1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1991, 1992 Linus Torvalds 4 * Copyright (C) 2001 Andrea Arcangeli <[email protected]> SuSE 5 * Copyright (C) 2016 - 2020 Christoph Hellwig 6 */ 7 8 #include <linux/init.h> 9 #include <linux/mm.h> 10 #include <linux/slab.h> 11 #include <linux/kmod.h> 12 #include <linux/major.h> 13 #include <linux/device_cgroup.h> 14 #include <linux/blkdev.h> 15 #include <linux/blk-integrity.h> 16 #include <linux/backing-dev.h> 17 #include <linux/module.h> 18 #include <linux/blkpg.h> 19 #include <linux/magic.h> 20 #include <linux/buffer_head.h> 21 #include <linux/swap.h> 22 #include <linux/writeback.h> 23 #include <linux/mount.h> 24 #include <linux/pseudo_fs.h> 25 #include <linux/uio.h> 26 #include <linux/namei.h> 27 #include <linux/part_stat.h> 28 #include <linux/uaccess.h> 29 #include <linux/stat.h> 30 #include "../fs/internal.h" 31 #include "blk.h" 32 33 /* Should we allow writing to mounted block devices? */ 34 static bool bdev_allow_write_mounted = IS_ENABLED(CONFIG_BLK_DEV_WRITE_MOUNTED); 35 36 struct bdev_inode { 37 struct block_device bdev; 38 struct inode vfs_inode; 39 }; 40 41 static inline struct bdev_inode *BDEV_I(struct inode *inode) 42 { 43 return container_of(inode, struct bdev_inode, vfs_inode); 44 } 45 46 struct block_device *I_BDEV(struct inode *inode) 47 { 48 return &BDEV_I(inode)->bdev; 49 } 50 EXPORT_SYMBOL(I_BDEV); 51 52 struct block_device *file_bdev(struct file *bdev_file) 53 { 54 return I_BDEV(bdev_file->f_mapping->host); 55 } 56 EXPORT_SYMBOL(file_bdev); 57 58 static void bdev_write_inode(struct block_device *bdev) 59 { 60 struct inode *inode = bdev->bd_inode; 61 int ret; 62 63 spin_lock(&inode->i_lock); 64 while (inode->i_state & I_DIRTY) { 65 spin_unlock(&inode->i_lock); 66 ret = write_inode_now(inode, true); 67 if (ret) 68 pr_warn_ratelimited( 69 "VFS: Dirty inode writeback failed for block device %pg (err=%d).\n", 70 bdev, ret); 71 spin_lock(&inode->i_lock); 72 } 73 spin_unlock(&inode->i_lock); 74 } 75 76 /* Kill _all_ buffers and pagecache , dirty or not.. */ 77 static void kill_bdev(struct block_device *bdev) 78 { 79 struct address_space *mapping = bdev->bd_inode->i_mapping; 80 81 if (mapping_empty(mapping)) 82 return; 83 84 invalidate_bh_lrus(); 85 truncate_inode_pages(mapping, 0); 86 } 87 88 /* Invalidate clean unused buffers and pagecache. */ 89 void invalidate_bdev(struct block_device *bdev) 90 { 91 struct address_space *mapping = bdev->bd_inode->i_mapping; 92 93 if (mapping->nrpages) { 94 invalidate_bh_lrus(); 95 lru_add_drain_all(); /* make sure all lru add caches are flushed */ 96 invalidate_mapping_pages(mapping, 0, -1); 97 } 98 } 99 EXPORT_SYMBOL(invalidate_bdev); 100 101 /* 102 * Drop all buffers & page cache for given bdev range. This function bails 103 * with error if bdev has other exclusive owner (such as filesystem). 104 */ 105 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode, 106 loff_t lstart, loff_t lend) 107 { 108 /* 109 * If we don't hold exclusive handle for the device, upgrade to it 110 * while we discard the buffer cache to avoid discarding buffers 111 * under live filesystem. 112 */ 113 if (!(mode & BLK_OPEN_EXCL)) { 114 int err = bd_prepare_to_claim(bdev, truncate_bdev_range, NULL); 115 if (err) 116 goto invalidate; 117 } 118 119 truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend); 120 if (!(mode & BLK_OPEN_EXCL)) 121 bd_abort_claiming(bdev, truncate_bdev_range); 122 return 0; 123 124 invalidate: 125 /* 126 * Someone else has handle exclusively open. Try invalidating instead. 127 * The 'end' argument is inclusive so the rounding is safe. 128 */ 129 return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping, 130 lstart >> PAGE_SHIFT, 131 lend >> PAGE_SHIFT); 132 } 133 134 static void set_init_blocksize(struct block_device *bdev) 135 { 136 unsigned int bsize = bdev_logical_block_size(bdev); 137 loff_t size = i_size_read(bdev->bd_inode); 138 139 while (bsize < PAGE_SIZE) { 140 if (size & bsize) 141 break; 142 bsize <<= 1; 143 } 144 bdev->bd_inode->i_blkbits = blksize_bits(bsize); 145 } 146 147 int set_blocksize(struct block_device *bdev, int size) 148 { 149 /* Size must be a power of two, and between 512 and PAGE_SIZE */ 150 if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size)) 151 return -EINVAL; 152 153 /* Size cannot be smaller than the size supported by the device */ 154 if (size < bdev_logical_block_size(bdev)) 155 return -EINVAL; 156 157 /* Don't change the size if it is same as current */ 158 if (bdev->bd_inode->i_blkbits != blksize_bits(size)) { 159 sync_blockdev(bdev); 160 bdev->bd_inode->i_blkbits = blksize_bits(size); 161 kill_bdev(bdev); 162 } 163 return 0; 164 } 165 166 EXPORT_SYMBOL(set_blocksize); 167 168 int sb_set_blocksize(struct super_block *sb, int size) 169 { 170 if (set_blocksize(sb->s_bdev, size)) 171 return 0; 172 /* If we get here, we know size is power of two 173 * and it's value is between 512 and PAGE_SIZE */ 174 sb->s_blocksize = size; 175 sb->s_blocksize_bits = blksize_bits(size); 176 return sb->s_blocksize; 177 } 178 179 EXPORT_SYMBOL(sb_set_blocksize); 180 181 int sb_min_blocksize(struct super_block *sb, int size) 182 { 183 int minsize = bdev_logical_block_size(sb->s_bdev); 184 if (size < minsize) 185 size = minsize; 186 return sb_set_blocksize(sb, size); 187 } 188 189 EXPORT_SYMBOL(sb_min_blocksize); 190 191 int sync_blockdev_nowait(struct block_device *bdev) 192 { 193 if (!bdev) 194 return 0; 195 return filemap_flush(bdev->bd_inode->i_mapping); 196 } 197 EXPORT_SYMBOL_GPL(sync_blockdev_nowait); 198 199 /* 200 * Write out and wait upon all the dirty data associated with a block 201 * device via its mapping. Does not take the superblock lock. 202 */ 203 int sync_blockdev(struct block_device *bdev) 204 { 205 if (!bdev) 206 return 0; 207 return filemap_write_and_wait(bdev->bd_inode->i_mapping); 208 } 209 EXPORT_SYMBOL(sync_blockdev); 210 211 int sync_blockdev_range(struct block_device *bdev, loff_t lstart, loff_t lend) 212 { 213 return filemap_write_and_wait_range(bdev->bd_inode->i_mapping, 214 lstart, lend); 215 } 216 EXPORT_SYMBOL(sync_blockdev_range); 217 218 /** 219 * bdev_freeze - lock a filesystem and force it into a consistent state 220 * @bdev: blockdevice to lock 221 * 222 * If a superblock is found on this device, we take the s_umount semaphore 223 * on it to make sure nobody unmounts until the snapshot creation is done. 224 * The reference counter (bd_fsfreeze_count) guarantees that only the last 225 * unfreeze process can unfreeze the frozen filesystem actually when multiple 226 * freeze requests arrive simultaneously. It counts up in bdev_freeze() and 227 * count down in bdev_thaw(). When it becomes 0, thaw_bdev() will unfreeze 228 * actually. 229 * 230 * Return: On success zero is returned, negative error code on failure. 231 */ 232 int bdev_freeze(struct block_device *bdev) 233 { 234 int error = 0; 235 236 mutex_lock(&bdev->bd_fsfreeze_mutex); 237 238 if (atomic_inc_return(&bdev->bd_fsfreeze_count) > 1) { 239 mutex_unlock(&bdev->bd_fsfreeze_mutex); 240 return 0; 241 } 242 243 mutex_lock(&bdev->bd_holder_lock); 244 if (bdev->bd_holder_ops && bdev->bd_holder_ops->freeze) { 245 error = bdev->bd_holder_ops->freeze(bdev); 246 lockdep_assert_not_held(&bdev->bd_holder_lock); 247 } else { 248 mutex_unlock(&bdev->bd_holder_lock); 249 error = sync_blockdev(bdev); 250 } 251 252 if (error) 253 atomic_dec(&bdev->bd_fsfreeze_count); 254 255 mutex_unlock(&bdev->bd_fsfreeze_mutex); 256 return error; 257 } 258 EXPORT_SYMBOL(bdev_freeze); 259 260 /** 261 * bdev_thaw - unlock filesystem 262 * @bdev: blockdevice to unlock 263 * 264 * Unlocks the filesystem and marks it writeable again after bdev_freeze(). 265 * 266 * Return: On success zero is returned, negative error code on failure. 267 */ 268 int bdev_thaw(struct block_device *bdev) 269 { 270 int error = -EINVAL, nr_freeze; 271 272 mutex_lock(&bdev->bd_fsfreeze_mutex); 273 274 /* 275 * If this returns < 0 it means that @bd_fsfreeze_count was 276 * already 0 and no decrement was performed. 277 */ 278 nr_freeze = atomic_dec_if_positive(&bdev->bd_fsfreeze_count); 279 if (nr_freeze < 0) 280 goto out; 281 282 error = 0; 283 if (nr_freeze > 0) 284 goto out; 285 286 mutex_lock(&bdev->bd_holder_lock); 287 if (bdev->bd_holder_ops && bdev->bd_holder_ops->thaw) { 288 error = bdev->bd_holder_ops->thaw(bdev); 289 lockdep_assert_not_held(&bdev->bd_holder_lock); 290 } else { 291 mutex_unlock(&bdev->bd_holder_lock); 292 } 293 294 if (error) 295 atomic_inc(&bdev->bd_fsfreeze_count); 296 out: 297 mutex_unlock(&bdev->bd_fsfreeze_mutex); 298 return error; 299 } 300 EXPORT_SYMBOL(bdev_thaw); 301 302 /* 303 * pseudo-fs 304 */ 305 306 static __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock); 307 static