1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * gendisk handling 4 * 5 * Portions Copyright (C) 2020 Christoph Hellwig 6 */ 7 8 #include <linux/module.h> 9 #include <linux/ctype.h> 10 #include <linux/fs.h> 11 #include <linux/genhd.h> 12 #include <linux/kdev_t.h> 13 #include <linux/kernel.h> 14 #include <linux/blkdev.h> 15 #include <linux/backing-dev.h> 16 #include <linux/init.h> 17 #include <linux/spinlock.h> 18 #include <linux/proc_fs.h> 19 #include <linux/seq_file.h> 20 #include <linux/slab.h> 21 #include <linux/kmod.h> 22 #include <linux/mutex.h> 23 #include <linux/idr.h> 24 #include <linux/log2.h> 25 #include <linux/pm_runtime.h> 26 #include <linux/badblocks.h> 27 28 #include "blk.h" 29 30 static struct kobject *block_depr; 31 32 /* 33 * Unique, monotonically increasing sequential number associated with block 34 * devices instances (i.e. incremented each time a device is attached). 35 * Associating uevents with block devices in userspace is difficult and racy: 36 * the uevent netlink socket is lossy, and on slow and overloaded systems has 37 * a very high latency. 38 * Block devices do not have exclusive owners in userspace, any process can set 39 * one up (e.g. loop devices). Moreover, device names can be reused (e.g. loop0 40 * can be reused again and again). 41 * A userspace process setting up a block device and watching for its events 42 * cannot thus reliably tell whether an event relates to the device it just set 43 * up or another earlier instance with the same name. 44 * This sequential number allows userspace processes to solve this problem, and 45 * uniquely associate an uevent to the lifetime to a device. 46 */ 47 static atomic64_t diskseq; 48 49 /* for extended dynamic devt allocation, currently only one major is used */ 50 #define NR_EXT_DEVT (1 << MINORBITS) 51 static DEFINE_IDA(ext_devt_ida); 52 53 void set_capacity(struct gendisk *disk, sector_t sectors) 54 { 55 struct block_device *bdev = disk->part0; 56 57 spin_lock(&bdev->bd_size_lock); 58 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 59 spin_unlock(&bdev->bd_size_lock); 60 } 61 EXPORT_SYMBOL(set_capacity); 62 63 /* 64 * Set disk capacity and notify if the size is not currently zero and will not 65 * be set to zero. Returns true if a uevent was sent, otherwise false. 66 */ 67 bool set_capacity_and_notify(struct gendisk *disk, sector_t size) 68 { 69 sector_t capacity = get_capacity(disk); 70 char *envp[] = { "RESIZE=1", NULL }; 71 72 set_capacity(disk, size); 73 74 /* 75 * Only print a message and send a uevent if the gendisk is user visible 76 * and alive. This avoids spamming the log and udev when setting the 77 * initial capacity during probing. 78 */ 79 if (size == capacity || 80 !disk_live(disk) || 81 (disk->flags & GENHD_FL_HIDDEN)) 82 return false; 83 84 pr_info("%s: detected capacity change from %lld to %lld\n", 85 disk->disk_name, capacity, size); 86 87 /* 88 * Historically we did not send a uevent for changes to/from an empty 89 * device. 90 */ 91 if (!capacity || !size) 92 return false; 93 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp); 94 return true; 95 } 96 EXPORT_SYMBOL_GPL(set_capacity_and_notify); 97 98 /* 99 * Format the device name of the indicated block device into the supplied buffer 100 * and return a pointer to that same buffer for convenience. 101 * 102 * Note: do not use this in new code, use the %pg specifier to sprintf and 103 * printk insted. 104 */ 105 const char *bdevname(struct block_device *bdev, char *buf) 106 { 107 struct gendisk *hd = bdev->bd_disk; 108 int partno = bdev->bd_partno; 109 110 if (!partno) 111 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name); 112 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1])) 113 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno); 114 else 115 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno); 116 117 return buf; 118 } 119 EXPORT_SYMBOL(bdevname); 120 121 static void part_stat_read_all(struct block_device *part, 122 struct disk_stats *stat) 123 { 124 int cpu; 125 126 memset(stat, 0, sizeof(struct disk_stats)); 127 for_each_possible_cpu(cpu) { 128 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu); 129 int group; 130 131 for (group = 0; group < NR_STAT_GROUPS; group++) { 132 stat->nsecs[group] += ptr->nsecs[group]; 133 stat->sectors[group] += ptr->sectors[group]; 134 stat->ios[group] += ptr->ios[group]; 135 stat->merges[group] += ptr->merges[group]; 136 } 137 138 stat->io_ticks += ptr->io_ticks; 139 } 140 } 141 142 static unsigned int part_in_flight(struct block_device *part) 143 { 144 unsigned int inflight = 0; 145 int cpu; 146 147 for_each_possible_cpu(cpu) { 148 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) + 149 part_stat_local_read_cpu(part, in_flight[1], cpu); 150 } 151 if ((int)inflight < 0) 152 inflight = 0; 153 154 return inflight; 155 } 156 157 static void part_in_flight_rw(struct block_device *part, 158 unsigned int inflight[2]) 159 { 160 int cpu; 161 162 inflight[0] = 0; 163 inflight[1] = 0; 164 for_each_possible_cpu(cpu) { 165 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu); 166 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu); 167 } 168 if ((int)inflight[0] < 0) 169 inflight[0] = 0; 170 if ((int)inflight[1] < 0) 171 inflight[1] = 0; 172 } 173 174 /* 175 * Can be deleted altogether. Later. 