1 /* 2 * Copyright (C) 2007 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 19 #include <linux/fs.h> 20 #include <linux/blkdev.h> 21 #include <linux/scatterlist.h> 22 #include <linux/swap.h> 23 #include <linux/radix-tree.h> 24 #include <linux/writeback.h> 25 #include <linux/buffer_head.h> 26 #include <linux/workqueue.h> 27 #include <linux/kthread.h> 28 #include <linux/freezer.h> 29 #include <linux/slab.h> 30 #include <linux/migrate.h> 31 #include <linux/ratelimit.h> 32 #include <linux/uuid.h> 33 #include <linux/semaphore.h> 34 #include <asm/unaligned.h> 35 #include "ctree.h" 36 #include "disk-io.h" 37 #include "hash.h" 38 #include "transaction.h" 39 #include "btrfs_inode.h" 40 #include "volumes.h" 41 #include "print-tree.h" 42 #include "async-thread.h" 43 #include "locking.h" 44 #include "tree-log.h" 45 #include "free-space-cache.h" 46 #include "inode-map.h" 47 #include "check-integrity.h" 48 #include "rcu-string.h" 49 #include "dev-replace.h" 50 #include "raid56.h" 51 #include "sysfs.h" 52 #include "qgroup.h" 53 54 #ifdef CONFIG_X86 55 #include <asm/cpufeature.h> 56 #endif 57 58 static struct extent_io_ops btree_extent_io_ops; 59 static void end_workqueue_fn(struct btrfs_work *work); 60 static void free_fs_root(struct btrfs_root *root); 61 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 62 int read_only); 63 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t, 64 struct btrfs_root *root); 65 static void btrfs_destroy_ordered_extents(struct btrfs_root *root); 66 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 67 struct btrfs_root *root); 68 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root); 69 static int btrfs_destroy_marked_extents(struct btrfs_root *root, 70 struct extent_io_tree *dirty_pages, 71 int mark); 72 static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 73 struct extent_io_tree *pinned_extents); 74 static int btrfs_cleanup_transaction(struct btrfs_root *root); 75 static void btrfs_error_commit_super(struct btrfs_root *root); 76 77 /* 78 * end_io_wq structs are used to do processing in task context when an IO is 79 * complete. This is used during reads to verify checksums, and it is used 80 * by writes to insert metadata for new file extents after IO is complete. 81 */ 82 struct end_io_wq { 83 struct bio *bio; 84 bio_end_io_t *end_io; 85 void *private; 86 struct btrfs_fs_info *info; 87 int error; 88 int metadata; 89 struct list_head list; 90 struct btrfs_work work; 91 }; 92 93 /* 94 * async submit bios are used to offload expensive checksumming 95 * onto the worker threads. They checksum file and metadata bios 96 * just before they are sent down the IO stack. 97 */ 98 struct async_submit_bio { 99 struct inode *inode; 100 struct bio *bio; 101 struct list_head list; 102 extent_submit_bio_hook_t *submit_bio_start; 103 extent_submit_bio_hook_t *submit_bio_done; 104 int rw; 105 int mirror_num; 106 unsigned long bio_flags; 107 /* 108 * bio_offset is optional, can be used if the pages in the bio 109 * can't tell us where in the file the bio should go 110 */ 111 u64 bio_offset; 112 struct btrfs_work work; 113 int error; 114 }; 115 116 /* 117 * Lockdep class keys for extent_buffer->lock's in this root. For a given 118 * eb, the lockdep key is determined by the btrfs_root it belongs to and 119 * the level the eb occupies in the tree. 120 * 121 * Different roots are used for different purposes and may nest inside each 122 * other and they require separate keysets. As lockdep keys should be 123 * static, assign keysets according to the purpose of the root as indicated 124 * by btrfs_root->objectid. This ensures that all special purpose roots 125 * have separate keysets. 126 * 127 * Lock-nesting across peer nodes is always done with the immediate parent 128 * node locked thus preventing deadlock. As lockdep doesn't know this, use 129 * subclass to avoid triggering lockdep warning in such cases. 130 * 131 * The key is set by the readpage_end_io_hook after the buffer has passed 132 * csum validation but before the pages are unlocked. It is also set by 133 * btrfs_init_new_buffer on freshly allocated blocks. 134 * 135 * We also add a check to make sure the highest level of the tree is the 136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code 137 * needs update as well. 138 */ 139 #ifdef CONFIG_DEBUG_LOCK_ALLOC 140 # if BTRFS_MAX_LEVEL != 8 141 # error 142 # endif 143 144 static struct btrfs_lockdep_keyset { 145 u64 id; /* root objectid */ 146 const char *name_stem; /* lock name stem */ 147 char names[BTRFS_MAX_LEVEL + 1][20]; 148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1]; 149 } btrfs_lockdep_keysets[] = { 150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" }, 151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" }, 152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" }, 153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" }, 154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" }, 155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" }, 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" }, 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" }, 158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" }, 159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" }, 160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" }, 161 { .id = 0, .name_stem = "tree" }, 162 }; 163 164 void __init btrfs_init_lockdep(void) 165 { 166 int i, j; 167 168 /* initialize lockdep class names */ 169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) { 170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i]; 171 172 for (j = 0; j < ARRAY_SIZE(ks->names); j++) 173 snprintf(ks->names[j], sizeof(ks->names[j]), 174 "btrfs-%s-%02d", ks->name_stem, j); 175 } 176 } 177 178 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, 179 int level) 180 { 181 struct btrfs_lockdep_keyset *ks; 182 183 BUG_ON(level >= ARRAY_SIZE(ks->keys)); 184 185 /* find the matching keyset, id 0 is the default entry */ 186 for (ks = btrfs_lockdep_keysets; ks->id; ks++) 187 if (ks->id == objectid) 188 break; 189 190 lockdep_set_class_and_name(&eb->lock, 191 &ks->keys[level], ks->names[level]); 192 } 193 194 #endif 195 196 /* 197 * extents on the btree inode are pretty simple, there's one extent 198 * that covers the entire device 199 */ 200 static struct extent_map *btree_get_extent(struct inode *inode, 201 struct page *page, size_t pg_offset, u64 start, u64 len, 202 int create) 203 { 204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 205 struct extent_map *em; 206 int ret; 207 208 read_lock(&em_tree->lock); 209 em = lookup_extent_mapping(em_tree, start, len); 210 if (em) { 211 em->bdev = 212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 213 read_unlock(&em_tree->lock); 214 goto out; 215 } 216 read_unlock(&em_tree->lock); 217 218 em = alloc_extent_map(); 219 if (!em) { 220 em = ERR_PTR(-ENOMEM); 221 goto out; 222 } 223 em->start = 0; 224 em->len = (u64)-1; 225 em->block_len = (u64)-1; 226 em->block_start = 0; 227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev; 228 229 write_lock(&em_tree->lock); 230 ret = add_extent_mapping(em_tree, em, 0); 231 if (ret == -EEXIST) { 232 free_extent_map(em); 233 em = lookup_extent_mapping(em_tree, start, len); 234 if (!em) 235 em = ERR_PTR(-EIO); 236 } else if (ret) { 237 free_extent_map(em); 238 em = ERR_PTR(ret); 239 } 240 write_unlock(&em_tree->lock); 241 242 out: 243 return em; 244 } 245 246 u32 btrfs_csum_data(char *data, u32 seed, size_t len) 247 { 248 return btrfs_crc32c(seed, data, len); 249 } 250 251 void btrfs_csum_final(u32 crc, char *result) 252 { 253 put_unaligned_le32(~crc, result); 254 } 255 256 /* 257 * compute the csum for a btree block, and either verify it or write it 258 * into the csum field of the block. 259 */ 260 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf, 261 int verify) 262 { 263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy); 264 char *result = NULL; 265 unsigned long len; 266 unsigned long cur_len; 267 unsigned long offset = BTRFS_CSUM_SIZE; 268 char *kaddr; 269 unsigned long map_start; 270 unsigned long map_len; 271 int err; 272 u32 crc = ~(u32)0; 273 unsigned long inline_result; 274 275 len = buf->len - offset; 276 while (len > 0) { 277 err = map_private_extent_buffer(buf, offset, 32, 278 &kaddr, &map_start, &map_len); 279 if (err) 280 return 1; 281 cur_len = min(len, map_len - (offset - map_start)); 282 crc = btrfs_csum_data(kaddr + offset - map_start, 283 crc, cur_len); 284 len -= cur_len; 285 offset += cur_len; 286 } 287 if (csum_size > sizeof(inline_result)) { 288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS); 289 if (!result) 290 return 1; 291 } else { 292 result = (char *)&inline_result; 293 } 294 295 btrfs_csum_final(crc, result); 296 297 if (verify) { 298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) { 299 u32 val; 300 u32 found = 0; 301 memcpy(&found, result, csum_size); 302 303 read_extent_buffer(buf, &val, 0, csum_size); 304 printk_ratelimited(KERN_INFO 305 "BTRFS: %s checksum verify failed on %llu wanted %X found %X " 306 "level %d\n", 307 root->fs_info->sb->s_id, buf->start, 308 val, found, btrfs_header_level(buf)); 309 if (result != (char *)&inline_result) 310 kfree(result); 311 return 1; 312 } 313 } else { 314 write_extent_buffer(buf, result, 0, csum_size); 315 } 316 if (result != (char *)&inline_result) 317 kfree(result); 318 return 0; 319 } 320 321 /* 322 * we can't consider a given block up to date unless the transid of the 323 * block matches the transid in the parent node's pointer. This is how we 324 * detect blocks that either didn't get written at all or got written 325 * in the wrong place. 326 */ 327 static int verify_parent_transid(struct extent_io_tree *io_tree, 328 struct extent_buffer *eb, u64 parent_transid, 329 int atomic) 330 { 331 struct extent_state *cached_state = NULL; 332 int ret; 333 bool need_lock = (current->journal_info == 334 (void *)BTRFS_SEND_TRANS_STUB); 335 336 if (!parent_transid || btrfs_header_generation(eb) == parent_transid) 337 return 0; 338 339 if (atomic) 340 return -EAGAIN; 341 342 if (need_lock) { 343 btrfs_tree_read_lock(eb); 344 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); 345 } 346 347 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1, 348 0, &cached_state); 349 if (extent_buffer_uptodate(eb) && 350 btrfs_header_generation(eb) == parent_transid) { 351 ret = 0; 352 goto out; 353 } 354 printk_ratelimited("parent transid verify failed on %llu wanted %llu " 355 "found %llu\n", 356 eb->start, parent_transid, btrfs_header_generation(eb)); 357 ret = 1; 358 359 /* 360 * Things reading via commit roots that don't have normal protection, 361 * like send, can have a really old block in cache that may point at a 362 * block that has been free'd and re-allocated. So don't clear uptodate 363 * if we find an eb that is under IO (dirty/writeback) because we could 364 * end up reading in the stale data and then writing it back out and 365 * making everybody very sad. 366 */ 367 if (!extent_buffer_under_io(eb)) 368 clear_extent_buffer_uptodate(eb); 369 out: 370 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1, 371 &cached_state, GFP_NOFS); 372 btrfs_tree_read_unlock_blocking(eb); 373 return ret; 374 } 375 376 /* 377 * Return 0 if the superblock checksum type matches the checksum value of that 378 * algorithm. Pass the raw disk superblock data. 379 */ 380 static int btrfs_check_super_csum(char *raw_disk_sb) 381 { 382 struct btrfs_super_block *disk_sb = 383 (struct btrfs_super_block *)raw_disk_sb; 384 u16 csum_type = btrfs_super_csum_type(disk_sb); 385 int ret = 0; 386 387 if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 388 u32 crc = ~(u32)0; 389 const int csum_size = sizeof(crc); 390 char result[csum_size]; 391 392 /* 393 * The super_block structure does not span the whole 394 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 395 * is filled with zeros and is included in the checkum. 396 */ 397 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 398 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 399 btrfs_csum_final(crc, result); 400 401 if (memcmp(raw_disk_sb, result, csum_size)) 402 ret = 1; 403 404 if (ret && btrfs_super_generation(disk_sb) < 10) { 405 printk(KERN_WARNING 406 "BTRFS: super block crcs don't match, older mkfs detected\n"); 407 ret = 0; 408 } 409 } 410 411 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 412 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n", 413 csum_type); 414 ret = 1; 415 } 416 417 return ret; 418 } 419 420 /* 421 * helper to read a given tree block, doing retries as required when 422 * the checksums don't match and we have alternate mirrors to try. 423 */ 424 static int btree_read_extent_buffer_pages(struct btrfs_root *root, 425 struct extent_buffer *eb, 426 u64 start, u64 parent_transid) 427 { 428 struct extent_io_tree *io_tree; 429 int failed = 0; 430 int ret; 431 int num_copies = 0; 432 int mirror_num = 0; 433 int failed_mirror = 0; 434 435 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 436 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 437 while (1) { 438 ret = read_extent_buffer_pages(io_tree, eb, start, 439 WAIT_COMPLETE, 440 btree_get_extent, mirror_num); 441 if (!ret) { 442 if (!verify_parent_transid(io_tree, eb, 443 parent_transid, 0)) 444 break; 445 else 446 ret = -EIO; 447 } 448 449 /* 450 * This buffer's crc is fine, but its contents are corrupted, so 451 * there is no reason to read the other copies, they won't be 452 * any less wrong. 453 */ 454 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 455 break; 456 457 num_copies = btrfs_num_copies(root->fs_info, 458 eb->start, eb->len); 459 if (num_copies == 1) 460 break; 461 462 if (!failed_mirror) { 463 failed = 1; 464 failed_mirror = eb->read_mirror; 465 } 466 467 mirror_num++; 468 if (mirror_num == failed_mirror) 469 mirror_num++; 470 471 if (mirror_num > num_copies) 472 break; 473 } 474 475 if (failed && !ret && failed_mirror) 476 repair_eb_io_failure(root, eb, failed_mirror); 477 478 return ret; 479 } 480 481 /* 482 * checksum a dirty tree block before IO. This has extra checks to make sure 483 * we only fill in the checksum field in the first page of a multi-page block 484 */ 485 486 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 487 { 488 u64 start = page_offset(page); 489 u64 found_start; 490 struct extent_buffer *eb; 491 492 eb = (struct extent_buffer *)page->private; 493 if (page != eb->pages[0]) 494 return 0; 495 found_start = btrfs_header_bytenr(eb); 496 if (WARN_ON(found_start != start || !PageUptodate(page))) 497 return 0; 498 csum_tree_block(root, eb, 0); 499 return 0; 500 } 501 502 static int check_tree_block_fsid(struct btrfs_root *root, 503 struct extent_buffer *eb) 504 { 505 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 506 u8 fsid[BTRFS_UUID_SIZE]; 507 int ret = 1; 508 509 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 510 while (fs_devices) { 511 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 512 ret = 0; 513 break; 514 } 515 fs_devices = fs_devices->seed; 516 } 517 return ret; 518 } 519 520 #define CORRUPT(reason, eb, root, slot) \ 521 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \ 522 "root=%llu, slot=%d", reason, \ 523 btrfs_header_bytenr(eb), root->objectid, slot) 524 525 static noinline int check_leaf(struct btrfs_root *root, 526 struct extent_buffer *leaf) 527 { 528 struct btrfs_key key; 529 struct btrfs_key leaf_key; 530 u32 nritems = btrfs_header_nritems(leaf); 531 int slot; 532 533 if (nritems == 0) 534 return 0; 535 536 /* Check the 0 item */ 537 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 538 BTRFS_LEAF_DATA_SIZE(root)) { 539 CORRUPT("invalid item offset size pair", leaf, root, 0); 540 return -EIO; 541 } 542 543 /* 544 * Check to make sure each items keys are in the correct order and their 545 * offsets make sense. We only have to loop through nritems-1 because 546 * we check the current slot against the next slot, which verifies the 547 * next slot's offset+size makes sense and that the current's slot 548 * offset is correct. 