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_hw_show(struct request_queue *q, char *page) 149 { 150 unsigned long long val; 151 152 val = q->limits.max_hw_discard_sectors << 9; 153 return sprintf(page, "%llu\n", val); 154 } 155 156 static ssize_t queue_discard_max_show(struct request_queue *q, char *page) 157 { 158 return sprintf(page, "%llu\n", 159 (unsigned long long)q->limits.max_discard_sectors << 9); 160 } 161 162 static ssize_t queue_discard_max_store(struct request_queue *q, 163 const char *page, size_t count) 164 { 165 unsigned long max_discard; 166 ssize_t ret = queue_var_store(&max_discard, page, count); 167 168 if (ret < 0) 169 return ret; 170 171 if (max_discard & (q->limits.discard_granularity - 1)) 172 return -EINVAL; 173 174 max_discard >>= 9; 175 if (max_discard > UINT_MAX) 176 return -EINVAL; 177 178 if (max_discard > q->limits.max_hw_discard_sectors) 179 max_discard = q->limits.max_hw_discard_sectors; 180 181 q->limits.max_discard_sectors = max_discard; 182 return ret; 183 } 184 185 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page) 186 { 187 return queue_var_show(queue_discard_zeroes_data(q), page); 188 } 189 190 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page) 191 { 192 return sprintf(page, "%llu\n", 193 (unsigned long long)q->limits.max_write_same_sectors << 9); 194 } 195 196 197 static ssize_t 198 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) 199 { 200 unsigned long max_sectors_kb, 201 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1, 202 page_kb = 1 << (PAGE_CACHE_SHIFT - 10); 203 ssize_t ret = queue_var_store(&max_sectors_kb, page, count); 204 205 if (ret < 0) 206 return ret; 207 208 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) 209 return -EINVAL; 210 211 spin_lock_irq(q->queue_lock); 212 q->limits.max_sectors = max_sectors_kb << 1; 213 spin_unlock_irq(q->queue_lock); 214 215 return ret; 216 } 217 218 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) 219 { 220 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1; 221 222 return queue_var_show(max_hw_sectors_kb, (page)); 223 } 224 225 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \ 226 static ssize_t \ 227 queue_show_##name(struct request_queue *q, char *page) \ 228 { \ 229 int bit; \ 230 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \ 231 return queue_var_show(neg ? !bit : bit, page); \ 232 } \ 233 static ssize_t \ 234 queue_store_##name(struct request_queue *q, const char *page, size_t count) \ 235 { \ 236 unsigned long val; \ 237 ssize_t ret; \ 238 ret = queue_var_store(&val, page, count); \ 239 if (ret < 0) \ 240 return ret; \ 241 if (neg) \ 242 val = !val; \ 243 \ 244 spin_lock_irq(q->queue_lock); \ 245 if (val) \ 246 queue_flag_set(QUEUE_FLAG_##flag, q); \ 247 else \ 248 queue_flag_clear(QUEUE_FLAG_##flag, q); \ 249 spin_unlock_irq(q->queue_lock); \ 250 return ret; \ 251 } 252 253 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1); 254 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0); 255 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0); 256 #undef QUEUE_SYSFS_BIT_FNS 257 258 static ssize_t queue_nomerges_show(struct request_queue *q, char *page) 259 { 260 return queue_var_show((blk_queue_nomerges(q) << 1) | 261 blk_queue_noxmerges(q), page); 262 } 263 264 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page, 265 size_t count) 266 { 267 unsigned long nm; 268 ssize_t ret = queue_var_store(&nm, page, count); 269 270 if (ret < 0) 271 return ret; 272 273 spin_lock_irq(q->queue_lock); 274 queue_flag_clear(QUEUE_FLAG_NOMERGES, q); 275 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q); 276 if (nm == 2) 277 queue_flag_set(QUEUE_FLAG_NOMERGES, q); 278 else if (nm) 279 queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 280 spin_unlock_irq(q->queue_lock); 281 282 return ret; 283 } 284 285 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page) 286 { 287 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags); 288 