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