549 */ 550 for (slot = 0; slot < nritems - 1; slot++) { 551 btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 552 btrfs_item_key_to_cpu(leaf, &key, slot + 1); 553 554 /* Make sure the keys are in the right order */ 555 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 556 CORRUPT("bad key order", leaf, root, slot); 557 return -EIO; 558 } 559 560 /* 561 * Make sure the offset and ends are right, remember that the 562 * item data starts at the end of the leaf and grows towards the 563 * front. 564 */ 565 if (btrfs_item_offset_nr(leaf, slot) != 566 btrfs_item_end_nr(leaf, slot + 1)) { 567 CORRUPT("slot offset bad", leaf, root, slot); 568 return -EIO; 569 } 570 571 /* 572 * Check to make sure that we don't point outside of the leaf, 573 * just incase all the items are consistent to eachother, but 574 * all point outside of the leaf. 575 */ 576 if (btrfs_item_end_nr(leaf, slot) > 577 BTRFS_LEAF_DATA_SIZE(root)) { 578 CORRUPT("slot end outside of leaf", leaf, root, slot); 579 return -EIO; 580 } 581 } 582 583 return 0; 584 } 585 586 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 587 u64 phy_offset, struct page *page, 588 u64 start, u64 end, int mirror) 589 { 590 u64 found_start; 591 int found_level; 592 struct extent_buffer *eb; 593 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 594 int ret = 0; 595 int reads_done; 596 597 if (!page->private) 598 goto out; 599 600 eb = (struct extent_buffer *)page->private; 601 602 /* the pending IO might have been the only thing that kept this buffer 603 * in memory. Make sure we have a ref for all this other checks 604 */ 605 extent_buffer_get(eb); 606 607 reads_done = atomic_dec_and_test(&eb->io_pages); 608 if (!reads_done) 609 goto err; 610 611 eb->read_mirror = mirror; 612 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { 613 ret = -EIO; 614 goto err; 615 } 616 617 found_start = btrfs_header_bytenr(eb); 618 if (found_start != eb->start) { 619 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start " 620 "%llu %llu\n", 621 found_start, eb->start); 622 ret = -EIO; 623 goto err; 624 } 625 if (check_tree_block_fsid(root, eb)) { 626 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n", 627 eb->start); 628 ret = -EIO; 629 goto err; 630 } 631 found_level = btrfs_header_level(eb); 632 if (found_level >= BTRFS_MAX_LEVEL) { 633 btrfs_info(root->fs_info, "bad tree block level %d", 634 (int)btrfs_header_level(eb)); 635 ret = -EIO; 636 goto err; 637 } 638 639 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 640 eb, found_level); 641 642 ret = csum_tree_block(root, eb, 1); 643 if (ret) { 644 ret = -EIO; 645 goto err; 646 } 647 648 /* 649 * If this is a leaf block and it is corrupt, set the corrupt bit so 650 * that we don't try and read the other copies of this block, just 651 * return -EIO. 652 */ 653 if (found_level == 0 && check_leaf(root, eb)) { 654 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 655 ret = -EIO; 656 } 657 658 if (!ret) 659 set_extent_buffer_uptodate(eb); 660 err: 661 if (reads_done && 662 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 663 btree_readahead_hook(root, eb, eb->start, ret); 664 665 if (ret) { 666 /* 667 * our io error hook is going to dec the io pages 668 * again, we have to make sure it has something 669 * to decrement 670 */ 671 atomic_inc(&eb->io_pages); 672 clear_extent_buffer_uptodate(eb); 673 } 674 free_extent_buffer(eb); 675 out: 676 return ret; 677 } 678 679 static int btree_io_failed_hook(struct page *page, int failed_mirror) 680 { 681 struct extent_buffer *eb; 682 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 683 684 eb = (struct extent_buffer *)page->private; 685 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); 686 eb->read_mirror = failed_mirror; 687 atomic_dec(&eb->io_pages); 688 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 689 btree_readahead_hook(root, eb, eb->start, -EIO); 690 return -EIO; /* we fixed nothing */ 691 } 692 693 static void end_workqueue_bio(struct bio *bio, int err) 694 { 695 struct end_io_wq *end_io_wq = bio->bi_private; 696 struct btrfs_fs_info *fs_info; 697 698 fs_info = end_io_wq->info; 699 end_io_wq->error = err; 700 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 701 702 if (bio->bi_rw & REQ_WRITE) { 703 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 704 btrfs_queue_work(fs_info->endio_meta_write_workers, 705 &end_io_wq->work); 706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 707 btrfs_queue_work(fs_info->endio_freespace_worker, 708 &end_io_wq->work); 709 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 710 btrfs_queue_work(fs_info->endio_raid56_workers, 711 &end_io_wq->work); 712 else 713 btrfs_queue_work(fs_info->endio_write_workers, 714 &end_io_wq->work); 715 } else { 716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 717 btrfs_queue_work(fs_info->endio_raid56_workers, 718 &end_io_wq->work); 719 else if (end_io_wq->metadata) 720 btrfs_queue_work(fs_info->endio_meta_workers, 721 &end_io_wq->work); 722 else 723 btrfs_queue_work(fs_info->endio_workers, 724 &end_io_wq->work); 725 } 726 } 727 728 /* 729 * For the metadata arg you want 730 * 731 * 0 - if data 732 * 1 - if normal metadta 733 * 2 - if writing to the free space cache area 734 * 3 - raid parity work 735 */ 736 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 737 int metadata) 738 { 739 struct end_io_wq *end_io_wq; 740 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 741 if (!end_io_wq) 742 return -ENOMEM; 743 744 end_io_wq->private = bio->bi_private; 745 end_io_wq->end_io = bio->bi_end_io; 746 end_io_wq->info = info; 747 end_io_wq->error = 0; 748 end_io_wq->bio = bio; 749 end_io_wq->metadata = metadata; 750 751 bio->bi_private = end_io_wq; 752 bio->bi_end_io = end_workqueue_bio; 753 return 0; 754 } 755 756 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 757 { 758 unsigned long limit = min_t(unsigned long, 759 info->thread_pool_size, 760 info->fs_devices->open_devices); 761 return 256 * limit; 762 } 763 764 static void run_one_async_start(struct btrfs_work *work) 765 { 766 struct async_submit_bio *async; 767 int ret; 768 769 async = container_of(work, struct async_submit_bio, work); 770 ret = async->submit_bio_start(async->inode, async->rw, async->bio, 771 async->mirror_num, async->bio_flags, 772 async->bio_offset); 773 if (ret) 774 async->error = ret; 775 } 776 777 static void run_one_async_done(struct btrfs_work *work) 778 { 779 struct btrfs_fs_info *fs_info; 780 struct async_submit_bio *async; 781 int limit; 782 783 async = container_of(work, struct async_submit_bio, work); 784 fs_info = BTRFS_I(async->inode)->root->fs_info; 785 786 limit = btrfs_async_submit_limit(fs_info); 787 limit = limit * 2 / 3; 788 789 if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 790 waitqueue_active(&fs_info->async_submit_wait)) 791 wake_up(&fs_info->async_submit_wait); 792 793 /* If an error occured we just want to clean up the bio and move on */ 794 if (async->error) { 795 bio_endio(async->bio, async->error); 796 return; 797 } 798 799 async->submit_bio_done(async->inode, async->rw, async->bio, 800 async->mirror_num, async->bio_flags, 801 async->bio_offset); 802 } 803 804 static void run_one_async_free(struct btrfs_work *work) 805 { 806 struct async_submit_bio *async; 807 808 async = container_of(work, struct async_submit_bio, work); 809 kfree(async); 810 } 811 812 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 813 int rw, struct bio *bio, int mirror_num, 814 unsigned long bio_flags, 815 u64 bio_offset, 816 extent_submit_bio_hook_t *submit_bio_start, 817 extent_submit_bio_hook_t *submit_bio_done) 818 { 819 struct async_submit_bio *async; 820 821 async = kmalloc(sizeof(*async), GFP_NOFS); 822 if (!async) 823 return -ENOMEM; 824 825 async->inode = inode; 826 async->rw = rw; 827 async->bio = bio; 828 async->mirror_num = mirror_num; 829 async->submit_bio_start = submit_bio_start; 830 async->submit_bio_done = submit_bio_done; 831 832 btrfs_init_work(&async->work, run_one_async_start, 833 run_one_async_done, run_one_async_free); 834 835 async->bio_flags = bio_flags; 836 async->bio_offset = bio_offset; 837 838 async->error = 0; 839 840 atomic_inc(&fs_info->nr_async_submits); 841 842 if (rw & REQ_SYNC) 843 btrfs_set_work_high_priority(&async->work); 844 845 btrfs_queue_work(fs_info->workers, &async->work); 846 847 while (atomic_read(&fs_info->async_submit_draining) && 848 atomic_read(&fs_info->nr_async_submits)) { 849 wait_event(fs_info->async_submit_wait, 850 (atomic_read(&fs_info->nr_async_submits) == 0)); 851 } 852 853 return 0; 854 } 855 856 static int btree_csum_one_bio(struct bio *bio) 857 { 858 struct bio_vec *bvec; 859 struct btrfs_root *root; 860 int i, ret = 0; 861 862 bio_for_each_segment_all(bvec, bio, i) { 863 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 864 ret = csum_dirty_buffer(root, bvec->bv_page); 865 if (ret) 866 break; 867 } 868 869 return ret; 870 } 871 872 static int __btree_submit_bio_start(struct inode *inode, int rw, 873 struct bio *bio, int mirror_num, 874 unsigned long bio_flags, 875 u64 bio_offset) 876 { 877 /* 878 * when we're called for a write, we're already in the async 879 * submission context. Just jump into btrfs_map_bio 880 */ 881 return btree_csum_one_bio(bio); 882 } 883 884 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 885 int mirror_num, unsigned long bio_flags, 886 u64 bio_offset) 887 { 888 int ret; 889 890 /* 891 * when we're called for a write, we're already in the async 892 * submission context. Just jump into btrfs_map_bio 893 */ 894 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 895 if (ret) 896 bio_endio(bio, ret); 897 return ret; 898 } 899 900 static int check_async_write(struct inode *inode, unsigned long bio_flags) 901 { 902 if (bio_flags & EXTENT_BIO_TREE_LOG) 903 return 0; 904 #ifdef CONFIG_X86 905 if (cpu_has_xmm4_2) 906 return 0; 907 #endif 908 return 1; 909 } 910 911 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 912 int mirror_num, unsigned long bio_flags, 913 u64 bio_offset) 914 { 915 int async = check_async_write(inode, bio_flags); 916 int ret; 917 918 if (!(rw & REQ_WRITE)) { 919 /* 920 * called for a read, do the setup so that checksum validation 921 * can happen in the async kernel threads 922 */ 923 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 924 bio, 1); 925 if (ret) 926 goto out_w_error; 927 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 928 mirror_num, 0); 929 } else if (!async) { 930 ret = btree_csum_one_bio(bio); 931 if (ret) 932 goto out_w_error; 933 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 934 mirror_num, 0); 935 } else { 936 /* 937 * kthread helpers are used to submit writes so that 938 * checksumming can happen in parallel across all CPUs 939 */ 940 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 941 inode, rw, bio, mirror_num, 0, 942 bio_offset, 943 __btree_submit_bio_start, 944 __btree_submit_bio_done); 945 } 946 947 if (ret) { 948 out_w_error: 949 bio_endio(bio, ret); 950 } 951 return ret; 952 } 953 954 #ifdef CONFIG_MIGRATION 955 static int btree_migratepage(struct address_space *mapping, 956 struct page *newpage, struct page *page, 957 enum migrate_mode mode) 958 { 959 /* 960 * we can't safely write a btree page from here, 961 * we haven't done the locking hook 962 */ 963 if (PageDirty(page)) 964 return -EAGAIN; 965 /* 966 * Buffers may be managed in a filesystem specific way. 967 * We must have no buffers or drop them. 968 */ 969 if (page_has_private(page) && 970 !try_to_release_page(page, GFP_KERNEL)) 971 return -EAGAIN; 972 return migrate_page(mapping, newpage, page, mode); 973 } 974 #endif 975 976 977 static int btree_writepages(struct address_space *mapping, 978 struct writeback_control *wbc) 979 { 980 struct btrfs_fs_info *fs_info; 981 int ret; 982 983 if (wbc->sync_mode == WB_SYNC_NONE) { 984 985 if (wbc->for_kupdate) 986 return 0; 987 988 fs_info = BTRFS_I(mapping->host)->root->fs_info; 989 /* this is a bit racy, but that's ok */ 990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 991 BTRFS_DIRTY_METADATA_THRESH); 992 if (ret < 0) 993 return 0; 994 } 995 return btree_write_cache_pages(mapping, wbc); 996 } 997 998 static int btree_readpage(struct file *file, struct page *page) 999 { 1000 struct extent_io_tree *tree; 1001 tree = &BTRFS_I(page->mapping->host)->io_tree; 1002 return extent_read_full_page(tree, page, btree_get_extent, 0); 1003 } 1004 1005 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 1006 { 1007 if (PageWriteback(page) || PageDirty(page)) 1008 return 0; 1009 1010 return try_release_extent_buffer(page); 1011 } 1012 1013 static void btree_invalidatepage(struct page *page, unsigned int offset, 1014 unsigned int length) 1015 { 1016 struct extent_io_tree *tree; 1017 tree = &BTRFS_I(page->mapping->host)->io_tree; 1018 extent_invalidatepage(tree, page, offset); 1019 btree_releasepage(page, GFP_NOFS); 1020 if (PagePrivate(page)) { 1021 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1022 "page private not zero on page %llu", 1023 (unsigned long long)page_offset(page)); 1024 ClearPagePrivate(page); 1025 set_page_private(page, 0); 1026 page_cache_release(page); 1027 } 1028 } 1029 1030 static int btree_set_page_dirty(struct page *page) 1031 { 1032 #ifdef DEBUG 1033 struct extent_buffer *eb; 1034 1035 BUG_ON(!PagePrivate(page)); 1036 eb = (struct extent_buffer *)page->private; 1037 BUG_ON(!eb); 1038 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1039 BUG_ON(!atomic_read(&eb->refs)); 1040 btrfs_assert_tree_locked(eb); 1041 #endif 1042 return __set_page_dirty_nobuffers(page); 1043 } 1044 1045 static const struct address_space_operations btree_aops = { 1046 .readpage = btree_readpage, 1047 .writepages = btree_writepages, 1048 .releasepage = btree_releasepage, 1049 .invalidatepage = btree_invalidatepage, 1050 #ifdef CONFIG_MIGRATION 1051 .migratepage = btree_migratepage, 1052 #endif 1053 .set_page_dirty = btree_set_page_dirty, 1054 }; 1055 1056 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1057 u64 parent_transid) 1058 { 1059 struct extent_buffer *buf = NULL; 1060 struct inode *btree_inode = root->fs_info->btree_inode; 1061 int ret = 0; 1062 1063 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1064 if (!buf) 1065 return 0; 1066 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1067 buf, 0, WAIT_NONE, btree_get_extent, 0); 1068 free_extent_buffer(buf); 1069 return ret; 1070 } 1071 1072 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1073 int mirror_num, struct extent_buffer **eb) 1074 { 1075 struct extent_buffer *buf = NULL; 1076 struct inode *btree_inode = root->fs_info->btree_inode; 1077 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1078 int ret; 1079 1080 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1081 if (!buf) 1082 return 0; 1083 1084 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1085 1086 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1087 btree_get_extent, mirror_num); 1088 if (ret) { 1089 free_extent_buffer(buf); 1090 return ret; 1091 } 1092 1093 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1094 free_extent_buffer(buf); 1095 return -EIO; 1096 } else if (extent_buffer_uptodate(buf)) { 1097 *eb = buf; 1098 } else { 1099 free_extent_buffer(buf); 1100 } 1101 return 0; 1102 } 1103 1104 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 1105 u64 bytenr, u32 blocksize) 1106 { 1107 return