struct kmem_cache *bdev_cachep __ro_after_init; 308 309 static struct inode *bdev_alloc_inode(struct super_block *sb) 310 { 311 struct bdev_inode *ei = alloc_inode_sb(sb, bdev_cachep, GFP_KERNEL); 312 313 if (!ei) 314 return NULL; 315 memset(&ei->bdev, 0, sizeof(ei->bdev)); 316 return &ei->vfs_inode; 317 } 318 319 static void bdev_free_inode(struct inode *inode) 320 { 321 struct block_device *bdev = I_BDEV(inode); 322 323 free_percpu(bdev->bd_stats); 324 kfree(bdev->bd_meta_info); 325 326 if (!bdev_is_partition(bdev)) { 327 if (bdev->bd_disk && bdev->bd_disk->bdi) 328 bdi_put(bdev->bd_disk->bdi); 329 kfree(bdev->bd_disk); 330 } 331 332 if (MAJOR(bdev->bd_dev) == BLOCK_EXT_MAJOR) 333 blk_free_ext_minor(MINOR(bdev->bd_dev)); 334 335 kmem_cache_free(bdev_cachep, BDEV_I(inode)); 336 } 337 338 static void init_once(void *data) 339 { 340 struct bdev_inode *ei = data; 341 342 inode_init_once(&ei->vfs_inode); 343 } 344 345 static void bdev_evict_inode(struct inode *inode) 346 { 347 truncate_inode_pages_final(&inode->i_data); 348 invalidate_inode_buffers(inode); /* is it needed here? */ 349 clear_inode(inode); 350 } 351 352 static const struct super_operations bdev_sops = { 353 .statfs = simple_statfs, 354 .alloc_inode = bdev_alloc_inode, 355 .free_inode = bdev_free_inode, 356 .drop_inode = generic_delete_inode, 357 .evict_inode = bdev_evict_inode, 358 }; 359 360 static int bd_init_fs_context(struct fs_context *fc) 361 { 362 struct pseudo_fs_context *ctx = init_pseudo(fc, BDEVFS_MAGIC); 363 if (!ctx) 364 return -ENOMEM; 365 fc->s_iflags |= SB_I_CGROUPWB; 366 ctx->ops = &bdev_sops; 367 return 0; 368 } 369 370 static struct file_system_type bd_type = { 371 .name = "bdev", 372 .init_fs_context = bd_init_fs_context, 373 .kill_sb = kill_anon_super, 374 }; 375 376 struct super_block *blockdev_superblock __ro_after_init; 377 struct vfsmount *blockdev_mnt __ro_after_init; 378 EXPORT_SYMBOL_GPL(blockdev_superblock); 379 380 void __init bdev_cache_init(void) 381 { 382 int err; 383 384 bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode), 385 0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT| 386 SLAB_ACCOUNT|SLAB_PANIC), 387 init_once); 388 err = register_filesystem(&bd_type); 389 if (err) 390 panic("Cannot register bdev pseudo-fs"); 391 blockdev_mnt = kern_mount(&bd_type); 392 if (IS_ERR(blockdev_mnt)) 393 panic("Cannot create bdev pseudo-fs"); 394 blockdev_superblock = blockdev_mnt->mnt_sb; /* For writeback */ 395 } 396 397 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno) 398 { 399 struct block_device *bdev; 400 struct inode *inode; 401 402 inode = new_inode(blockdev_superblock); 403 if (!inode) 404 return NULL; 405 inode->i_mode = S_IFBLK; 406 inode->i_rdev = 0; 407 inode->i_data.a_ops = &def_blk_aops; 408 mapping_set_gfp_mask(&inode->i_data, GFP_USER); 409 410 bdev = I_BDEV(inode); 411 mutex_init(&bdev->bd_fsfreeze_mutex); 412 spin_lock_init(&bdev->bd_size_lock); 413 mutex_init(&bdev->bd_holder_lock); 414 atomic_set(&bdev->__bd_flags, partno); 415 bdev->bd_inode = inode; 416 bdev->bd_queue = disk->queue; 417 if (partno && bdev_test_flag(disk->part0, BD_HAS_SUBMIT_BIO)) 418 bdev_set_flag(bdev, BD_HAS_SUBMIT_BIO); 419 bdev->bd_stats = alloc_percpu(struct disk_stats); 420 if (!bdev->bd_stats) { 421 iput(inode); 422 return NULL; 423 } 424 bdev->bd_disk = disk; 425 return bdev; 426 } 427 428 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors) 429 { 430 spin_lock(&bdev->bd_size_lock); 431 i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT); 432 bdev->bd_nr_sectors = sectors; 433 spin_unlock(&bdev->bd_size_lock); 434 } 435 436 void bdev_add(struct block_device *bdev, dev_t dev) 437 { 438 if (bdev_stable_writes(bdev)) 439 mapping_set_stable_writes(bdev->bd_inode->i_mapping); 440 bdev->bd_dev = dev; 441 bdev->bd_inode->i_rdev = dev; 442 bdev->bd_inode->i_ino = dev; 443 insert_inode_hash(bdev->bd_inode); 444 } 445 446 long nr_blockdev_pages(void) 447 { 448 struct inode *inode; 449 long ret = 0; 450 451 spin_lock(&blockdev_superblock->s_inode_list_lock); 452 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) 453 ret += inode->i_mapping->nrpages; 454 spin_unlock(&blockdev_superblock->s_inode_list_lock); 455 456 return ret; 457 } 458 459 /** 460 * bd_may_claim - test whether a block device can be claimed 461 * @bdev: block device of interest 462 * @holder: holder trying to claim @bdev 463 * @hops: holder ops 464 * 465 * Test whether @bdev can be claimed by @holder. 