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 DECLARE_RWSEM(bdev_lookup_sem); 33 34 /* for extended dynamic devt allocation, currently only one major is used */ 35 #define NR_EXT_DEVT (1 << MINORBITS) 36 static DEFINE_IDA(ext_devt_ida); 37 38 static void disk_check_events(struct disk_events *ev, 39 unsigned int *clearing_ptr); 40 static void disk_alloc_events(struct gendisk *disk); 41 static void disk_add_events(struct gendisk *disk); 42 static void disk_del_events(struct gendisk *disk); 43 static void disk_release_events(struct gendisk *disk); 44 45 void set_capacity(struct gendisk *disk, sector_t sectors) 46 { 47 struct block_device *bdev = disk->part0; 48 49 spin_lock(&bdev->bd_size_lock); 50 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 51 spin_unlock(&bdev->bd_size_lock); 52 } 53 EXPORT_SYMBOL(set_capacity); 54 55 /* 56 * Set disk capacity and notify if the size is not currently zero and will not 57 * be set to zero. Returns true if a uevent was sent, otherwise false. 58 */ 59 bool set_capacity_and_notify(struct gendisk *disk, sector_t size) 60 { 61 sector_t capacity = get_capacity(disk); 62 char *envp[] = { "RESIZE=1", NULL }; 63 64 set_capacity(disk, size); 65 66 /* 67 * Only print a message and send a uevent if the gendisk is user visible 68 * and alive. This avoids spamming the log and udev when setting the 69 * initial capacity during probing. 70 */ 71 if (size == capacity || 72 (disk->flags & (GENHD_FL_UP | GENHD_FL_HIDDEN)) != GENHD_FL_UP) 73 return false; 74 75 pr_info("%s: detected capacity change from %lld to %lld\n", 76 disk->disk_name, capacity, size); 77 78 /* 79 * Historically we did not send a uevent for changes to/from an empty 80 * device. 81 */ 82 if (!capacity || !size) 83 return false; 84 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp); 85 return true; 86 } 87 EXPORT_SYMBOL_GPL(set_capacity_and_notify); 88 89 /* 90 * Format the device name of the indicated disk into the supplied buffer and 91 * return a pointer to that same buffer for convenience. 92 */ 93 char *disk_name(struct gendisk *hd, int partno, char *buf) 94 { 95 if (!partno) 96 snprintf(buf, BDEVNAME_SIZE, "%s", hd->disk_name); 97 else if (isdigit(hd->disk_name[strlen(hd->disk_name)-1])) 98 snprintf(buf, BDEVNAME_SIZE, "%sp%d", hd->disk_name, partno); 99 else 100 snprintf(buf, BDEVNAME_SIZE, "%s%d", hd->disk_name, partno); 101 102 return buf; 103 } 104 105 const char *bdevname(struct block_device *bdev, char *buf) 106 { 107 return disk_name(bdev->bd_disk, bdev->bd_partno, buf); 108 } 109 EXPORT_SYMBOL(bdevname); 110 111 static void part_stat_read_all(struct block_device *part, 112 struct disk_stats *stat) 113 { 114 int cpu; 115 116 memset(stat, 0, sizeof(struct disk_stats)); 117 for_each_possible_cpu(cpu) { 118 struct disk_stats *ptr = per_cpu_ptr(part->bd_stats, cpu); 119 int group; 120 121 for (group = 0; group < NR_STAT_GROUPS; group++) { 122 stat->nsecs[group] += ptr->nsecs[group]; 123 stat->sectors[group] += ptr->sectors[group]; 124 stat->ios[group] += ptr->ios[group]; 125 stat->merges[group] += ptr->merges[group]; 126 } 127 128 stat->io_ticks += ptr->io_ticks; 129 } 130 } 131 132 static unsigned int part_in_flight(struct block_device *part) 133 { 134 unsigned int inflight = 0; 135 int cpu; 136 137 for_each_possible_cpu(cpu) { 138 inflight += part_stat_local_read_cpu(part, in_flight[0], cpu) + 139 part_stat_local_read_cpu(part, in_flight[1], cpu); 140 } 141 if ((int)inflight < 0) 142 inflight = 0; 143 144 return inflight; 145 } 146 147 static void part_in_flight_rw(struct block_device *part, 148 unsigned int inflight[2]) 149 { 150 int cpu; 151 152 inflight[0] = 0; 153 inflight[1] = 0; 154 for_each_possible_cpu(cpu) { 155 inflight[0] += part_stat_local_read_cpu(part, in_flight[0], cpu); 156 inflight[1] += part_stat_local_read_cpu(part, in_flight[1], cpu); 157 } 158 if ((int)inflight[0] < 0) 159 inflight[0] = 0; 160 if ((int)inflight[1] < 0) 161 inflight[1] = 0; 162 } 163 164 /** 165 * disk_part_iter_init - initialize partition iterator 166 * @piter: iterator to initialize 167 * @disk: disk to iterate over 168 * @flags: DISK_PITER_* flags 169 * 170 * Initialize @piter so that it iterates over partitions of @disk. 171 * 172 * CONTEXT: 173 * Don't care. 174 */ 175 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk, 176 unsigned int flags) 177 { 178 piter->disk = disk; 179 piter->part = NULL; 180 if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0)) 181 piter->idx = 0; 182 else 183 piter->idx = 1; 184 piter->flags = flags; 185 } 186 187 /** 188 * disk_part_iter_next - proceed iterator to the next partition and return it 189 * @piter: iterator of interest 190 * 191 * Proceed @piter to the next partition and return it. 192 * 193 * CONTEXT: 194 * Don't care. 195 */ 196 struct block_device *disk_part_iter_next(struct disk_part_iter *piter) 197 { 198 struct block_device *part; 199 unsigned long idx; 200 201 /* put the last partition */ 202 disk_part_iter_exit(piter); 203 204 rcu_read_lock(); 205 xa_for_each_start(&piter->disk->part_tbl, idx, part, piter->idx) { 206 if (!bdev_nr_sectors(part) && 207 !(piter->flags & DISK_PITER_INCL_EMPTY) && 208 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 && 209 piter->idx == 0)) 210 continue; 211 212 piter->part = bdgrab(part); 213 if (!piter->part) 214 continue; 215 piter->idx = idx + 1; 216 break; 217 } 218 rcu_read_unlock(); 219 220 return piter->part; 221 } 222 223 /** 224 * disk_part_iter_exit - finish up partition iteration 225 * @piter: iter of interest 226 * 227 * Called when iteration is over. Cleans up @piter. 228 * 229 * CONTEXT: 230 * Don't care. 231 */ 232 void disk_part_iter_exit(struct disk_part_iter *piter) 233 { 234 if (piter->part) 235 bdput(piter->part); 236 piter->part = NULL; 237 } 238 239 /* 240 * Can be deleted altogether. Later. 241 * 242 */ 243 #define BLKDEV_MAJOR_HASH_SIZE 255 244 static struct blk_major_name { 245 struct blk_major_name *next; 246 int major; 247 char name[16]; 248 void (*probe)(dev_t devt); 249 } *major_names[BLKDEV_MAJOR_HASH_SIZE]; 250 static DEFINE_MUTEX(major_names_lock); 251 252 /* index in the above - for now: assume no multimajor ranges */ 253 static inline int major_to_index(unsigned major) 254 { 255 return major % BLKDEV_MAJOR_HASH_SIZE; 256 } 257 258 #ifdef CONFIG_PROC_FS 259 void blkdev_show(struct seq_file *seqf, off_t offset) 260 { 261 struct blk_major_name *dp; 262 263 mutex_lock(&major_names_lock); 264 for (dp = major_names[major_to_index(offset)]; dp; dp = dp->next) 265 if (dp->major == offset) 266 seq_printf(seqf, "%3d %s\n", dp->major, dp->name); 267 mutex_unlock(&major_names_lock); 268 } 269 #endif /* CONFIG_PROC_FS */ 270 271 /** 272 * __register_blkdev - register a new block device 273 * 274 * @major: the requested major device number [1..BLKDEV_MAJOR_MAX-1]. If 275 * @major = 0, try to allocate any unused major number. 