1 /* 2 * gendisk handling 3 */ 4 5 #include <linux/module.h> 6 #include <linux/fs.h> 7 #include <linux/genhd.h> 8 #include <linux/kdev_t.h> 9 #include <linux/kernel.h> 10 #include <linux/blkdev.h> 11 #include <linux/backing-dev.h> 12 #include <linux/init.h> 13 #include <linux/spinlock.h> 14 #include <linux/proc_fs.h> 15 #include <linux/seq_file.h> 16 #include <linux/slab.h> 17 #include <linux/kmod.h> 18 #include <linux/kobj_map.h> 19 #include <linux/mutex.h> 20 #include <linux/idr.h> 21 #include <linux/log2.h> 22 #include <linux/pm_runtime.h> 23 24 #include "blk.h" 25 26 static DEFINE_MUTEX(block_class_lock); 27 struct kobject *block_depr; 28 29 /* for extended dynamic devt allocation, currently only one major is used */ 30 #define NR_EXT_DEVT (1 << MINORBITS) 31 32 /* For extended devt allocation. ext_devt_lock prevents look up 33 * results from going away underneath its user. 34 */ 35 static DEFINE_SPINLOCK(ext_devt_lock); 36 static DEFINE_IDR(ext_devt_idr); 37 38 static struct device_type disk_type; 39 40 static void disk_check_events(struct disk_events *ev, 41 unsigned int *clearing_ptr); 42 static void disk_alloc_events(struct gendisk *disk); 43 static void disk_add_events(struct gendisk *disk); 44 static void disk_del_events(struct gendisk *disk); 45 static void disk_release_events(struct gendisk *disk); 46 47 /** 48 * disk_get_part - get partition 49 * @disk: disk to look partition from 50 * @partno: partition number 51 * 52 * Look for partition @partno from @disk. If found, increment 53 * reference count and return it. 54 * 55 * CONTEXT: 56 * Don't care. 57 * 58 * RETURNS: 59 * Pointer to the found partition on success, NULL if not found. 60 */ 61 struct hd_struct *disk_get_part(struct gendisk *disk, int partno) 62 { 63 struct hd_struct *part = NULL; 64 struct disk_part_tbl *ptbl; 65 66 if (unlikely(partno < 0)) 67 return NULL; 68 69 rcu_read_lock(); 70 71 ptbl = rcu_dereference(disk->part_tbl); 72 if (likely(partno < ptbl->len)) { 73 part = rcu_dereference(ptbl->part[partno]); 74 if (part) 75 get_device(part_to_dev(part)); 76 } 77 78 rcu_read_unlock(); 79 80 return part; 81 } 82 EXPORT_SYMBOL_GPL(disk_get_part); 83 84 /** 85 * disk_part_iter_init - initialize partition iterator 86 * @piter: iterator to initialize 87 * @disk: disk to iterate over 88 * @flags: DISK_PITER_* flags 89 * 90 * Initialize @piter so that it iterates over partitions of @disk. 91 * 92 * CONTEXT: 93 * Don't care. 94 */ 95 void disk_part_iter_init(struct disk_part_iter *piter, struct gendisk *disk, 96 unsigned int flags) 97 { 98 struct disk_part_tbl *ptbl; 99 100 rcu_read_lock(); 101 ptbl = rcu_dereference(disk->part_tbl); 102 103 piter->disk = disk; 104 piter->part = NULL; 105 106 if (flags & DISK_PITER_REVERSE) 107 piter->idx = ptbl->len - 1; 108 else if (flags & (DISK_PITER_INCL_PART0 | DISK_PITER_INCL_EMPTY_PART0)) 109 piter->idx = 0; 110 else 111 piter->idx = 1; 112 113 piter->flags = flags; 114 115 rcu_read_unlock(); 116 } 117 EXPORT_SYMBOL_GPL(disk_part_iter_init); 118 119 /** 120 * disk_part_iter_next - proceed iterator to the next partition and return it 121 * @piter: iterator of interest 122 * 123 * Proceed @piter to the next partition and return it. 124 * 125 * CONTEXT: 126 * Don't care. 127 */ 128 struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter) 129 { 130 struct disk_part_tbl *ptbl; 131 int inc, end; 132 133 /* put the last partition */ 134 disk_put_part(piter->part); 135 piter->part = NULL; 136 137 /* get part_tbl */ 138 rcu_read_lock(); 139 ptbl = rcu_dereference(piter->disk->part_tbl); 140 141 /* determine iteration parameters */ 142 if (piter->flags & DISK_PITER_REVERSE) { 143 inc = -1; 144 if (piter->flags & (DISK_PITER_INCL_PART0 | 145 DISK_PITER_INCL_EMPTY_PART0)) 146 end = -1; 147 else 148 end = 0; 149 } else { 150 inc = 1; 151 end = ptbl->len; 152 } 153 154 /* iterate to the next partition */ 155 for (; piter->idx != end; piter->idx += inc) { 156 struct hd_struct *part; 157 158 part = rcu_dereference(ptbl->part[piter->idx]); 159 if (!part) 160 continue; 161 if (!part_nr_sects_read(part) && 162 !(piter->flags & DISK_PITER_INCL_EMPTY) && 163 !(piter->flags & DISK_PITER_INCL_EMPTY_PART0 && 164 piter->idx == 0)) 165 continue; 166 167 get_device(part_to_dev(part)); 168 piter->part = part; 169 piter->idx += inc; 170 break; 171 } 172 173 rcu_read_unlock(); 174 175 return piter->part; 176 } 177 EXPORT_SYMBOL_GPL(disk_part_iter_next); 178 179 /** 180 * disk_part_iter_exit - finish up partition iteration 181 * @piter: iter of interest 182 * 183 * Called when iteration is over. Cleans up @piter. 184 * 185 * CONTEXT: 186 * Don't care. 187 */ 188 void disk_part_iter_exit(struct disk_part_iter *piter) 189 { 190 disk_put_part(piter->part); 191 piter->part = NULL; 192 } 193 EXPORT_SYMBOL_GPL(disk_part_iter_exit); 194 195 static inline int sector_in_part(struct hd_struct *part, sector_t sector) 196 { 197 return part->start_sect <= sector && 198 sector < part->start_sect + part_nr_sects_read(part); 199 } 200 201 /** 202 * disk_map_sector_rcu - map sector to partition 203 * @disk: gendisk of interest 204 * @sector: sector to map 205 * 206 * Find out which partition @sector maps to on @disk. This is 207 * primarily used for stats accounting. 208 * 209 * CONTEXT: 210 * RCU read locked. The returned partition pointer is valid only 211 * while preemption is disabled. 212 * 213 * RETURNS: 214 * Found partition on success, part0 is returned if no partition matches 215 */ 216 struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector) 217 { 218 struct disk_part_tbl *ptbl; 219 struct hd_struct *part; 220 int i; 221 222 ptbl = rcu_dereference(disk->part_tbl); 223 224 part = rcu_dereference(ptbl->last_lookup); 225 if (part && sector_in_part(part, sector)) 226 return part; 227 228 for (i = 1; i < ptbl->len; i++) { 229 part = rcu_dereference(ptbl->part[i]); 230 231 if (part && sector_in_part(part, sector)) { 232 rcu_assign_pointer(ptbl->last_lookup, part); 233 return part; 234 } 235 } 236 return &disk->part0; 237 } 238 EXPORT_SYMBOL_GPL(disk_map_sector_rcu); 239 240 /* 241 * Can be deleted altogether. Later. 