466 * 467 * RETURNS: 468 * %true if @bdev can be claimed, %false otherwise. 469 */ 470 static bool bd_may_claim(struct block_device *bdev, void *holder, 471 const struct blk_holder_ops *hops) 472 { 473 struct block_device *whole = bdev_whole(bdev); 474 475 lockdep_assert_held(&bdev_lock); 476 477 if (bdev->bd_holder) { 478 /* 479 * The same holder can always re-claim. 480 */ 481 if (bdev->bd_holder == holder) { 482 if (WARN_ON_ONCE(bdev->bd_holder_ops != hops)) 483 return false; 484 return true; 485 } 486 return false; 487 } 488 489 /* 490 * If the whole devices holder is set to bd_may_claim, a partition on 491 * the device is claimed, but not the whole device. 492 */ 493 if (whole != bdev && 494 whole->bd_holder && whole->bd_holder != bd_may_claim) 495 return false; 496 return true; 497 } 498 499 /** 500 * bd_prepare_to_claim - claim a block device 501 * @bdev: block device of interest 502 * @holder: holder trying to claim @bdev 503 * @hops: holder ops. 504 * 505 * Claim @bdev. This function fails if @bdev is already claimed by another 506 * holder and waits if another claiming is in progress. return, the caller 507 * has ownership of bd_claiming and bd_holder[s]. 508 * 509 * RETURNS: 510 * 0 if @bdev can be claimed, -EBUSY otherwise. 511 */ 512 int bd_prepare_to_claim(struct block_device *bdev, void *holder, 513 const struct blk_holder_ops *hops) 514 { 515 struct block_device *whole = bdev_whole(bdev); 516 517 if (WARN_ON_ONCE(!holder)) 518 return -EINVAL; 519 retry: 520 mutex_lock(&bdev_lock); 521 /* if someone else claimed, fail */ 522 if (!bd_may_claim(bdev, holder, hops)) { 523 mutex_unlock(&bdev_lock); 524 return -EBUSY; 525 } 526 527 /* if claiming is already in progress, wait for it to finish */ 528 if (whole->bd_claiming) { 529 wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0); 530 DEFINE_WAIT(wait); 531 532 prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE); 533 mutex_unlock(&bdev_lock); 534 schedule(); 535 finish_wait(wq, &wait); 536 goto retry; 537 } 538 539 /* yay, all mine */ 540 whole->bd_claiming = holder; 541 mutex_unlock(&bdev_lock); 542 return 0; 543 } 544 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */ 545 546 static void bd_clear_claiming(struct block_device *whole, void *holder) 547 { 548 lockdep_assert_held(&bdev_lock); 549 /* tell others that we're done */ 550 BUG_ON(whole->bd_claiming != holder); 551 whole->bd_claiming = NULL; 552 wake_up_bit(&whole->bd_claiming, 0); 553 } 554 555 /** 556 * bd_finish_claiming - finish claiming of a block device 557 * @bdev: block device of interest 558 * @holder: holder that has claimed @bdev 559 * @hops: block device holder operations 560 * 561 * Finish exclusive open of a block device. Mark the device as exlusively 562 * open by the holder and wake up all waiters for exclusive open to finish. 563 */ 564 static void bd_finish_claiming(struct block_device *bdev, void *holder, 565 const struct blk_holder_ops *hops) 566 { 567 struct block_device *whole = bdev_whole(bdev); 568 569 mutex_lock(&bdev_lock); 570 BUG_ON(!bd_may_claim(bdev, holder, hops)); 571 /* 572 * Note that for a whole device bd_holders will be incremented twice, 573 * and bd_holder will be set to bd_may_claim before being set to holder 574 */ 575 whole->bd_holders++; 576 whole->bd_holder = bd_may_claim; 577 bdev->bd_holders++; 578 mutex_lock(&bdev->bd_holder_lock); 579 bdev->bd_holder = holder; 580 bdev->bd_holder_ops = hops; 581 mutex_unlock(&bdev->bd_holder_lock); 582 bd_clear_claiming(whole, holder); 583 mutex_unlock(&bdev_lock); 584 585 if (hops && hops->get_holder) 586 hops->get_holder(holder); 587 } 588 589 /** 590 * bd_abort_claiming - abort claiming of a block device 591 * @bdev: block device of interest 592 * @holder: holder that has claimed @bdev 593 * 594 * Abort claiming of a block device when the exclusive open failed. This can be 595 * also used when exclusive open is not actually desired and we just needed 596 * to block other exclusive openers for a while. 597 */ 598 void bd_abort_claiming(struct block_device *bdev, void *holder) 599 { 600 mutex_lock(&bdev_lock); 601 bd_clear_claiming(bdev_whole(bdev), holder); 602 mutex_unlock(&bdev_lock); 603 } 604 EXPORT_SYMBOL(bd_abort_claiming); 605 606 static void bd_end_claim(struct block_device *bdev, void *holder) 607 { 608 struct block_device *whole = bdev_whole(bdev); 609 const struct blk_holder_ops *hops = bdev->bd_holder_ops; 610 bool unblock = false; 611 612 /* 613 * Release a claim on the device. The holder fields are protected with 614 * bdev_lock. open_mutex is used to synchronize disk_holder unlinking. 