176 * 177 */ 178 #define BLKDEV_MAJOR_HASH_SIZE 255 179 static struct blk_major_name { 180 struct blk_major_name *next; 181 int major; 182 char name[16]; 183 void (*probe)(dev_t devt); 184 } *major_names[BLKDEV_MAJOR_HASH_SIZE]; 185 static DEFINE_MUTEX(major_names_lock); 186 187 /* index in the above - for now: assume no multimajor ranges */ 188 static inline int major_to_index(unsigned major) 189 { 190 return major % BLKDEV_MAJOR_HASH_SIZE; 191 } 192 193 #ifdef CONFIG_PROC_FS 194 void blkdev_show(struct seq_file *seqf, off_t offset) 195 { 196 struct blk_major_name *dp; 197 198 mutex_lock(&major_names_lock); 199 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next) 200 if (dp->major == offset) 201 seq_printf(seqf, "%3d %s\n", dp->major, dp->name); 202 mutex_unlock(&major_names_lock); 203 } 204 #endif /* CONFIG_PROC_FS */ 205 206 /** 207 * __register_blkdev - register a new block device 208 * 209 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If 210 * @major = 0, try to allocate any unused major number. 211 * @name: the name of the new block device as a zero terminated string 212 * @probe: allback that is called on access to any minor number of @major 213 * 214 * The @name must be unique within the system. 215 * 216 * The return value depends on the @major input parameter: 217 * 218 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1] 219 * then the function returns zero on success, or a negative error code 220 * - if any unused major number was requested with @major = 0 parameter 221 * then the return value is the allocated major number in range 222 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise 223 * 224 * See Documentation/admin-guide/devices.txt for the list of allocated 225 * major numbers. 226 * 227 * Use register_blkdev instead for any new code. 228 */ 229 int __register_blkdev(unsigned int major, const char *name, 230 void (*probe)(dev_t devt)) 231 { 232 struct blk_major_name **n, *p; 233 int index, ret = 0; 234 235 mutex_lock(&major_names_lock); 236 237 /* temporary */ 238 if (major == 0) { 239 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) { 240 if (major_names[index] == NULL) 241 break; 242 } 243 244 if (index == 0) { 245 printk("%s: failed to get major for %s\n", 246 __func__, name); 247 ret = -EBUSY; 248 goto out; 249 } 250 major = index; 251 ret = major; 252 } 253 254 if (major >= BLKDEV_MAJOR_MAX) { 255 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n", 256 __func__, major, BLKDEV_MAJOR_MAX-1, name); 257 258 ret = -EINVAL; 259 goto out; 260 } 261 262 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL); 263 if (p == NULL) { 264 ret = -ENOMEM; 265 goto out; 266 } 267 268 p->major = major; 269 p->probe = probe; 270 strlcpy(p->name, name, sizeof(p->name)); 271 p->next = NULL; 272 index = major_to_index(major); 273 274 for (n = &major_names[index]; *n; n = &(*n)->next) { 275 if ((*n)->major == major) 276 break; 277 } 278 if (!*n) 279 *n = p; 280 else 281 ret = -EBUSY; 282 283 if (ret < 0) { 284 printk("register_blkdev: cannot get major %u for %s\n", 285 major, name); 286 kfree(p); 287 } 288 out: 289 mutex_unlock(&major_names_lock); 290 return ret; 291 } 292 EXPORT_SYMBOL(__register_blkdev); 293 294 void unregister_blkdev(unsigned int major, const char *name) 295 { 296 struct blk_major_name **n; 297 struct blk_major_name *p = NULL; 298 int index = major_to_index(major); 299 300 mutex_lock(&major_names_lock); 301 for (n = &major_names[index]; *n; n = &(*n)->next) 302 if ((*n)->major == major) 303 break; 304 if (!*n || strcmp((*n)->name, name)) { 305 WARN_ON(1); 306 } else { 307 p = *n; 308 *n = p->next; 309 } 310 mutex_unlock(&major_names_lock); 311 kfree(p); 312 } 313 314 EXPORT_SYMBOL(unregister_blkdev); 315 316 int blk_alloc_ext_minor(void) 317 { 318 int idx; 319 320 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT, GFP_KERNEL); 321 if (idx == -ENOSPC) 322 return -EBUSY; 323 return idx; 324 } 325 326 void blk_free_ext_minor(unsigned int minor) 327 { 328 ida_free(&ext_devt_ida, minor); 329 } 330 331 static char *bdevt_str(dev_t devt, char *buf) 332 { 333 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) { 334 char tbuf[BDEVT_SIZE]; 335 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt)); 336 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf); 337 } else 338 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt)); 