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_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_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_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 { 150 151 return sprintf(page, "%llu\n", 152 (unsigned long long)q->limits.max_hw_discard_sectors << 9); 153 } 154 155 static ssize_t queue_discard_max_show(struct request_queue *q, char *page) 156 { 157 return sprintf(page, "%llu\n", 158 (unsigned long long)q->limits.max_discard_sectors << 9); 159 } 160 161 static ssize_t queue_discard_max_store(struct request_queue *q, 162 const char *page, size_t count) 163 { 164 unsigned long max_discard; 165 ssize_t ret = queue_var_store(&max_discard, page, count); 166 167 if (ret < 0) 168 return ret; 169 170 if (max_discard & (q->limits.discard_granularity - 1)) 171 return -EINVAL; 172 173 max_discard >>= 9; 174 if (max_discard > UINT_MAX) 175 return -EINVAL; 176 177 if (max_discard > q->limits.max_hw_discard_sectors) 178 max_discard = q->limits.max_hw_discard_sectors; 179 180 q->limits.max_discard_sectors = max_discard; 181 return ret; 182 } 183 184 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page) 185 { 186 return queue_var_show(queue_discard_zeroes_data(q), page); 187 } 188 189 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page) 190 { 191 return sprintf(page, "%llu\n", 192 (unsigned long long)q->limits.max_write_same_sectors << 9); 193 } 194 195 196 static ssize_t 197 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count) 198 { 199 unsigned long max_sectors_kb, 200 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1, 201 page_kb = 1 << (PAGE_SHIFT - 10); 202 ssize_t ret = queue_var_store(&max_sectors_kb, page, count); 203 204 if (ret < 0) 205 return ret; 206 207 max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long) 208 q->limits.max_dev_sectors >> 1); 209 210 if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb) 211 return -EINVAL; 212 213 spin_lock_irq(q->queue_lock); 214 q->limits.max_sectors = max_sectors_kb << 1; 215 spin_unlock_irq(q->queue_lock); 216 217 return ret; 218 } 219 220 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page) 221 { 222 int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1; 223 224 return queue_var_show(max_hw_sectors_kb, (page)); 225 } 226 227 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg) \ 228 static ssize_t \ 229 queue_show_##name(struct request_queue *q, char *page) \ 230 { \ 231 int bit; \ 232 bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags); \ 233 return queue_var_show(neg ? !bit : bit, page); \ 234 } \ 235 static ssize_t \ 236 queue_store_##name(struct request_queue *q, const char *page, size_t count) \ 237 { \ 238 unsigned long val; \ 239 ssize_t ret; \ 240 ret = queue_var_store(&val, page, count); \ 241 if (ret < 0) \ 242 return ret; \ 243 if (neg) \ 244 val = !val; \ 245 \ 246 spin_lock_irq(q->queue_lock); \ 247 if (val) \ 248 queue_flag_set(QUEUE_FLAG_##flag, q); \ 249 else \ 250 queue_flag_clear(QUEUE_FLAG_##flag, q); \ 251 spin_unlock_irq(q->queue_lock); \ 252 return ret; \ 253 } 254 255 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1); 256 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0); 257 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0); 258 #undef QUEUE_SYSFS_BIT_FNS 259 260 static ssize_t queue_nomerges_show(struct request_queue *q, char *page) 261 { 262 return queue_var_show((blk_queue_nomerges(q) << 1) | 263 blk_queue_noxmerges(q), page); 264 } 265 266 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page, 267 size_t count) 268 { 269 unsigned long nm; 270 ssize_t ret = queue_var_store(&nm, page, count); 271 272 if (ret < 0) 273 return ret; 274 275 spin_lock_irq(q->queue_lock); 276 queue_flag_clear(QUEUE_FLAG_NOMERGES, q); 277 queue_flag_clear(QUEUE_FLAG_NOXMERGES, q); 278 if (nm == 2) 279 queue_flag_set(QUEUE_FLAG_NOMERGES, q); 280 else if (nm) 281 queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 282 spin_unlock_irq(q->queue_lock); 283 284 return ret; 285 } 