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