242 * 243 */ 244 static struct blk_major_name { 245 struct blk_major_name *next; 246 int major; 247 char name[16]; 248 } *major_names[BLKDEV_MAJOR_HASH_SIZE]; 249 250 /* index in the above - for now: assume no multimajor ranges */ 251 static inline int major_to_index(unsigned major) 252 { 253 return major % BLKDEV_MAJOR_HASH_SIZE; 254 } 255 256 #ifdef CONFIG_PROC_FS 257 void blkdev_show(struct seq_file *seqf, off_t offset) 258 { 259 struct blk_major_name *dp; 260 261 if (offset < BLKDEV_MAJOR_HASH_SIZE) { 262 mutex_lock(&block_class_lock); 263 for (dp = major_names[offset]; dp; dp = dp->next) 264 seq_printf(seqf, "%3d %s\n", dp->major, dp->name); 265 mutex_unlock(&block_class_lock); 266 } 267 } 268 #endif /* CONFIG_PROC_FS */ 269 270 /** 271 * register_blkdev - register a new block device 272 * 273 * @major: the requested major device number [1..255]. If @major=0, try to 274 * allocate any unused major number. 275 * @name: the name of the new block device as a zero terminated string 276 * 277 * The @name must be unique within the system. 278 * 279 * The return value depends on the @major input parameter. 280 * - if a major device number was requested in range [1..255] then the 281 * function returns zero on success, or a negative error code 282 * - if any unused major number was requested with @major=0 parameter 283 * then the return value is the allocated major number in range 284 * [1..255] or a negative error code otherwise 285 */ 286 int register_blkdev(unsigned int major, const char *name) 287 { 288 struct blk_major_name **n, *p; 289 int index, ret = 0; 290 291 mutex_lock(&block_class_lock); 292 293 /* temporary */ 294 if (major == 0) { 295 for (index = ARRAY_SIZE(major_names)-1; index > 0; index--) { 296 if (major_names[index] == NULL) 297 break; 298 } 299 300 if (index == 0) { 301 printk("register_blkdev: failed to get major for %s\n", 302 name); 303 ret = -EBUSY; 304 goto out; 305 } 306 major = index; 307 ret = major; 308 } 309 310 p = kmalloc(sizeof(struct blk_major_name), GFP_KERNEL); 311 if (p == NULL) { 312 ret = -ENOMEM; 313 goto out; 314 } 315 316 p->major = major; 317 strlcpy(p->name, name, sizeof(p->name)); 318 p->next = NULL; 319 index = major_to_index(major); 320 321 for (n = &major_names[index]; *n; n = &(*n)->next) { 322 if ((*n)->major == major) 323 break; 324 } 325 if (!*n) 326 *n = p; 327 else 328 ret = -EBUSY; 329 330 if (ret < 0) { 331 printk("register_blkdev: cannot get major %d for %s\n", 332 major, name); 333 kfree(p); 334 } 335 out: 336 mutex_unlock(&block_class_lock); 337 return ret; 338 } 339 340 EXPORT_SYMBOL(register_blkdev); 341 342 void unregister_blkdev(unsigned int major, const char *name) 343 { 344 struct blk_major_name **n; 345 struct blk_major_name *p = NULL; 346 int index = major_to_index(major); 347 348 mutex_lock(&block_class_lock); 349 for (n = &major_names[index]; *n; n = &(*n)->next) 350 if ((*n)->major == major) 351 break; 352 if (!*n || strcmp((*n)->name, name)) { 353 WARN_ON(1); 354 } else { 355 p = *n; 356 *n = p->next; 357 } 358 mutex_unlock(&block_class_lock); 359 kfree(p); 360 } 361 362 EXPORT_SYMBOL(unregister_blkdev); 363 364 static struct kobj_map *bdev_map; 365 366 /** 367 * blk_mangle_minor - scatter minor numbers apart 368 * @minor: minor number to mangle 369 * 370 * Scatter consecutively allocated @minor number apart if MANGLE_DEVT 371 * is enabled. Mangling twice gives the original value. 372 * 373 * RETURNS: 374 * Mangled value. 375 * 376 * CONTEXT: 377 * Don't care. 378 */ 379 static int blk_mangle_minor(int minor) 380 { 381 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT 382 int i; 383 384 for (i = 0; i < MINORBITS / 2; i++) { 385 int low = minor & (1 << i); 386 int high = minor & (1 << (MINORBITS - 1 - i)); 387 int distance = MINORBITS - 1 - 2 * i; 388 389 minor ^= low | high; /* clear both bits */ 390 low <<= distance; /* swap the positions */ 391 high >>= distance; 392 minor |= low | high; /* and set */ 393 } 394 #endif 395 return minor; 396 } 397 398 /** 399 * blk_alloc_devt - allocate a dev_t for a partition 400 * @part: partition to allocate dev_t for 401 * @devt: out parameter for resulting dev_t 402 * 403 * Allocate a dev_t for block device. 404 * 405 * RETURNS: 406 * 0 on success, allocated dev_t is returned in *@devt. -errno on 407 * failure. 408 * 409 * CONTEXT: 410 * Might sleep. 411 */ 412 int blk_alloc_devt(struct hd_struct *part, dev_t *devt) 413 { 414 struct gendisk *disk = part_to_disk(part); 415 int idx; 416 417 /* in consecutive minor range? */ 418 if (part->partno < disk->minors) { 419 *devt = MKDEV(disk->major, disk->first_minor + part->partno); 420 return 0; 421 } 422 423 /* allocate ext devt */ 424 idr_preload(GFP_KERNEL); 425 426 spin_lock_bh(&ext_devt_lock); 427 idx = idr_alloc(&ext_devt_idr, part, 0, NR_EXT_DEVT, GFP_NOWAIT); 428 spin_unlock_bh(&ext_devt_lock); 429 430 idr_preload_end(); 431 if (idx < 0) 432 return idx == -ENOSPC ? -EBUSY : idx; 433 434 *devt = MKDEV(BLOCK_EXT_MAJOR, blk_mangle_minor(idx)); 435 return 0; 436 } 437 438 /** 439 * blk_free_devt - free a dev_t 440 * @devt: dev_t to free 441 * 442 * Free @devt which was allocated using blk_alloc_devt(). 443 * 444 * CONTEXT: 445 * Might sleep. 446 */ 447 void blk_free_devt(dev_t devt) 448 { 449 if (devt == MKDEV(0, 0)) 450 return; 451 452 if (MAJOR(devt) == BLOCK_EXT_MAJOR) { 453 spin_lock_bh(&ext_devt_lock); 454 idr_remove(&ext_devt_idr, blk_mangle_minor(MINOR(devt))); 455 spin_unlock_bh(&ext_devt_lock); 456 } 457 } 458 459 static char *bdevt_str(dev_t devt, char *buf) 460 { 461 if (MAJOR(devt) <= 0xff && MINOR(devt) <= 0xff) { 462 char tbuf[BDEVT_SIZE]; 463 snprintf(tbuf, BDEVT_SIZE, "%02x%02x", MAJOR(devt), MINOR(devt)); 464 snprintf(buf, BDEVT_SIZE, "%-9s", tbuf); 465 } else 466 snprintf(buf, BDEVT_SIZE, "%03x:%05x", MAJOR(devt), MINOR(devt)); 467 468 return buf; 469 } 470 471 /* 472 * Register device numbers dev..(dev+range-1) 473 * range must be nonzero 474 * The hash chain is sorted on range, so that subranges can override. 