286 287 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page) 288 { 289 bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags); 290 bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags); 291 292 return queue_var_show(set << force, page); 293 } 294 295 static ssize_t 296 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count) 297 { 298 ssize_t ret = -EINVAL; 299 #ifdef CONFIG_SMP 300 unsigned long val; 301 302 ret = queue_var_store(&val, page, count); 303 if (ret < 0) 304 return ret; 305 306 spin_lock_irq(q->queue_lock); 307 if (val == 2) { 308 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 309 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q); 310 } else if (val == 1) { 311 queue_flag_set(QUEUE_FLAG_SAME_COMP, q); 312 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 313 } else if (val == 0) { 314 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q); 315 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q); 316 } 317 spin_unlock_irq(q->queue_lock); 318 #endif 319 return ret; 320 } 321 322 static ssize_t queue_poll_show(struct request_queue *q, char *page) 323 { 324 return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page); 325 } 326 327 static ssize_t queue_poll_store(struct request_queue *q, const char *page, 328 size_t count) 329 { 330 unsigned long poll_on; 331 ssize_t ret; 332 333 if (!q->mq_ops || !q->mq_ops->poll) 334 return -EINVAL; 335 336 ret = queue_var_store(&poll_on, page, count); 337 if (ret < 0) 338 return ret; 339 340 spin_lock_irq(q->queue_lock); 341 if (poll_on) 342 queue_flag_set(QUEUE_FLAG_POLL, q); 343 else 344 queue_flag_clear(QUEUE_FLAG_POLL, q); 345 spin_unlock_irq(q->queue_lock); 346 347 return ret; 348 } 349 350 static struct queue_sysfs_entry queue_requests_entry = { 351 .attr = {.name = "nr_requests", .mode = S_IRUGO | S_IWUSR }, 352 .show = queue_requests_show, 353 .store = queue_requests_store, 354 }; 355 356 static struct queue_sysfs_entry queue_ra_entry = { 357 .attr = {.name = "read_ahead_kb", .mode = S_IRUGO | S_IWUSR }, 358 .show = queue_ra_show, 359 .store = queue_ra_store, 360 }; 361 362 static struct queue_sysfs_entry queue_max_sectors_entry = { 363 .attr = {.name = "max_sectors_kb", .mode = S_IRUGO | S_IWUSR }, 364 .show = queue_max_sectors_show, 365 .store = queue_max_sectors_store, 366 }; 367 368 static struct queue_sysfs_entry queue_max_hw_sectors_entry = { 369 .attr = {.name = "max_hw_sectors_kb", .mode = S_IRUGO }, 370 .show = queue_max_hw_sectors_show, 371 }; 372 373 static struct queue_sysfs_entry queue_max_segments_entry = { 374 .attr = {.name = "max_segments", .mode = S_IRUGO }, 375 .show = queue_max_segments_show, 376 }; 377 378 static struct queue_sysfs_entry queue_max_integrity_segments_entry = { 379 .attr = {.name = "max_integrity_segments", .mode = S_IRUGO }, 380 .show = queue_max_integrity_segments_show, 381 }; 382 383 static struct queue_sysfs_entry queue_max_segment_size_entry = { 384 .attr = {.name = "max_segment_size", .mode = S_IRUGO }, 385 .show = queue_max_segment_size_show, 386 }; 387 388 static struct queue_sysfs_entry queue_iosched_entry = { 389 .attr = {.name = "scheduler", .mode = S_IRUGO | S_IWUSR }, 390 .show = elv_iosched_show, 391 .store = elv_iosched_store, 392 }; 393 394 static struct queue_sysfs_entry queue_hw_sector_size_entry = { 395 .attr = {.name = "hw_sector_size", .mode = S_IRUGO }, 396 .show = queue_logical_block_size_show, 397 }; 398 399 static struct queue_sysfs_entry queue_logical_block_size_entry = { 400 .attr = {.name = "logical_block_size", .mode = S_IRUGO }, 401 .show = queue_logical_block_size_show, 402 }; 403 404 static struct queue_sysfs_entry queue_physical_block_size_entry = { 405 .attr = {.name = "physical_block_size", .mode = S_IRUGO }, 406 .show = queue_physical_block_size_show, 407 }; 408 409 static struct queue_sysfs_entry queue_io_min_entry = { 410 .attr = {.name = "minimum_io_size", .mode = S_IRUGO }, 411 .show = queue_io_min_show, 412 }; 413 414 static struct queue_sysfs_entry queue_io_opt_entry = { 415 .attr = {.name = "optimal_io_size", .mode = S_IRUGO }, 416 .show = queue_io_opt_show, 417 }; 418 419 static struct queue_sysfs_entry queue_discard_granularity_entry = { 420 .attr = {.name = "discard_granularity", .mode = S_IRUGO }, 421 .show = queue_discard_granularity_show, 422 }; 423 424 static struct queue_sysfs_entry queue_discard_max_hw_entry = { 425 .attr = {.name = "discard_max_hw_bytes", .mode = S_IRUGO }, 426 .show = queue_discard_max_hw_show, 427 }; 428 429 static struct queue_sysfs_entry queue_discard_max_entry = { 430 .attr = {.name = "discard_max_bytes", .mode = S_IRUGO | S_IWUSR }, 431 .show = queue_discard_max_show, 432 .store = queue_discard_max_store, 433 }; 434 435 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = { 436 .attr = {.name = "discard_zeroes_data", .mode = S_IRUGO }, 437 .show = queue_discard_zeroes_data_show, 438 }; 439 440 static struct queue_sysfs_entry queue_write_same_max_entry = { 441 .attr = {.name = "write_same_max_bytes", .mode = S_IRUGO }, 442 .show = queue_write_same_max_show, 443 }; 444 445 static struct queue_sysfs_entry queue_nonrot_entry = { 446 .attr = {.name = "rotational", .mode = S_IRUGO | S_IWUSR }, 447 .show = queue_show_nonrot, 448 .store = queue_store_nonrot, 449 }; 450 451 static struct queue_sysfs_entry queue_nomerges_entry = { 452 .attr = {.name = "nomerges", .mode = S_IRUGO | S_IWUSR }, 453 .show = queue_nomerges_show, 454 .store = queue_nomerges_store, 455 }; 456 457 static struct queue_sysfs_entry queue_rq_affinity_entry = { 458 .attr = {.name = "rq_affinity", .mode = S_IRUGO | S_IWUSR }, 459 .show = queue_rq_affinity_show, 460 .store = queue_rq_affinity_store, 461 }; 462 463 static struct queue_sysfs_entry queue_iostats_entry = { 464 .attr = {.name = "iostats", .mode = S_IRUGO | S_IWUSR }, 465 .show = queue_show_iostats, 466 .store = queue_store_iostats, 467 }; 468 469 static struct queue_sysfs_entry queue_random_entry = { 470 .attr = {.name = "add_random", .mode = S_IRUGO | S_IWUSR }, 471 .show = queue_show_random, 472 .store = queue_store_random, 473 }; 474 475 static struct queue_sysfs_entry queue_poll_entry = { 476 .attr = {.name = "io_poll", .mode = S_IRUGO | S_IWUSR }, 477 .show = queue_poll_show, 478 .store = queue_poll_store, 479 }; 480 481 static struct attribute *default_attrs[] = { 482 &queue_requests_entry.attr, 483 &queue_ra_entry.attr, 484 &queue_max_hw_sectors_entry.attr, 485 &queue_max_sectors_entry.attr, 486 &queue_max_segments_entry.attr, 487 &queue_max_integrity_segments_entry.attr, 488 &queue_max_segment_size_entry.attr, 489 &queue_iosched_entry.attr, 490 &queue_hw_sector_size_entry.attr, 491 &queue_logical_block_size_entry.attr, 492 &queue_physical_block_size_entry.attr, 493 &queue_io_min_entry.attr, 494 &queue_io_opt_entry.attr, 495 &queue_discard_granularity_entry.attr, 496 &queue_discard_max_entry.attr, 497 &queue_discard_max_hw_entry.attr, 498 &queue_discard_zeroes_data_entry.attr, 499 &queue_write_same_max_entry.attr, 500 &queue_nonrot_entry.attr, 501 &queue_nomerges_entry.attr, 502 &queue_rq_affinity_entry.attr, 503 &queue_iostats_entry.attr, 504 &queue_random_entry.attr, 505 &queue_poll_entry.attr, 506 NULL, 507 }; 508 509 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr) 510 511 static ssize_t 512 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page) 513 { 514 struct queue_sysfs_entry *entry = to_queue(attr); 515 struct request_queue *q = 516 container_of(kobj, struct request_queue, kobj); 517 ssize_t res; 518 519 if (!entry->show) 520 return -EIO; 521 mutex_lock(&q->sysfs_lock); 522 if (blk_queue_dying(q)) { 523 mutex_unlock(&q->sysfs_lock); 524 return -ENOENT; 525 } 526 res = entry->show(q, page); 527 mutex_unlock(&q->sysfs_lock); 528 return res; 529 } 530 531 static ssize_t 532 queue_attr_store(struct kobject *kobj, struct attribute *attr, 533 const char *page, size_t length) 534 { 535 struct queue_sysfs_entry *entry = to_queue(attr); 536 struct request_queue *q; 537 ssize_t res; 538 539 if (!entry->store) 540 return -EIO; 541 542 q = container_of(kobj, struct