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 15 struct queue_sysfs_entry { 16 struct attribute attr; 17 ssize_t (*show)(struct request_queue *, char *); 18 ssize_t (*store)(struct request_queue *, const char *, size_t); 19 }; 20 21 static ssize_t 22 queue_var_show(unsigned long var, char *page) 23 { 24 return sprintf(page, "%lu\n", var); 25 } 26 27 static ssize_t 28 queue_var_store(unsigned long *var, const char *page, size_t count) 29 { 30 int err; 31 unsigned long v; 32 33 err = kstrtoul(page, 10, &v); 34 if (err || v > UINT_MAX) 35 return -EINVAL; 36 37 *var = v; 38 39 return count; 40 } 41 42 static ssize_t queue_requests_show(struct request_queue *q, char *page) 43 { 44 return queue_var_show(q->nr_requests, (page)); 45 } 46 47 static ssize_t 48 queue_requests_store(struct request_queue *q, const char *page, size_t count) 49 { 50 struct request_list *rl; 51 unsigned long nr; 52 int ret; 53 54 if (!q->request_fn) 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 spin_lock_irq(q->queue_lock); 65 q->nr_requests = nr; 66 blk_queue_congestion_threshold(q); 67 68 /* congestion isn't cgroup aware and follows root blkcg for now */ 69 rl = &q->root_rl; 70 71 if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q)) 72 blk_set_queue_congested(q, BLK_RW_SYNC); 73 else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q)) 74 blk_clear_queue_congested(q, BLK_RW_SYNC); 75 76 if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q)) 77 blk_set_queue_congested(q, BLK_RW_ASYNC); 78 else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q)) 79 blk_clear_queue_congested(q, BLK_RW_ASYNC); 80 81 blk_queue_for_each_rl(rl, q) { 82 if (rl->count[BLK_RW_SYNC] >= q->nr_requests) { 83 blk_set_rl_full(rl, BLK_RW_SYNC); 84 } else { 85 blk_clear_rl_full(rl, BLK_RW_SYNC); 86 wake_up(&rl->wait[BLK_RW_SYNC]); 87 } 88 89 if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) { 90 blk_set_rl_full(rl, BLK_RW_ASYNC); 91 } else { 92 blk_clear_rl_full(rl, BLK_RW_ASYNC); 93 wake_up(&rl->wait[BLK_RW_ASYNC]); 94 } 95 } 96 97 spin_unlock_irq(q->queue_lock); 98 return ret; 99 } 100 101 static ssize_t queue_ra_show(struct request_queue *q, char *page) 102 { 103 unsigned long ra_kb = q->backing_dev_info.ra_pages << 104 (PAGE_CACHE_SHIFT - 10); 105 106 return queue_var_show(ra_kb, (page)); 107 } 108 109 static ssize_t 110 queue_ra_store(struct request_queue *q, const char *page, size_t count) 111 { 112 unsigned long ra_kb; 113 ssize_t ret = queue_var_store(&ra_kb, page, count); 114 115 if (ret < 0) 116 return ret; 117 118 q->backing_dev_info.ra_pages = ra_kb >> (PAGE_CACHE_SHIFT - 10); 119 120 return ret; 121 } 122 123 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page) 124 { 125 int max_sectors_kb = queue_max_sectors(q) >> 1; 126 127 return queue_var_show(max_sectors_kb, (page)); 128 } 129 130 static ssize_t queue_max_segments_show(struct request_queue *q, char *page) 131 { 132 return queue_var_show(queue_max_segments(q), (page)); 133 } 134 135 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page) 136 { 137 return queue_var_show(q->limits.max_integrity_segments, (page)); 138 } 139 140 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page) 141 { 142 if (blk_queue_cluster(q)) 143 return queue_var_show(queue_max_segment_size(q), (page)); 144 145 return queue_var_show(PAGE_CACHE_SIZE, (page)); 146 } 147 148 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page) 149 { 150 return queue_var_show(queue_logical_block_size(q), page); 151 } 152 153 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page) 154 { 155 return queue_var_show(queue_physical_block_size(q), page); 156 } 157 158 static ssize_t queue_io_min_show(struct request_queue *q, char *page) 159 { 160 return queue_var_show(queue_io_min(q), page); 161 } 162 163 static ssize_t queue_io_opt_show(struct request_queue *q, char *page) 164 { 165 return queue_var_show(queue_io_opt(q), page); 166 } 167 168 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page) 169 { 170 return queue_var_show(q->limits.discard_granularity, page); 171 } 172 173 static ssize_t queue_discard_max_show(struct request_queue *q, char *page) 174 { 175 return sprintf(page, "%llu\n", 176 (unsigned long long)q->limits.max_discard_sectors << 9); 177 } 178 179 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page) 180 { 181 return queue_var_show(queue_discard_zeroes_data(q), page); 