475 */ 476 void blk_register_region(dev_t devt, unsigned long range, struct module *module, 477 struct kobject *(*probe)(dev_t, int *, void *), 478 int (*lock)(dev_t, void *), void *data) 479 { 480 kobj_map(bdev_map, devt, range, module, probe, lock, data); 481 } 482 483 EXPORT_SYMBOL(blk_register_region); 484 485 void blk_unregister_region(dev_t devt, unsigned long range) 486 { 487 kobj_unmap(bdev_map, devt, range); 488 } 489 490 EXPORT_SYMBOL(blk_unregister_region); 491 492 static struct kobject *exact_match(dev_t devt, int *partno, void *data) 493 { 494 struct gendisk *p = data; 495 496 return &disk_to_dev(p)->kobj; 497 } 498 499 static int exact_lock(dev_t devt, void *data) 500 { 501 struct gendisk *p = data; 502 503 if (!get_disk(p)) 504 return -1; 505 return 0; 506 } 507 508 static void register_disk(struct gendisk *disk) 509 { 510 struct device *ddev = disk_to_dev(disk); 511 struct block_device *bdev; 512 struct disk_part_iter piter; 513 struct hd_struct *part; 514 int err; 515 516 ddev->parent = disk->driverfs_dev; 517 518 dev_set_name(ddev, "%s", disk->disk_name); 519 520 /* delay uevents, until we scanned partition table */ 521 dev_set_uevent_suppress(ddev, 1); 522 523 if (device_add(ddev)) 524 return; 525 if (!sysfs_deprecated) { 526 err = sysfs_create_link(block_depr, &ddev->kobj, 527 kobject_name(&ddev->kobj)); 528 if (err) { 529 device_del(ddev); 530 return; 531 } 532 } 533 534 /* 535 * avoid probable deadlock caused by allocating memory with 536 * GFP_KERNEL in runtime_resume callback of its all ancestor 537 * devices 538 */ 539 pm_runtime_set_memalloc_noio(ddev, true); 540 541 disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj); 542 disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj); 543 544 /* No minors to use for partitions */ 545 if (!disk_part_scan_enabled(disk)) 546 goto exit; 547 548 /* No such device (e.g., media were just removed) */ 549 if (!get_capacity(disk)) 550 goto exit; 551 552 bdev = bdget_disk(disk, 0); 553 if (!bdev) 554 goto exit; 555 556 bdev->bd_invalidated = 1; 557 err = blkdev_get(bdev, FMODE_READ, NULL); 558 if (err < 0) 559 goto exit; 560 blkdev_put(bdev, FMODE_READ); 561 562 exit: 563 /* announce disk after possible partitions are created */ 564 dev_set_uevent_suppress(ddev, 0); 565 kobject_uevent(&ddev->kobj, KOBJ_ADD); 566 567 /* announce possible partitions */ 568 disk_part_iter_init(&piter, disk, 0); 569 while ((part = disk_part_iter_next(&piter))) 570 kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD); 571 disk_part_iter_exit(&piter); 572 } 573 574 /** 575 * add_disk - add partitioning information to kernel list 576 * @disk: per-device partitioning information 577 * 578 * This function registers the partitioning information in @disk 579 * with the kernel. 580 * 581 * FIXME: error handling 582 */ 583 void add_disk(struct gendisk *disk) 584 { 585 struct backing_dev_info *bdi; 586 dev_t devt; 587 int retval; 588 589 /* minors == 0 indicates to use ext devt from part0 and should 590 * be accompanied with EXT_DEVT flag. Make sure all 591 * parameters make sense. 592 */ 593 WARN_ON(disk->minors && !(disk->major || disk->first_minor)); 594 WARN_ON(!disk->minors && !(disk->flags & GENHD_FL_EXT_DEVT)); 595 596 disk->flags |= GENHD_FL_UP; 597 598 retval = blk_alloc_devt(&disk->part0, &devt); 599 if (retval) { 600 WARN_ON(1); 601 return; 602 } 603 disk_to_dev(disk)->devt = devt; 604 605 /* ->major and ->first_minor aren't supposed to be 606 * dereferenced from here on, but set them just in case. 607 */ 608 disk->major = MAJOR(devt); 609 disk->first_minor = MINOR(devt); 610 611 disk_alloc_events(disk); 612 613 /* Register BDI before referencing it from bdev */ 614 bdi = &disk->queue->backing_dev_info; 615 bdi_register_dev(bdi, disk_devt(disk)); 616 617 blk_register_region(disk_devt(disk), disk->minors, NULL, 618 exact_match, exact_lock, disk); 619 register_disk(disk); 620 blk_register_queue(disk); 621 622 /* 623 * Take an extra ref on queue which will be put on disk_release() 624 * so that it sticks around as long as @disk is there. 625 */ 626 WARN_ON_ONCE(!blk_get_queue(disk->queue)); 627 628 retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj, 629 "bdi"); 630 WARN_ON(retval); 631 632 disk_add_events(disk); 633 blk_integrity_add(disk); 634 } 635 EXPORT_SYMBOL(add_disk); 636 637 void del_gendisk(struct gendisk *disk) 638 { 639 struct disk_part_iter piter; 640 struct hd_struct *part; 641 642 blk_integrity_del(disk); 643 disk_del_events(disk); 644 645 /* invalidate stuff */ 646 disk_part_iter_init(&piter, disk, 647 DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE); 648 while ((part = disk_part_iter_next(&piter))) { 649 invalidate_partition(disk, part->partno); 650 delete_partition(disk, part->partno); 651 } 652 disk_part_iter_exit(&piter); 653 654 invalidate_partition(disk, 0); 655 set_capacity(disk, 0); 656 disk->flags &= ~GENHD_FL_UP; 657 658 sysfs_remove_link(&disk_to_dev(disk)->kobj, "bdi"); 659 blk_unregister_queue(disk); 660 blk_unregister_region(disk_devt(disk), disk->minors); 661 662 part_stat_set_all(&disk->part0, 0); 663 disk->part0.stamp = 0; 664 665 kobject_put(disk->part0.holder_dir); 666 kobject_put(disk->slave_dir); 667 disk->driverfs_dev = NULL; 668 if (!sysfs_deprecated) 669 sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk))); 670 pm_runtime_set_memalloc_noio(disk_to_dev(disk), false); 671 device_del(disk_to_dev(disk)); 672 } 673 EXPORT_SYMBOL(del_gendisk); 674 675 /** 676 * get_gendisk - get partitioning information for a given device 677 * @devt: device to get partitioning information for 678 * @partno: returned partition index 679 * 680 * This function gets the structure containing partitioning 681 * information for the given device @devt. 682 */ 683 struct gendisk *get_gendisk(dev_t devt, int *partno) 684 { 685 struct gendisk *disk = NULL; 686 687 if (MAJOR(devt) != BLOCK_EXT_MAJOR) { 688 struct kobject *kobj; 689 690 kobj = kobj_lookup(bdev_map, devt, partno); 691 if (kobj) 692 disk = dev_to_disk(kobj_to_dev(kobj)); 693 } else { 694 struct hd_struct *part; 695 696 spin_lock_bh(&ext_devt_lock); 697 part = idr_find(&ext_devt_idr, blk_mangle_minor(MINOR(devt))); 698 if (part && get_disk(part_to_disk(part))) { 699 *partno = part->partno; 700 disk = part_to_disk(part); 701 } 702 spin_unlock_bh(&ext_devt_lock); 703 } 704 705 return disk; 706 } 707 EXPORT_SYMBOL(get_gendisk); 708 709 /** 710 * bdget_disk - do bdget() by gendisk and partition number 711 * @disk: gendisk of interest 712 * @partno: partition number 713 * 714 * Find partition @partno from @disk, do bdget() on it. 715 * 716 * CONTEXT: 717 * Don't care. 718 * 719 * RETURNS: 720 * Resulting block_device on success, NULL on failure. 