339 340 return buf; 341 } 342 343 void disk_uevent(struct gendisk *disk, enum kobject_action action) 344 { 345 struct block_device *part; 346 unsigned long idx; 347 348 rcu_read_lock(); 349 xa_for_each(&disk->part_tbl, idx, part) { 350 if (bdev_is_partition(part) && !bdev_nr_sectors(part)) 351 continue; 352 if (!kobject_get_unless_zero(&part->bd_device.kobj)) 353 continue; 354 355 rcu_read_unlock(); 356 kobject_uevent(bdev_kobj(part), action); 357 put_device(&part->bd_device); 358 rcu_read_lock(); 359 } 360 rcu_read_unlock(); 361 } 362 EXPORT_SYMBOL_GPL(disk_uevent); 363 364 static void disk_scan_partitions(struct gendisk *disk) 365 { 366 struct block_device *bdev; 367 368 if (!get_capacity(disk) || !disk_part_scan_enabled(disk)) 369 return; 370 371 set_bit(GD_NEED_PART_SCAN, &disk->state); 372 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL); 373 if (!IS_ERR(bdev)) 374 blkdev_put(bdev, FMODE_READ); 375 } 376 377 /** 378 * device_add_disk - add disk information to kernel list 379 * @parent: parent device for the disk 380 * @disk: per-device partitioning information 381 * @groups: Additional per-device sysfs groups 382 * 383 * This function registers the partitioning information in @disk 384 * with the kernel. 385 */ 386 int device_add_disk(struct device *parent, struct gendisk *disk, 387 const struct attribute_group **groups) 388 389 { 390 struct device *ddev = disk_to_dev(disk); 391 int ret; 392 393 /* 394 * The disk queue should now be all set with enough information about 395 * the device for the elevator code to pick an adequate default 396 * elevator if one is needed, that is, for devices requesting queue 397 * registration. 398 */ 399 elevator_init_mq(disk->queue); 400 401 /* 402 * If the driver provides an explicit major number it also must provide 403 * the number of minors numbers supported, and those will be used to 404 * setup the gendisk. 405 * Otherwise just allocate the device numbers for both the whole device 406 * and all partitions from the extended dev_t space. 407 */ 408 if (disk->major) { 409 if (WARN_ON(!disk->minors)) 410 return -EINVAL; 411 412 if (disk->minors > DISK_MAX_PARTS) { 413 pr_err("block: can't allocate more than %d partitions\n", 414 DISK_MAX_PARTS); 415 disk->minors = DISK_MAX_PARTS; 416 } 417 } else { 418 if (WARN_ON(disk->minors)) 419 return -EINVAL; 420 421 ret = blk_alloc_ext_minor(); 422 if (ret < 0) 423 return ret; 424 disk->major = BLOCK_EXT_MAJOR; 425 disk->first_minor = ret; 426 disk->flags |= GENHD_FL_EXT_DEVT; 427 } 428 429 ret = disk_alloc_events(disk); 430 if (ret) 431 goto out_free_ext_minor; 432 433 /* delay uevents, until we scanned partition table */ 434 dev_set_uevent_suppress(ddev, 1); 435 436 ddev->parent = parent; 437 ddev->groups = groups; 438 dev_set_name(ddev, "%s", disk->disk_name); 439 if (!(disk->flags & GENHD_FL_HIDDEN)) 440 ddev->devt = MKDEV(disk->major, disk->first_minor); 441 ret = device_add(ddev); 442 if (ret) 443 goto out_disk_release_events; 444 if (!sysfs_deprecated) { 445 ret = sysfs_create_link(block_depr, &ddev->kobj, 446 kobject_name(&ddev->kobj)); 447 if (ret) 448 goto out_device_del; 449 } 450 451 /* 452 * avoid probable deadlock caused by allocating memory with 453 * GFP_KERNEL in runtime_resume callback of its all ancestor 454 * devices 455 */ 456 pm_runtime_set_memalloc_noio(ddev, true); 457 458 ret = blk_integrity_add(disk); 459 if (ret) 460 goto out_del_block_link; 461 462 disk->part0->bd_holder_dir = 463 kobject_create_and_add("holders", &ddev->kobj); 464 if (!disk->part0->bd_holder_dir) 465 goto out_del_integrity; 466 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj); 467 if (!disk->slave_dir) 468 goto out_put_holder_dir; 469 470 ret = bd_register_pending_holders(disk); 471 if (ret < 0) 472 goto out_put_slave_dir; 473 474 ret = blk_register_queue(disk); 475 if (ret) 476 goto out_put_slave_dir; 477 478 if (disk->flags & GENHD_FL_HIDDEN) { 479 /* 480 * Don't let hidden disks show up in /proc/partitions, 481 * and don't bother scanning for partitions either. 482 */ 483 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO; 484 disk->flags |= GENHD_FL_NO_PART_SCAN; 485 } else { 486 ret = bdi_register(disk->bdi, "%u:%u", 487 disk->major, disk->first_minor); 488 if (ret) 489 goto out_unregister_queue; 490 bdi_set_owner(disk->bdi, ddev); 491 ret = sysfs_create_link(&ddev->kobj, 492 &disk->bdi->dev->kobj, "bdi"); 493 if (ret) 494 goto out_unregister_bdi; 495 496 bdev_add(disk->part0, ddev->devt); 497 disk_scan_partitions(disk); 498 499 /* 500 * Announce the disk and partitions after all partitions are 501 * created. (for hidden disks uevents remain suppressed forever) 502 */ 503 dev_set_uevent_suppress(ddev, 0); 504 disk_uevent(disk, KOBJ_ADD); 505 } 506 507 disk_update_readahead(disk); 508 disk_add_events(disk); 509 return 0; 510 511 out_unregister_bdi: 512 if (!