find_extent_buffer(root->fs_info, bytenr); 1108 } 1109 1110 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 1111 u64 bytenr, u32 blocksize) 1112 { 1113 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1114 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) 1115 return alloc_test_extent_buffer(root->fs_info, bytenr, 1116 blocksize); 1117 #endif 1118 return alloc_extent_buffer(root->fs_info, bytenr, blocksize); 1119 } 1120 1121 1122 int btrfs_write_tree_block(struct extent_buffer *buf) 1123 { 1124 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 1125 buf->start + buf->len - 1); 1126 } 1127 1128 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1129 { 1130 return filemap_fdatawait_range(buf->pages[0]->mapping, 1131 buf->start, buf->start + buf->len - 1); 1132 } 1133 1134 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1135 u32 blocksize, u64 parent_transid) 1136 { 1137 struct extent_buffer *buf = NULL; 1138 int ret; 1139 1140 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1141 if (!buf) 1142 return NULL; 1143 1144 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 1145 if (ret) { 1146 free_extent_buffer(buf); 1147 return NULL; 1148 } 1149 return buf; 1150 1151 } 1152 1153 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 1154 struct extent_buffer *buf) 1155 { 1156 struct btrfs_fs_info *fs_info = root->fs_info; 1157 1158 if (btrfs_header_generation(buf) == 1159 fs_info->running_transaction->transid) { 1160 btrfs_assert_tree_locked(buf); 1161 1162 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1163 __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1164 -buf->len, 1165 fs_info->dirty_metadata_batch); 1166 /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1167 btrfs_set_lock_blocking(buf); 1168 clear_extent_buffer_dirty(buf); 1169 } 1170 } 1171 } 1172 1173 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 1174 { 1175 struct btrfs_subvolume_writers *writers; 1176 int ret; 1177 1178 writers = kmalloc(sizeof(*writers), GFP_NOFS); 1179 if (!writers) 1180 return ERR_PTR(-ENOMEM); 1181 1182 ret = percpu_counter_init(&writers->counter, 0); 1183 if (ret < 0) { 1184 kfree(writers); 1185 return ERR_PTR(ret); 1186 } 1187 1188 init_waitqueue_head(&writers->wait); 1189 return writers; 1190 } 1191 1192 static void 1193 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 1194 { 1195 percpu_counter_destroy(&writers->counter); 1196 kfree(writers); 1197 } 1198 1199 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 1200 u32 stripesize, struct btrfs_root *root, 1201 struct btrfs_fs_info *fs_info, 1202 u64 objectid) 1203 { 1204 root->node = NULL; 1205 root->commit_root = NULL; 1206 root->sectorsize = sectorsize; 1207 root->nodesize = nodesize; 1208 root->leafsize = leafsize; 1209 root->stripesize = stripesize; 1210 root->state = 0; 1211 root->orphan_cleanup_state = 0; 1212 1213 root->objectid = objectid; 1214 root->last_trans = 0; 1215 root->highest_objectid = 0; 1216 root->nr_delalloc_inodes = 0; 1217 root->nr_ordered_extents = 0; 1218 root->name = NULL; 1219 root->inode_tree = RB_ROOT; 1220 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1221 root->block_rsv = NULL; 1222 root->orphan_block_rsv = NULL; 1223 1224 INIT_LIST_HEAD(&root->dirty_list); 1225 INIT_LIST_HEAD(&root->root_list); 1226 INIT_LIST_HEAD(&root->delalloc_inodes); 1227 INIT_LIST_HEAD(&root->delalloc_root); 1228 INIT_LIST_HEAD(&root->ordered_extents); 1229 INIT_LIST_HEAD(&root->ordered_root); 1230 INIT_LIST_HEAD(&root->logged_list[0]); 1231 INIT_LIST_HEAD(&root->logged_list[1]); 1232 spin_lock_init(&root->orphan_lock); 1233 spin_lock_init(&root->inode_lock); 1234 spin_lock_init(&root->delalloc_lock); 1235 spin_lock_init(&root->ordered_extent_lock); 1236 spin_lock_init(&root->accounting_lock); 1237 spin_lock_init(&root->log_extents_lock[0]); 1238 spin_lock_init(&root->log_extents_lock[1]); 1239 mutex_init(&root->objectid_mutex); 1240 mutex_init(&root->log_mutex); 1241 mutex_init(&root->ordered_extent_mutex); 1242 mutex_init(&root->delalloc_mutex); 1243 init_waitqueue_head(&root->log_writer_wait); 1244 init_waitqueue_head(&root->log_commit_wait[0]); 1245 init_waitqueue_head(&root->log_commit_wait[1]); 1246 INIT_LIST_HEAD(&root->log_ctxs[0]); 1247 INIT_LIST_HEAD(&root->log_ctxs[1]); 1248 atomic_set(&root->log_commit[0], 0); 1249 atomic_set(&root->log_commit[1], 0); 1250 atomic_set(&root->log_writers, 0); 1251 atomic_set(&root->log_batch, 0); 1252 atomic_set(&root->orphan_inodes, 0); 1253 atomic_set(&root->refs, 1); 1254 atomic_set(&root->will_be_snapshoted, 0); 1255 root->log_transid = 0; 1256 root->log_transid_committed = -1; 1257 root->last_log_commit = 0; 1258 if (fs_info) 1259 extent_io_tree_init(&root->dirty_log_pages, 1260 fs_info->btree_inode->i_mapping); 1261 1262 memset(&root->root_key, 0, sizeof(root->root_key)); 1263 memset(&root->root_item, 0, sizeof(root->root_item)); 1264 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 1265 memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 1266 if (fs_info) 1267 root->defrag_trans_start = fs_info->generation; 1268 else 1269 root->defrag_trans_start = 0; 1270 init_completion(&root->kobj_unregister); 1271 root->root_key.objectid = objectid; 1272 root->anon_dev = 0; 1273 1274 spin_lock_init(&root->root_item_lock); 1275 } 1276 1277 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 1278 { 1279 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 1280 if (root) 1281 root->fs_info = fs_info; 1282 return root; 1283 } 1284 1285 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1286 /* Should only be used by the testing infrastructure */ 1287 struct btrfs_root *btrfs_alloc_dummy_root(void) 1288 { 1289 struct btrfs_root *root; 1290 1291 root = btrfs_alloc_root(NULL); 1292 if (!root) 1293 return ERR_PTR(-ENOMEM); 1294 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1); 1295 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state); 1296 root->alloc_bytenr = 0; 1297 1298 return root; 1299 } 1300 #endif 1301 1302 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 1303 struct btrfs_fs_info *fs_info, 1304 u64 objectid) 1305 { 1306 struct extent_buffer *leaf; 1307 struct btrfs_root *tree_root = fs_info->tree_root; 1308 struct btrfs_root *root; 1309 struct btrfs_key key; 1310 int ret = 0; 1311 uuid_le uuid; 1312 1313 root = btrfs_alloc_root(fs_info); 1314 if (!root) 1315 return ERR_PTR(-ENOMEM); 1316 1317 __setup_root(tree_root->nodesize, tree_root->leafsize, 1318 tree_root->sectorsize, tree_root->stripesize, 1319 root, fs_info, objectid); 1320 root->root_key.objectid = objectid; 1321 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1322 root->root_key.offset = 0; 1323 1324 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 1325 0, objectid, NULL, 0, 0, 0); 1326 if (IS_ERR(leaf)) { 1327 ret = PTR_ERR(leaf); 1328 leaf = NULL; 1329 goto fail; 1330 } 1331 1332 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 1333 btrfs_set_header_bytenr(leaf, leaf->start); 1334 btrfs_set_header_generation(leaf, trans->transid); 1335 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1336 btrfs_set_header_owner(leaf, objectid); 1337 root->node = leaf; 1338 1339 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 1340 BTRFS_FSID_SIZE); 1341 write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1342 btrfs_header_chunk_tree_uuid(leaf), 1343 BTRFS_UUID_SIZE); 1344 btrfs_mark_buffer_dirty(leaf); 1345 1346 root->commit_root = btrfs_root_node(root); 1347 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 1348 1349 root->root_item.flags = 0; 1350 root->root_item.byte_limit = 0; 1351 btrfs_set_root_bytenr(&root->root_item, leaf->start); 1352 btrfs_set_root_generation(&root->root_item, trans->transid); 1353 btrfs_set_root_level(&root->root_item, 0); 1354 btrfs_set_root_refs(&root->root_item, 1); 1355 btrfs_set_root_used(&root->root_item, leaf->len); 1356 btrfs_set_root_last_snapshot(&root->root_item, 0); 1357 btrfs_set_root_dirid(&root->root_item, 0); 1358 uuid_le_gen(&uuid); 1359 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 1360 root->root_item.drop_level = 0; 1361 1362 key.objectid = objectid; 1363 key.type = BTRFS_ROOT_ITEM_KEY; 1364 key.offset = 0; 1365 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 1366 if (ret) 1367 goto fail; 1368 1369 btrfs_tree_unlock(leaf); 1370 1371 return root; 1372 1373 fail: 1374 if (leaf) { 1375 btrfs_tree_unlock(leaf); 1376 free_extent_buffer(root->commit_root); 1377 free_extent_buffer(leaf); 1378 } 1379 kfree(root); 1380 1381 return ERR_PTR(ret); 1382 } 1383 1384 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1385 struct btrfs_fs_info *fs_info) 1386 { 1387 struct btrfs_root *root; 1388 struct btrfs_root *tree_root = fs_info->tree_root; 1389 struct extent_buffer *leaf; 1390 1391 root = btrfs_alloc_root(fs_info); 1392 if (!root) 1393 return ERR_PTR(-ENOMEM); 1394 1395 __setup_root(tree_root->nodesize, tree_root->leafsize, 1396 tree_root->sectorsize, tree_root->stripesize, 1397 root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1398 1399 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1400 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1401 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1402 1403 /* 1404 * DON'T set REF_COWS for log trees 1405 * 1406 * log trees do not get reference counted because they go away 1407 * before a real commit is actually done. They do store pointers 1408 * to file data extents, and those reference counts still get 1409 * updated (along with back refs to the log tree). 1410 */ 1411 1412 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 1413 BTRFS_TREE_LOG_OBJECTID, NULL, 1414 0, 0, 0); 1415 if (IS_ERR(leaf)) { 1416 kfree(root); 1417 return ERR_CAST(leaf); 1418 } 1419 1420 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 1421 btrfs_set_header_bytenr(leaf, leaf->start); 1422 btrfs_set_header_generation(leaf, trans->transid); 1423 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1424 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 1425 root->node = leaf; 1426 1427 write_extent_buffer(root->node, root->fs_info->fsid, 1428 btrfs_header_fsid(), BTRFS_FSID_SIZE); 1429 btrfs_mark_buffer_dirty(root->node); 1430 btrfs_tree_unlock(root->node); 1431 return root; 1432 } 1433 1434 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1435 struct btrfs_fs_info *fs_info) 1436 { 1437 struct btrfs_root *log_root; 1438 1439 log_root = alloc_log_tree(trans, fs_info); 1440 if (IS_ERR(log_root)) 1441 return PTR_ERR(log_root); 1442 WARN_ON(fs_info->log_root_tree); 1443 fs_info->log_root_tree = log_root; 1444 return 0; 1445 } 1446 1447 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1448 struct btrfs_root *root) 1449 { 1450 struct btrfs_root *log_root; 1451 struct btrfs_inode_item *inode_item; 1452 1453 log_root = alloc_log_tree(trans, root->fs_info); 1454 if (IS_ERR(log_root)) 1455 return PTR_ERR(log_root); 1456 1457 log_root->last_trans = trans->transid; 1458 log_root->root_key.offset = root->root_key.objectid; 1459 1460 inode_item = &log_root->root_item.inode; 1461 btrfs_set_stack_inode_generation(inode_item, 1); 1462 btrfs_set_stack_inode_size(inode_item, 3); 1463 btrfs_set_stack_inode_nlink(inode_item, 1); 1464 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize); 1465 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1466 1467 btrfs_set_root_node(&log_root->root_item, log_root->node); 1468 1469 WARN_ON(root->log_root); 1470 root->log_root = log_root; 1471 root->log_transid = 0; 1472 root->log_transid_committed = -1; 1473 root->last_log_commit = 0; 1474 return 0; 1475 } 1476 1477 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1478 struct btrfs_key *key) 1479 { 1480 struct btrfs_root *root; 1481 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1482 struct btrfs_path *path; 1483 u64 generation; 1484 u32 blocksize; 1485 int ret; 1486 1487 path = btrfs_alloc_path(); 1488 if (!path) 1489 return ERR_PTR(-ENOMEM); 1490 1491 root = btrfs_alloc_root(fs_info); 1492 if (!root) { 1493 ret = -ENOMEM; 1494 goto alloc_fail; 1495 } 1496 1497 __setup_root(tree_root->nodesize, tree_root->leafsize, 1498 tree_root->sectorsize, tree_root->stripesize, 1499 root, fs_info, key->objectid); 1500 1501 ret = btrfs_find_root(tree_root, key, path, 1502 &root->root_item, &root->root_key); 1503 if (ret) { 1504 if (ret > 0) 1505 ret = -ENOENT; 1506 goto find_fail; 1507 } 1508 1509 generation = btrfs_root_generation(&root->root_item); 1510 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1511 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1512 blocksize, generation); 1513 if (!root->node) { 1514 ret = -ENOMEM; 1515 goto find_fail; 1516 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1517 ret = -EIO; 1518 goto read_fail; 1519 } 1520 root->commit_root = btrfs_root_node(root); 1521 out: 1522 btrfs_free_path(path); 1523 return root; 1524 1525 read_fail: 1526 free_extent_buffer(root->node); 1527 find_fail: 1528 kfree(root); 1529 alloc_fail: 1530 root = ERR_PTR(ret); 1531 goto out; 1532 } 1533 1534 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1535 struct btrfs_key *location) 1536 { 1537 struct btrfs_root *root; 1538 1539 root = btrfs_read_tree_root(tree_root, location); 1540 if (IS_ERR(root)) 1541 return root; 1542 1543 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 1544 set_bit(BTRFS_ROOT_REF_COWS, &root->state); 1545 btrfs_check_and_init_root_item(&root->root_item); 1546 } 1547 1548 return root; 1549 } 1550 1551 int btrfs_init_fs_root(struct btrfs_root *root) 1552 { 1553 int ret; 1554 struct btrfs_subvolume_writers *writers; 1555 1556 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1557 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1558 GFP_NOFS); 1559 if (!root->free_ino_pinned || !root->free_ino_ctl) { 1560 ret = -ENOMEM; 1561 goto fail; 1562 } 1563 1564 writers = btrfs_alloc_subvolume_writers(); 1565 if (IS_ERR(writers)) { 1566 ret = PTR_ERR(writers); 1567 goto fail; 1568 } 1569 root->subv_writers = writers; 1570 1571 btrfs_init_free_ino_ctl(root); 1572 spin_lock_init(&root->cache_lock); 1573 init_waitqueue_head(&root->cache_wait); 1574 1575 ret = get_anon_bdev(&root->anon_dev); 1576 if (ret) 1577 goto free_writers; 1578 return 0; 1579 1580 free_writers: 1581 btrfs_free_subvolume_writers(root->subv_writers); 1582 fail: 1583 kfree(root->free_ino_ctl); 1584 kfree(root->free_ino_pinned); 1585 return ret; 1586 } 1587 1588 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1589 u64 root_id) 1590 { 1591 struct btrfs_root *root; 1592 1593 spin_lock(&fs_info->fs_roots_radix_lock); 1594 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1595 (unsigned long)root_id); 1596 spin_unlock(&fs_info->fs_roots_radix_lock); 1597 return root; 1598 } 1599 1600 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1601 struct btrfs_root *root) 1602 { 1603 int ret; 1604 1605 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1606 if (ret) 1607 return ret; 1608 1609 spin_lock(&fs_info->fs_roots_radix_lock); 1610 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1611 (unsigned long)root->root_key.objectid, 1612 root); 1613 if (ret == 0) 1614 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1615 spin_unlock(&fs_info->fs_roots_radix_lock); 1616 radix_tree_preload_end(); 1617 1618 return ret; 1619 } 1620 1621 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1622 struct btrfs_key *location, 1623 bool check_ref) 1624 { 1625 struct btrfs_root *root; 1626 int ret; 1627 1628 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1629 return fs_info->tree_root; 1630 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1631 return fs_info->extent_root; 1632 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 1633 return fs_info->chunk_root; 1634 if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 1635 return fs_info->dev_root; 1636 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 1637 return fs_info->csum_root; 1638 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1639 return fs_info->quota_root ? fs_info->quota_root : 1640 ERR_PTR(-ENOENT); 1641 if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1642 return fs_info->uuid_root ? fs_info->uuid_root : 1643 ERR_PTR(-ENOENT); 1644 again: 1645 root = btrfs_lookup_fs_root(fs_info, location->objectid); 1646 if (root) { 1647 if (check_ref && btrfs_root_refs(&root->root_item) == 0) 1648 return ERR_PTR(-ENOENT); 1649 return root; 1650 } 1651 1652 root = btrfs_read_fs_root(fs_info->tree_root, location); 1653 if (IS_ERR(root)) 1654 return root; 1655 1656 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1657 ret = -ENOENT; 1658 goto fail; 1659 } 1660 1661 ret = btrfs_init_fs_root(root); 1662 if (ret) 1663 goto fail; 1664 1665 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID, 1666 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL); 1667 if (ret < 0) 1668 goto fail; 1669 if (ret == 0) 1670 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1671 1672 ret = btrfs_insert_fs_root(fs_info, root); 1673 if (ret) { 1674 if (ret == -EEXIST) { 1675 free_fs_root(root); 1676 goto again; 1677 } 1678 goto fail; 1679 } 1680 return root; 1681 fail: 1682 free_fs_root(root); 1683 return ERR_PTR(ret); 1684 } 1685 1686 static int btrfs_congested_fn(void *congested_data, int bdi_bits) 1687 { 1688 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 1689 int ret = 0; 1690 struct btrfs_device *device; 1691 struct backing_dev_info *bdi; 1692 1693 rcu_read_lock(); 1694 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1695 if (!device->bdev) 1696 continue; 1697 bdi = blk_get_backing_dev_info(device->bdev); 1698 if (bdi && bdi_congested(bdi, bdi_bits)) { 1699 ret = 1; 1700 break; 1701 } 1702 } 1703 rcu_read_unlock(); 1704 return ret; 1705 } 1706 1707 /* 1708 * If this fails, caller must call bdi_destroy() to get rid of the 1709 * bdi again. 