721 */ 722 struct block_device *bdget_disk(struct gendisk *disk, int partno) 723 { 724 struct hd_struct *part; 725 struct block_device *bdev = NULL; 726 727 part = disk_get_part(disk, partno); 728 if (part) 729 bdev = bdget(part_devt(part)); 730 disk_put_part(part); 731 732 return bdev; 733 } 734 EXPORT_SYMBOL(bdget_disk); 735 736 /* 737 * print a full list of all partitions - intended for places where the root 738 * filesystem can't be mounted and thus to give the victim some idea of what 739 * went wrong 740 */ 741 void __init printk_all_partitions(void) 742 { 743 struct class_dev_iter iter; 744 struct device *dev; 745 746 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 747 while ((dev = class_dev_iter_next(&iter))) { 748 struct gendisk *disk = dev_to_disk(dev); 749 struct disk_part_iter piter; 750 struct hd_struct *part; 751 char name_buf[BDEVNAME_SIZE]; 752 char devt_buf[BDEVT_SIZE]; 753 754 /* 755 * Don't show empty devices or things that have been 756 * suppressed 757 */ 758 if (get_capacity(disk) == 0 || 759 (disk->flags & GENHD_FL_SUPPRESS_PARTITION_INFO)) 760 continue; 761 762 /* 763 * Note, unlike /proc/partitions, I am showing the 764 * numbers in hex - the same format as the root= 765 * option takes. 766 */ 767 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0); 768 while ((part = disk_part_iter_next(&piter))) { 769 bool is_part0 = part == &disk->part0; 770 771 printk("%s%s %10llu %s %s", is_part0 ? "" : " ", 772 bdevt_str(part_devt(part), devt_buf), 773 (unsigned long long)part_nr_sects_read(part) >> 1 774 , disk_name(disk, part->partno, name_buf), 775 part->info ? part->info->uuid : ""); 776 if (is_part0) { 777 if (disk->driverfs_dev != NULL && 778 disk->driverfs_dev->driver != NULL) 779 printk(" driver: %s\n", 780 disk->driverfs_dev->driver->name); 781 else 782 printk(" (driver?)\n"); 783 } else 784 printk("\n"); 785 } 786 disk_part_iter_exit(&piter); 787 } 788 class_dev_iter_exit(&iter); 789 } 790 791 #ifdef CONFIG_PROC_FS 792 /* iterator */ 793 static void *disk_seqf_start(struct seq_file *seqf, loff_t *pos) 794 { 795 loff_t skip = *pos; 796 struct class_dev_iter *iter; 797 struct device *dev; 798 799 iter = kmalloc(sizeof(*iter), GFP_KERNEL); 800 if (!iter) 801 return ERR_PTR(-ENOMEM); 802 803 seqf->private = iter; 804 class_dev_iter_init(iter, &block_class, NULL, &disk_type); 805 do { 806 dev = class_dev_iter_next(iter); 807 if (!dev) 808 return NULL; 809 } while (skip--); 810 811 return dev_to_disk(dev); 812 } 813 814 static void *disk_seqf_next(struct seq_file *seqf, void *v, loff_t *pos) 815 { 816 struct device *dev; 817 818 (*pos)++; 819 dev = class_dev_iter_next(seqf->private); 820 if (dev) 821 return dev_to_disk(dev); 822 823 return NULL; 824 } 825 826 static void disk_seqf_stop(struct seq_file *seqf, void *v) 827 { 828 struct class_dev_iter *iter = seqf->private; 829 830 /* stop is called even after start failed :-( */ 831 if (iter) { 832 class_dev_iter_exit(iter); 833 kfree(iter); 834 } 835 } 836 837 static void *show_partition_start(struct seq_file *seqf, loff_t *pos) 838 { 839 void *p; 840 841 p = disk_seqf_start(seqf, pos); 842 if (!IS_ERR_OR_NULL(p) && !*pos) 843 seq_puts(seqf, "major minor #blocks name\n\n"); 844 return p; 845 } 846 847 static int show_partition(struct seq_file *seqf, void *v) 848 { 849 struct gendisk *sgp = v; 850 struct disk_part_iter piter; 851 struct hd_struct *part; 852 char buf[BDEVNAME_SIZE]; 853 854 /* Don't show non-partitionable removeable devices or empty devices */ 855 if (!get_capacity(sgp) || (!disk_max_parts(sgp) && 856 (sgp->flags & GENHD_FL_REMOVABLE))) 857 return 0; 858 if (sgp->flags & GENHD_FL_SUPPRESS_PARTITION_INFO) 859 return 0; 860 861 /* show the full disk and all non-0 size partitions of it */ 862 disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0); 863 while ((part = disk_part_iter_next(&piter))) 864 seq_printf(seqf, "%4d %7d %10llu %s\n", 865 MAJOR(part_devt(part)), MINOR(part_devt(part)), 866 (unsigned long long)part_nr_sects_read(part) >> 1, 867 disk_name(sgp, part->partno, buf)); 868 disk_part_iter_exit(&piter); 869 870 return 0; 871 } 872 873 static const struct seq_operations partitions_op = { 874 .start = show_partition_start, 875 .next = disk_seqf_next, 876 .stop = disk_seqf_stop, 877 .show = show_partition 878 }; 879 880 static int partitions_open(struct inode *inode, struct file *file) 881 { 882 return seq_open(file, &partitions_op); 883 } 884 885 static const struct file_operations proc_partitions_operations = { 886 .open = partitions_open, 887 .read = seq_read, 888 .llseek = seq_lseek, 889 .release = seq_release, 890 }; 891 #endif 892 893 894 static struct kobject *base_probe(dev_t devt, int *partno, void *data) 895 { 896 if (request_module("block-major-%d-%d", MAJOR(devt), MINOR(devt)) > 0) 897 /* Make old-style 2.4 aliases work */ 898 request_module("block-major-%d", MAJOR(devt)); 899 return NULL; 900 } 901 902 static int __init genhd_device_init(void) 903 { 904 int error; 905 906 block_class.dev_kobj = sysfs_dev_block_kobj; 907 error = class_register(&block_class); 908 if (unlikely(error)) 909 return error; 910 bdev_map = kobj_map_init(base_probe, &block_class_lock); 911 blk_dev_init(); 912 913 register_blkdev(BLOCK_EXT_MAJOR, "blkext"); 914 915 /* create top-level block dir */ 916 if (!sysfs_deprecated) 917 block_depr = kobject_create_and_add("block", NULL); 918 return 0; 919 } 920 921 subsys_initcall(genhd_device_init); 922 923 static ssize_t disk_range_show(struct device *dev, 924 struct device_attribute *attr, char *buf) 925 { 926 struct gendisk *disk = dev_to_disk(dev); 927 928 return sprintf(buf, "%d\n", disk->minors); 929 } 930 931 static ssize_t disk_ext_range_show(struct device *dev, 932 struct device_attribute *attr, char *buf) 933 { 934 struct gendisk *disk = dev_to_disk(dev); 935 936 return sprintf(buf, "%d\n", disk_max_parts(disk)); 937 } 938 939 static ssize_t disk_removable_show(struct device *dev, 940 struct device_attribute *attr, char *buf) 941 { 942 struct gendisk *disk = dev_to_disk(dev); 943 944 return sprintf(buf, "%d\n", 945 (disk->flags & GENHD_FL_REMOVABLE ? 1 : 0)); 946 } 947 948 static ssize_t disk_ro_show(struct device *dev, 949 struct device_attribute *attr, char *buf) 950 { 951 struct gendisk *disk = dev_to_disk(dev); 952 953 return sprintf(buf, "%d\n", get_disk_ro(disk) ? 