(disk->flags & GENHD_FL_HIDDEN)) 513 bdi_unregister(disk->bdi); 514 out_unregister_queue: 515 blk_unregister_queue(disk); 516 out_put_slave_dir: 517 kobject_put(disk->slave_dir); 518 out_put_holder_dir: 519 kobject_put(disk->part0->bd_holder_dir); 520 out_del_integrity: 521 blk_integrity_del(disk); 522 out_del_block_link: 523 if (!sysfs_deprecated) 524 sysfs_remove_link(block_depr, dev_name(ddev)); 525 out_device_del: 526 device_del(ddev); 527 out_disk_release_events: 528 disk_release_events(disk); 529 out_free_ext_minor: 530 if (disk->major == BLOCK_EXT_MAJOR) 531 blk_free_ext_minor(disk->first_minor); 532 return WARN_ON_ONCE(ret); /* keep until all callers handle errors */ 533 } 534 EXPORT_SYMBOL(device_add_disk); 535 536 /** 537 * del_gendisk - remove the gendisk 538 * @disk: the struct gendisk to remove 539 * 540 * Removes the gendisk and all its associated resources. This deletes the 541 * partitions associated with the gendisk, and unregisters the associated 542 * request_queue. 543 * 544 * This is the counter to the respective __device_add_disk() call. 545 * 546 * The final removal of the struct gendisk happens when its refcount reaches 0 547 * with put_disk(), which should be called after del_gendisk(), if 548 * __device_add_disk() was used. 549 * 550 * Drivers exist which depend on the release of the gendisk to be synchronous, 551 * it should not be deferred. 552 * 553 * Context: can sleep 554 */ 555 void del_gendisk(struct gendisk *disk) 556 { 557 might_sleep(); 558 559 if (WARN_ON_ONCE(!disk_live(disk) && !(disk->flags & GENHD_FL_HIDDEN))) 560 return; 561 562 blk_integrity_del(disk); 563 disk_del_events(disk); 564 565 mutex_lock(&disk->open_mutex); 566 remove_inode_hash(disk->part0->bd_inode); 567 blk_drop_partitions(disk); 568 mutex_unlock(&disk->open_mutex); 569 570 fsync_bdev(disk->part0); 571 __invalidate_device(disk->part0, true); 572 573 set_capacity(disk, 0); 574 575 if (!(disk->flags & GENHD_FL_HIDDEN)) { 576 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi"); 577 578 /* 579 * Unregister bdi before releasing device numbers (as they can 580 * get reused and we'd get clashes in sysfs). 581 */ 582 bdi_unregister(disk->bdi); 583 } 584 585 blk_unregister_queue(disk); 586 587 kobject_put(disk->part0->bd_holder_dir); 588 kobject_put(disk->slave_dir); 589 590 part_stat_set_all(disk->part0, 0); 591 disk->part0->bd_stamp = 0; 592 if (!sysfs_deprecated) 593 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk))); 594 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false); 595 device_del(disk_to_dev(disk)); 596 } 597 EXPORT_SYMBOL(del_gendisk); 598 599 /* sysfs access to bad-blocks list. */ 600 static ssize_t disk_badblocks_show(struct device *dev, 601 struct device_attribute *attr, 602 char *page) 603 { 604 struct gendisk *disk = dev_to_disk(dev); 605 606 if (!disk->bb) 607 return sprintf(page, "\n"); 608 609 return badblocks_show(disk->bb, page, 0); 610 } 611 612 static ssize_t disk_badblocks_store(struct device *dev, 613 struct device_attribute *attr, 614 const char *page, size_t len) 615 { 616 struct gendisk *disk = dev_to_disk(dev); 617 618 if (!disk->bb) 619 return -ENXIO; 620 621 return badblocks_store(disk->bb, page, len, 0); 622 } 623 624 void blk_request_module(dev_t devt) 625 { 626 unsigned int major = MAJOR(devt); 627 struct blk_major_name **n; 628 629 mutex_lock(&major_names_lock); 630 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) { 631 if ((*n)->major == major && (*n)->probe) { 632 (*n)->probe(devt); 633 mutex_unlock(&major_names_lock); 634 return; 635 } 636 } 637 mutex_unlock(&major_names_lock); 638 639 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0) 640 /* Make old-style 2.4 aliases work */ 641 request_module("block-major-%d", MAJOR(devt)); 642 } 643 644 /* 645 * print a full list of all partitions - intended for places where the root 646 * filesystem can't be mounted and thus to give the victim some idea of what 647 * went wrong 648 */ 649 void __init printk_all_partitions(void) 650 { 651 struct class_dev_iter iter; 652 struct device *dev; 653 654 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 655 while ((dev = class_dev_iter_next(&iter))) { 656 struct gendisk *disk = dev_to_disk(dev); 657 struct block_device *part; 658 char devt_buf[BDEVT_SIZE]; 659 unsigned long idx; 660 661 /* 662 * Don't show empty devices or things that have been 663 * suppressed 664 */ 665 if (get_capacity(disk) == 0 || 666 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) 667 continue; 668 669 /* 670 * Note, unlike /proc/partitions, I am showing the numbers in 671 * hex - the same format as the root= option takes. 672 */ 673 rcu_read_lock(); 674 xa_for_each(&disk->part_tbl, idx, part) { 675 if (!bdev_nr_sectors(part)) 676 continue; 677 printk("%s%s %10llu %pg %s", 678 bdev_is_partition(part) ? " " : "", 679 bdevt_str(part->bd_dev, devt_buf), 680 bdev_nr_sectors(part) >> 1, part, 681 part->bd_meta_info ? 