1710 */ 1711 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 1712 { 1713 int err; 1714 1715 bdi->capabilities = BDI_CAP_MAP_COPY; 1716 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1717 if (err) 1718 return err; 1719 1720 bdi->ra_pages = default_backing_dev_info.ra_pages; 1721 bdi->congested_fn = btrfs_congested_fn; 1722 bdi->congested_data = info; 1723 return 0; 1724 } 1725 1726 /* 1727 * called by the kthread helper functions to finally call the bio end_io 1728 * functions. This is where read checksum verification actually happens 1729 */ 1730 static void end_workqueue_fn(struct btrfs_work *work) 1731 { 1732 struct bio *bio; 1733 struct end_io_wq *end_io_wq; 1734 int error; 1735 1736 end_io_wq = container_of(work, struct end_io_wq, work); 1737 bio = end_io_wq->bio; 1738 1739 error = end_io_wq->error; 1740 bio->bi_private = end_io_wq->private; 1741 bio->bi_end_io = end_io_wq->end_io; 1742 kfree(end_io_wq); 1743 bio_endio_nodec(bio, error); 1744 } 1745 1746 static int cleaner_kthread(void *arg) 1747 { 1748 struct btrfs_root *root = arg; 1749 int again; 1750 1751 do { 1752 again = 0; 1753 1754 /* Make the cleaner go to sleep early. */ 1755 if (btrfs_need_cleaner_sleep(root)) 1756 goto sleep; 1757 1758 if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1759 goto sleep; 1760 1761 /* 1762 * Avoid the problem that we change the status of the fs 1763 * during the above check and trylock. 1764 */ 1765 if (btrfs_need_cleaner_sleep(root)) { 1766 mutex_unlock(&root->fs_info->cleaner_mutex); 1767 goto sleep; 1768 } 1769 1770 btrfs_run_delayed_iputs(root); 1771 again = btrfs_clean_one_deleted_snapshot(root); 1772 mutex_unlock(&root->fs_info->cleaner_mutex); 1773 1774 /* 1775 * The defragger has dealt with the R/O remount and umount, 1776 * needn't do anything special here. 1777 */ 1778 btrfs_run_defrag_inodes(root->fs_info); 1779 sleep: 1780 if (!try_to_freeze() && !again) { 1781 set_current_state(TASK_INTERRUPTIBLE); 1782 if (!kthread_should_stop()) 1783 schedule(); 1784 __set_current_state(TASK_RUNNING); 1785 } 1786 } while (!kthread_should_stop()); 1787 return 0; 1788 } 1789 1790 static int transaction_kthread(void *arg) 1791 { 1792 struct btrfs_root *root = arg; 1793 struct btrfs_trans_handle *trans; 1794 struct btrfs_transaction *cur; 1795 u64 transid; 1796 unsigned long now; 1797 unsigned long delay; 1798 bool cannot_commit; 1799 1800 do { 1801 cannot_commit = false; 1802 delay = HZ * root->fs_info->commit_interval; 1803 mutex_lock(&root->fs_info->transaction_kthread_mutex); 1804 1805 spin_lock(&root->fs_info->trans_lock); 1806 cur = root->fs_info->running_transaction; 1807 if (!cur) { 1808 spin_unlock(&root->fs_info->trans_lock); 1809 goto sleep; 1810 } 1811 1812 now = get_seconds(); 1813 if (cur->state < TRANS_STATE_BLOCKED && 1814 (now < cur->start_time || 1815 now - cur->start_time < root->fs_info->commit_interval)) { 1816 spin_unlock(&root->fs_info->trans_lock); 1817 delay = HZ * 5; 1818 goto sleep; 1819 } 1820 transid = cur->transid; 1821 spin_unlock(&root->fs_info->trans_lock); 1822 1823 /* If the file system is aborted, this will always fail. */ 1824 trans = btrfs_attach_transaction(root); 1825 if (IS_ERR(trans)) { 1826 if (PTR_ERR(trans) != -ENOENT) 1827 cannot_commit = true; 1828 goto sleep; 1829 } 1830 if (transid == trans->transid) { 1831 btrfs_commit_transaction(trans, root); 1832 } else { 1833 btrfs_end_transaction(trans, root); 1834 } 1835 sleep: 1836 wake_up_process(root->fs_info->cleaner_kthread); 1837 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1838 1839 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 1840 &root->fs_info->fs_state))) 1841 btrfs_cleanup_transaction(root); 1842 if (!try_to_freeze()) { 1843 set_current_state(TASK_INTERRUPTIBLE); 1844 if (!kthread_should_stop() && 1845 (!btrfs_transaction_blocked(root->fs_info) || 1846 cannot_commit)) 1847 schedule_timeout(delay); 1848 __set_current_state(TASK_RUNNING); 1849 } 1850 } while (!kthread_should_stop()); 1851 return 0; 1852 } 1853 1854 /* 1855 * this will find the highest generation in the array of 1856 * root backups. The index of the highest array is returned, 1857 * or -1 if we can't find anything. 1858 * 1859 * We check to make sure the array is valid by comparing the 1860 * generation of the latest root in the array with the generation 1861 * in the super block. If they don't match we pitch it. 1862 */ 1863 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1864 { 1865 u64 cur; 1866 int newest_index = -1; 1867 struct btrfs_root_backup *root_backup; 1868 int i; 1869 1870 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1871 root_backup = info->super_copy->super_roots + i; 1872 cur = btrfs_backup_tree_root_gen(root_backup); 1873 if (cur == newest_gen) 1874 newest_index = i; 1875 } 1876 1877 /* check to see if we actually wrapped around */ 1878 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1879 root_backup = info->super_copy->super_roots; 1880 cur = btrfs_backup_tree_root_gen(root_backup); 1881 if (cur == newest_gen) 1882 newest_index = 0; 1883 } 1884 return newest_index; 1885 } 1886 1887 1888 /* 1889 * find the oldest backup so we know where to store new entries 1890 * in the backup array. This will set the backup_root_index 1891 * field in the fs_info struct 1892 */ 1893 static void find_oldest_super_backup(struct btrfs_fs_info *info, 1894 u64 newest_gen) 1895 { 1896 int newest_index = -1; 1897 1898 newest_index = find_newest_super_backup(info, newest_gen); 1899 /* if there was garbage in there, just move along */ 1900 if (newest_index == -1) { 1901 info->backup_root_index = 0; 1902 } else { 1903 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1904 } 1905 } 1906 1907 /* 1908 * copy all the root pointers into the super backup array. 1909 * this will bump the backup pointer by one when it is 1910 * done 1911 */ 1912 static void backup_super_roots(struct btrfs_fs_info *info) 1913 { 1914 int next_backup; 1915 struct btrfs_root_backup *root_backup; 1916 int last_backup; 1917 1918 next_backup = info->backup_root_index; 1919 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1920 BTRFS_NUM_BACKUP_ROOTS; 1921 1922 /* 1923 * just overwrite the last backup if we're at the same generation 1924 * this happens only at umount 1925 */ 1926 root_backup = info->super_for_commit->super_roots + last_backup; 1927 if (btrfs_backup_tree_root_gen(root_backup) == 1928 btrfs_header_generation(info->tree_root->node)) 1929 next_backup = last_backup; 1930 1931 root_backup = info->super_for_commit->super_roots + next_backup; 1932 1933 /* 1934 * make sure all of our padding and empty slots get zero filled 1935 * regardless of which ones we use today 1936 */ 1937 memset(root_backup, 0, sizeof(*root_backup)); 1938 1939 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1940 1941 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1942 btrfs_set_backup_tree_root_gen(root_backup, 1943 btrfs_header_generation(info->tree_root->node)); 1944 1945 btrfs_set_backup_tree_root_level(root_backup, 1946 btrfs_header_level(info->tree_root->node)); 1947 1948 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1949 btrfs_set_backup_chunk_root_gen(root_backup, 1950 btrfs_header_generation(info->chunk_root->node)); 1951 btrfs_set_backup_chunk_root_level(root_backup, 1952 btrfs_header_level(info->chunk_root->node)); 1953 1954 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1955 btrfs_set_backup_extent_root_gen(root_backup, 1956 btrfs_header_generation(info->extent_root->node)); 1957 btrfs_set_backup_extent_root_level(root_backup, 1958 btrfs_header_level(info->extent_root->node)); 1959 1960 /* 1961 * we might commit during log recovery, which happens before we set 1962 * the fs_root. Make sure it is valid before we fill it in. 1963 */ 1964 if (info->fs_root && info->fs_root->node) { 1965 btrfs_set_backup_fs_root(root_backup, 1966 info->fs_root->node->start); 1967 btrfs_set_backup_fs_root_gen(root_backup, 1968 btrfs_header_generation(info->fs_root->node)); 1969 btrfs_set_backup_fs_root_level(root_backup, 1970 btrfs_header_level(info->fs_root->node)); 1971 } 1972 1973 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1974 btrfs_set_backup_dev_root_gen(root_backup, 1975 btrfs_header_generation(info->dev_root->node)); 1976 btrfs_set_backup_dev_root_level(root_backup, 1977 btrfs_header_level(info->dev_root->node)); 1978 1979 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1980 btrfs_set_backup_csum_root_gen(root_backup, 1981 btrfs_header_generation(info->csum_root->node)); 1982 btrfs_set_backup_csum_root_level(root_backup, 1983 btrfs_header_level(info->csum_root->node)); 1984 1985 btrfs_set_backup_total_bytes(root_backup, 1986 btrfs_super_total_bytes(info->super_copy)); 1987 btrfs_set_backup_bytes_used(root_backup, 1988 btrfs_super_bytes_used(info->super_copy)); 1989 btrfs_set_backup_num_devices(root_backup, 1990 btrfs_super_num_devices(info->super_copy)); 1991 1992 /* 1993 * if we don't copy this out to the super_copy, it won't get remembered 1994 * for the next commit 1995 */ 1996 memcpy(&info->super_copy->super_roots, 1997 &info->super_for_commit->super_roots, 1998 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 1999 } 2000 2001 /* 2002 * this copies info out of the root backup array and back into 2003 * the in-memory super block. It is meant to help iterate through 2004 * the array, so you send it the number of backups you've already 2005 * tried and the last backup index you used. 2006 * 2007 * this returns -1 when it has tried all the backups 2008 */ 2009 static noinline int next_root_backup(struct btrfs_fs_info *info, 2010 struct btrfs_super_block *super, 2011 int *num_backups_tried, int *backup_index) 2012 { 2013 struct btrfs_root_backup *root_backup; 2014 int newest = *backup_index; 2015 2016 if (*num_backups_tried == 0) { 2017 u64 gen = btrfs_super_generation(super); 2018 2019 newest = find_newest_super_backup(info, gen); 2020 if (newest == -1) 2021 return -1; 2022 2023 *backup_index = newest; 2024 *num_backups_tried = 1; 2025 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 2026 /* we've tried all the backups, all done */ 2027 return -1; 2028 } else { 2029 /* jump to the next oldest backup */ 2030 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2031 BTRFS_NUM_BACKUP_ROOTS; 2032 *backup_index = newest; 2033 *num_backups_tried += 1; 2034 } 2035 root_backup = super->super_roots + newest; 2036 2037 btrfs_set_super_generation(super, 2038 btrfs_backup_tree_root_gen(root_backup)); 2039 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2040 btrfs_set_super_root_level(super, 2041 btrfs_backup_tree_root_level(root_backup)); 2042 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2043 2044 /* 2045 * fixme: the total bytes and num_devices need to match or we should 2046 * need a fsck 2047 */ 2048 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2049 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2050 return 0; 2051 } 2052 2053 /* helper to cleanup workers */ 2054 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 2055 { 2056 btrfs_destroy_workqueue(fs_info->fixup_workers); 2057 btrfs_destroy_workqueue(fs_info->delalloc_workers); 2058 btrfs_destroy_workqueue(fs_info->workers); 2059 btrfs_destroy_workqueue(fs_info->endio_workers); 2060 btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2061 btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2062 btrfs_destroy_workqueue(fs_info->rmw_workers); 2063 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2064 btrfs_destroy_workqueue(fs_info->endio_write_workers); 2065 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2066 btrfs_destroy_workqueue(fs_info->submit_workers); 2067 btrfs_destroy_workqueue(fs_info->delayed_workers); 2068 btrfs_destroy_workqueue(fs_info->caching_workers); 2069 btrfs_destroy_workqueue(fs_info->readahead_workers); 2070 btrfs_destroy_workqueue(fs_info->flush_workers); 2071 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2072 btrfs_destroy_workqueue(fs_info->extent_workers); 2073 } 2074 2075 static void free_root_extent_buffers(struct btrfs_root *root) 2076 { 2077 if (root) { 2078 free_extent_buffer(root->node); 2079 free_extent_buffer(root->commit_root); 2080 root->node = NULL; 2081 root->commit_root = NULL; 2082 } 2083 } 2084 2085 /* helper to cleanup tree roots */ 2086 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2087 { 2088 free_root_extent_buffers(info->tree_root); 2089 2090 free_root_extent_buffers(info->dev_root); 2091 free_root_extent_buffers(info->extent_root); 2092 free_root_extent_buffers(info->csum_root); 2093 free_root_extent_buffers(info->quota_root); 2094 free_root_extent_buffers(info->uuid_root); 2095 if (chunk_root) 2096 free_root_extent_buffers(info->chunk_root); 2097 } 2098 2099 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2100 { 2101 int ret; 2102 struct btrfs_root *gang[8]; 2103 int i; 2104 2105 while (!list_empty(&fs_info->dead_roots)) { 2106 gang[0] = list_entry(fs_info->dead_roots.next, 2107 struct btrfs_root, root_list); 2108 list_del(&gang[0]->root_list); 2109 2110 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2111 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2112 } else { 2113 free_extent_buffer(gang[0]->node); 2114 free_extent_buffer(gang[0]->commit_root); 2115 btrfs_put_fs_root(gang[0]); 2116 } 2117 } 2118 2119 while (1) { 2120 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2121 (void **)gang, 0, 2122 ARRAY_SIZE(gang)); 2123 if (!ret) 2124 break; 2125 for (i = 0; i < ret; i++) 2126 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2127 } 2128 2129 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 2130 btrfs_free_log_root_tree(NULL, fs_info); 