276 * @name: the name of the new block device as a zero terminated string 277 * @probe: allback that is called on access to any minor number of @major 278 * 279 * The @name must be unique within the system. 280 * 281 * The return value depends on the @major input parameter: 282 * 283 * - if a major device number was requested in range [1..BLKDEV_MAJOR_MAX-1] 284 * then the function returns zero on success, or a negative error code 285 * - if any unused major number was requested with @major = 0 parameter 286 * then the return value is the allocated major number in range 287 * [1..BLKDEV_MAJOR_MAX-1] or a negative error code otherwise 288 * 289 * See Documentation/admin-guide/devices.txt for the list of allocated 290 * major numbers. 291 * 292 * Use register_blkdev instead for any new code. 293 */ 294 int __register_blkdev(unsigned int major, const char *name, 295 void (*probe)(dev_t devt)) 296 { 297 struct blk_major_name **n, *p; 298 int index, ret = 0; 299 300 mutex_lock(&major_names_lock); 301 302 /* temporary */ 303 if (major == 0) { 304 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) { 305 if (major_names[index] == NULL) 306 break; 307 } 308 309 if (index == 0) { 310 printk("%s: failed to get major for %s\n", 311 __func__, name); 312 ret = -EBUSY; 313 goto out; 314 } 315 major = index; 316 ret = major; 317 } 318 319 if (major >= BLKDEV_MAJOR_MAX) { 320 pr_err("%s: major requested (%u) is greater than the maximum (%u) for %s\n", 321 __func__, major, BLKDEV_MAJOR_MAX-1, name); 322 323 ret = -EINVAL; 324 goto out; 325 } 326 327 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL); 328 if (p == NULL) { 329 ret = -ENOMEM; 330 goto out; 331 } 332 333 p->major = major; 334 p->probe = probe; 335 strlcpy(p->name, name, sizeof(p->name)); 336 p->next = NULL; 337 index = major_to_index(major); 338 339 for (n = &major_names[index]; *n; n = &(*n)->next) { 340 if ((*n)->major == major) 341 break; 342 } 343 if (!*n) 344 *n = p; 345 else 346 ret = -EBUSY; 347 348 if (ret < 0) { 349 printk("register_blkdev: cannot get major %u for %s\n", 350 major, name); 351 kfree(p); 352 } 353 out: 354 mutex_unlock(&major_names_lock); 355 return ret; 356 } 357 EXPORT_SYMBOL(__register_blkdev); 358 359 void unregister_blkdev(unsigned int major, const char *name) 360 { 361 struct blk_major_name **n; 362 struct blk_major_name *p = NULL; 363 int index = major_to_index(major); 364 365 mutex_lock(&major_names_lock); 366 for (n = &major_names[index]; *n; n = &(*n)->next) 367 if ((*n)->major == major) 368 break; 369 if (!*n || strcmp((*n)->name, name)) { 370 WARN_ON(1); 371 } else { 372 p = *n; 373 *n = p->next; 374 } 375 mutex_unlock(&major_names_lock); 376 kfree(p); 377 } 378 379 EXPORT_SYMBOL(unregister_blkdev); 380 381 /** 382 * blk_mangle_minor - scatter minor numbers apart 383 * @minor: minor number to mangle 384 * 385 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT 386 * is enabled. Mangling twice gives the original value. 387 * 388 * RETURNS: 389 * Mangled value. 390 * 391 * CONTEXT: 392 * Don't care. 393 */ 394 static int blk_mangle_minor(int minor) 395 { 396 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT 397 int i; 398 399 for (i = 0; i < MINORBITS / 2; i++) { 400 int low = minor & (1 << i); 401 int high = minor & (1 << (MINORBITS - 1 - i)); 402 int distance = MINORBITS - 1 - 2 * i; 403 404 minor ^= low | high; /* clear both bits */ 405 low <<= distance; /* swap the positions */ 406 high >>= distance; 407 minor |= low | high; /* and set */ 408 } 409 #endif 410 return minor; 411 } 412 413 /** 414 * blk_alloc_devt - allocate a dev_t for a block device 415 * @bdev: block device to allocate dev_t for 416 * @devt: out parameter for resulting dev_t 417 * 418 * Allocate a dev_t for block device. 419 * 420 * RETURNS: 421 * 0 on success, allocated dev_t is returned in *@devt. -errno on 422 * failure. 423 * 424 * CONTEXT: 425 * Might sleep. 426 */ 427 int blk_alloc_devt(struct block_device *bdev, dev_t *devt) 428 { 429 struct gendisk *disk = bdev->bd_disk; 430 int idx; 431 432 /* in consecutive minor range? */ 433 if (bdev->bd_partno < disk->minors) { 434 *devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno); 435 return 0; 436 } 437 438 idx = ida_alloc_range(&ext_devt_ida, 0, NR_EXT_DEVT, GFP_KERNEL); 439 if (idx < 0) 440 return idx == -ENOSPC ? -EBUSY : idx; 441 442 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx)); 443 return 0; 444 } 445 446 /** 447 * blk_free_devt - free a dev_t 448 * @devt: dev_t to free 449 * 450 * Free @devt which was allocated using blk_alloc_devt(). 451 * 452 * CONTEXT: 453 * Might sleep. 454 */ 455 void blk_free_devt(dev_t devt) 456 { 457 if (MAJOR(devt) == BLOCK_EXT_MAJOR) 458 ida_free(&ext_devt_ida, blk_mangle_minor(MINOR(devt))); 459 } 460 461 static char *bdevt_str(dev_t devt, char *buf) 462 { 463 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) { 464 char tbuf[BDEVT_SIZE]; 465 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt)); 466 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf); 467 } else 468 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt)); 469 470 return buf; 471 } 472 473 void disk_uevent(struct gendisk *disk, enum kobject_action action) 474 { 475 struct disk_part_iter piter; 476 struct block_device *part; 477 478 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY_PART0); 479 while ((part = disk_part_iter_next(&piter))) 480 kobject_uevent(bdev_kobj(part), action); 481 disk_part_iter_exit(&piter); 482 } 483 EXPORT_SYMBOL_GPL(disk_uevent); 484 485 static void disk_scan_partitions(struct gendisk *disk) 486 { 487 struct block_device *bdev; 488 489 if (!get_capacity(disk) || !disk_part_scan_enabled(disk)) 490 return; 491 492 set_bit(GD_NEED_PART_SCAN, &disk->state); 493 bdev = blkdev_get_by_dev(disk_devt(disk), FMODE_READ, NULL); 494 if (!IS_ERR(bdev)) 495 blkdev_put(bdev, FMODE_READ); 496 } 497 498 static void register_disk(struct device *parent, struct gendisk *disk, 499 const struct attribute_group **groups) 500 { 501 struct device *ddev = disk_to_dev(disk); 502 int err; 503 504 ddev->parent = parent; 505 506 dev_set_name(ddev, "%s", disk->disk_name); 507 508 /* delay uevents, until we scanned partition table */ 509 dev_set_uevent_suppress(ddev, 1); 510 511 if (groups) { 512 WARN_ON(ddev->groups); 513 ddev->groups = groups; 514 } 515 if (device_add(ddev)) 516 return; 517 if (!sysfs_deprecated) { 518 err = sysfs_create_link(block_depr, &ddev->kobj, 519 kobject_name(&ddev->kobj)); 520 if (err) { 521 device_del(ddev); 522 return; 523 } 524 } 525 526 /* 527 * avoid probable deadlock caused by allocating memory with 528 * GFP_KERNEL in runtime_resume callback of its all ancestor 529 * devices 530 */ 531 pm_runtime_set_memalloc_noio(ddev, true); 532 533 disk->part0->bd_holder_dir = 534 kobject_create_and_add("holders", &ddev->kobj); 535 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj); 536 537 if (disk->flags & GENHD_FL_HIDDEN) 538 return; 539 540 disk_scan_partitions(disk); 541 542 /* announce the disk and partitions after all partitions are created */ 543 dev_set_uevent_suppress(ddev, 0); 544 disk_uevent(disk, KOBJ_ADD); 545 546 if (disk->queue->backing_dev_info->dev) { 547 err = sysfs_create_link(&ddev->kobj, 548 &disk->queue->backing_dev_info->dev->kobj, 549 "bdi"); 550 WARN_ON(err); 551 } 552 } 553 554 /** 555 * __device_add_disk - add disk information to kernel list 556 * @parent: parent device for the disk 557 * @disk: per-device partitioning information 558 * @groups: Additional per-device sysfs groups 559 * @register_queue: register the queue if set to true 560 * 561 * This function registers the partitioning information in @disk 562 * with the kernel. 