682 part->bd_meta_info->uuid : ""); 683 if (bdev_is_partition(part)) 684 printk("\n"); 685 else if (dev->parent && dev->parent->driver) 686 printk(" driver: %s\n", 687 dev->parent->driver->name); 688 else 689 printk(" (driver?)\n"); 690 } 691 rcu_read_unlock(); 692 } 693 class_dev_iter_exit(&iter); 694 } 695 696 #ifdef CONFIG_PROC_FS 697 /* iterator */ 698 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos) 699 { 700 loff_t skip = *pos; 701 struct class_dev_iter *iter; 702 struct device *dev; 703 704 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 705 if (!iter) 706 return ERR_PTR(-ENOMEM); 707 708 seqf->private = iter; 709 class_dev_iter_init(iter, &block_class, NULL, &disk_type); 710 do { 711 dev = class_dev_iter_next(iter); 712 if (!dev) 713 return NULL; 714 } while (skip--); 715 716 return dev_to_disk(dev); 717 } 718 719 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos) 720 { 721 struct device *dev; 722 723 (*pos)++; 724 dev = class_dev_iter_next(seqf->private); 725 if (dev) 726 return dev_to_disk(dev); 727 728 return NULL; 729 } 730 731 static void disk_seqf_stop(struct seq_file *seqf, void *v) 732 { 733 struct class_dev_iter *iter = seqf->private; 734 735 /* stop is called even after start failed :-( */ 736 if (iter) { 737 class_dev_iter_exit(iter); 738 kfree(iter); 739 seqf->private = NULL; 740 } 741 } 742 743 static void *show_partition_start(struct seq_file *seqf, loff_t *pos) 744 { 745 void *p; 746 747 p = disk_seqf_start(seqf, pos); 748 if (!IS_ERR_OR_NULL(p) && !*pos) 749 seq_puts(seqf, "major minor #blocks name\n\n"); 750 return p; 751 } 752 753 static int show_partition(struct seq_file *seqf, void *v) 754 { 755 struct gendisk *sgp = v; 756 struct block_device *part; 757 unsigned long idx; 758 759 /* Don't show non-partitionable removeable devices or empty devices */ 760 if (!get_capacity(sgp) || (!disk_max_parts(sgp) && 761 (sgp->flags & GENHD_FL_REMOVABLE))) 762 return 0; 763 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO) 764 return 0; 765 766 rcu_read_lock(); 767 xa_for_each(&sgp->part_tbl, idx, part) { 768 if (!bdev_nr_sectors(part)) 769 continue; 770 seq_printf(seqf, "%4d %7d %10llu %pg\n", 771 MAJOR(part->bd_dev), MINOR(part->bd_dev), 772 bdev_nr_sectors(part) >> 1, part); 773 } 774 rcu_read_unlock(); 775 return 0; 776 } 777 778 static const struct seq_operations partitions_op = { 779 .start = show_partition_start, 780 .next = disk_seqf_next, 781 .stop = disk_seqf_stop, 782 .show = show_partition 783 }; 784 #endif 785 786 static int __init genhd_device_init(void) 787 { 788 int error; 789 790 block_class.dev_kobj = sysfs_dev_block_kobj; 791 error = class_register(&block_class); 792 if (unlikely(error)) 793 return error; 794 blk_dev_init(); 795 796 register_blkdev(BLOCK_EXT_MAJOR, "blkext"); 797 798 /* create top-level block dir */ 799 if (!sysfs_deprecated) 800 block_depr = kobject_create_and_add("block", NULL); 801 return 0; 802 } 803 804 subsys_initcall(genhd_device_init); 805 806 static ssize_t disk_range_show(struct device *dev, 807 struct device_attribute *attr, char *buf) 808 { 809 struct gendisk *disk = dev_to_disk(dev); 810 811 return sprintf(buf, "%d\n", disk->minors); 812 } 813 814 static ssize_t disk_ext_range_show(struct device *dev, 815 struct device_attribute *attr, char *buf) 816 { 817 struct gendisk *disk = dev_to_disk(dev); 818 819 return sprintf(buf, "%d\n", disk_max_parts(disk)); 820 } 821 822 static ssize_t disk_removable_show(struct device *dev, 823 struct device_attribute *attr, char *buf) 824 { 825 struct gendisk *disk = dev_to_disk(dev); 826 827 return sprintf(buf, "%d\n", 828 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0)); 829 } 830 831 static ssize_t disk_hidden_show(struct device *dev, 832 struct device_attribute *attr, char *buf) 833 { 834 struct gendisk *disk = dev_to_disk(dev); 835 836 return sprintf(buf, "%d\n", 837 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0)); 838 } 839 840 static ssize_t disk_ro_show(struct device *dev, 841 struct device_attribute *attr, char *buf) 842 { 843 struct gendisk *disk = dev_to_disk(dev); 844 845 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0); 846 } 847 848 ssize_t part_size_show(struct device *dev, 849 struct device_attribute *attr, char *buf) 850 { 851 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev))); 852 } 853 854 ssize_t part_stat_show(struct device *dev, 855 struct device_attribute *attr, char *buf) 856 { 857 struct block_device *bdev = dev_to_bdev(dev); 858 struct request_queue *q = bdev->bd_disk->queue; 859 struct disk_stats stat; 860 unsigned int inflight; 861 862 part_stat_read_all(bdev, &stat); 863 if (queue_is_mq(q)) 864 inflight = blk_mq_in_flight(q, bdev); 865 else 866 inflight = part_in_flight(bdev); 867 868 