2131 btrfs_destroy_pinned_extent(fs_info->tree_root, 2132 fs_info->pinned_extents); 2133 } 2134 } 2135 2136 int open_ctree(struct super_block *sb, 2137 struct btrfs_fs_devices *fs_devices, 2138 char *options) 2139 { 2140 u32 sectorsize; 2141 u32 nodesize; 2142 u32 leafsize; 2143 u32 blocksize; 2144 u32 stripesize; 2145 u64 generation; 2146 u64 features; 2147 struct btrfs_key location; 2148 struct buffer_head *bh; 2149 struct btrfs_super_block *disk_super; 2150 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2151 struct btrfs_root *tree_root; 2152 struct btrfs_root *extent_root; 2153 struct btrfs_root *csum_root; 2154 struct btrfs_root *chunk_root; 2155 struct btrfs_root *dev_root; 2156 struct btrfs_root *quota_root; 2157 struct btrfs_root *uuid_root; 2158 struct btrfs_root *log_tree_root; 2159 int ret; 2160 int err = -EINVAL; 2161 int num_backups_tried = 0; 2162 int backup_index = 0; 2163 int max_active; 2164 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 2165 bool create_uuid_tree; 2166 bool check_uuid_tree; 2167 2168 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 2169 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2170 if (!tree_root || !chunk_root) { 2171 err = -ENOMEM; 2172 goto fail; 2173 } 2174 2175 ret = init_srcu_struct(&fs_info->subvol_srcu); 2176 if (ret) { 2177 err = ret; 2178 goto fail; 2179 } 2180 2181 ret = setup_bdi(fs_info, &fs_info->bdi); 2182 if (ret) { 2183 err = ret; 2184 goto fail_srcu; 2185 } 2186 2187 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0); 2188 if (ret) { 2189 err = ret; 2190 goto fail_bdi; 2191 } 2192 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2193 (1 + ilog2(nr_cpu_ids)); 2194 2195 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0); 2196 if (ret) { 2197 err = ret; 2198 goto fail_dirty_metadata_bytes; 2199 } 2200 2201 ret = percpu_counter_init(&fs_info->bio_counter, 0); 2202 if (ret) { 2203 err = ret; 2204 goto fail_delalloc_bytes; 2205 } 2206 2207 fs_info->btree_inode = new_inode(sb); 2208 if (!fs_info->btree_inode) { 2209 err = -ENOMEM; 2210 goto fail_bio_counter; 2211 } 2212 2213 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2214 2215 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2216 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 2217 INIT_LIST_HEAD(&fs_info->trans_list); 2218 INIT_LIST_HEAD(&fs_info->dead_roots); 2219 INIT_LIST_HEAD(&fs_info->delayed_iputs); 2220 INIT_LIST_HEAD(&fs_info->delalloc_roots); 2221 INIT_LIST_HEAD(&fs_info->caching_block_groups); 2222 spin_lock_init(&fs_info->delalloc_root_lock); 2223 spin_lock_init(&fs_info->trans_lock); 2224 spin_lock_init(&fs_info->fs_roots_radix_lock); 2225 spin_lock_init(&fs_info->delayed_iput_lock); 2226 spin_lock_init(&fs_info->defrag_inodes_lock); 2227 spin_lock_init(&fs_info->free_chunk_lock); 2228 spin_lock_init(&fs_info->tree_mod_seq_lock); 2229 spin_lock_init(&fs_info->super_lock); 2230 spin_lock_init(&fs_info->qgroup_op_lock); 2231 spin_lock_init(&fs_info->buffer_lock); 2232 rwlock_init(&fs_info->tree_mod_log_lock); 2233 mutex_init(&fs_info->reloc_mutex); 2234 mutex_init(&fs_info->delalloc_root_mutex); 2235 seqlock_init(&fs_info->profiles_lock); 2236 2237 init_completion(&fs_info->kobj_unregister); 2238 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 2239 INIT_LIST_HEAD(&fs_info->space_info); 2240 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 2241 btrfs_mapping_init(&fs_info->mapping_tree); 2242 btrfs_init_block_rsv(&fs_info->global_block_rsv, 2243 BTRFS_BLOCK_RSV_GLOBAL); 2244 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 2245 BTRFS_BLOCK_RSV_DELALLOC); 2246 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 2247 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 2248 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 2249 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 2250 BTRFS_BLOCK_RSV_DELOPS); 2251 atomic_set(&fs_info->nr_async_submits, 0); 2252 atomic_set(&fs_info->async_delalloc_pages, 0); 2253 atomic_set(&fs_info->async_submit_draining, 0); 2254 atomic_set(&fs_info->nr_async_bios, 0); 2255 atomic_set(&fs_info->defrag_running, 0); 2256 atomic_set(&fs_info->qgroup_op_seq, 0); 2257 atomic64_set(&fs_info->tree_mod_seq, 0); 2258 fs_info->sb = sb; 2259 fs_info->max_inline = 8192 * 1024; 2260 fs_info->metadata_ratio = 0; 2261 fs_info->defrag_inodes = RB_ROOT; 2262 fs_info->free_chunk_space = 0; 2263 fs_info->tree_mod_log = RB_ROOT; 2264 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2265 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64); 2266 /* readahead state */ 2267 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 2268 spin_lock_init(&fs_info->reada_lock); 2269 2270 fs_info->thread_pool_size = min_t(unsigned long, 2271 num_online_cpus() + 2, 8); 2272 2273 INIT_LIST_HEAD(&fs_info->ordered_roots); 2274 spin_lock_init(&fs_info->ordered_root_lock); 2275 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2276 GFP_NOFS); 2277 if (!fs_info->delayed_root) { 2278 err = -ENOMEM; 2279 goto fail_iput; 2280 } 2281 btrfs_init_delayed_root(fs_info->delayed_root); 2282 2283 mutex_init(&fs_info->scrub_lock); 2284 atomic_set(&fs_info->scrubs_running, 0); 2285 atomic_set(&fs_info->scrub_pause_req, 0); 2286 atomic_set(&fs_info->scrubs_paused, 0); 2287 atomic_set(&fs_info->scrub_cancel_req, 0); 2288 init_waitqueue_head(&fs_info->replace_wait); 2289 init_waitqueue_head(&fs_info->scrub_pause_wait); 2290 fs_info->scrub_workers_refcnt = 0; 2291 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2292 fs_info->check_integrity_print_mask = 0; 2293 #endif 2294 2295 spin_lock_init(&fs_info->balance_lock); 2296 mutex_init(&fs_info->balance_mutex); 2297 atomic_set(&fs_info->balance_running, 0); 2298 atomic_set(&fs_info->balance_pause_req, 0); 2299 atomic_set(&fs_info->balance_cancel_req, 0); 2300 fs_info->balance_ctl = NULL; 2301 init_waitqueue_head(&fs_info->balance_wait_q); 2302 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2303 2304 sb->s_blocksize = 4096; 2305 sb->s_blocksize_bits = blksize_bits(4096); 2306 sb->s_bdi = &fs_info->bdi; 2307 2308 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2309 set_nlink(fs_info->btree_inode, 1); 2310 /* 2311 * we set the i_size on the btree inode to the max possible int. 2312 * the real end of the address space is determined by all of 2313 * the devices in the system 2314 */ 2315 fs_info->btree_inode->i_size = OFFSET_MAX; 2316 fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 2317 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 2318 2319 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2320 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2321 fs_info->btree_inode->i_mapping); 2322 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2323 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 2324 2325 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2326 2327 BTRFS_I(fs_info->btree_inode)->root = tree_root; 2328 memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 2329 sizeof(struct btrfs_key)); 2330 set_bit(BTRFS_INODE_DUMMY, 2331 &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2332 btrfs_insert_inode_hash(fs_info->btree_inode); 2333 2334 spin_lock_init(&fs_info->block_group_cache_lock); 2335 fs_info->block_group_cache_tree = RB_ROOT; 2336 fs_info->first_logical_byte = (u64)-1; 2337 2338 extent_io_tree_init(&fs_info->freed_extents[0], 2339 fs_info->btree_inode->i_mapping); 2340 extent_io_tree_init(&fs_info->freed_extents[1], 2341 fs_info->btree_inode->i_mapping); 2342 fs_info->pinned_extents = &fs_info->freed_extents[0]; 2343 fs_info->do_barriers = 1; 2344 2345 2346 mutex_init(&fs_info->ordered_operations_mutex); 2347 mutex_init(&fs_info->ordered_extent_flush_mutex); 2348 mutex_init(&fs_info->tree_log_mutex); 2349 mutex_init(&fs_info->chunk_mutex); 2350 mutex_init(&fs_info->transaction_kthread_mutex); 2351 mutex_init(&fs_info->cleaner_mutex); 2352 mutex_init(&fs_info->volume_mutex); 2353 init_rwsem(&fs_info->commit_root_sem); 2354 init_rwsem(&fs_info->cleanup_work_sem); 2355 init_rwsem(&fs_info->subvol_sem); 2356 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2357 fs_info->dev_replace.lock_owner = 0; 2358 atomic_set(&fs_info->dev_replace.nesting_level, 0); 2359 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2360 mutex_init(&fs_info->dev_replace.lock_management_lock); 2361 mutex_init(&fs_info->dev_replace.lock); 2362 2363 spin_lock_init(&fs_info->qgroup_lock); 2364 mutex_init(&fs_info->qgroup_ioctl_lock); 2365 fs_info->qgroup_tree = RB_ROOT; 2366 fs_info->qgroup_op_tree = RB_ROOT; 2367 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2368 fs_info->qgroup_seq = 1; 2369 fs_info->quota_enabled = 0; 2370 fs_info->pending_quota_state = 0; 2371 fs_info->qgroup_ulist = NULL; 2372 mutex_init(&fs_info->qgroup_rescan_lock); 2373 2374 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2375 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2376 2377 init_waitqueue_head(&fs_info->transaction_throttle); 2378 init_waitqueue_head(&fs_info->transaction_wait); 2379 init_waitqueue_head(&fs_info->transaction_blocked_wait); 2380 init_waitqueue_head(&fs_info->async_submit_wait); 2381 2382 ret = btrfs_alloc_stripe_hash_table(fs_info); 2383 if (ret) { 2384 err = ret; 2385 goto fail_alloc; 2386 } 2387 2388 __setup_root(4096, 4096, 4096, 4096, tree_root, 2389 fs_info, BTRFS_ROOT_TREE_OBJECTID); 2390 2391 invalidate_bdev(fs_devices->latest_bdev); 2392 2393 /* 2394 * Read super block and check the signature bytes only 2395 */ 2396 bh = btrfs_read_dev_super(fs_devices->latest_bdev); 2397 if (!bh) { 2398 err = -EINVAL; 2399 goto fail_alloc; 2400 } 2401 2402 /* 2403 * We want to check superblock checksum, the type is stored inside. 2404 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 2405 */ 2406 if (btrfs_check_super_csum(bh->b_data)) { 2407 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n"); 2408 err = -EINVAL; 2409 goto fail_alloc; 2410 } 2411 2412 /* 2413 * super_copy is zeroed at allocation time and we never touch the 2414 * following bytes up to INFO_SIZE, the checksum is calculated from 2415 * the whole block of INFO_SIZE 2416 */ 2417 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2418 memcpy(fs_info->super_for_commit, fs_info->super_copy, 2419 sizeof(*fs_info->super_for_commit)); 2420 brelse(bh); 2421 2422 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 2423 2424 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 2425 if (ret) { 2426 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n"); 2427 err = -EINVAL; 2428 goto fail_alloc; 2429 } 2430 2431 disk_super = fs_info->super_copy; 2432 if (!btrfs_super_root(disk_super)) 2433 goto fail_alloc; 2434 2435 /* check FS state, whether FS is broken. */ 2436 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 2437 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2438 2439 /* 2440 * run through our array of backup supers and setup 2441 * our ring pointer to the oldest one 2442 */ 2443 generation = btrfs_super_generation(disk_super); 2444 find_oldest_super_backup(fs_info, generation); 2445 2446 /* 2447 * In the long term, we'll store the compression type in the super 2448 * block, and it'll be used for per file compression control. 2449 */ 2450 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 2451 2452 ret = btrfs_parse_options(tree_root, options); 2453 if (ret) { 2454 err = ret; 2455 goto fail_alloc; 2456 } 2457 2458 features = btrfs_super_incompat_flags(disk_super) & 2459 ~BTRFS_FEATURE_INCOMPAT_SUPP; 2460 if (features) { 2461 printk(KERN_ERR "BTRFS: couldn't mount because of " 2462 "unsupported optional features (%Lx).\n", 2463 features); 2464 err = -EINVAL; 2465 goto fail_alloc; 2466 } 2467 2468 if (btrfs_super_leafsize(disk_super) != 2469 btrfs_super_nodesize(disk_super)) { 2470 printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2471 "blocksizes don't match. node %d leaf %d\n", 2472 btrfs_super_nodesize(disk_super), 2473 btrfs_super_leafsize(disk_super)); 2474 err = -EINVAL; 2475 goto fail_alloc; 2476 } 2477 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2478 printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2479 "blocksize (%d) was too large\n", 2480 btrfs_super_leafsize(disk_super)); 2481 err = -EINVAL; 2482 goto fail_alloc; 2483 } 2484 2485 features = btrfs_super_incompat_flags(disk_super); 2486 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 2487 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2488 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2489 2490 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 2491 printk(KERN_ERR "BTRFS: has skinny extents\n"); 2492 2493 /* 2494 * flag our filesystem as having big metadata blocks if 2495 * they are bigger than the page size 2496 */ 2497 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) { 2498 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2499 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n"); 2500 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2501 } 2502 2503 nodesize = btrfs_super_nodesize(disk_super); 2504 leafsize = btrfs_super_leafsize(disk_super); 2505 sectorsize = btrfs_super_sectorsize(disk_super); 2506 stripesize = btrfs_super_stripesize(disk_super); 2507 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids)); 2508 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2509 2510 /* 2511 * mixed block groups end up with duplicate but slightly offset 2512 * extent buffers for the same range. It leads to corruptions 2513 */ 2514 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2515 (sectorsize != leafsize)) { 2516 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes " 2517 "are not allowed for mixed block groups on %s\n", 2518 sb->s_id); 2519 goto fail_alloc; 2520 } 2521 2522 /* 2523 * Needn't use the lock because there is no other task which will 2524 * update the flag. 2525 */ 2526 btrfs_set_super_incompat_flags(disk_super, features); 2527 2528 features = btrfs_super_compat_ro_flags(disk_super) & 2529 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2530 if (!(sb->s_flags & MS_RDONLY) && features) { 2531 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2532 "unsupported option features (%Lx).\n", 2533 features); 2534 err = -EINVAL; 2535 goto fail_alloc; 2536 } 2537 2538 max_active = fs_info->thread_pool_size; 2539 2540 fs_info->workers = 2541 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI, 2542 max_active, 16); 2543 2544 fs_info->delalloc_workers = 2545 btrfs_alloc_workqueue("delalloc", flags, max_active, 2); 2546 2547 fs_info->flush_workers = 2548 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0); 2549 2550 fs_info->caching_workers = 2551 btrfs_alloc_workqueue("cache", flags, max_active, 0); 2552 2553 /* 2554 * a higher idle thresh on the submit workers makes it much more 2555 * likely that bios will be send down in a sane order to the 2556 * devices 2557 */ 2558 fs_info->submit_workers = 2559 btrfs_alloc_workqueue("submit", flags, 2560 min_t(u64, fs_devices->num_devices, 2561 max_active), 64); 2562 2563 fs_info->fixup_workers = 2564 btrfs_alloc_workqueue("fixup", flags, 1, 0); 2565 2566 /* 2567 * endios are largely parallel and should have a very 2568 * low idle thresh 2569 */ 2570 fs_info->endio_workers = 2571 btrfs_alloc_workqueue("endio", flags, max_active, 4); 2572 fs_info->endio_meta_workers = 2573 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4); 2574 fs_info->endio_meta_write_workers = 2575 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2); 2576 fs_info->endio_raid56_workers = 2577 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4); 2578 fs_info->rmw_workers = 2579 btrfs_alloc_workqueue("rmw", flags, max_active, 2); 2580 fs_info->endio_write_workers = 2581 btrfs_alloc_workqueue("endio-write", flags, max_active, 2); 2582 fs_info->endio_freespace_worker = 2583 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0); 2584 fs_info->delayed_workers = 2585 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0); 2586 fs_info->readahead_workers = 2587 btrfs_alloc_workqueue("readahead", flags, max_active, 2); 2588 fs_info->qgroup_rescan_workers = 2589 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0); 2590 fs_info->extent_workers = 2591 btrfs_alloc_workqueue("extent-refs", flags, 2592 min_t(u64, fs_devices->num_devices, 2593 max_active), 8); 2594 2595 if (!