1 : 0); 954 } 955 956 static ssize_t disk_capability_show(struct device *dev, 957 struct device_attribute *attr, char *buf) 958 { 959 struct gendisk *disk = dev_to_disk(dev); 960 961 return sprintf(buf, "%x\n", disk->flags); 962 } 963 964 static ssize_t disk_alignment_offset_show(struct device *dev, 965 struct device_attribute *attr, 966 char *buf) 967 { 968 struct gendisk *disk = dev_to_disk(dev); 969 970 return sprintf(buf, "%d\n", queue_alignment_offset(disk->queue)); 971 } 972 973 static ssize_t disk_discard_alignment_show(struct device *dev, 974 struct device_attribute *attr, 975 char *buf) 976 { 977 struct gendisk *disk = dev_to_disk(dev); 978 979 return sprintf(buf, "%d\n", queue_discard_alignment(disk->queue)); 980 } 981 982 static DEVICE_ATTR(range, S_IRUGO, disk_range_show, NULL); 983 static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL); 984 static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL); 985 static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL); 986 static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL); 987 static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL); 988 static DEVICE_ATTR(discard_alignment, S_IRUGO, disk_discard_alignment_show, 989 NULL); 990 static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL); 991 static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL); 992 static DEVICE_ATTR(inflight, S_IRUGO, part_inflight_show, NULL); 993 #ifdef CONFIG_FAIL_MAKE_REQUEST 994 static struct device_attribute dev_attr_fail = 995 __ATTR(make-it-fail, S_IRUGO|S_IWUSR, part_fail_show, part_fail_store); 996 #endif 997 #ifdef CONFIG_FAIL_IO_TIMEOUT 998 static struct device_attribute dev_attr_fail_timeout = 999 __ATTR(io-timeout-fail, S_IRUGO|S_IWUSR, part_timeout_show, 1000 part_timeout_store); 1001 #endif 1002 1003 static struct attribute *disk_attrs[] = { 1004 &dev_attr_range.attr, 1005 &dev_attr_ext_range.attr, 1006 &dev_attr_removable.attr, 1007 &dev_attr_ro.attr, 1008 &dev_attr_size.attr, 1009 &dev_attr_alignment_offset.attr, 1010 &dev_attr_discard_alignment.attr, 1011 &dev_attr_capability.attr, 1012 &dev_attr_stat.attr, 1013 &dev_attr_inflight.attr, 1014 #ifdef CONFIG_FAIL_MAKE_REQUEST 1015 &dev_attr_fail.attr, 1016 #endif 1017 #ifdef CONFIG_FAIL_IO_TIMEOUT 1018 &dev_attr_fail_timeout.attr, 1019 #endif 1020 NULL 1021 }; 1022 1023 static struct attribute_group disk_attr_group = { 1024 .attrs = disk_attrs, 1025 }; 1026 1027 static const struct attribute_group *disk_attr_groups[] = { 1028 &disk_attr_group, 1029 NULL 1030 }; 1031 1032 /** 1033 * disk_replace_part_tbl - replace disk->part_tbl in RCU-safe way 1034 * @disk: disk to replace part_tbl for 1035 * @new_ptbl: new part_tbl to install 1036 * 1037 * Replace disk->part_tbl with @new_ptbl in RCU-safe way. The 1038 * original ptbl is freed using RCU callback. 1039 * 1040 * LOCKING: 1041 * Matching bd_mutx locked. 1042 */ 1043 static void disk_replace_part_tbl(struct gendisk *disk, 1044 struct disk_part_tbl *new_ptbl) 1045 { 1046 struct disk_part_tbl *old_ptbl = disk->part_tbl; 1047 1048 rcu_assign_pointer(disk->part_tbl, new_ptbl); 1049 1050 if (old_ptbl) { 1051 rcu_assign_pointer(old_ptbl->last_lookup, NULL); 1052 kfree_rcu(old_ptbl, rcu_head); 1053 } 1054 } 1055 1056 /** 1057 * disk_expand_part_tbl - expand disk->part_tbl 1058 * @disk: disk to expand part_tbl for 1059 * @partno: expand such that this partno can fit in 1060 * 1061 * Expand disk->part_tbl such that @partno can fit in. disk->part_tbl 1062 * uses RCU to allow unlocked dereferencing for stats and other stuff. 1063 * 1064 * LOCKING: 1065 * Matching bd_mutex locked, might sleep. 1066 * 1067 * RETURNS: 1068 * 0 on success, -errno on failure. 1069 */ 1070 int disk_expand_part_tbl(struct gendisk *disk, int partno) 1071 { 1072 struct disk_part_tbl *old_ptbl = disk->part_tbl; 1073 struct disk_part_tbl *new_ptbl; 1074 int len = old_ptbl ? old_ptbl->len : 0; 1075 int i, target; 1076 size_t size; 1077 1078 /* 1079 * check for int overflow, since we can get here from blkpg_ioctl() 1080 * with a user passed 'partno'. 1081 */ 1082 target = partno + 1; 1083 if (target < 0) 1084 return -EINVAL; 1085 1086 /* disk_max_parts() is zero during initialization, ignore if so */ 1087 if (disk_max_parts(disk) && target > disk_max_parts(disk)) 1088 return -EINVAL; 1089 1090 if (target <= len) 1091 return 0; 1092 1093 size = sizeof(*new_ptbl) + target * sizeof(new_ptbl->part[0]); 1094 new_ptbl = kzalloc_node(size, GFP_KERNEL, disk->node_id); 1095 if (!new_ptbl) 1096 return -ENOMEM; 1097 1098 new_ptbl->len = target; 1099 1100 for (i = 0; i < len; i++) 1101 rcu_assign_pointer(new_ptbl->part[i], old_ptbl->part[i]); 1102 1103 disk_replace_part_tbl(disk, new_ptbl); 1104 return 0; 1105 } 1106 1107 static void disk_release(struct device *dev) 1108 { 1109 struct gendisk *disk = dev_to_disk(dev); 1110 1111 blk_free_devt(dev->devt); 1112 disk_release_events(disk); 1113 kfree(disk->random); 1114 disk_replace_part_tbl(disk, NULL); 1115 hd_free_part(&disk->part0); 1116 if (disk->queue) 1117 blk_put_queue(disk->queue); 1118 kfree(disk); 1119 } 1120 struct class block_class = { 1121 .name = "block", 1122 }; 1123 1124 static char *block_devnode(struct device *dev, umode_t *mode, 1125 kuid_t *uid, kgid_t *gid) 1126 { 1127 struct gendisk *disk = dev_to_disk(dev); 1128 1129 if (disk->devnode) 1130 return disk->devnode(disk, mode); 1131 return NULL; 1132 } 1133 1134 static struct device_type disk_type = { 1135 .name = "disk", 1136 .groups = disk_attr_groups, 1137 .release = disk_release, 1138 .devnode = block_devnode, 1139 }; 1140 1141 #ifdef CONFIG_PROC_FS 1142 /* 1143 * aggregate disk stat collector. Uses the same stats that the sysfs 1144 * entries do, above, but makes them available through one seq_file. 1145 * 1146 * The output looks suspiciously like /proc/partitions with a bunch of 1147 * extra fields. 1148 */ 1149 static int diskstats_show(struct seq_file *seqf, void *v) 1150 { 1151 struct gendisk *gp = v; 1152 struct disk_part_iter piter; 1153 struct hd_struct *hd; 1154 char buf[BDEVNAME_SIZE]; 1155 int cpu; 1156 1157 /* 1158 if (&disk_to_dev(gp)->kobj.entry == block_class.devices.next) 1159 seq_puts(seqf, "major minor name" 1160 " rio rmerge rsect ruse wio wmerge " 1161 "wsect wuse running use aveq" 1162 "\n\n"); 1163 */ 1164 1165 disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0); 1166 while ((hd = disk_part_iter_next(&piter))) { 1167 cpu = part_stat_lock(); 1168 part_round_stats(cpu, hd); 1169 part_stat_unlock(); 1170 seq_printf(seqf, "%4d %7d %s %lu %lu %lu " 1171 "%u %lu %lu %lu %u %u %u %u\n", 1172 MAJOR(part_devt(hd)), MINOR(part_devt(hd)), 1173 disk_name(gp, hd->partno, buf), 1174 part_stat_read(hd, ios[READ]), 1175 part_stat_read(hd, merges[READ]), 1176 part_stat_read(hd, sectors[READ]), 1177 jiffies_to_msecs(part_stat_read(hd, ticks[READ])), 1178 part_stat_read(hd, ios[WRITE]), 