return sprintf(buf, 869 "%8lu %8lu %8llu %8u " 870 "%8lu %8lu %8llu %8u " 871 "%8u %8u %8u " 872 "%8lu %8lu %8llu %8u " 873 "%8lu %8u" 874 "\n", 875 stat.ios[STAT_READ], 876 stat.merges[STAT_READ], 877 (unsigned long long)stat.sectors[STAT_READ], 878 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC), 879 stat.ios[STAT_WRITE], 880 stat.merges[STAT_WRITE], 881 (unsigned long long)stat.sectors[STAT_WRITE], 882 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC), 883 inflight, 884 jiffies_to_msecs(stat.io_ticks), 885 (unsigned int)div_u64(stat.nsecs[STAT_READ] + 886 stat.nsecs[STAT_WRITE] + 887 stat.nsecs[STAT_DISCARD] + 888 stat.nsecs[STAT_FLUSH], 889 NSEC_PER_MSEC), 890 stat.ios[STAT_DISCARD], 891 stat.merges[STAT_DISCARD], 892 (unsigned long long)stat.sectors[STAT_DISCARD], 893 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC), 894 stat.ios[STAT_FLUSH], 895 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC)); 896 } 897 898 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr, 899 char *buf) 900 { 901 struct block_device *bdev = dev_to_bdev(dev); 902 struct request_queue *q = bdev->bd_disk->queue; 903 unsigned int inflight[2]; 904 905 if (queue_is_mq(q)) 906 blk_mq_in_flight_rw(q, bdev, inflight); 907 else 908 part_in_flight_rw(bdev, inflight); 909 910 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]); 911 } 912 913 static ssize_t disk_capability_show(struct device *dev, 914 struct device_attribute *attr, char *buf) 915 { 916 struct gendisk *disk = dev_to_disk(dev); 917 918 return sprintf(buf, "%x\n", disk->flags); 919 } 920 921 static ssize_t disk_alignment_offset_show(struct device *dev, 922 struct device_attribute *attr, 923 char *buf) 924 { 925 struct gendisk *disk = dev_to_disk(dev); 926 927 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue)); 928 } 929 930 static ssize_t disk_discard_alignment_show(struct device *dev, 931 struct device_attribute *attr, 932 char *buf) 933 { 934 struct gendisk *disk = dev_to_disk(dev); 935 936 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue)); 937 } 938 939 static ssize_t diskseq_show(struct device *dev, 940 struct device_attribute *attr, char *buf) 941 { 942 struct gendisk *disk = dev_to_disk(dev); 943 944 return sprintf(buf, "%llu\n", disk->diskseq); 945 } 946 947 static DEVICE_ATTR(range, 0444, disk_range_show, NULL); 948 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL); 949 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL); 950 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL); 951 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL); 952 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 953 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL); 954 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL); 955 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL); 956 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 957 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 958 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store); 959 static DEVICE_ATTR(diskseq, 0444, diskseq_show, NULL); 960 961 #ifdef CONFIG_FAIL_MAKE_REQUEST 962 ssize_t part_fail_show(struct device *dev, 963 struct device_attribute *attr, char *buf) 964 { 965 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail); 966 } 967 968 ssize_t part_fail_store(struct device *dev, 969 struct device_attribute *attr, 970 const char *buf, size_t count) 971 { 972 int i; 973 974 if (count > 0 && sscanf(buf, "%d", &i) > 0) 975 dev_to_bdev(dev)->bd_make_it_fail = i; 976 977 return count; 978 } 979 980 static struct device_attribute dev_attr_fail = 981 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 982 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 983 984 #ifdef CONFIG_FAIL_IO_TIMEOUT 985 static struct device_attribute dev_attr_fail_timeout = 986 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store); 987 #endif 988 989 static struct attribute *disk_attrs[] = { 990 &dev_attr_range.attr, 991 &dev_attr_ext_range.attr, 992 &dev_attr_removable.attr, 993 &dev_attr_hidden.attr, 994 &dev_attr_ro.attr, 995 &dev_attr_size.attr, 996 &dev_attr_alignment_offset.attr, 997 &dev_attr_discard_alignment.attr, 998 &dev_attr_capability.attr, 999 &dev_attr_stat.attr, 1000 &dev_attr_inflight.attr, 1001 &dev_attr_badblocks.attr, 1002 &dev_attr_events.attr, 1003 &dev_attr_events_async.attr, 1004 &dev_attr_events_poll_msecs.attr, 1005 &dev_attr_diskseq.attr, 1006 #ifdef CONFIG_FAIL_MAKE_REQUEST 1007 &dev_attr_fail.attr, 1008 #endif 1009 #ifdef CONFIG_FAIL_IO_TIMEOUT 1010 &dev_attr_fail_timeout.attr, 1011 #endif 1012 NULL 1013 }; 1014 1015 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n) 1016 { 1017 struct device *dev = container_of(kobj, typeof(*dev), kobj); 1018 struct gendisk *disk = dev_to_disk(dev); 1019 1020 if (a == &dev_attr_badblocks.attr && !disk->bb) 1021 return 0; 1022 return a->mode; 1023 } 1024 1025 static struct attribute_group disk_attr_group = { 1026 .attrs = disk_attrs, 1027 .is_visible = disk_visible, 1028 }; 1029 1030 static const struct attribute_group *disk_attr_groups[] = { 1031 &disk_attr_group, 1032 NULL 1033 }; 1034 1035 /** 1036 * disk_release - releases all allocated resources of the gendisk 1037 * @dev: the device representing this disk 1038 * 1039 * This function releases all allocated resources of the gendisk. 