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