182 } 183 184 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page) 185 { 186 return sprintf(page, "%llu\n", 187 (unsigned long long)q->limits.max_write_same_sectors << 9); 188 } 189 190 191 static ssize_t 192 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) 193 { 194 unsigned long max_sectors_kb, 195 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1, 196 page_kb = 1 << (PAGE_CACHE_SHIFT - 10); 197 ssize_t ret = queue_var_store(&max_sectors_kb, page, count); 198 199 if (ret < 0) 200 return ret; 201 202 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) 203 return -EINVAL; 204 205 spin_lock_irq(q->queue_lock); 206 q->limits.max_sectors = max_sectors_kb << 1; 207 spin_unlock_irq(q->queue_lock); 208 209 return ret; 210 } 211 212 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) 213 { 214 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1; 215 216 return queue_var_show(max_hw_sectors_kb, (page)); 217 } 218 219 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \ 220 static ssize_t \ 221 queue_show_##name(struct request_queue *q, char *page) \ 222 { \ 223 int bit; \ 224 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \ 225 return queue_var_show(neg ? !bit : bit, page); \ 226 } \ 227 static ssize_t \ 228 queue_store_##name(struct request_queue *q, const char *page, size_t count) \ 229 { \ 230 unsigned long val; \ 231 ssize_t ret; \ 232 ret = queue_var_store(&val, page, count); \ 233 if (ret < 0) \ 234 return ret; \ 235 if (neg) \ 236 val = !val; \ 237 \ 238 spin_lock_irq(q->queue_lock); \ 239 if (val) \ 240 queue_flag_set(QUEUE_FLAG_##flag, q); \ 241 else \ 242 queue_flag_clear(QUEUE_FLAG_##flag, q); \ 243 spin_unlock_irq(q->queue_lock); \ 244 return ret; \ 245 } 246 247 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1); 248 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0); 249 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0); 250 #undef QUEUE_SYSFS_BIT_FNS 251 252 static ssize_t queue_nomerges_show(struct request_queue *q, char *page) 253 { 254 return queue_var_show((blk_queue_nomerges(q) << 1) | 255 blk_queue_noxmerges(q), page); 256 } 257 258 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page, 259 size_t count) 260 { 261 unsigned long nm; 262 ssize_t ret = queue_var_store(&nm, page, count); 263 264 if (ret < 0) 265 return ret; 266 267 spin_lock_irq(q->queue_lock); 268 queue_flag_clear(QUEUE_FLAG_NOMERGES, q); 269 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q); 270 if (nm == 2) 271 queue_flag_set(QUEUE_FLAG_NOMERGES, q); 272 else if (nm) 273 queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 274 spin_unlock_irq(q->queue_lock); 275 276 return ret; 277 } 278 279 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page) 280 { 281 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags); 282 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags); 283 284 return queue_var_show(set << force, page); 285 } 286 287 static ssize_t 288 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count) 289 { 290 ssize_t ret = -EINVAL; 291 #ifdef CONFIG_SMP 292 unsigned long val; 293 294 ret = queue_var_store(&val, page, count); 295 if (ret < 0) 296 return ret; 297 298 spin_lock_irq(q->queue_lock); 299 if (val == 2) { 300 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 301 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q); 302 } else if (val == 1) { 303 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 304 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 305 } else if (val == 0) { 306 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q); 307 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 308 } 309 spin_unlock_irq(q->queue_lock); 310 #endif 311 return ret; 312 } 313 314 static struct queue_sysfs_entry queue_requests_entry = { 315 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, 316 .show = queue_requests_show, 317 .store = queue_requests_store, 318 }; 319 320 static struct queue_sysfs_entry queue_ra_entry = { 321 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, 322 .show = queue_ra_show, 323 .store = queue_ra_store, 324 }; 325 326 static struct queue_sysfs_entry queue_max_sectors_entry = { 327 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, 328 .show = queue_max_sectors_show, 329 .store = queue_max_sectors_store, 330 }; 331 332 static struct queue_sysfs_entry