1040 * 1041 * Drivers which used __device_add_disk() have a gendisk with a request_queue 1042 * assigned. Since the request_queue sits on top of the gendisk for these 1043 * drivers we also call blk_put_queue() for them, and we expect the 1044 * request_queue refcount to reach 0 at this point, and so the request_queue 1045 * will also be freed prior to the disk. 1046 * 1047 * Context: can sleep 1048 */ 1049 static void disk_release(struct device *dev) 1050 { 1051 struct gendisk *disk = dev_to_disk(dev); 1052 1053 might_sleep(); 1054 1055 disk_release_events(disk); 1056 kfree(disk->random); 1057 xa_destroy(&disk->part_tbl); 1058 disk->queue->disk = NULL; 1059 blk_put_queue(disk->queue); 1060 iput(disk->part0->bd_inode); /* frees the disk */ 1061 } 1062 1063 static int block_uevent(struct device *dev, struct kobj_uevent_env *env) 1064 { 1065 struct gendisk *disk = dev_to_disk(dev); 1066 1067 return add_uevent_var(env, "DISKSEQ=%llu", disk->diskseq); 1068 } 1069 1070 struct class block_class = { 1071 .name = "block", 1072 .dev_uevent = block_uevent, 1073 }; 1074 1075 static char *block_devnode(struct device *dev, umode_t *mode, 1076 kuid_t *uid, kgid_t *gid) 1077 { 1078 struct gendisk *disk = dev_to_disk(dev); 1079 1080 if (disk->fops->devnode) 1081 return disk->fops->devnode(disk, mode); 1082 return NULL; 1083 } 1084 1085 const struct device_type disk_type = { 1086 .name = "disk", 1087 .groups = disk_attr_groups, 1088 .release = disk_release, 1089 .devnode = block_devnode, 1090 }; 1091 1092 #ifdef CONFIG_PROC_FS 1093 /* 1094 * aggregate disk stat collector. Uses the same stats that the sysfs 1095 * entries do, above, but makes them available through one seq_file. 1096 * 1097 * The output looks suspiciously like /proc/partitions with a bunch of 1098 * extra fields. 1099 */ 1100 static int diskstats_show(struct seq_file *seqf, void *v) 1101 { 1102 struct gendisk *gp = v; 1103 struct block_device *hd; 1104 unsigned int inflight; 1105 struct disk_stats stat; 1106 unsigned long idx; 1107 1108 /* 1109 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next) 1110 seq_puts(seqf, "major minor name" 1111 " rio rmerge rsect ruse wio wmerge " 1112 "wsect wuse running use aveq" 1113 "\n\n"); 1114 */ 1115 1116 rcu_read_lock(); 1117 xa_for_each(&gp->part_tbl, idx, hd) { 1118 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd)) 1119 continue; 1120 part_stat_read_all(hd, &stat); 1121 if (queue_is_mq(gp->queue)) 1122 inflight = blk_mq_in_flight(gp->queue, hd); 1123 else 1124 inflight = part_in_flight(hd); 1125 1126 seq_printf(seqf, "%4d %7d %pg " 1127 "%lu %lu %lu %u " 1128 "%lu %lu %lu %u " 1129 "%u %u %u " 1130 "%lu %lu %lu %u " 1131 "%lu %u" 1132 "\n", 1133 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), hd, 1134 stat.ios[STAT_READ], 1135 stat.merges[STAT_READ], 1136 stat.sectors[STAT_READ], 1137 (unsigned int)div_u64(stat.nsecs[STAT_READ], 1138 NSEC_PER_MSEC), 1139 stat.ios[STAT_WRITE], 1140 stat.merges[STAT_WRITE], 1141 stat.sectors[STAT_WRITE], 1142 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], 1143 NSEC_PER_MSEC), 1144 inflight, 1145 jiffies_to_msecs(stat.io_ticks), 1146 (unsigned int)div_u64(stat.nsecs[STAT_READ] + 1147 stat.nsecs[STAT_WRITE] + 1148 stat.nsecs[STAT_DISCARD] + 1149 stat.nsecs[STAT_FLUSH], 1150 NSEC_PER_MSEC), 1151 stat.ios[STAT_DISCARD], 1152 stat.merges[STAT_DISCARD], 1153 stat.sectors[STAT_DISCARD], 1154 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], 1155 NSEC_PER_MSEC), 1156 stat.ios[STAT_FLUSH], 1157 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], 1158 NSEC_PER_MSEC) 1159 ); 1160 } 1161 rcu_read_unlock(); 1162 1163 return 0; 1164 } 1165 1166 static const struct seq_operations diskstats_op = { 1167 .start = disk_seqf_start, 1168 .next = disk_seqf_next, 1169 .stop = disk_seqf_stop, 1170 .show = diskstats_show 1171 }; 1172 1173 static int __init proc_genhd_init(void) 1174 { 1175 proc_create_seq("diskstats", 0, NULL, &diskstats_op); 1176 proc_create_seq("partitions", 0, NULL, &partitions_op); 1177 return 0; 1178 } 1179 module_init(proc_genhd_init); 1180 #endif /* CONFIG_PROC_FS */ 1181 1182 dev_t part_devt(struct gendisk *disk, u8 partno) 1183 { 1184 struct block_device *part; 1185 dev_t devt = 0; 1186 1187 rcu_read_lock(); 1188 part = xa_load(&disk->part_tbl, partno); 1189 if (part) 1190 devt = part->bd_dev; 1191 rcu_read_unlock(); 1192 1193 