(fs_info->workers && fs_info->delalloc_workers && 2596 fs_info->submit_workers && fs_info->flush_workers && 2597 fs_info->endio_workers && fs_info->endio_meta_workers && 2598 fs_info->endio_meta_write_workers && 2599 fs_info->endio_write_workers && fs_info->endio_raid56_workers && 2600 fs_info->endio_freespace_worker && fs_info->rmw_workers && 2601 fs_info->caching_workers && fs_info->readahead_workers && 2602 fs_info->fixup_workers && fs_info->delayed_workers && 2603 fs_info->fixup_workers && fs_info->extent_workers && 2604 fs_info->qgroup_rescan_workers)) { 2605 err = -ENOMEM; 2606 goto fail_sb_buffer; 2607 } 2608 2609 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2610 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2611 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 2612 2613 tree_root->nodesize = nodesize; 2614 tree_root->leafsize = leafsize; 2615 tree_root->sectorsize = sectorsize; 2616 tree_root->stripesize = stripesize; 2617 2618 sb->s_blocksize = sectorsize; 2619 sb->s_blocksize_bits = blksize_bits(sectorsize); 2620 2621 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2622 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id); 2623 goto fail_sb_buffer; 2624 } 2625 2626 if (sectorsize != PAGE_SIZE) { 2627 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) " 2628 "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2629 goto fail_sb_buffer; 2630 } 2631 2632 mutex_lock(&fs_info->chunk_mutex); 2633 ret = btrfs_read_sys_array(tree_root); 2634 mutex_unlock(&fs_info->chunk_mutex); 2635 if (ret) { 2636 printk(KERN_WARNING "BTRFS: failed to read the system " 2637 "array on %s\n", sb->s_id); 2638 goto fail_sb_buffer; 2639 } 2640 2641 blocksize = btrfs_level_size(tree_root, 2642 btrfs_super_chunk_root_level(disk_super)); 2643 generation = btrfs_super_chunk_root_generation(disk_super); 2644 2645 __setup_root(nodesize, leafsize, sectorsize, stripesize, 2646 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 2647 2648 chunk_root->node = read_tree_block(chunk_root, 2649 btrfs_super_chunk_root(disk_super), 2650 blocksize, generation); 2651 if (!chunk_root->node || 2652 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 2653 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n", 2654 sb->s_id); 2655 goto fail_tree_roots; 2656 } 2657 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 2658 chunk_root->commit_root = btrfs_root_node(chunk_root); 2659 2660 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2661 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2662 2663 ret = btrfs_read_chunk_tree(chunk_root); 2664 if (ret) { 2665 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n", 2666 sb->s_id); 2667 goto fail_tree_roots; 2668 } 2669 2670 /* 2671 * keep the device that is marked to be the target device for the 2672 * dev_replace procedure 2673 */ 2674 btrfs_close_extra_devices(fs_info, fs_devices, 0); 2675 2676 if (!fs_devices->latest_bdev) { 2677 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n", 2678 sb->s_id); 2679 goto fail_tree_roots; 2680 } 2681 2682 retry_root_backup: 2683 blocksize = btrfs_level_size(tree_root, 2684 btrfs_super_root_level(disk_super)); 2685 generation = btrfs_super_generation(disk_super); 2686 2687 tree_root->node = read_tree_block(tree_root, 2688 btrfs_super_root(disk_super), 2689 blocksize, generation); 2690 if (!tree_root->node || 2691 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 2692 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n", 2693 sb->s_id); 2694 2695 goto recovery_tree_root; 2696 } 2697 2698 btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2699 tree_root->commit_root = btrfs_root_node(tree_root); 2700 btrfs_set_root_refs(&tree_root->root_item, 1); 2701 2702 location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 2703 location.type = BTRFS_ROOT_ITEM_KEY; 2704 location.offset = 0; 2705 2706 extent_root = btrfs_read_tree_root(tree_root, &location); 2707 if (IS_ERR(extent_root)) { 2708 ret = PTR_ERR(extent_root); 2709 goto recovery_tree_root; 2710 } 2711 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state); 2712 fs_info->extent_root = extent_root; 2713 2714 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2715 dev_root = btrfs_read_tree_root(tree_root, &location); 2716 if (IS_ERR(dev_root)) { 2717 ret = PTR_ERR(dev_root); 2718 goto recovery_tree_root; 2719 } 2720 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state); 2721 fs_info->dev_root = dev_root; 2722 btrfs_init_devices_late(fs_info); 2723 2724 location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2725 csum_root = btrfs_read_tree_root(tree_root, &location); 2726 if (IS_ERR(csum_root)) { 2727 ret = PTR_ERR(csum_root); 2728 goto recovery_tree_root; 2729 } 2730 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state); 2731 fs_info->csum_root = csum_root; 2732 2733 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2734 quota_root = btrfs_read_tree_root(tree_root, &location); 2735 if (!IS_ERR(quota_root)) { 2736 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state); 2737 fs_info->quota_enabled = 1; 2738 fs_info->pending_quota_state = 1; 2739 fs_info->quota_root = quota_root; 2740 } 2741 2742 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2743 uuid_root = btrfs_read_tree_root(tree_root, &location); 2744 if (IS_ERR(uuid_root)) { 2745 ret = PTR_ERR(uuid_root); 2746 if (ret != -ENOENT) 2747 goto recovery_tree_root; 2748 create_uuid_tree = true; 2749 check_uuid_tree = false; 2750 } else { 2751 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state); 2752 fs_info->uuid_root = uuid_root; 2753 create_uuid_tree = false; 2754 check_uuid_tree = 2755 generation != btrfs_super_uuid_tree_generation(disk_super); 2756 } 2757 2758 fs_info->generation = generation; 2759 fs_info->last_trans_committed = generation; 2760 2761 ret = btrfs_recover_balance(fs_info); 2762 if (ret) { 2763 printk(KERN_WARNING "BTRFS: failed to recover balance\n"); 2764 goto fail_block_groups; 2765 } 2766 2767 ret = btrfs_init_dev_stats(fs_info); 2768 if (ret) { 2769 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n", 2770 ret); 2771 goto fail_block_groups; 2772 } 2773 2774 ret = btrfs_init_dev_replace(fs_info); 2775 if (ret) { 2776 pr_err("BTRFS: failed to init dev_replace: %d\n", ret); 2777 goto fail_block_groups; 2778 } 2779 2780 btrfs_close_extra_devices(fs_info, fs_devices, 1); 2781 2782 ret = btrfs_sysfs_add_one(fs_info); 2783 if (ret) { 2784 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret); 2785 goto fail_block_groups; 2786 } 2787 2788 ret = btrfs_init_space_info(fs_info); 2789 if (ret) { 2790 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret); 2791 goto fail_sysfs; 2792 } 2793 2794 ret = btrfs_read_block_groups(extent_root); 2795 if (ret) { 2796 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret); 2797 goto fail_sysfs; 2798 } 2799 fs_info->num_tolerated_disk_barrier_failures = 2800 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2801 if (fs_info->fs_devices->missing_devices > 2802 fs_info->num_tolerated_disk_barrier_failures && 2803 !(sb->s_flags & MS_RDONLY)) { 2804 printk(KERN_WARNING "BTRFS: " 2805 "too many missing devices, writeable mount is not allowed\n"); 2806 goto fail_sysfs; 2807 } 2808 2809 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2810 "btrfs-cleaner"); 2811 if (IS_ERR(fs_info->cleaner_kthread)) 2812 goto fail_sysfs; 2813 2814 fs_info->transaction_kthread = kthread_run(transaction_kthread, 2815 tree_root, 2816 "btrfs-transaction"); 2817 if (IS_ERR(fs_info->transaction_kthread)) 2818 goto fail_cleaner; 2819 2820 if (!btrfs_test_opt(tree_root, SSD) && 2821 !btrfs_test_opt(tree_root, NOSSD) && 2822 !fs_info->fs_devices->rotating) { 2823 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD " 2824 "mode\n"); 2825 btrfs_set_opt(fs_info->mount_opt, SSD); 2826 } 2827 2828 /* Set the real inode map cache flag */ 2829 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE)) 2830 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE); 2831 2832 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2833 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 2834 ret = btrfsic_mount(tree_root, fs_devices, 2835 btrfs_test_opt(tree_root, 2836 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 2837 1 : 0, 2838 fs_info->check_integrity_print_mask); 2839 if (ret) 2840 printk(KERN_WARNING "BTRFS: failed to initialize" 2841 " integrity check module %s\n", sb->s_id); 2842 } 2843 #endif 2844 ret = btrfs_read_qgroup_config(fs_info); 2845 if (ret) 2846 goto fail_trans_kthread; 2847 2848 /* do not make disk changes in broken FS */ 2849 if (btrfs_super_log_root(disk_super) != 0) { 2850 u64 bytenr = btrfs_super_log_root(disk_super); 2851 2852 if (fs_devices->rw_devices == 0) { 2853 printk(KERN_WARNING "BTRFS: log replay required " 2854 "on RO media\n"); 2855 err = -EIO; 2856 goto fail_qgroup; 2857 } 2858 blocksize = 2859 btrfs_level_size(tree_root, 2860 btrfs_super_log_root_level(disk_super)); 2861 2862 log_tree_root = btrfs_alloc_root(fs_info); 2863 if (!log_tree_root) { 2864 err = -ENOMEM; 2865 goto fail_qgroup; 2866 } 2867 2868 __setup_root(nodesize, leafsize, sectorsize, stripesize, 2869 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2870 2871 log_tree_root->node = read_tree_block(tree_root, bytenr, 2872 blocksize, 2873 generation + 1); 2874 if (!log_tree_root->node || 2875 !extent_buffer_uptodate(log_tree_root->node)) { 2876 printk(KERN_ERR "BTRFS: failed to read log tree\n"); 2877 free_extent_buffer(log_tree_root->node); 2878 kfree(log_tree_root); 2879 goto fail_qgroup; 2880 } 2881 /* returns with log_tree_root freed on success */ 2882 ret = btrfs_recover_log_trees(log_tree_root); 2883 if (ret) { 2884 btrfs_error(tree_root->fs_info, ret, 2885 "Failed to recover log tree"); 2886 free_extent_buffer(log_tree_root->node); 2887 kfree(log_tree_root); 2888 goto fail_qgroup; 2889 } 2890 2891 if (sb->s_flags & MS_RDONLY) { 2892 ret = btrfs_commit_super(tree_root); 2893 if (ret) 2894 goto fail_qgroup; 2895 } 2896 } 2897 2898 ret = btrfs_find_orphan_roots(tree_root); 2899 if (ret) 2900 goto fail_qgroup; 2901 2902 if (!(sb->s_flags & MS_RDONLY)) { 2903 ret = btrfs_cleanup_fs_roots(fs_info); 2904 if (ret) 2905 goto fail_qgroup; 2906 2907 ret = btrfs_recover_relocation(tree_root); 2908 if (ret < 0) { 2909 printk(KERN_WARNING 2910 "BTRFS: failed to recover relocation\n"); 2911 err = -EINVAL; 2912 goto fail_qgroup; 2913 } 2914 } 2915 2916 location.objectid = BTRFS_FS_TREE_OBJECTID; 2917 location.type = BTRFS_ROOT_ITEM_KEY; 2918 location.offset = 0; 2919 2920 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 2921 if (IS_ERR(fs_info->fs_root)) { 2922 err = PTR_ERR(fs_info->fs_root); 2923 goto fail_qgroup; 2924 } 2925 2926 if (sb->s_flags & MS_RDONLY) 2927 return 0; 2928 2929 down_read(&fs_info->cleanup_work_sem); 2930 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 2931 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 2932 up_read(&fs_info->cleanup_work_sem); 2933 close_ctree(tree_root); 2934 return ret; 2935 } 2936 up_read(&fs_info->cleanup_work_sem); 2937 2938 ret = btrfs_resume_balance_async(fs_info); 2939 if (ret) { 2940 printk(KERN_WARNING "BTRFS: failed to resume balance\n"); 2941 close_ctree(tree_root); 2942 return ret; 2943 } 2944 2945 ret = btrfs_resume_dev_replace_async(fs_info); 2946 if (ret) { 2947 pr_warn("BTRFS: failed to resume dev_replace\n"); 2948 close_ctree(tree_root); 2949 return ret; 2950 } 2951 2952 btrfs_qgroup_rescan_resume(fs_info); 2953 2954 if (create_uuid_tree) { 2955 pr_info("BTRFS: creating UUID tree\n"); 2956 ret = btrfs_create_uuid_tree(fs_info); 2957 if (ret) { 2958 pr_warn("BTRFS: failed to create the UUID tree %d\n", 2959 ret); 2960 close_ctree(tree_root); 2961 return ret; 2962 } 2963 } else if (check_uuid_tree || 2964 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) { 2965 pr_info("BTRFS: checking UUID tree\n"); 2966 ret = btrfs_check_uuid_tree(fs_info); 2967 if (ret) { 2968 pr_warn("BTRFS: failed to check the UUID tree %d\n", 2969 ret); 2970 close_ctree(tree_root); 2971 return ret; 2972 } 2973 } else { 2974 fs_info->update_uuid_tree_gen = 1; 2975 } 2976 2977 return 0; 2978 2979 fail_qgroup: 2980 btrfs_free_qgroup_config(fs_info); 2981 fail_trans_kthread: 2982 kthread_stop(fs_info->transaction_kthread); 2983 btrfs_cleanup_transaction(fs_info->tree_root); 2984 btrfs_free_fs_roots(fs_info); 2985 fail_cleaner: 2986 kthread_stop(fs_info->cleaner_kthread); 2987 2988 /* 2989 * make sure we're done with the btree inode before we stop our 2990 * kthreads 2991 */ 2992 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 2993 2994 fail_sysfs: 2995 btrfs_sysfs_remove_one(fs_info); 2996 2997 fail_block_groups: 2998 btrfs_put_block_group_cache(fs_info); 2999 btrfs_free_block_groups(fs_info); 3000 3001 fail_tree_roots: 3002 free_root_pointers(fs_info, 1); 3003 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3004 3005 fail_sb_buffer: 3006 btrfs_stop_all_workers(fs_info); 3007 fail_alloc: 3008 fail_iput: 3009 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3010 3011 iput(fs_info->btree_inode); 3012 fail_bio_counter: 3013 percpu_counter_destroy(&fs_info->bio_counter); 3014 fail_delalloc_bytes: 3015 percpu_counter_destroy(&fs_info->delalloc_bytes); 3016 fail_dirty_metadata_bytes: 3017 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3018 fail_bdi: 3019 bdi_destroy(&fs_info->bdi); 3020 fail_srcu: 3021 cleanup_srcu_struct(&fs_info->subvol_srcu); 3022 fail: 3023 btrfs_free_stripe_hash_table(fs_info); 3024 btrfs_close_devices(fs_info->fs_devices); 3025 return err; 3026 3027 recovery_tree_root: 3028 if (!btrfs_test_opt(tree_root, RECOVERY)) 3029 goto fail_tree_roots; 3030 3031 free_root_pointers(fs_info, 0); 3032 3033 /* don't use the log in recovery mode, it won't be valid */ 3034 btrfs_set_super_log_root(disk_super, 0); 3035 3036 /* we can't trust the free space cache either */ 3037 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3038 3039 ret = next_root_backup(fs_info, fs_info->super_copy, 3040 &num_backups_tried, &backup_index); 3041 if (ret == -1) 3042 goto fail_block_groups; 3043 goto retry_root_backup; 3044 } 3045 3046 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3047 { 3048 if (uptodate) { 3049 set_buffer_uptodate(bh); 3050 } else { 3051 struct btrfs_device *device = (struct btrfs_device *) 3052 bh->b_private; 3053 3054 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to " 3055 "I/O error on %s\n", 3056 rcu_str_deref(device->name)); 3057 /* note, we dont' set_buffer_write_io_error because we have 3058 * our own ways of dealing with the IO errors 3059 */ 3060 clear_buffer_uptodate(bh); 3061 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3062 } 3063 unlock_buffer(bh); 3064 put_bh(bh); 3065 } 3066 3067 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3068 { 3069 struct buffer_head *bh; 3070 struct buffer_head *latest = NULL; 3071 struct btrfs_super_block *super; 3072 int i; 3073 u64 transid = 0; 3074 u64 bytenr; 3075 3076 /* we would like to check all the supers, but that would make 3077 * a btrfs mount succeed after a mkfs from a different FS. 3078 * So, we need to add a special mount option to scan for 3079 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3080 */ 3081 for (i = 0; i < 1; i++) { 3082 bytenr = btrfs_sb_offset(i); 3083 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3084 i_size_read(bdev->bd_inode)) 3085 break; 3086 bh = __bread(bdev, bytenr / 4096, 3087 BTRFS_SUPER_INFO_SIZE); 3088 if (!bh) 3089 continue; 3090 3091 super = (struct btrfs_super_block *)bh->b_data; 3092 if (btrfs_super_bytenr(super) != bytenr || 3093 btrfs_super_magic(super) != BTRFS_MAGIC) { 3094 brelse(bh); 3095 continue; 3096 } 3097 3098 if (!latest || btrfs_super_generation(super) > transid) { 3099 brelse(latest); 3100 latest = bh; 3101 transid = btrfs_super_generation(super); 3102 } else { 3103 brelse(bh); 3104 } 3105 } 3106 return latest; 3107 } 3108 3109 /* 3110 * this should be called twice, once with wait == 0 and 3111 * once with wait == 1. When wait == 0 is done, all the buffer heads 3112 * we write are pinned. 