queue_max_hw_sectors_entry = { 333 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, 334 .show = queue_max_hw_sectors_show, 335 }; 336 337 static struct queue_sysfs_entry queue_max_segments_entry = { 338 .attr = {.name = "max_segments", .mode = S_IRUGO }, 339 .show = queue_max_segments_show, 340 }; 341 342 static struct queue_sysfs_entry queue_max_integrity_segments_entry = { 343 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO }, 344 .show = queue_max_integrity_segments_show, 345 }; 346 347 static struct queue_sysfs_entry queue_max_segment_size_entry = { 348 .attr = {.name = "max_segment_size", .mode = S_IRUGO }, 349 .show = queue_max_segment_size_show, 350 }; 351 352 static struct queue_sysfs_entry queue_iosched_entry = { 353 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, 354 .show = elv_iosched_show, 355 .store = elv_iosched_store, 356 }; 357 358 static struct queue_sysfs_entry queue_hw_sector_size_entry = { 359 .attr = {.name = "hw_sector_size", .mode = S_IRUGO }, 360 .show = queue_logical_block_size_show, 361 }; 362 363 static struct queue_sysfs_entry queue_logical_block_size_entry = { 364 .attr = {.name = "logical_block_size", .mode = S_IRUGO }, 365 .show = queue_logical_block_size_show, 366 }; 367 368 static struct queue_sysfs_entry queue_physical_block_size_entry = { 369 .attr = {.name = "physical_block_size", .mode = S_IRUGO }, 370 .show = queue_physical_block_size_show, 371 }; 372 373 static struct queue_sysfs_entry queue_io_min_entry = { 374 .attr = {.name = "minimum_io_size", .mode = S_IRUGO }, 375 .show = queue_io_min_show, 376 }; 377 378 static struct queue_sysfs_entry queue_io_opt_entry = { 379 .attr = {.name = "optimal_io_size", .mode = S_IRUGO }, 380 .show = queue_io_opt_show, 381 }; 382 383 static struct queue_sysfs_entry queue_discard_granularity_entry = { 384 .attr = {.name = "discard_granularity", .mode = S_IRUGO }, 385 .show = queue_discard_granularity_show, 386 }; 387 388 static struct queue_sysfs_entry queue_discard_max_entry = { 389 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO }, 390 .show = queue_discard_max_show, 391 }; 392 393 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = { 394 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO }, 395 .show = queue_discard_zeroes_data_show, 396 }; 397 398 static struct queue_sysfs_entry queue_write_same_max_entry = { 399 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO }, 400 .show = queue_write_same_max_show, 401 }; 402 403 static struct queue_sysfs_entry queue_nonrot_entry = { 404 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR }, 405 .show = queue_show_nonrot, 406 .store = queue_store_nonrot, 407 }; 408 409 static struct queue_sysfs_entry queue_nomerges_entry = { 410 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR }, 411 .show = queue_nomerges_show, 412 .store = queue_nomerges_store, 413 }; 414 415 static struct queue_sysfs_entry queue_rq_affinity_entry = { 416 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR }, 417 .show = queue_rq_affinity_show, 418 .store = queue_rq_affinity_store, 419 }; 420 421 static struct queue_sysfs_entry queue_iostats_entry = { 422 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR }, 423 .show = queue_show_iostats, 424 .store = queue_store_iostats, 425 }; 426 427 static struct queue_sysfs_entry queue_random_entry = { 428 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR }, 429 .show = queue_show_random, 430 .store = queue_store_random, 431 }; 432 433 static struct attribute *default_attrs[] = { 434 &queue_requests_entry.attr, 435 &queue_ra_entry.attr, 436 &queue_max_hw_sectors_entry.attr, 437 &queue_max_sectors_entry.attr, 438 &queue_max_segments_entry.attr, 439 &queue_max_integrity_segments_entry.attr, 440 &queue_max_segment_size_entry.attr, 441 &queue_iosched_entry.attr, 442 &queue_hw_sector_size_entry.attr, 443 &queue_logical_block_size_entry.attr, 444 &queue_physical_block_size_entry.attr, 445 &queue_io_min_entry.attr, 446 &queue_io_opt_entry.attr, 447 &queue_discard_granularity_entry.attr, 448 &queue_discard_max_entry.attr, 449 &queue_discard_zeroes_data_entry.attr, 450 &queue_write_same_max_entry.attr, 451 &queue_nonrot_entry.attr, 452 &queue_nomerges_entry.attr, 453 &queue_rq_affinity_entry.attr, 454 &queue_iostats_entry.attr, 455 &queue_random_entry.attr, 456 NULL, 457 }; 458 459 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) 