563 * 564 * FIXME: error handling 565 */ 566 static void __device_add_disk(struct device *parent, struct gendisk *disk, 567 const struct attribute_group **groups, 568 bool register_queue) 569 { 570 dev_t devt; 571 int retval; 572 573 /* 574 * The disk queue should now be all set with enough information about 575 * the device for the elevator code to pick an adequate default 576 * elevator if one is needed, that is, for devices requesting queue 577 * registration. 578 */ 579 if (register_queue) 580 elevator_init_mq(disk->queue); 581 582 /* minors == 0 indicates to use ext devt from part0 and should 583 * be accompanied with EXT_DEVT flag. Make sure all 584 * parameters make sense. 585 */ 586 WARN_ON(disk->minors && !(disk->major || disk->first_minor)); 587 WARN_ON(!disk->minors && 588 !(disk->flags & (GENHD_FL_EXT_DEVT | GENHD_FL_HIDDEN))); 589 590 disk->flags |= GENHD_FL_UP; 591 592 retval = blk_alloc_devt(disk->part0, &devt); 593 if (retval) { 594 WARN_ON(1); 595 return; 596 } 597 disk->major = MAJOR(devt); 598 disk->first_minor = MINOR(devt); 599 600 disk_alloc_events(disk); 601 602 if (disk->flags & GENHD_FL_HIDDEN) { 603 /* 604 * Don't let hidden disks show up in /proc/partitions, 605 * and don't bother scanning for partitions either. 606 */ 607 disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO; 608 disk->flags |= GENHD_FL_NO_PART_SCAN; 609 } else { 610 struct backing_dev_info *bdi = disk->queue->backing_dev_info; 611 struct device *dev = disk_to_dev(disk); 612 int ret; 613 614 /* Register BDI before referencing it from bdev */ 615 dev->devt = devt; 616 ret = bdi_register(bdi, "%u:%u", MAJOR(devt), MINOR(devt)); 617 WARN_ON(ret); 618 bdi_set_owner(bdi, dev); 619 bdev_add(disk->part0, devt); 620 } 621 register_disk(parent, disk, groups); 622 if (register_queue) 623 blk_register_queue(disk); 624 625 /* 626 * Take an extra ref on queue which will be put on disk_release() 627 * so that it sticks around as long as @disk is there. 628 */ 629 WARN_ON_ONCE(!blk_get_queue(disk->queue)); 630 631 disk_add_events(disk); 632 blk_integrity_add(disk); 633 } 634 635 void device_add_disk(struct device *parent, struct gendisk *disk, 636 const struct attribute_group **groups) 637 638 { 639 __device_add_disk(parent, disk, groups, true); 640 } 641 EXPORT_SYMBOL(device_add_disk); 642 643 void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk) 644 { 645 __device_add_disk(parent, disk, NULL, false); 646 } 647 EXPORT_SYMBOL(device_add_disk_no_queue_reg); 648 649 /** 650 * del_gendisk - remove the gendisk 651 * @disk: the struct gendisk to remove 652 * 653 * Removes the gendisk and all its associated resources. This deletes the 654 * partitions associated with the gendisk, and unregisters the associated 655 * request_queue. 656 * 657 * This is the counter to the respective __device_add_disk() call. 658 * 659 * The final removal of the struct gendisk happens when its refcount reaches 0 660 * with put_disk(), which should be called after del_gendisk(), if 661 * __device_add_disk() was used. 662 * 663 * Drivers exist which depend on the release of the gendisk to be synchronous, 664 * it should not be deferred. 665 * 666 * Context: can sleep 667 */ 668 void del_gendisk(struct gendisk *disk) 669 { 670 might_sleep(); 671 672 if (WARN_ON_ONCE(!disk->queue)) 673 return; 674 675 blk_integrity_del(disk); 676 disk_del_events(disk); 677 678 /* 679 * Block lookups of the disk until all bdevs are unhashed and the 680 * disk is marked as dead (GENHD_FL_UP cleared). 681 */ 682 down_write(&bdev_lookup_sem); 683 684 mutex_lock(&disk->part0->bd_mutex); 685 blk_drop_partitions(disk); 686 mutex_unlock(&disk->part0->bd_mutex); 687 688 fsync_bdev(disk->part0); 689 __invalidate_device(disk->part0, true); 690 691 /* 692 * Unhash the bdev inode for this device so that it can't be looked 693 * up any more even if openers still hold references to it. 694 */ 695 remove_inode_hash(disk->part0->bd_inode); 696 697 set_capacity(disk, 0); 698 disk->flags &= ~GENHD_FL_UP; 699 up_write(&bdev_lookup_sem); 700 701 if (!(disk->flags & GENHD_FL_HIDDEN)) { 702 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi"); 703 704 /* 705 * Unregister bdi before releasing device numbers (as they can 706 * get reused and we'd get clashes in sysfs). 707 */ 708 bdi_unregister(disk->queue->backing_dev_info); 709 } 710 711 blk_unregister_queue(disk); 712 713 kobject_put(disk->part0->bd_holder_dir); 714 kobject_put(disk->slave_dir); 715 716 part_stat_set_all(disk->part0, 0); 717 disk->part0->bd_stamp = 0; 718 if (!sysfs_deprecated) 719 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk))); 720 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false); 721 device_del(disk_to_dev(disk)); 722 } 723 EXPORT_SYMBOL(del_gendisk); 724 725 /* sysfs access to bad-blocks list. */ 726 static ssize_t disk_badblocks_show(struct device *dev, 727 struct device_attribute *attr, 728 char *page) 729 { 730 struct gendisk *disk = dev_to_disk(dev); 731 732 if (!disk->bb) 733 return sprintf(page, "\n"); 734 735 return badblocks_show(disk->bb, page, 0); 736 } 737 738 static ssize_t disk_badblocks_store(struct device *dev, 739 struct device_attribute *attr, 740 const char *page, size_t len) 741 { 742 struct gendisk *disk = dev_to_disk(dev); 743 744 if (!disk->bb) 745 return -ENXIO; 746 747 return badblocks_store(disk->bb, page, len, 0); 748 } 749 750 void blk_request_module(dev_t devt) 751 { 752 unsigned int major = MAJOR(devt); 753 struct blk_major_name **n; 754 755 mutex_lock(&major_names_lock); 756 for (n = &major_names[major_to_index(major)]; *n; n = &(*n)->next) { 757 if ((*n)->major == major && (*n)->probe) { 758 (*n)->probe(devt); 759 mutex_unlock(&major_names_lock); 760 return; 761 } 762 } 763 mutex_unlock(&major_names_lock); 764 765 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0) 766 /* Make old-style 2.4 aliases work */ 767 request_module("block-major-%d", MAJOR(devt)); 768 } 769 770 /** 771 * bdget_disk - do bdget() by gendisk and partition number 772 * @disk: gendisk of interest 773 * @partno: partition number 774 * 775 * Find partition @partno from @disk, do bdget() on it. 776 * 777 * CONTEXT: 778 * Don't care. 779 * 780 * RETURNS: 781 * Resulting block_device on success, NULL on failure. 