615 */ 616 mutex_lock(&bdev_lock); 617 WARN_ON_ONCE(bdev->bd_holder != holder); 618 WARN_ON_ONCE(--bdev->bd_holders < 0); 619 WARN_ON_ONCE(--whole->bd_holders < 0); 620 if (!bdev->bd_holders) { 621 mutex_lock(&bdev->bd_holder_lock); 622 bdev->bd_holder = NULL; 623 bdev->bd_holder_ops = NULL; 624 mutex_unlock(&bdev->bd_holder_lock); 625 if (bdev_test_flag(bdev, BD_WRITE_HOLDER)) 626 unblock = true; 627 } 628 if (!whole->bd_holders) 629 whole->bd_holder = NULL; 630 mutex_unlock(&bdev_lock); 631 632 if (hops && hops->put_holder) 633 hops->put_holder(holder); 634 635 /* 636 * If this was the last claim, remove holder link and unblock evpoll if 637 * it was a write holder. 638 */ 639 if (unblock) { 640 disk_unblock_events(bdev->bd_disk); 641 bdev_clear_flag(bdev, BD_WRITE_HOLDER); 642 } 643 } 644 645 static void blkdev_flush_mapping(struct block_device *bdev) 646 { 647 WARN_ON_ONCE(bdev->bd_holders); 648 sync_blockdev(bdev); 649 kill_bdev(bdev); 650 bdev_write_inode(bdev); 651 } 652 653 static int blkdev_get_whole(struct block_device *bdev, blk_mode_t mode) 654 { 655 struct gendisk *disk = bdev->bd_disk; 656 int ret; 657 658 if (disk->fops->open) { 659 ret = disk->fops->open(disk, mode); 660 if (ret) { 661 /* avoid ghost partitions on a removed medium */ 662 if (ret == -ENOMEDIUM && 663 test_bit(GD_NEED_PART_SCAN, &disk->state)) 664 bdev_disk_changed(disk, true); 665 return ret; 666 } 667 } 668 669 if (!atomic_read(&bdev->bd_openers)) 670 set_init_blocksize(bdev); 671 if (test_bit(GD_NEED_PART_SCAN, &disk->state)) 672 bdev_disk_changed(disk, false); 673 atomic_inc(&bdev->bd_openers); 674 return 0; 675 } 676 677 static void blkdev_put_whole(struct block_device *bdev) 678 { 679 if (atomic_dec_and_test(&bdev->bd_openers)) 680 blkdev_flush_mapping(bdev); 681 if (bdev->bd_disk->fops->release) 682 bdev->bd_disk->fops->release(bdev->bd_disk); 683 } 684 685 static int blkdev_get_part(struct block_device *part, blk_mode_t mode) 686 { 687 struct gendisk *disk = part->bd_disk; 688 int ret; 689 690 ret = blkdev_get_whole(bdev_whole(part), mode); 691 if (ret) 692 return ret; 693 694 ret = -ENXIO; 695 if (!bdev_nr_sectors(part)) 696 goto out_blkdev_put; 697 698 if (!atomic_read(&part->bd_openers)) { 699 disk->open_partitions++; 700 set_init_blocksize(part); 701 } 702 atomic_inc(&part->bd_openers); 703 return 0; 704 705 out_blkdev_put: 706 blkdev_put_whole(bdev_whole(part)); 707 return ret; 708 } 709 710 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder) 711 { 712 int ret; 713 714 ret = devcgroup_check_permission(DEVCG_DEV_BLOCK, 715 MAJOR(dev), MINOR(dev), 716 ((mode & BLK_OPEN_READ) ? DEVCG_ACC_READ : 0) | 717 ((mode & BLK_OPEN_WRITE) ? DEVCG_ACC_WRITE : 0)); 718 if (ret) 719 return ret; 720 721 /* Blocking writes requires exclusive opener */ 722 if (mode & BLK_OPEN_RESTRICT_WRITES && !holder) 723 return -EINVAL; 724 725 /* 726 * We're using error pointers to indicate to ->release() when we 727 * failed to open that block device. Also this doesn't make sense. 728 */ 729 if (WARN_ON_ONCE(IS_ERR(holder))) 730 return -EINVAL; 731 732 return 0; 733 } 734 735 static void blkdev_put_part(struct block_device *part) 736 { 737 struct block_device *whole = bdev_whole(part); 738 739 if (atomic_dec_and_test(&part->bd_openers)) { 740 blkdev_flush_mapping(part); 741 whole->bd_disk->open_partitions--; 742 } 743 blkdev_put_whole(whole); 744 } 745 746 struct block_device *blkdev_get_no_open(dev_t dev) 747 { 748 struct block_device *bdev; 749 struct inode *inode; 750 751 inode = ilookup(blockdev_superblock, dev); 752 if (!inode && IS_ENABLED(CONFIG_BLOCK_LEGACY_AUTOLOAD)) { 753 blk_request_module(dev); 754 inode = ilookup(blockdev_superblock, dev); 755 if (inode) 756 pr_warn_ratelimited( 757 "block device autoloading is deprecated and will be removed.