460 461 static ssize_t 462 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 463 { 464 struct queue_sysfs_entry *entry = to_queue(attr); 465 struct request_queue *q = 466 container_of(kobj, struct request_queue, kobj); 467 ssize_t res; 468 469 if (!entry->show) 470 return -EIO; 471 mutex_lock(&q->sysfs_lock); 472 if (blk_queue_dying(q)) { 473 mutex_unlock(&q->sysfs_lock); 474 return -ENOENT; 475 } 476 res = entry->show(q, page); 477 mutex_unlock(&q->sysfs_lock); 478 return res; 479 } 480 481 static ssize_t 482 queue_attr_store(struct kobject *kobj, struct attribute *attr, 483 const char *page, size_t length) 484 { 485 struct queue_sysfs_entry *entry = to_queue(attr); 486 struct request_queue *q; 487 ssize_t res; 488 489 if (!entry->store) 490 return -EIO; 491 492 q = container_of(kobj, struct request_queue, kobj); 493 mutex_lock(&q->sysfs_lock); 494 if (blk_queue_dying(q)) { 495 mutex_unlock(&q->sysfs_lock); 496 return -ENOENT; 497 } 498 res = entry->store(q, page, length); 499 mutex_unlock(&q->sysfs_lock); 500 return res; 501 } 502 503 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 504 { 505 struct request_queue *q = container_of(rcu_head, struct request_queue, 506 rcu_head); 507 kmem_cache_free(blk_requestq_cachep, q); 508 } 509 510 /** 511 * blk_release_queue: - release a &struct request_queue when it is no longer needed 512 * @kobj: the kobj belonging to the request queue to be released 513 * 514 * Description: 515 * blk_release_queue is the pair to blk_init_queue() or 516 * blk_queue_make_request(). It should be called when a request queue is 517 * being released; typically when a block device is being de-registered. 518 * Currently, its primary task it to free all the &struct request 519 * structures that were allocated to the queue and the queue itself. 520 * 521 * Caveat: 522 * Hopefully the low level driver will have finished any 523 * outstanding requests first... 524 **/ 525 static void blk_release_queue(struct kobject *kobj) 526 { 527 struct request_queue *q = 528 container_of(kobj, struct request_queue, kobj); 529 530 blk_sync_queue(q); 531 532 blkcg_exit_queue(q); 533 534 if (q->elevator) { 535 spin_lock_irq(q->queue_lock); 536 ioc_clear_queue(q); 537 spin_unlock_irq(q->queue_lock); 538 elevator_exit(q->elevator); 539 } 540 541 blk_exit_rl(&q->root_rl); 542 543 if (q->queue_tags) 544 __blk_queue_free_tags(q); 545 546 percpu_counter_destroy(&q->mq_usage_counter); 547 548 if (q->mq_ops) 549 blk_mq_free_queue(q); 550 551 blk_trace_shutdown(q); 552 553 bdi_destroy(&q->backing_dev_info); 554 555 ida_simple_remove(&blk_queue_ida, q->id); 556 call_rcu(&q->rcu_head, blk_free_queue_rcu); 557 } 558 559 static const struct sysfs_ops queue_sysfs_ops = { 560 .show = queue_attr_show, 561 .store = queue_attr_store, 562 }; 563 564 struct kobj_type blk_queue_ktype = { 565 .sysfs_ops = &queue_sysfs_ops, 566 .default_attrs = default_attrs, 567 .release = blk_release_queue, 568 }; 569 570 int blk_register_queue(struct gendisk *disk) 571 { 572 int ret; 573 struct device *dev = disk_to_dev(disk); 574 struct request_queue *q = disk->queue; 575 576 if (WARN_ON(!q)) 577 return -ENXIO; 578 579 /* 580 * Initialization must be complete by now. Finish the initial 581 * bypass from queue allocation. 582 */ 583 blk_queue_bypass_end(q); 584 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q); 585 586 ret = blk_trace_init_sysfs(dev); 587 if (ret) 588 return ret; 589 590 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue"); 591 if (ret < 0) { 592 blk_trace_remove_sysfs(dev); 593 return ret; 594 } 595 596 kobject_uevent(&q->kobj, KOBJ_ADD); 597 598 if (q->mq_ops) 599 blk_mq_register_disk(disk); 600 601 if (!q->request_fn) 602 return 0; 603 604 ret = elv_register_queue(q); 605 if (ret) { 606 kobject_uevent(&q->kobj, KOBJ_REMOVE); 607 kobject_del(&q->kobj); 608 blk_trace_remove_sysfs(dev); 609 kobject_put(&dev->kobj); 610 return ret; 611 } 612 613 return 0; 614 } 615 616 void blk_unregister_queue(struct gendisk *disk) 617 { 618 struct request_queue *q = disk->queue; 619 620 if (WARN_ON(!q)) 621 return; 622 623 if (q->mq_ops) 624 blk_mq_unregister_disk(disk); 625 626 if (q->request_fn) 627 elv_unregister_queue(q); 628 629 kobject_uevent(&q->kobj, KOBJ_REMOVE); 630 kobject_del(&q->kobj); 631 blk_trace_remove_sysfs(disk_to_dev(disk)); 632 kobject_put(&disk_to_dev(disk)->kobj); 633 } 634