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags); 289 290 return queue_var_show(set << force, page); 291 } 292 293 static ssize_t 294 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count) 295 { 296 ssize_t ret = -EINVAL; 297 #ifdef CONFIG_SMP 298 unsigned long val; 299 300 ret = queue_var_store(&val, page, count); 301 if (ret < 0) 302 return ret; 303 304 spin_lock_irq(q->queue_lock); 305 if (val == 2) { 306 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 307 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q); 308 } else if (val == 1) { 309 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 310 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 311 } else if (val == 0) { 312 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q); 313 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 314 } 315 spin_unlock_irq(q->queue_lock); 316 #endif 317 return ret; 318 } 319 320 static ssize_t queue_poll_show(struct request_queue *q, char *page) 321 { 322 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page); 323 } 324 325 static ssize_t queue_poll_store(struct request_queue *q, const char *page, 326 size_t count) 327 { 328 unsigned long poll_on; 329 ssize_t ret; 330 331 if (!q->mq_ops || !q->mq_ops->poll) 332 return -EINVAL; 333 334 ret = queue_var_store(&poll_on, page, count); 335 if (ret < 0) 336 return ret; 337 338 spin_lock_irq(q->queue_lock); 339 if (poll_on) 340 queue_flag_set(QUEUE_FLAG_POLL, q); 341 else 342 queue_flag_clear(QUEUE_FLAG_POLL, q); 343 spin_unlock_irq(q->queue_lock); 344 345 return ret; 346 } 347 348 static struct queue_sysfs_entry queue_requests_entry = { 349 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, 350 .show = queue_requests_show, 351 .store = queue_requests_store, 352 }; 353 354 static struct queue_sysfs_entry queue_ra_entry = { 355 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, 356 .show = queue_ra_show, 357 .store = queue_ra_store, 358 }; 359 360 static struct queue_sysfs_entry queue_max_sectors_entry = { 361 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, 362 .show = queue_max_sectors_show, 363 .store = queue_max_sectors_store, 364 }; 365 366 static struct queue_sysfs_entry queue_max_hw_sectors_entry = { 367 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, 368 .show = queue_max_hw_sectors_show, 369 }; 370 371 static struct queue_sysfs_entry queue_max_segments_entry = { 372 .attr = {.name = "max_segments", .mode = S_IRUGO }, 373 .show = queue_max_segments_show, 374 }; 375 376 static struct queue_sysfs_entry queue_max_integrity_segments_entry = { 377 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO }, 378 .show = queue_max_integrity_segments_show, 379 }; 380 381 static struct queue_sysfs_entry queue_max_segment_size_entry = { 382 .attr = {.name = "max_segment_size", .mode = S_IRUGO }, 383 .show = queue_max_segment_size_show, 384 }; 385 386 static struct queue_sysfs_entry queue_iosched_entry = { 387 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, 388 .show = elv_iosched_show, 389 .store = elv_iosched_store, 390 }; 391 392 static struct queue_sysfs_entry queue_hw_sector_size_entry = { 393 .attr = {.name = "hw_sector_size", .mode = S_IRUGO }, 394 .show = queue_logical_block_size_show, 395 }; 396 397 static struct queue_sysfs_entry queue_logical_block_size_entry = { 398 .attr = {.name = "logical_block_size", .mode = S_IRUGO }, 399 .show = queue_logical_block_size_show, 400 }; 401 402 static struct queue_sysfs_entry queue_physical_block_size_entry = { 403 .attr = {.name = "physical_block_size", .mode = S_IRUGO }, 404 .show = queue_physical_block_size_show, 405 }; 406 407 static struct queue_sysfs_entry queue_io_min_entry = { 408 .attr = {.name = "minimum_io_size", .mode = S_IRUGO }, 409 .show = queue_io_min_show, 410 }; 411 412 static struct queue_sysfs_entry queue_io_opt_entry = { 413 .attr = {.name = "optimal_io_size", .mode = S_IRUGO }, 414 .show = queue_io_opt_show, 415 }; 416 417 static struct