782 */ 783 struct block_device *bdget_disk(struct gendisk *disk, int partno) 784 { 785 struct block_device *bdev = NULL; 786 787 rcu_read_lock(); 788 bdev = xa_load(&disk->part_tbl, partno); 789 if (bdev && !bdgrab(bdev)) 790 bdev = NULL; 791 rcu_read_unlock(); 792 793 return bdev; 794 } 795 796 /* 797 * print a full list of all partitions - intended for places where the root 798 * filesystem can't be mounted and thus to give the victim some idea of what 799 * went wrong 800 */ 801 void __init printk_all_partitions(void) 802 { 803 struct class_dev_iter iter; 804 struct device *dev; 805 806 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 807 while ((dev = class_dev_iter_next(&iter))) { 808 struct gendisk *disk = dev_to_disk(dev); 809 struct block_device *part; 810 char name_buf[BDEVNAME_SIZE]; 811 char devt_buf[BDEVT_SIZE]; 812 unsigned long idx; 813 814 /* 815 * Don't show empty devices or things that have been 816 * suppressed 817 */ 818 if (get_capacity(disk) == 0 || 819 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) 820 continue; 821 822 /* 823 * Note, unlike /proc/partitions, I am showing the numbers in 824 * hex - the same format as the root= option takes. 825 */ 826 rcu_read_lock(); 827 xa_for_each(&disk->part_tbl, idx, part) { 828 if (!bdev_nr_sectors(part)) 829 continue; 830 printk("%s%s %10llu %s %s", 831 bdev_is_partition(part) ? " " : "", 832 bdevt_str(part->bd_dev, devt_buf), 833 bdev_nr_sectors(part) >> 1, 834 disk_name(disk, part->bd_partno, name_buf), 835 part->bd_meta_info ? 836 part->bd_meta_info->uuid : ""); 837 if (bdev_is_partition(part)) 838 printk("\n"); 839 else if (dev->parent && dev->parent->driver) 840 printk(" driver: %s\n", 841 dev->parent->driver->name); 842 else 843 printk(" (driver?)\n"); 844 } 845 rcu_read_unlock(); 846 } 847 class_dev_iter_exit(&iter); 848 } 849 850 #ifdef CONFIG_PROC_FS 851 /* iterator */ 852 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos) 853 { 854 loff_t skip = *pos; 855 struct class_dev_iter *iter; 856 struct device *dev; 857 858 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 859 if (!iter) 860 return ERR_PTR(-ENOMEM); 861 862 seqf->private = iter; 863 class_dev_iter_init(iter, &block_class, NULL, &disk_type); 864 do { 865 dev = class_dev_iter_next(iter); 866 if (!dev) 867 return NULL; 868 } while (skip--); 869 870 return dev_to_disk(dev); 871 } 872 873 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos) 874 { 875 struct device *dev; 876 877 (*pos)++; 878 dev = class_dev_iter_next(seqf->private); 879 if (dev) 880 return dev_to_disk(dev); 881 882 return NULL; 883 } 884 885 static void disk_seqf_stop(struct seq_file *seqf, void *v) 886 { 887 struct class_dev_iter *iter = seqf->private; 888 889 /* stop is called even after start failed :-( */ 890 if (iter) { 891 class_dev_iter_exit(iter); 892 kfree(iter); 893 seqf->private = NULL; 894 } 895 } 896 897 static void *show_partition_start(struct seq_file *seqf, loff_t *pos) 898 { 899 void *p; 900 901 p = disk_seqf_start(seqf, pos); 902 if (!IS_ERR_OR_NULL(p) && !*pos) 903 seq_puts(seqf, "major minor #blocks name\n\n"); 904 return p; 905 } 906 907 static int show_partition(struct seq_file *seqf, void *v) 908 { 909 struct gendisk *sgp = v; 910 struct block_device *part; 911 unsigned long idx; 912 char buf[BDEVNAME_SIZE]; 913 914 /* Don't show non-partitionable removeable devices or empty devices */ 915 if (!get_capacity(sgp) || (!disk_max_parts(sgp) && 916 (sgp->flags & GENHD_FL_REMOVABLE))) 917 return 0; 918 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO) 919 return 0; 920 921 rcu_read_lock(); 922 xa_for_each(&sgp->part_tbl, idx, part) { 923 if (!bdev_nr_sectors(part)) 924 continue; 925 seq_printf(seqf, "%4d %7d %10llu %s\n", 926 MAJOR(part->bd_dev), MINOR(part->bd_dev), 927 bdev_nr_sectors(part) >> 1, 928 disk_name(sgp, part->bd_partno, buf)); 929 } 930 rcu_read_unlock(); 931 return 0; 932 } 933 934 static const struct seq_operations partitions_op = { 935 .start = show_partition_start, 936 .next = disk_seqf_next, 937 .stop = disk_seqf_stop, 938 .show = show_partition 939 }; 940 #endif 941 942 static int __init genhd_device_init(void) 943 { 944 int error; 945 946 block_class.dev_kobj = sysfs_dev_block_kobj; 947 error = class_register(&block_class); 948 if (unlikely(error)) 949 return error; 950 blk_dev_init(); 951 952 register_blkdev(BLOCK_EXT_MAJOR, "blkext"); 953 954 /* create top-level block dir */ 955 if (!sysfs_deprecated) 956 block_depr = kobject_create_and_add("block", NULL); 957 return 0; 958 } 959 960 subsys_initcall(genhd_device_init); 961 962 static ssize_t disk_range_show(struct device *dev, 963 struct device_attribute *attr, char *buf) 964 { 965 struct gendisk *disk = dev_to_disk(dev); 966 967 return sprintf(buf, "%d\n", disk->minors); 968 } 969 970 static ssize_t disk_ext_range_show(struct device *dev, 971 struct device_attribute *attr, char *buf) 972 { 973 struct gendisk *disk = dev_to_disk(dev); 974 975 return sprintf(buf, "%d\n", disk_max_parts(disk)); 976 } 977 978 static ssize_t disk_removable_show(struct device *dev, 979 struct device_attribute *attr, char *buf) 980 { 981 struct gendisk *disk = dev_to_disk(dev); 982 983 return sprintf(buf, "%d\n", 984 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0)); 985 } 986 987 static ssize_t disk_hidden_show(struct device *dev, 988 struct device_attribute *attr, char *buf) 989 { 990 struct gendisk *disk = dev_to_disk(dev); 991 992 return sprintf(buf, "%d\n", 993 (disk->flags & GENHD_FL_HIDDEN ? 