_show,
 392         .store = queue_rq_affinity_store,
 393 };
 394
 395 static struct queue_sysfs_entry queue_iostats_entry = {
 396         .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR },
 397         .show = queue_show_iostats,
 398         .store = queue_store_iostats,
 399 };
 400
 401 static struct queue_sysfs_entry queue_random_entry = {
 402         .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR },
 403         .show = queue_show_random,
 404         .store = queue_store_random,
 405 };
 406
 407 static struct attribute *default_attrs[] = {
 408         &queue_requests_entry.attr,
 409         &queue_ra_entry.attr,
 410         &queue_max_hw_sectors_entry.attr,
 411         &queue_max_sectors_entry.attr,
 412         &queue_max_segments_entry.attr,
 413         &queue_max_integrity_segments_entry.attr,
 414         &queue_max_segment_size_entry.attr,
 415         &queue_iosched_entry.attr,
 416         &queue_hw_sector_size_entry.attr,
 417         &queue_logical_block_size_entry.attr,
 418         &queue_physical_block_size_entry.attr,
 419         &queue_io_min_entry.attr,
 420         &queue_io_opt_entry.attr,
 421         &queue_discard_granularity_entry.attr,
 422         &queue_discard_max_entry.attr,
 423         &queue_discard_zeroes_data_entry.attr,
 424         &queue_write_same_max_entry.attr,
 425         &queue_nonrot_entry.attr,
 426         &queue_nomerges_entry.attr,
 427         &queue_rq_affinity_entry.attr,
 428         &queue_iostats_entry.attr,
 429         &queue_random_entry.attr,
 430         NULL,
 431 };
 432
 433 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 434
 435 static ssize_t
 436 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 437 {
 438         struct queue_sysfs_entry *entry = to_queue(attr);
 439         struct request_queue *q =
 440                 container_of(kobj, struct request_queue, kobj);
 441         ssize_t res;
 442
 443         if (!entry->show)
 444                 return -EIO;
 445         mutex_lock(&q->sysfs_lock);
 446         if (blk_queue_dying(q)) {
 447                 mutex_unlock(&q->sysfs_lock);
 448                 return -ENOENT;
 449         }
 450         res = entry->show(q, page);
 451         mutex_unlock(&q->sysfs_lock);
 452         return res;
 453 }
 454
 455 static ssize_t
 456 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 457                     const char *page, size_t length)
 458 {
 459         struct queue_sysfs_entry *entry = to_queue(attr);
 460         struct request_queue *q;
 461         ssize_t res;
 462
 463         if (!entry->store)
 464                 return -EIO;
 465
 466         q = container_of(kobj, struct request_queue, kobj);
 467         mutex_lock(&q->sysfs_lock);
 468         if (blk_queue_dying(q)) {
 469                 mutex_unlock(&q->sysfs_lock);
 470                 return -ENOENT;
 471         }
 472         res = entry->store(q, page, length);
 473         mutex_unlock(&q->sysfs_lock);
 474         return res;
 475 }
 476
 477 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 478 {
 479         struct request_queue *q = container_of(rcu_head, struct request_queue,
 480                                                rcu_head);
 481         kmem_cache_free(blk_requestq_cachep, q);
 482 }
 483
 484 /**
 485  * blk_release_queue: - release a &struct request_queue when it is no longer needed
 486  * @kobj:    the kobj belonging to the request queue to be released
 487  *
 488  * Description:
 489  *     blk_release_queue is the pair to blk_init_queue() or
 490  *     blk_queue_make_request().  It should be called when a request queue is
 491  *     being released; typically when a block device is being de-registered.
 492  *     Currently, its primary task it to free all the &struct request
 493  *     structures that were allocated to the queue and the queue itself.
 494  *
 495  * Caveat:
 496  *     Hopefully the low level driver will have finished any
 497  *     outstanding requests first...
 498  **/
 499 static void blk_release_queue(struct kobject *kobj)
 500 {
 501         struct request_queue *q =
 502                 container_of(kobj, struct request_queue, kobj);
 503
 504         blk_sync_queue(q);
 505
 506         blkcg_exit_queue(q);
 507
 508         if (q->elevator) {
 509                 spin_lock_irq(q->queue_lock);
 510                 ioc_clear_queue(q);
 511                 spin_unlock_irq(q->queue_lock);
 512                 elevator_exit(q->elevator);
 513         }
 514
 515         blk_exit_rl(&q->root_rl);
 516
 517         if (q->queue_tags)
 518                 __blk_queue_free_tags(q);
 519
 520         if (q->mq_ops)
 521                 blk_mq_free_queue(q);
 522         else
 523                 blk_free_flush_queue(q->fq);
 524
 525         blk_trace_shutdown(q);
 526
 527         bdi_destroy(&q->backing_dev_info);
 528
 529         ida_simple_remove(&blk_queue_ida, q->id);
 530         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 531 }
 532
 533 static const struct sysfs_ops queue_sysfs_ops = {
 534         .show   = queue_attr_show,
 535         .store  = queue_attr_store,
 536 };
 537
 538 struct kobj_type blk_queue_ktype = {
 539         .sysfs_ops      = &queue_sysfs_ops,
 540         .default_attrs  = default_attrs,
 541         .release        = blk_release_queue,
 542 };
 543
 544 int blk_register_queue(struct gendisk *disk)
 545 {
 546         int ret;
 547         struct device *dev = disk_to_dev(disk);
 548         struct request_queue *q = disk->queue;
 549
 550         if (WARN_ON(!q))
 551                 return -ENXIO;
 552
 553         /*
 554          * SCSI probing may synchronously create and destroy a lot of
 555          * request_queues for non-existent devices.  Shutting down a fully
 556          * functional queue takes measureable wallclock time as RCU grace
 557          * periods are involved.  To avoid excessive latency in these
 558          * cases, a request_queue starts out in a degraded mode which is
 559          * faster to shut down and is made fully functional here as
 560          * request_queues for non-existent devices never get registered.
 561          */
 562         if (!blk_queue_init_done(q)) {
 563                 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
 564                 blk_queue_bypass_end(q);
 565                 if (q->mq_ops)
 566                         blk_mq_finish_init(q);
 567         }
 568
 569         ret = blk_trace_init_sysfs(dev);
 570         if (ret)
 571                 return ret;
 572
 573         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 574         if (ret < 0) {
 575                 blk_trace_remove_sysfs(dev);
 576                 return ret;
 577         }
 578
 579         kobject_uevent(&q->kobj, KOBJ_ADD);
 580
 581         if (q->mq_ops)
 582                 blk_mq_register_disk(disk);
 583
 584         if (!q->request_fn)
 585                 return 0;
 586
 587         ret = elv_register_queue(q);
 588         if (ret) {
 589                 kobject_uevent(&q->kobj, KOBJ_REMOVE);
 590                 kobject_del(&q->kobj);
 591                 blk_trace_remove_sysfs(dev);
 592                 kobject_put(&dev->kobj);
 593                 return ret;
 594         }
 595
 596         return 0;
 597 }
 598
 599 void blk_unregister_queue(struct gendisk *disk)
 600 {
 601         struct request_queue *q = disk->queue;
 602
 603         if (WARN_ON(!q))
 604                 return;
 605
 606         if (q->mq_ops)
 607                 blk_mq_unregister_disk(disk);
 608
 609         if (q->request_fn)
 610                 elv_unregister_queue(q);
 611
 612         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 613         kobject_del(&q->kobj);
 614         blk_trace_remove_sysfs(disk_to_dev(disk));
 615         kobject_put(&disk_to_dev(disk)->kobj);
 616 }