3113 * 3114 * They are released when wait == 1 is done. 3115 * max_mirrors must be the same for both runs, and it indicates how 3116 * many supers on this one device should be written. 3117 * 3118 * max_mirrors == 0 means to write them all. 3119 */ 3120 static int write_dev_supers(struct btrfs_device *device, 3121 struct btrfs_super_block *sb, 3122 int do_barriers, int wait, int max_mirrors) 3123 { 3124 struct buffer_head *bh; 3125 int i; 3126 int ret; 3127 int errors = 0; 3128 u32 crc; 3129 u64 bytenr; 3130 3131 if (max_mirrors == 0) 3132 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3133 3134 for (i = 0; i < max_mirrors; i++) { 3135 bytenr = btrfs_sb_offset(i); 3136 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 3137 break; 3138 3139 if (wait) { 3140 bh = __find_get_block(device->bdev, bytenr / 4096, 3141 BTRFS_SUPER_INFO_SIZE); 3142 if (!bh) { 3143 errors++; 3144 continue; 3145 } 3146 wait_on_buffer(bh); 3147 if (!buffer_uptodate(bh)) 3148 errors++; 3149 3150 /* drop our reference */ 3151 brelse(bh); 3152 3153 /* drop the reference from the wait == 0 run */ 3154 brelse(bh); 3155 continue; 3156 } else { 3157 btrfs_set_super_bytenr(sb, bytenr); 3158 3159 crc = ~(u32)0; 3160 crc = btrfs_csum_data((char *)sb + 3161 BTRFS_CSUM_SIZE, crc, 3162 BTRFS_SUPER_INFO_SIZE - 3163 BTRFS_CSUM_SIZE); 3164 btrfs_csum_final(crc, sb->csum); 3165 3166 /* 3167 * one reference for us, and we leave it for the 3168 * caller 3169 */ 3170 bh = __getblk(device->bdev, bytenr / 4096, 3171 BTRFS_SUPER_INFO_SIZE); 3172 if (!bh) { 3173 printk(KERN_ERR "BTRFS: couldn't get super " 3174 "buffer head for bytenr %Lu\n", bytenr); 3175 errors++; 3176 continue; 3177 } 3178 3179 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3180 3181 /* one reference for submit_bh */ 3182 get_bh(bh); 3183 3184 set_buffer_uptodate(bh); 3185 lock_buffer(bh); 3186 bh->b_end_io = btrfs_end_buffer_write_sync; 3187 bh->b_private = device; 3188 } 3189 3190 /* 3191 * we fua the first super. The others we allow 3192 * to go down lazy. 3193 */ 3194 if (i == 0) 3195 ret = btrfsic_submit_bh(WRITE_FUA, bh); 3196 else 3197 ret = btrfsic_submit_bh(WRITE_SYNC, bh); 3198 if (ret) 3199 errors++; 3200 } 3201 return errors < i ? 0 : -1; 3202 } 3203 3204 /* 3205 * endio for the write_dev_flush, this will wake anyone waiting 3206 * for the barrier when it is done 3207 */ 3208 static void btrfs_end_empty_barrier(struct bio *bio, int err) 3209 { 3210 if (err) { 3211 if (err == -EOPNOTSUPP) 3212 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 3213 clear_bit(BIO_UPTODATE, &bio->bi_flags); 3214 } 3215 if (bio->bi_private) 3216 complete(bio->bi_private); 3217 bio_put(bio); 3218 } 3219 3220 /* 3221 * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3222 * sent down. With wait == 1, it waits for the previous flush. 3223 * 3224 * any device where the flush fails with eopnotsupp are flagged as not-barrier 3225 * capable 3226 */ 3227 static int write_dev_flush(struct btrfs_device *device, int wait) 3228 { 3229 struct bio *bio; 3230 int ret = 0; 3231 3232 if (device->nobarriers) 3233 return 0; 3234 3235 if (wait) { 3236 bio = device->flush_bio; 3237 if (!bio) 3238 return 0; 3239 3240 wait_for_completion(&device->flush_wait); 3241 3242 if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 3243 printk_in_rcu("BTRFS: disabling barriers on dev %s\n", 3244 rcu_str_deref(device->name)); 3245 device->nobarriers = 1; 3246 } else if (!bio_flagged(bio, BIO_UPTODATE)) { 3247 ret = -EIO; 3248 btrfs_dev_stat_inc_and_print(device, 3249 BTRFS_DEV_STAT_FLUSH_ERRS); 3250 } 3251 3252 /* drop the reference from the wait == 0 run */ 3253 bio_put(bio); 3254 device->flush_bio = NULL; 3255 3256 return ret; 3257 } 3258 3259 /* 3260 * one reference for us, and we leave it for the 3261 * caller 3262 */ 3263 device->flush_bio = NULL; 3264 bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3265 if (!bio) 3266 return -ENOMEM; 3267 3268 bio->bi_end_io = btrfs_end_empty_barrier; 3269 bio->bi_bdev = device->bdev; 3270 init_completion(&device->flush_wait); 3271 bio->bi_private = &device->flush_wait; 3272 device->flush_bio = bio; 3273 3274 bio_get(bio); 3275 btrfsic_submit_bio(WRITE_FLUSH, bio); 3276 3277 return 0; 3278 } 3279 3280 /* 3281 * send an empty flush down to each device in parallel, 3282 * then wait for them 3283 */ 3284 static int barrier_all_devices(struct btrfs_fs_info *info) 3285 { 3286 struct list_head *head; 3287 struct btrfs_device *dev; 3288 int errors_send = 0; 3289 int errors_wait = 0; 3290 int ret; 3291 3292 /* send down all the barriers */ 3293 head = &info->fs_devices->devices; 3294 list_for_each_entry_rcu(dev, head, dev_list) { 3295 if (dev->missing) 3296 continue; 3297 if (!dev->bdev) { 3298 errors_send++; 3299 continue; 3300 } 3301 if (!dev->in_fs_metadata || !dev->writeable) 3302 continue; 3303 3304 ret = write_dev_flush(dev, 0); 3305 if (ret) 3306 errors_send++; 3307 } 3308 3309 /* wait for all the barriers */ 3310 list_for_each_entry_rcu(dev, head, dev_list) { 3311 if (dev->missing) 3312 continue; 3313 if (!dev->bdev) { 3314 errors_wait++; 3315 continue; 3316 } 3317 if (!dev->in_fs_metadata || !dev->writeable) 3318 continue; 3319 3320 ret = write_dev_flush(dev, 1); 3321 if (ret) 3322 errors_wait++; 3323 } 3324 if (errors_send > info->num_tolerated_disk_barrier_failures || 3325 errors_wait > info->num_tolerated_disk_barrier_failures) 3326 return -EIO; 3327 return 0; 3328 } 3329 3330 int btrfs_calc_num_tolerated_disk_barrier_failures( 3331 struct btrfs_fs_info *fs_info) 3332 { 3333 struct btrfs_ioctl_space_info space; 3334 struct btrfs_space_info *sinfo; 3335 u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 3336 BTRFS_BLOCK_GROUP_SYSTEM, 3337 BTRFS_BLOCK_GROUP_METADATA, 3338 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 3339 int num_types = 4; 3340 int i; 3341 int c; 3342 int num_tolerated_disk_barrier_failures = 3343 (int)fs_info->fs_devices->num_devices; 3344 3345 for (i = 0; i < num_types; i++) { 3346 struct btrfs_space_info *tmp; 3347 3348 sinfo = NULL; 3349 rcu_read_lock(); 3350 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 3351 if (tmp->flags == types[i]) { 3352 sinfo = tmp; 3353 break; 3354 } 3355 } 3356 rcu_read_unlock(); 3357 3358 if (!sinfo) 3359 continue; 3360 3361 down_read(&sinfo->groups_sem); 3362 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3363 if (!list_empty(&sinfo->block_groups[c])) { 3364 u64 flags; 3365 3366 btrfs_get_block_group_info( 3367 &sinfo->block_groups[c], &space); 3368 if (space.total_bytes == 0 || 3369 space.used_bytes == 0) 3370 continue; 3371 flags = space.flags; 3372 /* 3373 * return 3374 * 0: if dup, single or RAID0 is configured for 3375 * any of metadata, system or data, else 3376 * 1: if RAID5 is configured, or if RAID1 or 3377 * RAID10 is configured and only two mirrors 3378 * are used, else 3379 * 2: if RAID6 is configured, else 3380 * num_mirrors - 1: if RAID1 or RAID10 is 3381 * configured and more than 3382 * 2 mirrors are used. 3383 */ 3384 if (num_tolerated_disk_barrier_failures > 0 && 3385 ((flags & (BTRFS_BLOCK_GROUP_DUP | 3386 BTRFS_BLOCK_GROUP_RAID0)) || 3387 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) 3388 == 0))) 3389 num_tolerated_disk_barrier_failures = 0; 3390 else if (num_tolerated_disk_barrier_failures > 1) { 3391 if (flags & (BTRFS_BLOCK_GROUP_RAID1 | 3392 BTRFS_BLOCK_GROUP_RAID5 | 3393 BTRFS_BLOCK_GROUP_RAID10)) { 3394 num_tolerated_disk_barrier_failures = 1; 3395 } else if (flags & 3396 BTRFS_BLOCK_GROUP_RAID6) { 3397 num_tolerated_disk_barrier_failures = 2; 3398 } 3399 } 3400 } 3401 } 3402 up_read(&sinfo->groups_sem); 3403 } 3404 3405 return num_tolerated_disk_barrier_failures; 3406 } 3407 3408 static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3409 { 3410 struct list_head *head; 3411 struct btrfs_device *dev; 3412 struct btrfs_super_block *sb; 3413 struct btrfs_dev_item *dev_item; 3414 int ret; 3415 int do_barriers; 3416 int max_errors; 3417 int total_errors = 0; 3418 u64 flags; 3419 3420 do_barriers = !btrfs_test_opt(root, NOBARRIER); 3421 backup_super_roots(root->fs_info); 3422 3423 sb = root->fs_info->super_for_commit; 3424 dev_item = &sb->dev_item; 3425 3426 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3427 head = &root->fs_info->fs_devices->devices; 3428 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3429 3430 if (do_barriers) { 3431 ret = barrier_all_devices(root->fs_info); 3432 if (ret) { 3433 mutex_unlock( 3434 &root->fs_info->fs_devices->device_list_mutex); 3435 btrfs_error(root->fs_info, ret, 3436 "errors while submitting device barriers."); 3437 return ret; 3438 } 3439 } 3440 3441 list_for_each_entry_rcu(dev, head, dev_list) { 3442 if (!dev->bdev) { 3443 total_errors++; 3444 continue; 3445 } 3446 if (!dev->in_fs_metadata || !dev->writeable) 3447 continue; 3448 3449 btrfs_set_stack_device_generation(dev_item, 0); 3450 btrfs_set_stack_device_type(dev_item, dev->type); 3451 btrfs_set_stack_device_id(dev_item, dev->devid); 3452 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 3453 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 3454 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3455 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3456 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3457 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 3458 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3459 3460 flags = btrfs_super_flags(sb); 3461 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3462 3463 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3464 if (ret) 3465 total_errors++; 3466 } 3467 if (total_errors > max_errors) { 3468 btrfs_err(root->fs_info, "%d errors while writing supers", 3469 total_errors); 3470 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3471 3472 /* FUA is masked off if unsupported and can't be the reason */ 3473 btrfs_error(root->fs_info, -EIO, 3474 "%d errors while writing supers", total_errors); 3475 return -EIO; 3476 } 3477 3478 total_errors = 0; 3479 list_for_each_entry_rcu(dev, head, dev_list) { 3480 if (!dev->bdev) 3481 continue; 3482 if (!dev->in_fs_metadata || !dev->writeable) 3483 continue; 3484 3485 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3486 if (ret) 3487 total_errors++; 3488 } 3489 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3490 if (total_errors > max_errors) { 3491 btrfs_error(root->fs_info, -EIO, 3492 "%d errors while writing supers", total_errors); 3493 return -EIO; 3494 } 3495 return 0; 3496 } 3497 3498 int write_ctree_super(struct btrfs_trans_handle *trans, 3499 struct btrfs_root *root, int max_mirrors) 3500 { 3501 return write_all_supers(root, max_mirrors); 3502 } 3503 3504 /* Drop a fs root from the radix tree and free it. */ 3505 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3506 struct btrfs_root *root) 3507 { 3508 spin_lock(&fs_info->fs_roots_radix_lock); 3509 radix_tree_delete(&fs_info->fs_roots_radix, 3510 (unsigned long)root->root_key.objectid); 3511 spin_unlock(&fs_info->fs_roots_radix_lock); 3512 3513 if (btrfs_root_refs(&root->root_item) == 0) 3514 synchronize_srcu(&fs_info->subvol_srcu); 3515 3516 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 3517 btrfs_free_log(NULL, root); 3518 3519 if (root->free_ino_pinned) 3520 __btrfs_remove_free_space_cache(root->free_ino_pinned); 3521 if (root->free_ino_ctl) 3522 __btrfs_remove_free_space_cache(root->free_ino_ctl); 3523 free_fs_root(root); 3524 } 3525 3526 static void free_fs_root(struct btrfs_root *root) 3527 { 3528 iput(root->cache_inode); 3529 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 3530 btrfs_free_block_rsv(root, root->orphan_block_rsv); 3531 root->orphan_block_rsv = NULL; 3532 if (root->anon_dev) 3533 free_anon_bdev(root->anon_dev); 3534 if (root->subv_writers) 3535 btrfs_free_subvolume_writers(root->subv_writers); 3536 free_extent_buffer(root->node); 3537 free_extent_buffer(root->commit_root); 3538 kfree(root->free_ino_ctl); 3539 kfree(root->free_ino_pinned); 3540 kfree(root->name); 3541 btrfs_put_fs_root(root); 3542 } 3543 3544 void btrfs_free_fs_root(struct btrfs_root *root) 3545 { 3546 free_fs_root(root); 3547 } 3548 3549 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3550 { 3551 u64 root_objectid = 0; 3552 struct btrfs_root *gang[8]; 3553 int i = 0; 3554 int err = 0; 3555 unsigned int ret = 0; 3556 int index; 3557 3558 while (1) { 3559 index = srcu_read_lock(&fs_info->subvol_srcu); 3560 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3561 (void **)gang, root_objectid, 3562 ARRAY_SIZE(gang)); 3563 if (!ret) { 3564 srcu_read_unlock(&fs_info->subvol_srcu, index); 3565 break; 3566 } 3567 root_objectid = gang[ret - 1]->root_key.objectid + 1; 3568 3569 for (i = 0; i < ret; i++) { 3570 /* Avoid to grab roots in dead_roots */ 3571 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 3572 gang[i] = NULL; 3573 continue; 3574 } 3575 /* grab all the search result for later use */ 3576 gang[i] = btrfs_grab_fs_root(gang[i]); 3577 } 3578 srcu_read_unlock(&fs_info->subvol_srcu, index); 3579 3580 for (i = 0; i < ret; i++) { 3581 if (!gang[i]) 3582 continue; 3583 root_objectid = gang[i]->root_key.objectid; 3584 err = btrfs_orphan_cleanup(gang[i]); 3585 if (err) 3586 break; 3587 btrfs_put_fs_root(gang[i]); 3588 } 3589 root_objectid++; 3590 } 3591 3592 /* release the uncleaned roots due to error */ 3593 for (; i < ret; i++) { 3594 if (gang[i]) 3595 btrfs_put_fs_root(gang[i]); 3596 } 3597 return err; 3598 } 3599 3600 int btrfs_commit_super(struct btrfs_root *root) 3601 { 3602 struct btrfs_trans_handle *trans; 3603 3604 mutex_lock(&root->fs_info->cleaner_mutex); 3605 btrfs_run_delayed_iputs(root); 3606 mutex_unlock(&root->fs_info->cleaner_mutex); 3607 wake_up_process(root->fs_info->cleaner_kthread); 3608 3609 /* wait until ongoing cleanup work done */ 3610 down_write(&root->fs_info->cleanup_work_sem); 3611 up_write(&root->fs_info->cleanup_work_sem); 3612 3613 trans = btrfs_join_transaction(root); 3614 if (IS_ERR(trans)) 3615 return PTR_ERR(trans); 3616 return btrfs_commit_transaction(trans, root); 3617 } 3618 3619 int close_ctree(struct btrfs_root *root) 3620 { 3621 struct btrfs_fs_info *fs_info = root->fs_info; 3622 int ret; 3623 3624 fs_info->closing = 1; 3625 smp_mb(); 3626 3627 /* wait for the uuid_scan task to finish */ 3628 down(&fs_info->uuid_tree_rescan_sem); 3629 /* avoid complains from lockdep et al., set sem back to initial state */ 3630 up(&fs_info->uuid_tree_rescan_sem); 3631 3632 /* pause restriper - we want to resume on mount */ 3633 btrfs_pause_balance(fs_info); 3634 3635 btrfs_dev_replace_suspend_for_unmount(fs_info); 3636 3637 btrfs_scrub_cancel(fs_info); 3638 3639 /* wait for any defraggers to finish */ 3640 wait_event(fs_info->transaction_wait, 3641 (atomic_read(&fs_info->defrag_running) == 0)); 3642 3643 /* clear out the rbtree of defraggable inodes */ 3644 btrfs_cleanup_defrag_inodes(fs_info); 3645 3646 cancel_work_sync(&fs_info->async_reclaim_work); 3647 3648 if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3649 ret = btrfs_commit_super(root); 3650 if (ret) 3651 btrfs_err(root->fs_info, "commit super ret %d", ret); 3652 } 3653 3654 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 3655 btrfs_error_commit_super(root); 3656 3657 kthread_stop(fs_info->transaction_kthread); 3658 kthread_stop(fs_info->cleaner_kthread); 3659 3660 fs_info->closing = 2; 3661 smp_mb(); 3662 3663 btrfs_free_qgroup_config(root->fs_info); 3664 3665 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 3666 btrfs_info(root->fs_info, "at unmount delalloc count %lld", 3667 percpu_counter_sum(&fs_info->delalloc_bytes)); 3668 } 3669 3670 btrfs_sysfs_remove_one(fs_info); 3671 3672 btrfs_free_fs_roots(fs_info); 3673 3674 btrfs_put_block_group_cache(fs_info); 3675 3676 btrfs_free_block_groups(fs_info); 3677 3678 /* 3679 * we must make sure there is not any read request to 3680 * submit after we stopping all workers. 