1179 part_stat_read(hd, merges[WRITE]), 1180 part_stat_read(hd, sectors[WRITE]), 1181 jiffies_to_msecs(part_stat_read(hd, ticks[WRITE])), 1182 part_in_flight(hd), 1183 jiffies_to_msecs(part_stat_read(hd, io_ticks)), 1184 jiffies_to_msecs(part_stat_read(hd, time_in_queue)) 1185 ); 1186 } 1187 disk_part_iter_exit(&piter); 1188 1189 return 0; 1190 } 1191 1192 static const struct seq_operations diskstats_op = { 1193 .start = disk_seqf_start, 1194 .next = disk_seqf_next, 1195 .stop = disk_seqf_stop, 1196 .show = diskstats_show 1197 }; 1198 1199 static int diskstats_open(struct inode *inode, struct file *file) 1200 { 1201 return seq_open(file, &diskstats_op); 1202 } 1203 1204 static const struct file_operations proc_diskstats_operations = { 1205 .open = diskstats_open, 1206 .read = seq_read, 1207 .llseek = seq_lseek, 1208 .release = seq_release, 1209 }; 1210 1211 static int __init proc_genhd_init(void) 1212 { 1213 proc_create("diskstats", 0, NULL, &proc_diskstats_operations); 1214 proc_create("partitions", 0, NULL, &proc_partitions_operations); 1215 return 0; 1216 } 1217 module_init(proc_genhd_init); 1218 #endif /* CONFIG_PROC_FS */ 1219 1220 dev_t blk_lookup_devt(const char *name, int partno) 1221 { 1222 dev_t devt = MKDEV(0, 0); 1223 struct class_dev_iter iter; 1224 struct device *dev; 1225 1226 class_dev_iter_init(&iter, &block_class, NULL, &disk_type); 1227 while ((dev = class_dev_iter_next(&iter))) { 1228 struct gendisk *disk = dev_to_disk(dev); 1229 struct hd_struct *part; 1230 1231 if (strcmp(dev_name(dev), name)) 1232 continue; 1233 1234 if (partno < disk->minors) { 1235 /* We need to return the right devno, even 1236 * if the partition doesn't exist yet. 1237 */ 1238 devt = MKDEV(MAJOR(dev->devt), 1239 MINOR(dev->devt) + partno); 1240 break; 1241 } 1242 part = disk_get_part(disk, partno); 1243 if (part) { 1244 devt = part_devt(part); 1245 disk_put_part(part); 1246 break; 1247 } 1248 disk_put_part(part); 1249 } 1250 class_dev_iter_exit(&iter); 1251 return devt; 1252 } 1253 EXPORT_SYMBOL(blk_lookup_devt); 1254 1255 struct gendisk *alloc_disk(int minors) 1256 { 1257 return alloc_disk_node(minors, NUMA_NO_NODE); 1258 } 1259 EXPORT_SYMBOL(alloc_disk); 1260 1261 struct gendisk *alloc_disk_node(int minors, int node_id) 1262 { 1263 struct gendisk *disk; 1264 1265 disk = kzalloc_node(sizeof(struct gendisk), GFP_KERNEL, node_id); 1266 if (disk) { 1267 if (!init_part_stats(&disk->part0)) { 1268 kfree(disk); 1269 return NULL; 1270 } 1271 disk->node_id = node_id; 1272 if (disk_expand_part_tbl(disk, 0)) { 1273 free_part_stats(&disk->part0); 1274 kfree(disk); 1275 return NULL; 1276 } 1277 disk->part_tbl->part[0] = &disk->part0; 1278 1279 /* 1280 * set_capacity() and get_capacity() currently don't use 1281 * seqcounter to read/update the part0->nr_sects. Still init 1282 * the counter as we can read the sectors in IO submission 1283 * patch using seqence counters. 1284 * 1285 * TODO: Ideally set_capacity() and get_capacity() should be 1286 * converted to make use of bd_mutex and sequence counters. 1287 */ 1288 seqcount_init(&disk->part0.nr_sects_seq); 1289 if (hd_ref_init(&disk->part0)) { 1290 hd_free_part(&disk->part0); 1291 kfree(disk); 1292 return NULL; 1293 } 1294 1295 disk->minors = minors; 1296 rand_initialize_disk(disk); 1297 disk_to_dev(disk)->class = &block_class; 1298 disk_to_dev(disk)->type = &disk_type; 1299 device_initialize(disk_to_dev(disk)); 1300 } 1301 return disk; 1302 } 1303 EXPORT_SYMBOL(alloc_disk_node); 1304 1305 struct kobject *get_disk(struct gendisk *disk) 1306 { 1307 struct module *owner; 1308 struct kobject *kobj; 1309 1310 if (!disk->fops) 1311 return NULL; 1312 owner = disk->fops->owner; 1313 if (owner && !try_module_get(owner)) 1314 return NULL; 1315 kobj = kobject_get(&disk_to_dev(disk)->kobj); 1316 if (kobj == NULL) { 1317 module_put(owner); 1318 return NULL; 1319 } 1320 return kobj; 1321 1322 } 1323 1324 EXPORT_SYMBOL(get_disk); 1325 1326 void put_disk(struct gendisk *disk) 1327 { 1328 if (disk) 1329 kobject_put(&disk_to_dev(disk)->kobj); 1330 } 1331 1332 EXPORT_SYMBOL(put_disk); 1333 1334 static void set_disk_ro_uevent(struct gendisk *gd, int ro) 1335 { 1336 char event[] = "DISK_RO=1"; 1337 char *envp[] = { event, NULL }; 1338 1339 if (!ro) 1340 event[8] = '0'; 1341 kobject_uevent_env(&disk_to_dev(gd)->kobj, KOBJ_CHANGE, envp); 1342 } 1343 1344 void set_device_ro(struct block_device *bdev, int flag) 1345 { 1346 bdev->bd_part->policy = flag; 1347 } 1348 1349 EXPORT_SYMBOL(set_device_ro); 1350 1351 void set_disk_ro(struct gendisk *disk, int flag) 1352 { 1353 struct disk_part_iter piter; 1354 struct hd_struct *part; 1355 1356 if (disk->part0.policy != flag) { 1357 set_disk_ro_uevent(disk, flag); 1358 disk->part0.policy = flag; 1359 } 1360 1361 disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY); 1362 while ((part = disk_part_iter_next(&piter))) 1363 part->policy = flag; 1364 disk_part_iter_exit(&piter); 1365 } 1366 1367 EXPORT_SYMBOL(set_disk_ro); 1368 1369 int bdev_read_only(struct block_device *bdev) 1370 { 1371 if (!bdev) 1372 return 0; 1373 return bdev->bd_part->policy; 1374 } 1375 1376 EXPORT_SYMBOL(bdev_read_only); 1377 1378 int invalidate_partition(struct gendisk *disk, int partno) 1379 { 1380 int res = 0; 1381 struct block_device *bdev = bdget_disk(disk, partno); 1382 if (bdev) { 1383 fsync_bdev(bdev); 1384 res = __invalidate_device(bdev, true); 1385 bdput(bdev); 1386 } 1387 return res; 1388 } 1389 1390 EXPORT_SYMBOL(invalidate_partition); 1391 1392 /* 1393 * Disk events - monitor disk events like media change and eject request. 1394 */ 1395 struct disk_events { 1396 struct list_head node; /* all disk_event's */ 1397 struct gendisk *disk; /* the associated disk */ 1398 spinlock_t lock; 1399 1400 struct mutex block_mutex; /* protects blocking */ 1401 int block; /* event blocking depth */ 1402 unsigned int pending; /* events already sent out */ 1403 unsigned int clearing; /* events being cleared */ 1404 1405 long poll_msecs; /* interval, -1 for default */ 1406 struct delayed_work dwork; 1407 }; 1408 1409 static const char *disk_events_strs[] = { 1410 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "media_change", 1411 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "eject_request", 1412 }; 1413 1414 static char *disk_uevents[] = { 1415 [ilog2(DISK_EVENT_MEDIA_CHANGE)] = "DISK_MEDIA_CHANGE=1", 1416 [ilog2(DISK_EVENT_EJECT_REQUEST)] = "DISK_EJECT_REQUEST=1", 1417 }; 1418 1419 /* list of all disk_events */ 1420 static DEFINE_MUTEX(disk_events_mutex); 1421 static LIST_HEAD(disk_events); 1422 1423 /* disable in-kernel polling by default */ 1424 static unsigned long disk_events_dfl_poll_msecs = 0; 1425 1426 static unsigned long disk_events_poll_jiffies(struct gendisk *disk) 1427 { 1428 struct disk_events *ev = disk->ev; 1429 long intv_msecs = 0; 1430 1431 /* 1432 * If device-specific poll interval is set, always use it. If 1433 * the default is being used, poll iff there are events which 1434 * can't be monitored asynchronously. 