1 : 0)); 994 } 995 996 static ssize_t disk_ro_show(struct device *dev, 997 struct device_attribute *attr, char *buf) 998 { 999 struct gendisk *disk = dev_to_disk(dev); 1000 1001 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0); 1002 } 1003 1004 ssize_t part_size_show(struct device *dev, 1005 struct device_attribute *attr, char *buf) 1006 { 1007 return sprintf(buf, "%llu\n", bdev_nr_sectors(dev_to_bdev(dev))); 1008 } 1009 1010 ssize_t part_stat_show(struct device *dev, 1011 struct device_attribute *attr, char *buf) 1012 { 1013 struct block_device *bdev = dev_to_bdev(dev); 1014 struct request_queue *q = bdev->bd_disk->queue; 1015 struct disk_stats stat; 1016 unsigned int inflight; 1017 1018 part_stat_read_all(bdev, &stat); 1019 if (queue_is_mq(q)) 1020 inflight = blk_mq_in_flight(q, bdev); 1021 else 1022 inflight = part_in_flight(bdev); 1023 1024 return sprintf(buf, 1025 "%8lu %8lu %8llu %8u " 1026 "%8lu %8lu %8llu %8u " 1027 "%8u %8u %8u " 1028 "%8lu %8lu %8llu %8u " 1029 "%8lu %8u" 1030 "\n", 1031 stat.ios[STAT_READ], 1032 stat.merges[STAT_READ], 1033 (unsigned long long)stat.sectors[STAT_READ], 1034 (unsigned int)div_u64(stat.nsecs[STAT_READ], NSEC_PER_MSEC), 1035 stat.ios[STAT_WRITE], 1036 stat.merges[STAT_WRITE], 1037 (unsigned long long)stat.sectors[STAT_WRITE], 1038 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], NSEC_PER_MSEC), 1039 inflight, 1040 jiffies_to_msecs(stat.io_ticks), 1041 (unsigned int)div_u64(stat.nsecs[STAT_READ] + 1042 stat.nsecs[STAT_WRITE] + 1043 stat.nsecs[STAT_DISCARD] + 1044 stat.nsecs[STAT_FLUSH], 1045 NSEC_PER_MSEC), 1046 stat.ios[STAT_DISCARD], 1047 stat.merges[STAT_DISCARD], 1048 (unsigned long long)stat.sectors[STAT_DISCARD], 1049 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], NSEC_PER_MSEC), 1050 stat.ios[STAT_FLUSH], 1051 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], NSEC_PER_MSEC)); 1052 } 1053 1054 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr, 1055 char *buf) 1056 { 1057 struct block_device *bdev = dev_to_bdev(dev); 1058 struct request_queue *q = bdev->bd_disk->queue; 1059 unsigned int inflight[2]; 1060 1061 if (queue_is_mq(q)) 1062 blk_mq_in_flight_rw(q, bdev, inflight); 1063 else 1064 part_in_flight_rw(bdev, inflight); 1065 1066 return sprintf(buf, "%8u %8u\n", inflight[0], inflight[1]); 1067 } 1068 1069 static ssize_t disk_capability_show(struct device *dev, 1070 struct device_attribute *attr, char *buf) 1071 { 1072 struct gendisk *disk = dev_to_disk(dev); 1073 1074 return sprintf(buf, "%x\n", disk->flags); 1075 } 1076 1077 static ssize_t disk_alignment_offset_show(struct device *dev, 1078 struct device_attribute *attr, 1079 char *buf) 1080 { 1081 struct gendisk *disk = dev_to_disk(dev); 1082 1083 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue)); 1084 } 1085 1086 static ssize_t disk_discard_alignment_show(struct device *dev, 1087 struct device_attribute *attr, 1088 char *buf) 1089 { 1090 struct gendisk *disk = dev_to_disk(dev); 1091 1092 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue)); 1093 } 1094 1095 static DEVICE_ATTR(range, 0444, disk_range_show, NULL); 1096 static DEVICE_ATTR(ext_range, 0444, disk_ext_range_show, NULL); 1097 static DEVICE_ATTR(removable, 0444, disk_removable_show, NULL); 1098 static DEVICE_ATTR(hidden, 0444, disk_hidden_show, NULL); 1099 static DEVICE_ATTR(ro, 0444, disk_ro_show, NULL); 1100 static DEVICE_ATTR(size, 0444, part_size_show, NULL); 1101 static DEVICE_ATTR(alignment_offset, 0444, disk_alignment_offset_show, NULL); 1102 static DEVICE_ATTR(discard_alignment, 0444, disk_discard_alignment_show, NULL); 1103 static DEVICE_ATTR(capability, 0444, disk_capability_show, NULL); 1104 static DEVICE_ATTR(stat, 0444, part_stat_show, NULL); 1105 static DEVICE_ATTR(inflight, 0444, part_inflight_show, NULL); 1106 static DEVICE_ATTR(badblocks, 0644, disk_badblocks_show, disk_badblocks_store); 1107 1108 #ifdef CONFIG_FAIL_MAKE_REQUEST 1109 ssize_t part_fail_show(struct device *dev, 1110 struct device_attribute *attr, char *buf) 1111 { 1112 return sprintf(buf, "%d\n", dev_to_bdev(dev)->bd_make_it_fail); 1113 } 1114 1115 ssize_t part_fail_store(struct device *dev, 1116 struct device_attribute *attr, 1117 const char *buf, size_t count) 1118 { 1119 int i; 1120 1121 if (count > 0 && sscanf(buf, "%d", &i) > 0) 1122 dev_to_bdev(dev)->bd_make_it_fail = i; 1123 1124 return count; 1125 } 1126 1127 static struct device_attribute dev_attr_fail = 1128 __ATTR(make-it-fail, 0644, part_fail_show, part_fail_store); 1129 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1130 1131 #ifdef CONFIG_FAIL_IO_TIMEOUT 1132 static struct device_attribute dev_attr_fail_timeout = 1133 __ATTR(io-timeout-fail, 0644, part_timeout_show, part_timeout_store); 1134 #endif 1135 1136 static struct attribute *disk_attrs[] = { 1137 &dev_attr_range.attr, 1138 &dev_attr_ext_range.attr, 1139 &dev_attr_removable.attr, 1140 &dev_attr_hidden.attr, 1141 &dev_attr_ro.attr, 1142 &dev_attr_size.attr, 1143 &dev_attr_alignment_offset.attr, 1144 &dev_attr_discard_alignment.attr, 1145 &dev_attr_capability.attr, 1146 &dev_attr_stat.attr, 1147 &dev_attr_inflight.attr, 1148 &dev_attr_badblocks.attr, 1149 #ifdef CONFIG_FAIL_MAKE_REQUEST 1150 &dev_attr_fail.attr, 1151 #endif 1152 #ifdef CONFIG_FAIL_IO_TIMEOUT 1153 &dev_attr_fail_timeout.attr, 1154 #endif 1155 NULL 1156 }; 1157 1158 static umode_t disk_visible(struct kobject *kobj, struct attribute *a, int n) 1159 { 1160 struct device *dev = container_of(kobj, typeof(*dev), kobj); 1161 struct gendisk *disk = dev_to_disk(dev); 1162 1163 if (a == &dev_attr_badblocks.attr && !disk->bb) 1164 return 0; 1165 return a->mode; 1166 } 1167 1168 static struct attribute_group disk_attr_group = { 1169 .attrs = disk_attrs, 1170 .is_visible = disk_visible, 1171 }; 1172 1173 static const struct attribute_group *disk_attr_groups[] = { 1174 &disk_attr_group, 1175 NULL 1176 }; 1177 1178 /** 1179 * disk_release - releases all allocated resources of the gendisk 1180 * @dev: the device representing this disk 1181 * 1182 * This function releases all allocated resources of the gendisk. 1183 * 1184 * Drivers which used __device_add_disk() have a gendisk with a request_queue 1185 * assigned. Since the request_queue sits on top of the gendisk for these 1186 * drivers we also call blk_put_queue() for them, and we expect the 1187 * request_queue refcount to reach 0 at this point, and so the request_queue 1188 * will also be freed prior to the disk. 1189 * 1190 * Context: can sleep 1191 */ 1192 static void disk_release(struct device *dev) 1193 { 1194 struct gendisk *disk = dev_to_disk(dev); 1195 1196 might_sleep(); 1197 1198 blk_free_devt(dev->devt); 1199 disk_release_events(disk); 1200 kfree(disk->random); 1201 xa_destroy(&disk->part_tbl); 1202 bdput(disk->part0); 1203 if (disk->queue) 1204 blk_put_queue(disk->queue); 1205 kfree(disk); 1206 } 1207 struct class block_class = { 1208 .name = "block", 1209 }; 1210 1211 static char *block_devnode(struct device *dev, umode_t *mode, 1212 kuid_t *uid, kgid_t *gid) 1213 { 1214 struct gendisk *disk = dev_to_disk(dev); 1215 1216 if (disk->fops->devnode) 1217 return disk->fops->devnode(disk, mode); 1218 return NULL; 1219 } 1220 1221 const struct device_type disk_type = { 1222 .name = "disk", 1223 .groups = disk_attr_groups, 1224 .release = disk_release, 1225 .devnode = block_devnode, 1226 }; 1227 1228 #ifdef CONFIG_PROC_FS 1229 /* 1230 * aggregate disk stat collector. Uses the same stats that the sysfs 1231 * entries do, above, but makes them available through one seq_file. 1232 * 1233 * The output looks suspiciously like /proc/partitions with a bunch of 1234 * extra fields. 