return devt; 1194 } 1195 1196 dev_t blk_lookup_devt(const char *name, int partno) 1197 { 1198 dev_t devt = MKDEV(0, 0); 1199 struct class_dev_iter iter; 1200 struct device *dev; 1201 1202 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 1203 while ((dev = class_dev_iter_next(&iter))) { 1204 struct gendisk *disk = dev_to_disk(dev); 1205 1206 if (strcmp(dev_name(dev), name)) 1207 continue; 1208 1209 if (partno < disk->minors) { 1210 /* We need to return the right devno, even 1211 * if the partition doesn't exist yet. 1212 */ 1213 devt = MKDEV(MAJOR(dev->devt), 1214 MINOR(dev->devt) + partno); 1215 } else { 1216 devt = part_devt(disk, partno); 1217 if (devt) 1218 break; 1219 } 1220 } 1221 class_dev_iter_exit(&iter); 1222 return devt; 1223 } 1224 1225 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id, 1226 struct lock_class_key *lkclass) 1227 { 1228 struct gendisk *disk; 1229 1230 if (!blk_get_queue(q)) 1231 return NULL; 1232 1233 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id); 1234 if (!disk) 1235 goto out_put_queue; 1236 1237 disk->bdi = bdi_alloc(node_id); 1238 if (!disk->bdi) 1239 goto out_free_disk; 1240 1241 disk->part0 = bdev_alloc(disk, 0); 1242 if (!disk->part0) 1243 goto out_free_bdi; 1244 1245 disk->node_id = node_id; 1246 mutex_init(&disk->open_mutex); 1247 xa_init(&disk->part_tbl); 1248 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL)) 1249 goto out_destroy_part_tbl; 1250 1251 rand_initialize_disk(disk); 1252 disk_to_dev(disk)->class = &block_class; 1253 disk_to_dev(disk)->type = &disk_type; 1254 device_initialize(disk_to_dev(disk)); 1255 inc_diskseq(disk); 1256 disk->queue = q; 1257 q->disk = disk; 1258 lockdep_init_map(&disk->lockdep_map, "(bio completion)", lkclass, 0); 1259 #ifdef CONFIG_BLOCK_HOLDER_DEPRECATED 1260 INIT_LIST_HEAD(&disk->slave_bdevs); 1261 #endif 1262 return disk; 1263 1264 out_destroy_part_tbl: 1265 xa_destroy(&disk->part_tbl); 1266 iput(disk->part0->bd_inode); 1267 out_free_bdi: 1268 bdi_put(disk->bdi); 1269 out_free_disk: 1270 kfree(disk); 1271 out_put_queue: 1272 blk_put_queue(q); 1273 return NULL; 1274 } 1275 EXPORT_SYMBOL(__alloc_disk_node); 1276 1277 struct gendisk *__blk_alloc_disk(int node, struct lock_class_key *lkclass) 1278 { 1279 struct request_queue *q; 1280 struct gendisk *disk; 1281 1282 q = blk_alloc_queue(node); 1283 if (!q) 1284 return NULL; 1285 1286 disk = __alloc_disk_node(q, node, lkclass); 1287 if (!disk) { 1288 blk_cleanup_queue(q); 1289 return NULL; 1290 } 1291 return disk; 1292 } 1293 EXPORT_SYMBOL(__blk_alloc_disk); 1294 1295 /** 1296 * put_disk - decrements the gendisk refcount 1297 * @disk: the struct gendisk to decrement the refcount for 1298 * 1299 * This decrements the refcount for the struct gendisk. When this reaches 0 1300 * we'll have disk_release() called. 1301 * 1302 * Context: Any context, but the last reference must not be dropped from 1303 * atomic context. 1304 */ 1305 void put_disk(struct gendisk *disk) 1306 { 1307 if (disk) 1308 put_device(disk_to_dev(disk)); 1309 } 1310 EXPORT_SYMBOL(put_disk); 1311 1312 /** 1313 * blk_cleanup_disk - shutdown a gendisk allocated by blk_alloc_disk 1314 * @disk: gendisk to shutdown 1315 * 1316 * Mark the queue hanging off @disk DYING, drain all pending requests, then mark 1317 * the queue DEAD, destroy and put it and the gendisk structure. 1318 * 1319 * Context: can sleep 1320 */ 1321 void blk_cleanup_disk(struct gendisk *disk) 1322 { 1323 blk_cleanup_queue(disk->queue); 1324 put_disk(disk); 1325 } 1326 EXPORT_SYMBOL(blk_cleanup_disk); 1327 1328 static void set_disk_ro_uevent(struct gendisk *gd, int ro) 1329 { 1330 char event[] = "DISK_RO=1"; 1331 char *envp[] = { event, NULL }; 1332 1333 if (!ro) 1334 event[8] = '0'; 1335 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp); 1336 } 1337 1338 /** 1339 * set_disk_ro - set a gendisk read-only 1340 * @disk: gendisk to operate on 1341 * @read_only: %true to set the disk read-only, %false set the disk read/write 1342 * 1343 * This function is used to indicate whether a given disk device should have its 1344 * read-only flag set. set_disk_ro() is typically used by device drivers to 1345 * indicate whether the underlying physical device is write-protected. 1346 */ 1347 void set_disk_ro(struct gendisk *disk, bool read_only) 1348 { 1349 if (read_only) { 1350 if (test_and_set_bit(GD_READ_ONLY, &disk->state)) 1351 return; 1352 } else { 1353 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state)) 1354 return; 1355 } 1356 set_disk_ro_uevent(disk, read_only); 1357 } 1358 EXPORT_SYMBOL(set_disk_ro); 1359 1360 int bdev_read_only(struct block_device *bdev) 1361 { 1362 return bdev->bd_read_only || get_disk_ro(bdev->bd_disk); 1363 } 1364 EXPORT_SYMBOL(bdev_read_only); 1365 1366 void inc_diskseq(struct gendisk *disk) 1367 { 1368 disk->diskseq = atomic64_inc_return(&diskseq); 1369 } 1370