1435 */ 1436 if (ev->poll_msecs >= 0) 1437 intv_msecs = ev->poll_msecs; 1438 else if (disk->events & ~disk->async_events) 1439 intv_msecs = disk_events_dfl_poll_msecs; 1440 1441 return msecs_to_jiffies(intv_msecs); 1442 } 1443 1444 /** 1445 * disk_block_events - block and flush disk event checking 1446 * @disk: disk to block events for 1447 * 1448 * On return from this function, it is guaranteed that event checking 1449 * isn't in progress and won't happen until unblocked by 1450 * disk_unblock_events(). Events blocking is counted and the actual 1451 * unblocking happens after the matching number of unblocks are done. 1452 * 1453 * Note that this intentionally does not block event checking from 1454 * disk_clear_events(). 1455 * 1456 * CONTEXT: 1457 * Might sleep. 1458 */ 1459 void disk_block_events(struct gendisk *disk) 1460 { 1461 struct disk_events *ev = disk->ev; 1462 unsigned long flags; 1463 bool cancel; 1464 1465 if (!ev) 1466 return; 1467 1468 /* 1469 * Outer mutex ensures that the first blocker completes canceling 1470 * the event work before further blockers are allowed to finish. 1471 */ 1472 mutex_lock(&ev->block_mutex); 1473 1474 spin_lock_irqsave(&ev->lock, flags); 1475 cancel = !ev->block++; 1476 spin_unlock_irqrestore(&ev->lock, flags); 1477 1478 if (cancel) 1479 cancel_delayed_work_sync(&disk->ev->dwork); 1480 1481 mutex_unlock(&ev->block_mutex); 1482 } 1483 1484 static void __disk_unblock_events(struct gendisk *disk, bool check_now) 1485 { 1486 struct disk_events *ev = disk->ev; 1487 unsigned long intv; 1488 unsigned long flags; 1489 1490 spin_lock_irqsave(&ev->lock, flags); 1491 1492 if (WARN_ON_ONCE(ev->block <= 0)) 1493 goto out_unlock; 1494 1495 if (--ev->block) 1496 goto out_unlock; 1497 1498 /* 1499 * Not exactly a latency critical operation, set poll timer 1500 * slack to 25% and kick event check. 1501 */ 1502 intv = disk_events_poll_jiffies(disk); 1503 set_timer_slack(&ev->dwork.timer, intv / 4); 1504 if (check_now) 1505 queue_delayed_work(system_freezable_power_efficient_wq, 1506 &ev->dwork, 0); 1507 else if (intv) 1508 queue_delayed_work(system_freezable_power_efficient_wq, 1509 &ev->dwork, intv); 1510 out_unlock: 1511 spin_unlock_irqrestore(&ev->lock, flags); 1512 } 1513 1514 /** 1515 * disk_unblock_events - unblock disk event checking 1516 * @disk: disk to unblock events for 1517 * 1518 * Undo disk_block_events(). When the block count reaches zero, it 1519 * starts events polling if configured. 1520 * 1521 * CONTEXT: 1522 * Don't care. Safe to call from irq context. 1523 */ 1524 void disk_unblock_events(struct gendisk *disk) 1525 { 1526 if (disk->ev) 1527 __disk_unblock_events(disk, false); 1528 } 1529 1530 /** 1531 * disk_flush_events - schedule immediate event checking and flushing 1532 * @disk: disk to check and flush events for 1533 * @mask: events to flush 1534 * 1535 * Schedule immediate event checking on @disk if not blocked. Events in 1536 * @mask are scheduled to be cleared from the driver. Note that this 1537 * doesn't clear the events from @disk->ev. 1538 * 1539 * CONTEXT: 1540 * If @mask is non-zero must be called with bdev->bd_mutex held. 1541 */ 1542 void disk_flush_events(struct gendisk *disk, unsigned int mask) 1543 { 1544 struct disk_events *ev = disk->ev; 1545 1546 if (!ev) 1547 return; 1548 1549 spin_lock_irq(&ev->lock); 1550 ev->clearing |= mask; 1551 if (!ev->block) 1552 mod_delayed_work(system_freezable_power_efficient_wq, 1553 &ev->dwork, 0); 1554 spin_unlock_irq(&ev->lock); 1555 } 1556 1557 /** 1558 * disk_clear_events - synchronously check, clear and return pending events 1559 * @disk: disk to fetch and clear events from 1560 * @mask: mask of events to be fetched and cleared 1561 * 1562 * Disk events are synchronously checked and pending events in @mask 1563 * are cleared and returned. This ignores the block count. 1564 * 1565 * CONTEXT: 1566 * Might sleep. 1567 */ 1568 unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask) 1569 { 1570 const struct block_device_operations *bdops = disk->fops; 1571 struct disk_events *ev = disk->ev; 1572 unsigned int pending; 1573 unsigned int clearing = mask; 1574 1575 if (!ev) { 1576 /* for drivers still using the old ->media_changed method */ 1577 if ((mask & DISK_EVENT_MEDIA_CHANGE) && 1578 bdops->media_changed && bdops->media_changed(disk)) 1579 return DISK_EVENT_MEDIA_CHANGE; 1580 return 0; 1581 } 1582 1583 disk_block_events(disk); 1584 1585 /* 1586 * store the union of mask and ev->clearing on the stack so that the 1587 * race with disk_flush_events does not cause ambiguity (ev->clearing 1588 * can still be modified even if events are blocked). 1589 */ 1590 spin_lock_irq(&ev->lock); 1591 clearing |= ev->clearing; 1592 ev->clearing = 0; 1593 spin_unlock_irq(&ev->lock); 1594 1595 disk_check_events(ev, &clearing); 1596 /* 1597 * if ev->clearing is not 0, the disk_flush_events got called in the 1598 * middle of this function, so we want to run the workfn without delay. 1599 */ 1600 __disk_unblock_events(disk, ev->clearing ? true : false); 1601 1602 /* then, fetch and clear pending events */ 1603 spin_lock_irq(&ev->lock); 1604 pending = ev->pending & mask; 1605 ev->pending &= ~mask; 1606 spin_unlock_irq(&ev->lock); 1607 WARN_ON_ONCE(clearing & mask); 1608 1609 return pending; 1610 } 1611 1612 /* 1613 * Separate this part out so that a different pointer for clearing_ptr can be 1614 * passed in for disk_clear_events. 