\n"); 758 } 759 if (!inode) 760 return NULL; 761 762 /* switch from the inode reference to a device mode one: */ 763 bdev = &BDEV_I(inode)->bdev; 764 if (!kobject_get_unless_zero(&bdev->bd_device.kobj)) 765 bdev = NULL; 766 iput(inode); 767 return bdev; 768 } 769 770 void blkdev_put_no_open(struct block_device *bdev) 771 { 772 put_device(&bdev->bd_device); 773 } 774 775 static bool bdev_writes_blocked(struct block_device *bdev) 776 { 777 return bdev->bd_writers == -1; 778 } 779 780 static void bdev_block_writes(struct block_device *bdev) 781 { 782 bdev->bd_writers = -1; 783 } 784 785 static void bdev_unblock_writes(struct block_device *bdev) 786 { 787 bdev->bd_writers = 0; 788 } 789 790 static bool bdev_may_open(struct block_device *bdev, blk_mode_t mode) 791 { 792 if (bdev_allow_write_mounted) 793 return true; 794 /* Writes blocked? */ 795 if (mode & BLK_OPEN_WRITE && bdev_writes_blocked(bdev)) 796 return false; 797 if (mode & BLK_OPEN_RESTRICT_WRITES && bdev->bd_writers > 0) 798 return false; 799 return true; 800 } 801 802 static void bdev_claim_write_access(struct block_device *bdev, blk_mode_t mode) 803 { 804 if (bdev_allow_write_mounted) 805 return; 806 807 /* Claim exclusive or shared write access. */ 808 if (mode & BLK_OPEN_RESTRICT_WRITES) 809 bdev_block_writes(bdev); 810 else if (mode & BLK_OPEN_WRITE) 811 bdev->bd_writers++; 812 } 813 814 static void bdev_yield_write_access(struct file *bdev_file) 815 { 816 struct block_device *bdev; 817 818 if (bdev_allow_write_mounted) 819 return; 820 821 bdev = file_bdev(bdev_file); 822 /* Yield exclusive or shared write access. */ 823 if (bdev_file->f_mode & FMODE_WRITE) { 824 if (bdev_writes_blocked(bdev)) 825 bdev_unblock_writes(bdev); 826 else 827 bdev->bd_writers--; 828 } 829 } 830 831 /** 832 * bdev_open - open a block device 833 * @bdev: block device to open 834 * @mode: open mode (BLK_OPEN_*) 835 * @holder: exclusive holder identifier 836 * @hops: holder operations 837 * @bdev_file: file for the block device 838 * 839 * Open the block device. If @holder is not %NULL, the block device is opened 840 * with exclusive access. Exclusive opens may nest for the same @holder. 841 * 842 * CONTEXT: 843 * Might sleep. 844 * 845 * RETURNS: 846 * zero on success, -errno on failure. 847 */ 848 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder, 849 const struct blk_holder_ops *hops, struct file *bdev_file) 850 { 851 bool unblock_events = true; 852 struct gendisk *disk = bdev->bd_disk; 853 int ret; 854 855 if (holder) { 856 mode |= BLK_OPEN_EXCL; 857 ret = bd_prepare_to_claim(bdev, holder, hops); 858 if (ret) 859 return ret; 860 } else { 861 if (WARN_ON_ONCE(mode & BLK_OPEN_EXCL)) 862 return -EIO; 863 } 864 865 disk_block_events(disk); 866 867 mutex_lock(&disk->open_mutex); 868 ret = -ENXIO; 869 if (!disk_live(disk)) 870 goto abort_claiming; 871 if (!try_module_get(disk->fops->owner)) 872 goto abort_claiming; 873 ret = -EBUSY; 874 if (!bdev_may_open(bdev, mode)) 875 goto abort_claiming; 876 if (bdev_is_partition(bdev)) 877 ret = blkdev_get_part(bdev, mode); 878 else 879 ret = blkdev_get_whole(bdev, mode); 880 if (ret) 881 goto put_module; 882 bdev_claim_write_access(bdev, mode); 883 if (holder) { 884 bd_finish_claiming(bdev, holder, hops); 885 886 /* 887 * Block event polling for write claims if requested. Any write 888 * holder makes the write_holder state stick until all are 889 * released. This is good enough and tracking individual 890 * writeable reference is too fragile given the way @mode is 891 * used in blkdev_get/put(). 892 */ 893 if ((mode & BLK_OPEN_WRITE) && 894 !bdev_test_flag(bdev, BD_WRITE_HOLDER) && 895 (disk->event_flags & DISK_EVENT_FLAG_BLOCK_ON_EXCL_WRITE)) { 896 bdev_set_flag(bdev, BD_WRITE_HOLDER); 897 unblock_events = false; 898 } 899 } 900 mutex_unlock(&disk->open_mutex); 901 902 if (unblock_events) 903 disk_unblock_events(disk); 904 905 bdev_file->f_flags |= O_LARGEFILE; 906 bdev_file->f_mode |= FMODE_BUF_RASYNC | FMODE_CAN_ODIRECT; 907 if (bdev_nowait(bdev)) 908 bdev_file->f_mode |= FMODE_NOWAIT; 909 bdev_file->f_mapping = bdev->bd_inode->i_mapping; 910 bdev_file->f_wb_err = filemap_sample_wb_err(bdev_file->f_mapping); 911 bdev_file->private_data = holder; 912 913 return 0; 914 put_module: 915 module_put(disk->fops->owner); 916 abort_claiming: 917 if (holder) 918 bd_abort_claiming(bdev, holder); 919 mutex_unlock(&disk->open_mutex); 920 disk_unblock_events(disk); 921 return ret; 922 } 923 924 /* 925 * If BLK_OPEN_WRITE_IOCTL is set then this is a historical quirk 926 * associated with the floppy driver where it has allowed ioctls if the 927 * file was opened for writing, but does not allow reads or writes. 