1235 */ 1236 static int diskstats_show(struct seq_file *seqf, void *v) 1237 { 1238 struct gendisk *gp = v; 1239 struct block_device *hd; 1240 char buf[BDEVNAME_SIZE]; 1241 unsigned int inflight; 1242 struct disk_stats stat; 1243 unsigned long idx; 1244 1245 /* 1246 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next) 1247 seq_puts(seqf, "major minor name" 1248 " rio rmerge rsect ruse wio wmerge " 1249 "wsect wuse running use aveq" 1250 "\n\n"); 1251 */ 1252 1253 rcu_read_lock(); 1254 xa_for_each(&gp->part_tbl, idx, hd) { 1255 if (bdev_is_partition(hd) && !bdev_nr_sectors(hd)) 1256 continue; 1257 part_stat_read_all(hd, &stat); 1258 if (queue_is_mq(gp->queue)) 1259 inflight = blk_mq_in_flight(gp->queue, hd); 1260 else 1261 inflight = part_in_flight(hd); 1262 1263 seq_printf(seqf, "%4d %7d %s " 1264 "%lu %lu %lu %u " 1265 "%lu %lu %lu %u " 1266 "%u %u %u " 1267 "%lu %lu %lu %u " 1268 "%lu %u" 1269 "\n", 1270 MAJOR(hd->bd_dev), MINOR(hd->bd_dev), 1271 disk_name(gp, hd->bd_partno, buf), 1272 stat.ios[STAT_READ], 1273 stat.merges[STAT_READ], 1274 stat.sectors[STAT_READ], 1275 (unsigned int)div_u64(stat.nsecs[STAT_READ], 1276 NSEC_PER_MSEC), 1277 stat.ios[STAT_WRITE], 1278 stat.merges[STAT_WRITE], 1279 stat.sectors[STAT_WRITE], 1280 (unsigned int)div_u64(stat.nsecs[STAT_WRITE], 1281 NSEC_PER_MSEC), 1282 inflight, 1283 jiffies_to_msecs(stat.io_ticks), 1284 (unsigned int)div_u64(stat.nsecs[STAT_READ] + 1285 stat.nsecs[STAT_WRITE] + 1286 stat.nsecs[STAT_DISCARD] + 1287 stat.nsecs[STAT_FLUSH], 1288 NSEC_PER_MSEC), 1289 stat.ios[STAT_DISCARD], 1290 stat.merges[STAT_DISCARD], 1291 stat.sectors[STAT_DISCARD], 1292 (unsigned int)div_u64(stat.nsecs[STAT_DISCARD], 1293 NSEC_PER_MSEC), 1294 stat.ios[STAT_FLUSH], 1295 (unsigned int)div_u64(stat.nsecs[STAT_FLUSH], 1296 NSEC_PER_MSEC) 1297 ); 1298 } 1299 rcu_read_unlock(); 1300 1301 return 0; 1302 } 1303 1304 static const struct seq_operations diskstats_op = { 1305 .start = disk_seqf_start, 1306 .next = disk_seqf_next, 1307 .stop = disk_seqf_stop, 1308 .show = diskstats_show 1309 }; 1310 1311 static int __init proc_genhd_init(void) 1312 { 1313 proc_create_seq("diskstats", 0, NULL, &diskstats_op); 1314 proc_create_seq("partitions", 0, NULL, &partitions_op); 1315 return 0; 1316 } 1317 module_init(proc_genhd_init); 1318 #endif /* CONFIG_PROC_FS */ 1319 1320 dev_t blk_lookup_devt(const char *name, int partno) 1321 { 1322 dev_t devt = MKDEV(0, 0); 1323 struct class_dev_iter iter; 1324 struct device *dev; 1325 1326 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 1327 while ((dev = class_dev_iter_next(&iter))) { 1328 struct gendisk *disk = dev_to_disk(dev); 1329 struct block_device *part; 1330 1331 if (strcmp(dev_name(dev), name)) 1332 continue; 1333 1334 if (partno < disk->minors) { 1335 /* We need to return the right devno, even 1336 * if the partition doesn't exist yet. 1337 */ 1338 devt = MKDEV(MAJOR(dev->devt), 1339 MINOR(dev->devt) + partno); 1340 break; 1341 } 1342 part = bdget_disk(disk, partno); 1343 if (part) { 1344 devt = part->bd_dev; 1345 bdput(part); 1346 break; 1347 } 1348 } 1349 class_dev_iter_exit(&iter); 1350 return devt; 1351 } 1352 1353 struct gendisk *__alloc_disk_node(int minors, int node_id) 1354 { 1355 struct gendisk *disk; 1356 1357 if (minors > DISK_MAX_PARTS) { 1358 printk(KERN_ERR 1359 "block: can't allocate more than %d partitions\n", 1360 DISK_MAX_PARTS); 1361 minors = DISK_MAX_PARTS; 1362 } 1363 1364 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id); 1365 if (!disk) 1366 return NULL; 1367 1368 disk->part0 = bdev_alloc(disk, 0); 1369 if (!disk->part0) 1370 goto out_free_disk; 1371 1372 disk->node_id = node_id; 1373 xa_init(&disk->part_tbl); 1374 if (xa_insert(&disk->part_tbl, 0, disk->part0, GFP_KERNEL)) 1375 goto out_destroy_part_tbl; 1376 1377 disk->minors = minors; 1378 rand_initialize_disk(disk); 1379 disk_to_dev(disk)->class = &block_class; 1380 disk_to_dev(disk)->type = &disk_type; 1381 device_initialize(disk_to_dev(disk)); 1382 return disk; 1383 1384 out_destroy_part_tbl: 1385 xa_destroy(&disk->part_tbl); 1386 bdput(disk->part0); 1387 out_free_disk: 1388 kfree(disk); 1389 return NULL; 1390 } 1391 EXPORT_SYMBOL(__alloc_disk_node); 1392 1393 /** 1394 * put_disk - decrements the gendisk refcount 1395 * @disk: the struct gendisk to decrement the refcount for 1396 * 1397 * This decrements the refcount for the struct gendisk. When this reaches 0 1398 * we'll have disk_release() called. 1399 * 1400 * Context: Any context, but the last reference must not be dropped from 1401 * atomic context. 1402 */ 1403 void put_disk(struct gendisk *disk) 1404 { 1405 if (disk) 1406 put_device(disk_to_dev(disk)); 1407 } 1408 EXPORT_SYMBOL(put_disk); 1409 1410 static void set_disk_ro_uevent(struct gendisk *gd, int ro) 1411 { 1412 char event[] = "DISK_RO=1"; 1413 char *envp[] = { event, NULL }; 1414 1415 if (!ro) 1416 event[8] = '0'; 1417 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp); 1418 } 1419 1420 /** 1421 * set_disk_ro - set a gendisk read-only 1422 * @disk: gendisk to operate on 1423 * @read_only: %true to set the disk read-only, %false set the disk read/write 1424 * 1425 * This function is used to indicate whether a given disk device should have its 1426 * read-only flag set. set_disk_ro() is typically used by device drivers to 1427 * indicate whether the underlying physical device is write-protected. 1428 */ 1429 void set_disk_ro(struct gendisk *disk, bool read_only) 1430 { 1431 if (read_only) { 1432 if (test_and_set_bit(GD_READ_ONLY, &disk->state)) 1433 return; 1434 } else { 1435 if (!test_and_clear_bit(GD_READ_ONLY, &disk->state)) 1436 return; 1437 } 1438 set_disk_ro_uevent(disk, read_only); 1439 } 1440 EXPORT_SYMBOL(set_disk_ro); 1441 1442 int bdev_read_only(struct block_device *bdev) 1443 { 1444 return bdev->bd_read_only || get_disk_ro(bdev->bd_disk); 1445 } 1446 EXPORT_SYMBOL(bdev_read_only); 1447 1448 /* 1449 * Disk events - monitor disk events like media change and eject request. 