3681 */ 3682 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3683 btrfs_stop_all_workers(fs_info); 3684 3685 free_root_pointers(fs_info, 1); 3686 3687 iput(fs_info->btree_inode); 3688 3689 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3690 if (btrfs_test_opt(root, CHECK_INTEGRITY)) 3691 btrfsic_unmount(root, fs_info->fs_devices); 3692 #endif 3693 3694 btrfs_close_devices(fs_info->fs_devices); 3695 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3696 3697 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3698 percpu_counter_destroy(&fs_info->delalloc_bytes); 3699 percpu_counter_destroy(&fs_info->bio_counter); 3700 bdi_destroy(&fs_info->bdi); 3701 cleanup_srcu_struct(&fs_info->subvol_srcu); 3702 3703 btrfs_free_stripe_hash_table(fs_info); 3704 3705 btrfs_free_block_rsv(root, root->orphan_block_rsv); 3706 root->orphan_block_rsv = NULL; 3707 3708 return 0; 3709 } 3710 3711 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3712 int atomic) 3713 { 3714 int ret; 3715 struct inode *btree_inode = buf->pages[0]->mapping->host; 3716 3717 ret = extent_buffer_uptodate(buf); 3718 if (!ret) 3719 return ret; 3720 3721 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3722 parent_transid, atomic); 3723 if (ret == -EAGAIN) 3724 return ret; 3725 return !ret; 3726 } 3727 3728 int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 3729 { 3730 return set_extent_buffer_uptodate(buf); 3731 } 3732 3733 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 3734 { 3735 struct btrfs_root *root; 3736 u64 transid = btrfs_header_generation(buf); 3737 int was_dirty; 3738 3739 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 3740 /* 3741 * This is a fast path so only do this check if we have sanity tests 3742 * enabled. Normal people shouldn't be marking dummy buffers as dirty 3743 * outside of the sanity tests. 3744 */ 3745 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 3746 return; 3747 #endif 3748 root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3749 btrfs_assert_tree_locked(buf); 3750 if (transid != root->fs_info->generation) 3751 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3752 "found %llu running %llu\n", 3753 buf->start, transid, root->fs_info->generation); 3754 was_dirty = set_extent_buffer_dirty(buf); 3755 if (!was_dirty) 3756 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3757 buf->len, 3758 root->fs_info->dirty_metadata_batch); 3759 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3760 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) { 3761 btrfs_print_leaf(root, buf); 3762 ASSERT(0); 3763 } 3764 #endif 3765 } 3766 3767 static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3768 int flush_delayed) 3769 { 3770 /* 3771 * looks as though older kernels can get into trouble with 3772 * this code, they end up stuck in balance_dirty_pages forever 3773 */ 3774 int ret; 3775 3776 if (current->flags & PF_MEMALLOC) 3777 return; 3778 3779 if (flush_delayed) 3780 btrfs_balance_delayed_items(root); 3781 3782 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3783 BTRFS_DIRTY_METADATA_THRESH); 3784 if (ret > 0) { 3785 balance_dirty_pages_ratelimited( 3786 root->fs_info->btree_inode->i_mapping); 3787 } 3788 return; 3789 } 3790 3791 void btrfs_btree_balance_dirty(struct btrfs_root *root) 3792 { 3793 __btrfs_btree_balance_dirty(root, 1); 3794 } 3795 3796 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3797 { 3798 __btrfs_btree_balance_dirty(root, 0); 3799 } 3800 3801 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 3802 { 3803 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3804 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 3805 } 3806 3807 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3808 int read_only) 3809 { 3810 /* 3811 * Placeholder for checks 3812 */ 3813 return 0; 3814 } 3815 3816 static void btrfs_error_commit_super(struct btrfs_root *root) 3817 { 3818 mutex_lock(&root->fs_info->cleaner_mutex); 3819 btrfs_run_delayed_iputs(root); 3820 mutex_unlock(&root->fs_info->cleaner_mutex); 3821 3822 down_write(&root->fs_info->cleanup_work_sem); 3823 up_write(&root->fs_info->cleanup_work_sem); 3824 3825 /* cleanup FS via transaction */ 3826 btrfs_cleanup_transaction(root); 3827 } 3828 3829 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t, 3830 struct btrfs_root *root) 3831 { 3832 struct btrfs_inode *btrfs_inode; 3833 struct list_head splice; 3834 3835 INIT_LIST_HEAD(&splice); 3836 3837 mutex_lock(&root->fs_info->ordered_operations_mutex); 3838 spin_lock(&root->fs_info->ordered_root_lock); 3839 3840 list_splice_init(&t->ordered_operations, &splice); 3841 while (!list_empty(&splice)) { 3842 btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3843 ordered_operations); 3844 3845 list_del_init(&btrfs_inode->ordered_operations); 3846 spin_unlock(&root->fs_info->ordered_root_lock); 3847 3848 btrfs_invalidate_inodes(btrfs_inode->root); 3849 3850 spin_lock(&root->fs_info->ordered_root_lock); 3851 } 3852 3853 spin_unlock(&root->fs_info->ordered_root_lock); 3854 mutex_unlock(&root->fs_info->ordered_operations_mutex); 3855 } 3856 3857 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 3858 { 3859 struct btrfs_ordered_extent *ordered; 3860 3861 spin_lock(&root->ordered_extent_lock); 3862 /* 3863 * This will just short circuit the ordered completion stuff which will 3864 * make sure the ordered extent gets properly cleaned up. 3865 */ 3866 list_for_each_entry(ordered, &root->ordered_extents, 3867 root_extent_list) 3868 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 3869 spin_unlock(&root->ordered_extent_lock); 3870 } 3871 3872 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 3873 { 3874 struct btrfs_root *root; 3875 struct list_head splice; 3876 3877 INIT_LIST_HEAD(&splice); 3878 3879 spin_lock(&fs_info->ordered_root_lock); 3880 list_splice_init(&fs_info->ordered_roots, &splice); 3881 while (!list_empty(&splice)) { 3882 root = list_first_entry(&splice, struct btrfs_root, 3883 ordered_root); 3884 list_move_tail(&root->ordered_root, 3885 &fs_info->ordered_roots); 3886 3887 spin_unlock(&fs_info->ordered_root_lock); 3888 btrfs_destroy_ordered_extents(root); 3889 3890 cond_resched(); 3891 spin_lock(&fs_info->ordered_root_lock); 3892 } 3893 spin_unlock(&fs_info->ordered_root_lock); 3894 } 3895 3896 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 3897 struct btrfs_root *root) 3898 { 3899 struct rb_node *node; 3900 struct btrfs_delayed_ref_root *delayed_refs; 3901 struct btrfs_delayed_ref_node *ref; 3902 int ret = 0; 3903 3904 delayed_refs = &trans->delayed_refs; 3905 3906 spin_lock(&delayed_refs->lock); 3907 if (atomic_read(&delayed_refs->num_entries) == 0) { 3908 spin_unlock(&delayed_refs->lock); 3909 btrfs_info(root->fs_info, "delayed_refs has NO entry"); 3910 return ret; 3911 } 3912 3913 while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 3914 struct btrfs_delayed_ref_head *head; 3915 bool pin_bytes = false; 3916 3917 head = rb_entry(node, struct btrfs_delayed_ref_head, 3918 href_node); 3919 if (!mutex_trylock(&head->mutex)) { 3920 atomic_inc(&head->node.refs); 3921 spin_unlock(&delayed_refs->lock); 3922 3923 mutex_lock(&head->mutex); 3924 mutex_unlock(&head->mutex); 3925 btrfs_put_delayed_ref(&head->node); 3926 spin_lock(&delayed_refs->lock); 3927 continue; 3928 } 3929 spin_lock(&head->lock); 3930 while ((node = rb_first(&head->ref_root)) != NULL) { 3931 ref = rb_entry(node, struct btrfs_delayed_ref_node, 3932 rb_node); 3933 ref->in_tree = 0; 3934 rb_erase(&ref->rb_node, &head->ref_root); 3935 atomic_dec(&delayed_refs->num_entries); 3936 btrfs_put_delayed_ref(ref); 3937 } 3938 if (head->must_insert_reserved) 3939 pin_bytes = true; 3940 btrfs_free_delayed_extent_op(head->extent_op); 3941 delayed_refs->num_heads--; 3942 if (head->processing == 0) 3943 delayed_refs->num_heads_ready--; 3944 atomic_dec(&delayed_refs->num_entries); 3945 head->node.in_tree = 0; 3946 rb_erase(&head->href_node, &delayed_refs->href_root); 3947 spin_unlock(&head->lock); 3948 spin_unlock(&delayed_refs->lock); 3949 mutex_unlock(&head->mutex); 3950 3951 if (pin_bytes) 3952 btrfs_pin_extent(root, head->node.bytenr, 3953 head->node.num_bytes, 1); 3954 btrfs_put_delayed_ref(&head->node); 3955 cond_resched(); 3956 spin_lock(&delayed_refs->lock); 3957 } 3958 3959 spin_unlock(&delayed_refs->lock); 3960 3961 return ret; 3962 } 3963 3964 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 3965 { 3966 struct btrfs_inode *btrfs_inode; 3967 struct list_head splice; 3968 3969 INIT_LIST_HEAD(&splice); 3970 3971 spin_lock(&root->delalloc_lock); 3972 list_splice_init(&root->delalloc_inodes, &splice); 3973 3974 while (!list_empty(&splice)) { 3975 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 3976 delalloc_inodes); 3977 3978 list_del_init(&btrfs_inode->delalloc_inodes); 3979 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 3980 &btrfs_inode->runtime_flags); 3981 spin_unlock(&root->delalloc_lock); 3982 3983 btrfs_invalidate_inodes(btrfs_inode->root); 3984 3985 spin_lock(&root->delalloc_lock); 3986 } 3987 3988 spin_unlock(&root->delalloc_lock); 3989 } 3990 3991 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 3992 { 3993 struct btrfs_root *root; 3994 struct list_head splice; 3995 3996 INIT_LIST_HEAD(&splice); 3997 3998 spin_lock(&fs_info->delalloc_root_lock); 3999 list_splice_init(&fs_info->delalloc_roots, &splice); 4000 while (!list_empty(&splice)) { 4001 root = list_first_entry(&splice, struct btrfs_root, 4002 delalloc_root); 4003 list_del_init(&root->delalloc_root); 4004 root = btrfs_grab_fs_root(root); 4005 BUG_ON(!root); 4006 spin_unlock(&fs_info->delalloc_root_lock); 4007 4008 btrfs_destroy_delalloc_inodes(root); 4009 btrfs_put_fs_root(root); 4010 4011 spin_lock(&fs_info->delalloc_root_lock); 4012 } 4013 spin_unlock(&fs_info->delalloc_root_lock); 4014 } 4015 4016 static int btrfs_destroy_marked_extents(struct btrfs_root *root, 4017 struct extent_io_tree *dirty_pages, 4018 int mark) 4019 { 4020 int ret; 4021 struct extent_buffer *eb; 4022 u64 start = 0; 4023 u64 end; 4024 4025 while (1) { 4026 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4027 mark, NULL); 4028 if (ret) 4029 break; 4030 4031 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 4032 while (start <= end) { 4033 eb = btrfs_find_tree_block(root, start, 4034 root->leafsize); 4035 start += root->leafsize; 4036 if (!eb) 4037 continue; 4038 wait_on_extent_buffer_writeback(eb); 4039 4040 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4041 &eb->bflags)) 4042 clear_extent_buffer_dirty(eb); 4043 free_extent_buffer_stale(eb); 4044 } 4045 } 4046 4047 return ret; 4048 } 4049 4050 static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4051 struct extent_io_tree *pinned_extents) 4052 { 4053 struct extent_io_tree *unpin; 4054 u64 start; 4055 u64 end; 4056 int ret; 4057 bool loop = true; 4058 4059 unpin = pinned_extents; 4060 again: 4061 while (1) { 4062 ret = find_first_extent_bit(unpin, 0, &start, &end, 4063 EXTENT_DIRTY, NULL); 4064 if (ret) 4065 break; 4066 4067 /* opt_discard */ 4068 if (btrfs_test_opt(root, DISCARD)) 4069 ret = btrfs_error_discard_extent(root, start, 4070 end + 1 - start, 4071 NULL); 4072 4073 clear_extent_dirty(unpin, start, end, GFP_NOFS); 4074 btrfs_error_unpin_extent_range(root, start, end); 4075 cond_resched(); 4076 } 4077 4078 if (loop) { 4079 if (unpin == &root->fs_info->freed_extents[0]) 4080 unpin = &root->fs_info->freed_extents[1]; 4081 else 4082 unpin = &root->fs_info->freed_extents[0]; 4083 loop = false; 4084 goto again; 4085 } 4086 4087 return 0; 4088 } 4089 4090 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 4091 struct btrfs_root *root) 4092 { 4093 btrfs_destroy_ordered_operations(cur_trans, root); 4094 4095 btrfs_destroy_delayed_refs(cur_trans, root); 4096 4097 cur_trans->state = TRANS_STATE_COMMIT_START; 4098 wake_up(&root->fs_info->transaction_blocked_wait); 4099 4100 cur_trans->state = TRANS_STATE_UNBLOCKED; 4101 wake_up(&root->fs_info->transaction_wait); 4102 4103 btrfs_destroy_delayed_inodes(root); 4104 btrfs_assert_delayed_root_empty(root); 4105 4106 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 4107 EXTENT_DIRTY); 4108 btrfs_destroy_pinned_extent(root, 4109 root->fs_info->pinned_extents); 4110 4111 cur_trans->state =TRANS_STATE_COMPLETED; 4112 wake_up(&cur_trans->commit_wait); 4113 4114 /* 4115 memset(cur_trans, 0, sizeof(*cur_trans)); 4116 kmem_cache_free(btrfs_transaction_cachep, cur_trans); 4117 */ 4118 } 4119 4120 static int btrfs_cleanup_transaction(struct btrfs_root *root) 4121 { 4122 struct btrfs_transaction *t; 4123 4124 mutex_lock(&root->fs_info->transaction_kthread_mutex); 4125 4126 spin_lock(&root->fs_info->trans_lock); 4127 while (!list_empty(&root->fs_info->trans_list)) { 4128 t = list_first_entry(&root->fs_info->trans_list, 4129 struct btrfs_transaction, list); 4130 if (t->state >= TRANS_STATE_COMMIT_START) { 4131 atomic_inc(&t->use_count); 4132 spin_unlock(&root->fs_info->trans_lock); 4133 btrfs_wait_for_commit(root, t->transid); 4134 btrfs_put_transaction(t); 4135 spin_lock(&root->fs_info->trans_lock); 4136 continue; 4137 } 4138 if (t == root->fs_info->running_transaction) { 4139 t->state = TRANS_STATE_COMMIT_DOING; 4140 spin_unlock(&root->fs_info->trans_lock); 4141 /* 4142 * We wait for 0 num_writers since we don't hold a trans 4143 * handle open currently for this transaction. 4144 */ 4145 wait_event(t->writer_wait, 4146 atomic_read(&t->num_writers) == 0); 4147 } else { 4148 spin_unlock(&root->fs_info->trans_lock); 4149 } 4150 btrfs_cleanup_one_transaction(t, root); 4151 4152 spin_lock(&root->fs_info->trans_lock); 4153 if (t == root->fs_info->running_transaction) 4154 root->fs_info->running_transaction = NULL; 4155 list_del_init(&t->list); 4156 spin_unlock(&root->fs_info->trans_lock); 4157 4158 btrfs_put_transaction(t); 4159 trace_btrfs_transaction_commit(root); 4160 spin_lock(&root->fs_info->trans_lock); 4161 } 4162 spin_unlock(&root->fs_info->trans_lock); 4163 btrfs_destroy_all_ordered_extents(root->fs_info); 4164 btrfs_destroy_delayed_inodes(root); 4165 btrfs_assert_delayed_root_empty(root); 4166 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4167 btrfs_destroy_all_delalloc_inodes(root->fs_info); 4168 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4169 4170 return 0; 4171 } 4172 4173 static struct extent_io_ops btree_extent_io_ops = { 4174 .readpage_end_io_hook = btree_readpage_end_io_hook, 4175 .readpage_io_failed_hook = btree_io_failed_hook, 4176 .submit_bio_hook = btree_submit_bio_hook, 4177 /* note we're sharing with inode.c for the merge bio hook */ 4178 .merge_bio_hook = btrfs_merge_bio_hook, 4179 }; 4180