1615 */ 1616 static void disk_events_workfn(struct work_struct *work) 1617 { 1618 struct delayed_work *dwork = to_delayed_work(work); 1619 struct disk_events *ev = container_of(dwork, struct disk_events, dwork); 1620 1621 disk_check_events(ev, &ev->clearing); 1622 } 1623 1624 static void disk_check_events(struct disk_events *ev, 1625 unsigned int *clearing_ptr) 1626 { 1627 struct gendisk *disk = ev->disk; 1628 char *envp[ARRAY_SIZE(disk_uevents) + 1] = { }; 1629 unsigned int clearing = *clearing_ptr; 1630 unsigned int events; 1631 unsigned long intv; 1632 int nr_events = 0, i; 1633 1634 /* check events */ 1635 events = disk->fops->check_events(disk, clearing); 1636 1637 /* accumulate pending events and schedule next poll if necessary */ 1638 spin_lock_irq(&ev->lock); 1639 1640 events &= ~ev->pending; 1641 ev->pending |= events; 1642 *clearing_ptr &= ~clearing; 1643 1644 intv = disk_events_poll_jiffies(disk); 1645 if (!ev->block && intv) 1646 queue_delayed_work(system_freezable_power_efficient_wq, 1647 &ev->dwork, intv); 1648 1649 spin_unlock_irq(&ev->lock); 1650 1651 /* 1652 * Tell userland about new events. Only the events listed in 1653 * @disk->events are reported. Unlisted events are processed the 1654 * same internally but never get reported to userland. 1655 */ 1656 for (i = 0; i < ARRAY_SIZE(disk_uevents); i++) 1657 if (events & disk->events & (1 << i)) 1658 envp[nr_events++] = disk_uevents[i]; 1659 1660 if (nr_events) 1661 kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp); 1662 } 1663 1664 /* 1665 * A disk events enabled device has the following sysfs nodes under 1666 * its /sys/block/X/ directory. 1667 * 1668 * events : list of all supported events 1669 * events_async : list of events which can be detected w/o polling 1670 * events_poll_msecs : polling interval, 0: disable, -1: system default 1671 */ 1672 static ssize_t __disk_events_show(unsigned int events, char *buf) 1673 { 1674 const char *delim = ""; 1675 ssize_t pos = 0; 1676 int i; 1677 1678 for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++) 1679 if (events & (1 << i)) { 1680 pos += sprintf(buf + pos, "%s%s", 1681 delim, disk_events_strs[i]); 1682 delim = " "; 1683 } 1684 if (pos) 1685 pos += sprintf(buf + pos, "\n"); 1686 return pos; 1687 } 1688 1689 static ssize_t disk_events_show(struct device *dev, 1690 struct device_attribute *attr, char *buf) 1691 { 1692 struct gendisk *disk = dev_to_disk(dev); 1693 1694 return __disk_events_show(disk->events, buf); 1695 } 1696 1697 static ssize_t disk_events_async_show(struct device *dev, 1698 struct device_attribute *attr, char *buf) 1699 { 1700 struct gendisk *disk = dev_to_disk(dev); 1701 1702 return __disk_events_show(disk->async_events, buf); 1703 } 1704 1705 static ssize_t disk_events_poll_msecs_show(struct device *dev, 1706 struct device_attribute *attr, 1707 char *buf) 1708 { 1709 struct gendisk *disk = dev_to_disk(dev); 1710 1711 return sprintf(buf, "%ld\n", disk->ev->poll_msecs); 1712 } 1713 1714 static ssize_t disk_events_poll_msecs_store(struct device *dev, 1715 struct device_attribute *attr, 1716 const char *buf, size_t count) 1717 { 1718 struct gendisk *disk = dev_to_disk(dev); 1719 long intv; 1720 1721 if (!count || !sscanf(buf, "%ld", &intv)) 1722 return -EINVAL; 1723 1724 if (intv < 0 && intv != -1) 1725 return -EINVAL; 1726 1727 disk_block_events(disk); 1728 disk->ev->poll_msecs = intv; 1729 __disk_unblock_events(disk, true); 1730 1731 return count; 1732 } 1733 1734 static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL); 1735 static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL); 1736 static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR, 1737 disk_events_poll_msecs_show, 1738 disk_events_poll_msecs_store); 1739 1740 static const struct attribute *disk_events_attrs[] = { 1741 &dev_attr_events.attr, 1742 &dev_attr_events_async.attr, 1743 &dev_attr_events_poll_msecs.attr, 1744 NULL, 1745 }; 1746 1747 /* 1748 * The default polling interval can be specified by the kernel 1749 * parameter block.events_dfl_poll_msecs which defaults to 0 1750 * (disable). This can also be modified runtime by writing to 1751 * /sys/module/block/events_dfl_poll_msecs. 1752 */ 1753 static int disk_events_set_dfl_poll_msecs(const char *val, 1754 const struct kernel_param *kp) 1755 { 1756 struct disk_events *ev; 1757 int ret; 1758 1759 ret = param_set_ulong(val, kp); 1760 if (ret < 0) 1761 return ret; 1762 1763 mutex_lock(&disk_events_mutex); 1764 1765 list_for_each_entry(ev, &disk_events, node) 1766 disk_flush_events(ev->disk, 0); 1767 1768 mutex_unlock(&disk_events_mutex); 1769 1770 return 0; 1771 } 1772 1773 static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = { 1774 .set = disk_events_set_dfl_poll_msecs, 1775 .get = param_get_ulong, 1776 }; 1777 1778 #undef MODULE_PARAM_PREFIX 1779 #define MODULE_PARAM_PREFIX "block." 1780 1781 module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops, 1782 &disk_events_dfl_poll_msecs, 0644); 1783 1784 /* 1785 * disk_{alloc|add|del|release}_events - initialize and destroy disk_events. 1786 */ 1787 static void disk_alloc_events(struct gendisk *disk) 1788 { 1789 struct disk_events *ev; 1790 1791 if (!disk->fops->check_events) 1792 return; 1793 1794 ev = kzalloc(sizeof(*ev), GFP_KERNEL); 1795 if (!ev) { 1796 pr_warn("%s: failed to initialize events\n", disk->disk_name); 1797 return; 1798 } 1799 1800 INIT_LIST_HEAD(&ev->node); 1801 ev->disk = disk; 1802 spin_lock_init(&ev->lock); 1803 mutex_init(&ev->block_mutex); 1804 ev->block = 1; 1805 ev->poll_msecs = -1; 1806 INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn); 1807 1808 disk->ev = ev; 1809 } 1810 1811 static void disk_add_events(struct gendisk *disk) 1812 { 1813 if (!disk->ev) 1814 return; 1815 1816 /* FIXME: error handling */ 1817 if (sysfs_create_files(&disk_to_dev(disk)->kobj, disk_events_attrs) < 0) 1818 pr_warn("%s: failed to create sysfs files for events\n", 1819 disk->disk_name); 1820 1821 mutex_lock(&disk_events_mutex); 1822 list_add_tail(&disk->ev->node, &disk_events); 1823 mutex_unlock(&disk_events_mutex); 1824 1825 /* 1826 * Block count is initialized to 1 and the following initial 1827 * unblock kicks it into action. 1828 */ 1829 __disk_unblock_events(disk, true); 1830 } 1831 1832 static void disk_del_events(struct gendisk *disk) 1833 { 1834 if (!disk->ev) 1835 return; 1836 1837 disk_block_events(disk); 1838 1839 mutex_lock(&disk_events_mutex); 1840 list_del_init(&disk->ev->node); 1841 mutex_unlock(&disk_events_mutex); 1842 1843 sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs); 1844 } 1845 1846 static void disk_release_events(struct gendisk *disk) 1847 { 1848 /* the block count should be 1 from disk_del_events() */ 1849 WARN_ON_ONCE(disk->ev && disk->ev->block != 1); 1850 kfree(disk->ev); 1851 } 1852