1450 */ 1451 struct disk_events { 1452 struct list_head node; /* all disk_event's */ 1453 struct gendisk *disk; /* the associated disk */ 1454 spinlock_t lock; 1455 1456 struct mutex block_mutex; /* protects blocking */ 1457 int block; /* event blocking depth */ 1458 unsigned int pending; /* events already sent out */ 1459 unsigned int clearing; /* events being cleared */ 1460 1461 long poll_msecs; /* interval, -1 for default */ 1462 struct delayed_work dwork; 1463 }; 1464 1465 static const char *disk_events_strs[] = { 1466 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change", 1467 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request", 1468 }; 1469 1470 static char *disk_uevents[] = { 1471 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1", 1472 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1", 1473 }; 1474 1475 /* list of all disk_events */ 1476 static DEFINE_MUTEX(disk_events_mutex); 1477 static LIST_HEAD(disk_events); 1478 1479 /* disable in-kernel polling by default */ 1480 static unsigned long disk_events_dfl_poll_msecs; 1481 1482 static unsigned long disk_events_poll_jiffies(struct gendisk *disk) 1483 { 1484 struct disk_events *ev = disk->ev; 1485 long intv_msecs = 0; 1486 1487 /* 1488 * If device-specific poll interval is set, always use it. If 1489 * the default is being used, poll if the POLL flag is set. 1490 */ 1491 if (ev->poll_msecs >= 0) 1492 intv_msecs = ev->poll_msecs; 1493 else if (disk->event_flags & DISK_EVENT_FLAG_POLL) 1494 intv_msecs = disk_events_dfl_poll_msecs; 1495 1496 return msecs_to_jiffies(intv_msecs); 1497 } 1498 1499 /** 1500 * disk_block_events - block and flush disk event checking 1501 * @disk: disk to block events for 1502 * 1503 * On return from this function, it is guaranteed that event checking 1504 * isn't in progress and won't happen until unblocked by 1505 * disk_unblock_events(). Events blocking is counted and the actual 1506 * unblocking happens after the matching number of unblocks are done. 1507 * 1508 * Note that this intentionally does not block event checking from 1509 * disk_clear_events(). 1510 * 1511 * CONTEXT: 1512 * Might sleep. 1513 */ 1514 void disk_block_events(struct gendisk *disk) 1515 { 1516 struct disk_events *ev = disk->ev; 1517 unsigned long flags; 1518 bool cancel; 1519 1520 if (!ev) 1521 return; 1522 1523 /* 1524 * Outer mutex ensures that the first blocker completes canceling 1525 * the event work before further blockers are allowed to finish. 1526 */ 1527 mutex_lock(&ev->block_mutex); 1528 1529 spin_lock_irqsave(&ev->lock, flags); 1530 cancel = !ev->block++; 1531 spin_unlock_irqrestore(&ev->lock, flags); 1532 1533 if (cancel) 1534 cancel_delayed_work_sync(&disk->ev->dwork); 1535 1536 mutex_unlock(&ev->block_mutex); 1537 } 1538 1539 static void __disk_unblock_events(struct gendisk *disk, bool check_now) 1540 { 1541 struct disk_events *ev = disk->ev; 1542 unsigned long intv; 1543 unsigned long flags; 1544 1545 spin_lock_irqsave(&ev->lock, flags); 1546 1547 if (WARN_ON_ONCE(ev->block <= 0)) 1548 goto out_unlock; 1549 1550 if (--ev->block) 1551 goto out_unlock; 1552 1553 intv = disk_events_poll_jiffies(disk); 1554 if (check_now) 1555 queue_delayed_work(system_freezable_power_efficient_wq, 1556 &ev->dwork, 0); 1557 else if (intv) 1558 queue_delayed_work(system_freezable_power_efficient_wq, 1559 &ev->dwork, intv); 1560 out_unlock: 1561 spin_unlock_irqrestore(&ev->lock, flags); 1562 } 1563 1564 /** 1565 * disk_unblock_events - unblock disk event checking 1566 * @disk: disk to unblock events for 1567 * 1568 * Undo disk_block_events(). When the block count reaches zero, it 1569 * starts events polling if configured. 1570 * 1571 * CONTEXT: 1572 * Don't care. Safe to call from irq context. 1573 */ 1574 void disk_unblock_events(struct gendisk *disk) 1575 { 1576 if (disk->ev) 1577 __disk_unblock_events(disk, false); 1578 } 1579 1580 /** 1581 * disk_flush_events - schedule immediate event checking and flushing 1582 * @disk: disk to check and flush events for 1583 * @mask: events to flush 1584 * 1585 * Schedule immediate event checking on @disk if not blocked. Events in 1586 * @mask are scheduled to be cleared from the driver. Note that this 1587 * doesn't clear the events from @disk->ev. 1588 * 1589 * CONTEXT: 1590 * If @mask is non-zero must be called with bdev->bd_mutex held. 1591 */ 1592 void disk_flush_events(struct gendisk *disk, unsigned int mask) 1593 { 1594 struct disk_events *ev = disk->ev; 1595 1596 if (!ev) 1597 return; 1598 1599 spin_lock_irq(&ev->lock); 1600 ev->clearing |= mask; 1601 if (!ev->block) 1602 mod_delayed_work(system_freezable_power_efficient_wq, 1603 &ev->dwork, 0); 1604 spin_unlock_irq(&ev->lock); 1605 } 1606 1607 /** 1608 * disk_clear_events - synchronously check, clear and return pending events 1609 * @disk: disk to fetch and clear events from 1610 * @mask: mask of events to be fetched and cleared 1611 * 1612 * Disk events are synchronously checked and pending events in @mask 1613 * are cleared and returned. This ignores the block count. 1614 * 1615 * CONTEXT: 1616 * Might sleep. 1617 */ 1618 static unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask) 1619 { 1620 struct disk_events *ev = disk->ev; 1621 unsigned int pending; 1622 unsigned int clearing = mask; 1623 1624 if (!ev) 1625 return 0; 1626 1627 disk_block_events(disk); 1628 1629 /* 1630 * store the union of mask and ev->clearing on the stack so that the 1631 * race with disk_flush_events does not cause ambiguity (ev->clearing 1632 * can still be modified even if events are blocked). 1633 */ 1634 spin_lock_irq(&ev->lock); 1635 clearing |= ev->clearing; 1636 ev->clearing = 0; 1637 spin_unlock_irq(&ev->lock); 1638 1639 disk_check_events(ev, &clearing); 1640 /* 1641 * if ev->clearing is not 0, the disk_flush_events got called in the 1642 * middle of this function, so we want to run the workfn without delay. 1643 */ 1644 __disk_unblock_events(disk, ev->clearing ? true : false); 1645 1646 /* then, fetch and clear pending events */ 1647 spin_lock_irq(&ev->lock); 1648 pending = ev->pending & mask; 1649 ev->pending &= ~mask; 1650 spin_unlock_irq(&ev->lock); 1651 WARN_ON_ONCE(clearing & mask); 1652 1653 return pending; 1654 } 1655 1656 /** 1657 * bdev_check_media_change - check if a removable media has been changed 1658 * @bdev: block device to check 1659 * 1660 * Check whether a removable media has been changed, and attempt to free all 1661 * dentries and inodes and invalidates all block device page cache entries in 1662 * that case. 1663 * 1664 * Returns %true if the block device changed, or %false if not. 1665 */ 1666 bool bdev_check_media_change(struct block_device *bdev) 1667 { 1668 unsigned int events; 1669 1670 events = disk_clear_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE | 1671 DISK_EVENT_EJECT_REQUEST); 1672 if (!(events & DISK_EVENT_MEDIA_CHANGE)) 1673 return false; 1674 1675 if (__invalidate_device(bdev, true)) 1676 pr_warn("VFS: busy inodes on changed media %s\n", 1677 bdev->bd_disk->disk_name); 1678 set_bit(GD_NEED_PART_SCAN, &bdev->bd_disk->state); 1679 return true; 1680 } 1681 EXPORT_SYMBOL(bdev_check_media_change); 1682 1683 /* 1684 * Separate this part out so that a different pointer for clearing_ptr can be 1685 * passed in for disk_clear_events. 