queue_sysfs_entry queue_discard_granularity_entry = { 418 .attr = {.name = "discard_granularity", .mode = S_IRUGO }, 419 .show = queue_discard_granularity_show, 420 }; 421 422 static struct queue_sysfs_entry queue_discard_max_hw_entry = { 423 .attr = {.name = "discard_max_hw_bytes", .mode = S_IRUGO }, 424 .show = queue_discard_max_hw_show, 425 }; 426 427 static struct queue_sysfs_entry queue_discard_max_entry = { 428 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO | S_IWUSR }, 429 .show = queue_discard_max_show, 430 .store = queue_discard_max_store, 431 }; 432 433 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = { 434 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO }, 435 .show = queue_discard_zeroes_data_show, 436 }; 437 438 static struct queue_sysfs_entry queue_write_same_max_entry = { 439 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO }, 440 .show = queue_write_same_max_show, 441 }; 442 443 static struct queue_sysfs_entry queue_nonrot_entry = { 444 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR }, 445 .show = queue_show_nonrot, 446 .store = queue_store_nonrot, 447 }; 448 449 static struct queue_sysfs_entry queue_nomerges_entry = { 450 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR }, 451 .show = queue_nomerges_show, 452 .store = queue_nomerges_store, 453 }; 454 455 static struct queue_sysfs_entry queue_rq_affinity_entry = { 456 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR }, 457 .show = queue_rq_affinity_show, 458 .store = queue_rq_affinity_store, 459 }; 460 461 static struct queue_sysfs_entry queue_iostats_entry = { 462 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR }, 463 .show = queue_show_iostats, 464 .store = queue_store_iostats, 465 }; 466 467 static struct queue_sysfs_entry queue_random_entry = { 468 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR }, 469 .show = queue_show_random, 470 .store = queue_store_random, 471 }; 472 473 static struct queue_sysfs_entry queue_poll_entry = { 474 .attr = {.name = "io_poll", .mode = S_IRUGO | S_IWUSR }, 475 .show = queue_poll_show, 476 .store = queue_poll_store, 477 }; 478 479 static struct attribute *default_attrs[] = { 480 &queue_requests_entry.attr, 481 &queue_ra_entry.attr, 482 &queue_max_hw_sectors_entry.attr, 483 &queue_max_sectors_entry.attr, 484 &queue_max_segments_entry.attr, 485 &queue_max_integrity_segments_entry.attr, 486 &queue_max_segment_size_entry.attr, 487 &queue_iosched_entry.attr, 488 &queue_hw_sector_size_entry.attr, 489 &queue_logical_block_size_entry.attr, 490 &queue_physical_block_size_entry.attr, 491 &queue_io_min_entry.attr, 492 &queue_io_opt_entry.attr, 493 &queue_discard_granularity_entry.attr, 494 &queue_discard_max_entry.attr, 495 &queue_discard_max_hw_entry.attr, 496 &queue_discard_zeroes_data_entry.attr, 497 &queue_write_same_max_entry.attr, 498 &queue_nonrot_entry.attr, 499 &queue_nomerges_entry.attr, 500 &queue_rq_affinity_entry.attr, 501 &queue_iostats_entry.attr, 502 &queue_random_entry.attr, 503 &queue_poll_entry.attr, 504 NULL, 505 }; 506 507 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) 508 509 static ssize_t 510 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 511 { 512 struct queue_sysfs_entry *entry = to_queue(attr); 513 struct request_queue *q = 514 container_of(kobj, struct request_queue, kobj); 515 ssize_t res; 516 517 if (!entry->show) 518 return -EIO; 519 mutex_lock(&q->sysfs_lock); 520 if (blk_queue_dying(q)) { 521 mutex_unlock(&q->sysfs_lock); 522 return -ENOENT; 523 } 524 res = entry->show(q, page); 525 mutex_unlock(&q->sysfs_lock); 526 return res; 527 } 528 529 static ssize_t 530 queue_attr_store(struct kobject *kobj, struct attribute *attr, 531 const char *page, size_t length) 532 { 533 struct queue_sysfs_entry *entry = to_queue(attr); 534 struct request_queue *q; 535 ssize_t res; 536 537 if (!entry->store) 538 return -EIO; 539 540 q = container_of(kobj, struct request_queue, kobj); 541 mutex_lock(&q->sysfs_lock); 542 if (blk_queue_dying(q)) { 543 mutex_unlock(&q->sysfs_lock); 544 return -ENOENT; 545 } 546 res = entry->store(q, page, length); 547 mutex_unlock(&q->sysfs_lock); 548 return res; 549 } 550 551 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 552 { 553 struct request_queue *q = container_of(rcu_head, struct request_queue, 554 rcu_head); 555 kmem_cache_free(blk_requestq_cachep, q); 556 } 557 558 /** 559 * blk_release_queue: - release a &struct request_queue when it is no longer needed 560 * @kobj: the kobj belonging to the request queue to be released 561 * 562 * Description: 563 * blk_release_queue is the pair to blk_init_queue() or 564 * blk_queue_make_request(). It should be called when a request queue is 565 * being released; typically when a block device is being de-registered. 566 * Currently, its primary task it to free all the &struct request 567 * structures that were allocated to the queue and the queue itself. 568 * 569 * Note: 570 * The low level driver must have finished any outstanding requests first 571 * via blk_cleanup_queue(). 572 **/ 573 static void blk_release_queue(struct kobject *kobj) 574 { 575 struct request_queue *q = 576 container_of(kobj, struct request_queue, kobj); 577 578 bdi_exit(&q->backing_dev_info); 579 blkcg_exit_queue(q); 580 581 if (q->elevator) { 582 spin_lock_irq(q->queue_lock); 583 ioc_clear_queue(q); 584 spin_unlock_irq(q->queue_lock); 585 elevator_exit(q->elevator); 586 } 587 588 blk_exit_rl(&q->root_rl); 589 590 if (q->queue_tags) 591 __blk_queue_free_tags(q); 592 593 if (!q->mq_ops) 594 blk_free_flush_queue(q->fq); 595 else 596 blk_mq_release(q); 597 598 blk_trace_shutdown(q); 599 600 if (q->bio_split) 601 bioset_free(q->bio_split); 602 603 ida_simple_remove(&blk_queue_ida, q->id); 604 call_rcu(&q->rcu_head, blk_free_queue_rcu); 605 } 606 607 static const struct sysfs_ops queue_sysfs_ops = { 608 .show = queue_attr_show, 609 .store = queue_attr_store, 610 }; 611 612 struct kobj_type blk_queue_ktype = { 613 .sysfs_ops = &queue_sysfs_ops, 614 .default_attrs = default_attrs, 615 .release = blk_release_queue, 616 }; 617 618 int blk_register_queue(struct gendisk *disk) 619 { 620 int ret; 621 struct device *dev = disk_to_dev(disk); 622 struct request_queue *q = disk->queue; 623 624 if (WARN_ON(!q)) 625 return -ENXIO; 626 627 /* 628 * SCSI probing may synchronously create and destroy a lot of 629 * request_queues for non-existent devices. Shutting down a fully 630 * functional queue takes measureable wallclock time as RCU grace 631 * periods are involved. To avoid excessive latency in these 632 * cases, a request_queue starts out in a degraded mode which is 633 * faster to shut down and is made fully functional here as 634 * request_queues for non-existent devices never get registered. 635 */ 636 if (!blk_queue_init_done(q)) { 637 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q); 638 percpu_ref_switch_to_percpu(&q->q_usage_counter); 639 blk_queue_bypass_end(q); 640 } 641 642 ret = blk_trace_init_sysfs(dev); 643 if (ret) 644 return ret; 645 646 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue"); 647 if (ret < 0) { 648 blk_trace_remove_sysfs(dev); 649 return ret; 650 } 651 652 kobject_uevent(&q->kobj, KOBJ_ADD); 653 654 if (q->mq_ops) 655 blk_mq_register_disk(disk); 656 657 if (!q->request_fn) 658 return 0; 659 660 ret = elv_register_queue(q); 661 if (ret) { 662 kobject_uevent(&q->kobj, KOBJ_REMOVE); 663 kobject_del(&q->kobj); 664 blk_trace_remove_sysfs(dev); 665 kobject_put(&dev->kobj); 666 return ret; 667 } 668 669 return 0; 670 } 671 672 void blk_unregister_queue(struct gendisk *disk) 673 { 674 struct request_queue *q = disk->queue; 675 676 if (WARN_ON(!q)) 677 return; 678 679 if (q->mq_ops) 680 blk_mq_unregister_disk(disk); 681 682 if (q->request_fn) 683 elv_unregister_queue(q); 684 685 kobject_uevent(&q->kobj, KOBJ_REMOVE); 686 kobject_del(&q->kobj); 687 blk_trace_remove_sysfs(disk_to_dev(disk)); 688 kobject_put(&disk_to_dev(disk)->kobj); 689 } 690