928 * Make sure that this quirk is reflected in @f_flags. 929 * 930 * It can also happen if a block device is opened as O_RDWR | O_WRONLY. 931 */ 932 static unsigned blk_to_file_flags(blk_mode_t mode) 933 { 934 unsigned int flags = 0; 935 936 if ((mode & (BLK_OPEN_READ | BLK_OPEN_WRITE)) == 937 (BLK_OPEN_READ | BLK_OPEN_WRITE)) 938 flags |= O_RDWR; 939 else if (mode & BLK_OPEN_WRITE_IOCTL) 940 flags |= O_RDWR | O_WRONLY; 941 else if (mode & BLK_OPEN_WRITE) 942 flags |= O_WRONLY; 943 else if (mode & BLK_OPEN_READ) 944 flags |= O_RDONLY; /* homeopathic, because O_RDONLY is 0 */ 945 else 946 WARN_ON_ONCE(true); 947 948 if (mode & BLK_OPEN_NDELAY) 949 flags |= O_NDELAY; 950 951 return flags; 952 } 953 954 struct file *bdev_file_open_by_dev(dev_t dev, blk_mode_t mode, void *holder, 955 const struct blk_holder_ops *hops) 956 { 957 struct file *bdev_file; 958 struct block_device *bdev; 959 unsigned int flags; 960 int ret; 961 962 ret = bdev_permission(dev, mode, holder); 963 if (ret) 964 return ERR_PTR(ret); 965 966 bdev = blkdev_get_no_open(dev); 967 if (!bdev) 968 return ERR_PTR(-ENXIO); 969 970 flags = blk_to_file_flags(mode); 971 bdev_file = alloc_file_pseudo_noaccount(bdev->bd_inode, 972 blockdev_mnt, "", flags | O_LARGEFILE, &def_blk_fops); 973 if (IS_ERR(bdev_file)) { 974 blkdev_put_no_open(bdev); 975 return bdev_file; 976 } 977 ihold(bdev->bd_inode); 978 979 ret = bdev_open(bdev, mode, holder, hops, bdev_file); 980 if (ret) { 981 /* We failed to open the block device. Let ->release() know. */ 982 bdev_file->private_data = ERR_PTR(ret); 983 fput(bdev_file); 984 return ERR_PTR(ret); 985 } 986 return bdev_file; 987 } 988 EXPORT_SYMBOL(bdev_file_open_by_dev); 989 990 struct file *bdev_file_open_by_path(const char *path, blk_mode_t mode, 991 void *holder, 992 const struct blk_holder_ops *hops) 993 { 994 struct file *file; 995 dev_t dev; 996 int error; 997 998 error = lookup_bdev(path, &dev); 999 if (error) 1000 return ERR_PTR(error); 1001 1002 file = bdev_file_open_by_dev(dev, mode, holder, hops); 1003 if (!IS_ERR(file) && (mode & BLK_OPEN_WRITE)) { 1004 if (bdev_read_only(file_bdev(file))) { 1005 fput(file); 1006 file = ERR_PTR(-EACCES); 1007 } 1008 } 1009 1010 return file; 1011 } 1012 EXPORT_SYMBOL(bdev_file_open_by_path); 1013 1014 void bdev_release(struct file *bdev_file) 1015 { 1016 struct block_device *bdev = file_bdev(bdev_file); 1017 void *holder = bdev_file->private_data; 1018 struct gendisk *disk = bdev->bd_disk; 1019 1020 /* We failed to open that block device. */ 1021 if (IS_ERR(holder)) 1022 goto put_no_open; 1023 1024 /* 1025 * Sync early if it looks like we're the last one. If someone else 1026 * opens the block device between now and the decrement of bd_openers 1027 * then we did a sync that we didn't need to, but that's not the end 1028 * of the world and we want to avoid long (could be several minute) 1029 * syncs while holding the mutex. 1030 */ 1031 if (atomic_read(&bdev->bd_openers) == 1) 1032 sync_blockdev(bdev); 1033 1034 mutex_lock(&disk->open_mutex); 1035 bdev_yield_write_access(bdev_file); 1036 1037 if (holder) 1038 bd_end_claim(bdev, holder); 1039 1040 /* 1041 * Trigger event checking and tell drivers to flush MEDIA_CHANGE 1042 * event. This is to ensure detection of media removal commanded 1043 * from userland - e.g. eject(1). 1044 */ 1045 disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE); 1046 1047 if (bdev_is_partition(bdev)) 1048 blkdev_put_part(bdev); 1049 else 1050 blkdev_put_whole(bdev); 1051 mutex_unlock(&disk->open_mutex); 1052 1053 module_put(disk->fops->owner); 1054 put_no_open: 1055 blkdev_put_no_open(bdev); 1056 } 1057 1058 /** 1059 * lookup_bdev() - Look up a struct block_device by name. 1060 * @pathname: Name of the block device in the filesystem. 1061 * @dev: Pointer to the block device's dev_t, if found. 1062 * 1063 * Lookup the block device's dev_t at @pathname in the current 1064 * namespace if possible and return it in @dev. 1065 * 1066 * Context: May sleep. 1067 * Return: 0 if succeeded, negative errno otherwise. 1068 */ 1069 int lookup_bdev(const char *pathname, dev_t *dev) 1070 { 1071 struct inode *inode; 1072 struct path path; 1073 int error; 1074 1075 if (!pathname || !