1686 */ 1687 static void disk_events_workfn(struct work_struct *work) 1688 { 1689 struct delayed_work *dwork = to_delayed_work(work); 1690 struct disk_events *ev = container_of(dwork, struct disk_events, dwork); 1691 1692 disk_check_events(ev, &ev->clearing); 1693 } 1694 1695 static void disk_check_events(struct disk_events *ev, 1696 unsigned int *clearing_ptr) 1697 { 1698 struct gendisk *disk = ev->disk; 1699 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { }; 1700 unsigned int clearing = *clearing_ptr; 1701 unsigned int events; 1702 unsigned long intv; 1703 int nr_events = 0, i; 1704 1705 /* check events */ 1706 events = disk->fops->check_events(disk, clearing); 1707 1708 /* accumulate pending events and schedule next poll if necessary */ 1709 spin_lock_irq(&ev->lock); 1710 1711 events &= ~ev->pending; 1712 ev->pending |= events; 1713 *clearing_ptr &= ~clearing; 1714 1715 intv = disk_events_poll_jiffies(disk); 1716 if (!ev->block && intv) 1717 queue_delayed_work(system_freezable_power_efficient_wq, 1718 &ev->dwork, intv); 1719 1720 spin_unlock_irq(&ev->lock); 1721 1722 /* 1723 * Tell userland about new events. Only the events listed in 1724 * @disk->events are reported, and only if DISK_EVENT_FLAG_UEVENT 1725 * is set. Otherwise, events are processed internally but never 1726 * get reported to userland. 1727 */ 1728 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++) 1729 if ((events & disk->events & (1 << i)) && 1730 (disk->event_flags & DISK_EVENT_FLAG_UEVENT)) 1731 envp[nr_events++] = disk_uevents[i]; 1732 1733 if (nr_events) 1734 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp); 1735 } 1736 1737 /* 1738 * A disk events enabled device has the following sysfs nodes under 1739 * its /sys/block/X/ directory. 1740 * 1741 * events : list of all supported events 1742 * events_async : list of events which can be detected w/o polling 1743 * (always empty, only for backwards compatibility) 1744 * events_poll_msecs : polling interval, 0: disable, -1: system default 1745 */ 1746 static ssize_t __disk_events_show(unsigned int events, char *buf) 1747 { 1748 const char *delim = ""; 1749 ssize_t pos = 0; 1750 int i; 1751 1752 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++) 1753 if (events & (1 << i)) { 1754 pos += sprintf(buf + pos, "%s%s", 1755 delim, disk_events_strs[i]); 1756 delim = " "; 1757 } 1758 if (pos) 1759 pos += sprintf(buf + pos, "\n"); 1760 return pos; 1761 } 1762 1763 static ssize_t disk_events_show(struct device *dev, 1764 struct device_attribute *attr, char *buf) 1765 { 1766 struct gendisk *disk = dev_to_disk(dev); 1767 1768 if (!(disk->event_flags & DISK_EVENT_FLAG_UEVENT)) 1769 return 0; 1770 1771 return __disk_events_show(disk->events, buf); 1772 } 1773 1774 static ssize_t disk_events_async_show(struct device *dev, 1775 struct device_attribute *attr, char *buf) 1776 { 1777 return 0; 1778 } 1779 1780 static ssize_t disk_events_poll_msecs_show(struct device *dev, 1781 struct device_attribute *attr, 1782 char *buf) 1783 { 1784 struct gendisk *disk = dev_to_disk(dev); 1785 1786 if (!disk->ev) 1787 return sprintf(buf, "-1\n"); 1788 1789 return sprintf(buf, "%ld\n", disk->ev->poll_msecs); 1790 } 1791 1792 static ssize_t disk_events_poll_msecs_store(struct device *dev, 1793 struct device_attribute *attr, 1794 const char *buf, size_t count) 1795 { 1796 struct gendisk *disk = dev_to_disk(dev); 1797 long intv; 1798 1799 if (!count || !sscanf(buf, "%ld", &intv)) 1800 return -EINVAL; 1801 1802 if (intv < 0 && intv != -1) 1803 return -EINVAL; 1804 1805 if (!disk->ev) 1806 return -ENODEV; 1807 1808 disk_block_events(disk); 1809 disk->ev->poll_msecs = intv; 1810 __disk_unblock_events(disk, true); 1811 1812 return count; 1813 } 1814 1815 static const DEVICE_ATTR(events, 0444, disk_events_show, NULL); 1816 static const DEVICE_ATTR(events_async, 0444, disk_events_async_show, NULL); 1817 static const DEVICE_ATTR(events_poll_msecs, 0644, 1818 disk_events_poll_msecs_show, 1819 disk_events_poll_msecs_store); 1820 1821 static const struct attribute *disk_events_attrs[] = { 1822 &dev_attr_events.attr, 1823 &dev_attr_events_async.attr, 1824 &dev_attr_events_poll_msecs.attr, 1825 NULL, 1826 }; 1827 1828 /* 1829 * The default polling interval can be specified by the kernel 1830 * parameter block.events_dfl_poll_msecs which defaults to 0 1831 * (disable). This can also be modified runtime by writing to 1832 * /sys/module/block/parameters/events_dfl_poll_msecs. 1833 */ 1834 static int disk_events_set_dfl_poll_msecs(const char *val, 1835 const struct kernel_param *kp) 1836 { 1837 struct disk_events *ev; 1838 int ret; 1839 1840 ret = param_set_ulong(val, kp); 1841 if (ret < 0) 1842 return ret; 1843 1844 mutex_lock(&disk_events_mutex); 1845 1846 list_for_each_entry(ev, &disk_events, node) 1847 disk_flush_events(ev->disk, 0); 1848 1849 mutex_unlock(&disk_events_mutex); 1850 1851 return 0; 1852 } 1853 1854 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = { 1855 .set = disk_events_set_dfl_poll_msecs, 1856 .get = param_get_ulong, 1857 }; 1858 1859 #undef MODULE_PARAM_PREFIX 1860 #define MODULE_PARAM_PREFIX "block." 1861 1862 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops, 1863 &disk_events_dfl_poll_msecs, 0644); 1864 1865 /* 1866 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events. 1867 */ 1868 static void disk_alloc_events(struct gendisk *disk) 1869 { 1870 struct disk_events *ev; 1871 1872 if (!disk->fops->check_events || !disk->events) 1873 return; 1874 1875 ev = kzalloc(sizeof(*ev), GFP_KERNEL); 1876 if (!ev) { 1877 pr_warn("%s: failed to initialize events\n", disk->disk_name); 1878 return; 1879 } 1880 1881 INIT_LIST_HEAD(&ev->node); 1882 ev->disk = disk; 1883 spin_lock_init(&ev->lock); 1884 mutex_init(&ev->block_mutex); 1885 ev->block = 1; 1886 ev->poll_msecs = -1; 1887 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn); 1888 1889 disk->ev = ev; 1890 } 1891 1892 static void disk_add_events(struct gendisk *disk) 1893 { 1894 /* FIXME: error handling */ 1895 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0) 1896 pr_warn("%s: failed to create sysfs files for events\n", 1897 disk->disk_name); 1898 1899 if (!disk->ev) 1900 return; 1901 1902 mutex_lock(&disk_events_mutex); 1903 list_add_tail(&disk->ev->node, &disk_events); 1904 mutex_unlock(&disk_events_mutex); 1905 1906 /* 1907 * Block count is initialized to 1 and the following initial 1908 * unblock kicks it into action. 1909 */ 1910 __disk_unblock_events(disk, true); 1911 } 1912 1913 static void disk_del_events(struct gendisk *disk) 1914 { 1915 if (disk->ev) { 1916 disk_block_events(disk); 1917 1918 mutex_lock(&disk_events_mutex); 1919 list_del_init(&disk->ev->node); 1920 mutex_unlock(&disk_events_mutex); 1921 } 1922 1923 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs); 1924 } 1925 1926 static void disk_release_events(struct gendisk *disk) 1927 { 1928 /* the block count should be 1 from disk_del_events() */ 1929 WARN_ON_ONCE(disk->ev && disk->ev->block != 1); 1930 kfree(disk->ev); 1931 } 1932