request_queue, kobj); 543 mutex_lock(&q->sysfs_lock); 544 if (blk_queue_dying(q)) { 545 mutex_unlock(&q->sysfs_lock); 546 return -ENOENT; 547 } 548 res = entry->store(q, page, length); 549 mutex_unlock(&q->sysfs_lock); 550 return res; 551 } 552 553 static void blk_free_queue_rcu(struct rcu_head *rcu_head) 554 { 555 struct request_queue *q = container_of(rcu_head, struct request_queue, 556 rcu_head); 557 kmem_cache_free(blk_requestq_cachep, q); 558 } 559 560 /** 561 * blk_release_queue: - release a &struct request_queue when it is no longer needed 562 * @kobj: the kobj belonging to the request queue to be released 563 * 564 * Description: 565 * blk_release_queue is the pair to blk_init_queue() or 566 * blk_queue_make_request(). It should be called when a request queue is 567 * being released; typically when a block device is being de-registered. 568 * Currently, its primary task it to free all the &struct request 569 * structures that were allocated to the queue and the queue itself. 570 * 571 * Note: 572 * The low level driver must have finished any outstanding requests first 573 * via blk_cleanup_queue(). 574 **/ 575 static void blk_release_queue(struct kobject *kobj) 576 { 577 struct request_queue *q = 578 container_of(kobj, struct request_queue, kobj); 579 580 bdi_exit(&q->backing_dev_info); 581 blkcg_exit_queue(q); 582 583 if (q->elevator) { 584 spin_lock_irq(q->queue_lock); 585 ioc_clear_queue(q); 586 spin_unlock_irq(q->queue_lock); 587 elevator_exit(q->elevator); 588 } 589 590 blk_exit_rl(&q->root_rl); 591 592 if (q->queue_tags) 593 __blk_queue_free_tags(q); 594 595 if (!q->mq_ops) 596 blk_free_flush_queue(q->fq); 597 else 598 blk_mq_release(q); 599 600 blk_trace_shutdown(q); 601 602 if (q->bio_split) 603 bioset_free(q->bio_split); 604 605 ida_simple_remove(&blk_queue_ida, q->id); 606 call_rcu(&q->rcu_head, blk_free_queue_rcu); 607 } 608 609 static const struct sysfs_ops queue_sysfs_ops = { 610 .show = queue_attr_show, 611 .store = queue_attr_store, 612 }; 613 614 struct kobj_type blk_queue_ktype = { 615 .sysfs_ops = &queue_sysfs_ops, 616 .default_attrs = default_attrs, 617 .release = blk_release_queue, 618 }; 619 620 int blk_register_queue(struct gendisk *disk) 621 { 622 int ret; 623 struct device *dev = disk_to_dev(disk); 624 struct request_queue *q = disk->queue; 625 626 if (WARN_ON(!q)) 627 return -ENXIO; 628 629 /* 630 * SCSI probing may synchronously create and destroy a lot of 631 * request_queues for non-existent devices. Shutting down a fully 632 * functional queue takes measureable wallclock time as RCU grace 633 * periods are involved. To avoid excessive latency in these 634 * cases, a request_queue starts out in a degraded mode which is 635 * faster to shut down and is made fully functional here as 636 * request_queues for non-existent devices never get registered. 637 */ 638 if (!blk_queue_init_done(q)) { 639 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q); 640 percpu_ref_switch_to_percpu(&q->q_usage_counter); 641 blk_queue_bypass_end(q); 642 } 643 644 ret = blk_trace_init_sysfs(dev); 645 if (ret) 646 return ret; 647 648 ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue"); 649 if (ret < 0) { 650 blk_trace_remove_sysfs(dev); 651 return ret; 652 } 653 654 kobject_uevent(&q->kobj, KOBJ_ADD); 655 656 if (q->mq_ops) 657 blk_mq_register_disk(disk); 658 659 if (!q->request_fn) 660 return 0; 661 662 ret = elv_register_queue(q); 663 if (ret) { 664 kobject_uevent(&q->kobj, KOBJ_REMOVE); 665 kobject_del(&q->kobj); 666 blk_trace_remove_sysfs(dev); 667 kobject_put(&dev->kobj); 668 return ret; 669 } 670 671 return 0; 672 } 673 674 void blk_unregister_queue(struct gendisk *disk) 675 { 676 struct request_queue *q = disk->queue; 677 678 if (WARN_ON(!q)) 679 return; 680 681 if (q->mq_ops) 682 blk_mq_unregister_disk(disk); 683 684 if (q->request_fn) 685 elv_unregister_queue(q); 686 687 kobject_uevent(&q->kobj, KOBJ_REMOVE); 688 kobject_del(&q->kobj); 689 blk_trace_remove_sysfs(disk_to_dev(disk)); 690 kobject_put(&disk_to_dev(disk)->kobj); 691 } 692