*pathname) 1076 return -EINVAL; 1077 1078 error = kern_path(pathname, LOOKUP_FOLLOW, &path); 1079 if (error) 1080 return error; 1081 1082 inode = d_backing_inode(path.dentry); 1083 error = -ENOTBLK; 1084 if (!S_ISBLK(inode->i_mode)) 1085 goto out_path_put; 1086 error = -EACCES; 1087 if (!may_open_dev(&path)) 1088 goto out_path_put; 1089 1090 *dev = inode->i_rdev; 1091 error = 0; 1092 out_path_put: 1093 path_put(&path); 1094 return error; 1095 } 1096 EXPORT_SYMBOL(lookup_bdev); 1097 1098 /** 1099 * bdev_mark_dead - mark a block device as dead 1100 * @bdev: block device to operate on 1101 * @surprise: indicate a surprise removal 1102 * 1103 * Tell the file system that this devices or media is dead. If @surprise is set 1104 * to %true the device or media is already gone, if not we are preparing for an 1105 * orderly removal. 1106 * 1107 * This calls into the file system, which then typicall syncs out all dirty data 1108 * and writes back inodes and then invalidates any cached data in the inodes on 1109 * the file system. In addition we also invalidate the block device mapping. 1110 */ 1111 void bdev_mark_dead(struct block_device *bdev, bool surprise) 1112 { 1113 mutex_lock(&bdev->bd_holder_lock); 1114 if (bdev->bd_holder_ops && bdev->bd_holder_ops->mark_dead) 1115 bdev->bd_holder_ops->mark_dead(bdev, surprise); 1116 else { 1117 mutex_unlock(&bdev->bd_holder_lock); 1118 sync_blockdev(bdev); 1119 } 1120 1121 invalidate_bdev(bdev); 1122 } 1123 /* 1124 * New drivers should not use this directly. There are some drivers however 1125 * that needs this for historical reasons. For example, the DASD driver has 1126 * historically had a shutdown to offline mode that doesn't actually remove the 1127 * gendisk that otherwise looks a lot like a safe device removal. 1128 */ 1129 EXPORT_SYMBOL_GPL(bdev_mark_dead); 1130 1131 void sync_bdevs(bool wait) 1132 { 1133 struct inode *inode, *old_inode = NULL; 1134 1135 spin_lock(&blockdev_superblock->s_inode_list_lock); 1136 list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) { 1137 struct address_space *mapping = inode->i_mapping; 1138 struct block_device *bdev; 1139 1140 spin_lock(&inode->i_lock); 1141 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) || 1142 mapping->nrpages == 0) { 1143 spin_unlock(&inode->i_lock); 1144 continue; 1145 } 1146 __iget(inode); 1147 spin_unlock(&inode->i_lock); 1148 spin_unlock(&blockdev_superblock->s_inode_list_lock); 1149 /* 1150 * We hold a reference to 'inode' so it couldn't have been 1151 * removed from s_inodes list while we dropped the 1152 * s_inode_list_lock We cannot iput the inode now as we can 1153 * be holding the last reference and we cannot iput it under 1154 * s_inode_list_lock. So we keep the reference and iput it 1155 * later. 1156 */ 1157 iput(old_inode); 1158 old_inode = inode; 1159 bdev = I_BDEV(inode); 1160 1161 mutex_lock(&bdev->bd_disk->open_mutex); 1162 if (!atomic_read(&bdev->bd_openers)) { 1163 ; /* skip */ 1164 } else if (wait) { 1165 /* 1166 * We keep the error status of individual mapping so 1167 * that applications can catch the writeback error using 1168 * fsync(2). See filemap_fdatawait_keep_errors() for 1169 * details. 1170 */ 1171 filemap_fdatawait_keep_errors(inode->i_mapping); 1172 } else { 1173 filemap_fdatawrite(inode->i_mapping); 1174 } 1175 mutex_unlock(&bdev->bd_disk->open_mutex); 1176 1177 spin_lock(&blockdev_superblock->s_inode_list_lock); 1178 } 1179 spin_unlock(&blockdev_superblock->s_inode_list_lock); 1180 iput(old_inode); 1181 } 1182 1183 /* 1184 * Handle STATX_DIOALIGN for block devices. 1185 * 1186 * Note that the inode passed to this is the inode of a block device node file, 1187 * not the block device's internal inode. Therefore it is *not* valid to use 1188 * I_BDEV() here; the block device has to be looked up by i_rdev instead. 1189 */ 1190 void bdev_statx_dioalign(struct inode *inode, struct kstat *stat) 1191 { 1192 struct block_device *bdev; 1193 1194 bdev = blkdev_get_no_open(inode->i_rdev); 1195 if (!bdev) 1196 return; 1197 1198 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 1199 stat->dio_offset_align = bdev_logical_block_size(bdev); 1200 stat->result_mask |= STATX_DIOALIGN; 1201 1202 blkdev_put_no_open(bdev); 1203 } 1204 1205 static int __init setup_bdev_allow_write_mounted(char *str) 1206 { 1207 if (kstrtobool(str, &bdev_allow_write_mounted)) 1208 pr_warn("Invalid option string for bdev_allow_write_mounted:" 1209 " '%s'\n", str); 1210 return 1; 1211 } 1212 __setup("bdev_allow_write_mounted=", setup_bdev_allow_write_mounted); 1213