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 if (need_lock) 373 btrfs_tree_read_unlock_blocking(eb); 374 return ret; 375 } 376 377 /* 378 * Return 0 if the superblock checksum type matches the checksum value of that 379 * algorithm. Pass the raw disk superblock data. 380 */ 381 static int btrfs_check_super_csum(char *raw_disk_sb) 382 { 383 struct btrfs_super_block *disk_sb = 384 (struct btrfs_super_block *)raw_disk_sb; 385 u16 csum_type = btrfs_super_csum_type(disk_sb); 386 int ret = 0; 387 388 if (csum_type == BTRFS_CSUM_TYPE_CRC32) { 389 u32 crc = ~(u32)0; 390 const int csum_size = sizeof(crc); 391 char result[csum_size]; 392 393 /* 394 * The super_block structure does not span the whole 395 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space 396 * is filled with zeros and is included in the checkum. 397 */ 398 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE, 399 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE); 400 btrfs_csum_final(crc, result); 401 402 if (memcmp(raw_disk_sb, result, csum_size)) 403 ret = 1; 404 405 if (ret && btrfs_super_generation(disk_sb) < 10) { 406 printk(KERN_WARNING 407 "BTRFS: super block crcs don't match, older mkfs detected\n"); 408 ret = 0; 409 } 410 } 411 412 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) { 413 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n", 414 csum_type); 415 ret = 1; 416 } 417 418 return ret; 419 } 420 421 /* 422 * helper to read a given tree block, doing retries as required when 423 * the checksums don't match and we have alternate mirrors to try. 424 */ 425 static int btree_read_extent_buffer_pages(struct btrfs_root *root, 426 struct extent_buffer *eb, 427 u64 start, u64 parent_transid) 428 { 429 struct extent_io_tree *io_tree; 430 int failed = 0; 431 int ret; 432 int num_copies = 0; 433 int mirror_num = 0; 434 int failed_mirror = 0; 435 436 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 437 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree; 438 while (1) { 439 ret = read_extent_buffer_pages(io_tree, eb, start, 440 WAIT_COMPLETE, 441 btree_get_extent, mirror_num); 442 if (!ret) { 443 if (!verify_parent_transid(io_tree, eb, 444 parent_transid, 0)) 445 break; 446 else 447 ret = -EIO; 448 } 449 450 /* 451 * This buffer's crc is fine, but its contents are corrupted, so 452 * there is no reason to read the other copies, they won't be 453 * any less wrong. 454 */ 455 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags)) 456 break; 457 458 num_copies = btrfs_num_copies(root->fs_info, 459 eb->start, eb->len); 460 if (num_copies == 1) 461 break; 462 463 if (!failed_mirror) { 464 failed = 1; 465 failed_mirror = eb->read_mirror; 466 } 467 468 mirror_num++; 469 if (mirror_num == failed_mirror) 470 mirror_num++; 471 472 if (mirror_num > num_copies) 473 break; 474 } 475 476 if (failed && !ret && failed_mirror) 477 repair_eb_io_failure(root, eb, failed_mirror); 478 479 return ret; 480 } 481 482 /* 483 * checksum a dirty tree block before IO. This has extra checks to make sure 484 * we only fill in the checksum field in the first page of a multi-page block 485 */ 486 487 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page) 488 { 489 u64 start = page_offset(page); 490 u64 found_start; 491 struct extent_buffer *eb; 492 493 eb = (struct extent_buffer *)page->private; 494 if (page != eb->pages[0]) 495 return 0; 496 found_start = btrfs_header_bytenr(eb); 497 if (WARN_ON(found_start != start || !PageUptodate(page))) 498 return 0; 499 csum_tree_block(root, eb, 0); 500 return 0; 501 } 502 503 static int check_tree_block_fsid(struct btrfs_root *root, 504 struct extent_buffer *eb) 505 { 506 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; 507 u8 fsid[BTRFS_UUID_SIZE]; 508 int ret = 1; 509 510 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE); 511 while (fs_devices) { 512 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) { 513 ret = 0; 514 break; 515 } 516 fs_devices = fs_devices->seed; 517 } 518 return ret; 519 } 520 521 #define CORRUPT(reason, eb, root, slot) \ 522 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \ 523 "root=%llu, slot=%d", reason, \ 524 btrfs_header_bytenr(eb), root->objectid, slot) 525 526 static noinline int check_leaf(struct btrfs_root *root, 527 struct extent_buffer *leaf) 528 { 529 struct btrfs_key key; 530 struct btrfs_key leaf_key; 531 u32 nritems = btrfs_header_nritems(leaf); 532 int slot; 533 534 if (nritems == 0) 535 return 0; 536 537 /* Check the 0 item */ 538 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) != 539 BTRFS_LEAF_DATA_SIZE(root)) { 540 CORRUPT("invalid item offset size pair", leaf, root, 0); 541 return -EIO; 542 } 543 544 /* 545 * Check to make sure each items keys are in the correct order and their 546 * offsets make sense. We only have to loop through nritems-1 because 547 * we check the current slot against the next slot, which verifies the 548 * next slot's offset+size makes sense and that the current's slot 549 * offset is correct. 550 */ 551 for (slot = 0; slot < nritems - 1; slot++) { 552 btrfs_item_key_to_cpu(leaf, &leaf_key, slot); 553 btrfs_item_key_to_cpu(leaf, &key, slot + 1); 554 555 /* Make sure the keys are in the right order */ 556 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) { 557 CORRUPT("bad key order", leaf, root, slot); 558 return -EIO; 559 } 560 561 /* 562 * Make sure the offset and ends are right, remember that the 563 * item data starts at the end of the leaf and grows towards the 564 * front. 565 */ 566 if (btrfs_item_offset_nr(leaf, slot) != 567 btrfs_item_end_nr(leaf, slot + 1)) { 568 CORRUPT("slot offset bad", leaf, root, slot); 569 return -EIO; 570 } 571 572 /* 573 * Check to make sure that we don't point outside of the leaf, 574 * just incase all the items are consistent to eachother, but 575 * all point outside of the leaf. 576 */ 577 if (btrfs_item_end_nr(leaf, slot) > 578 BTRFS_LEAF_DATA_SIZE(root)) { 579 CORRUPT("slot end outside of leaf", leaf, root, slot); 580 return -EIO; 581 } 582 } 583 584 return 0; 585 } 586 587 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio, 588 u64 phy_offset, struct page *page, 589 u64 start, u64 end, int mirror) 590 { 591 u64 found_start; 592 int found_level; 593 struct extent_buffer *eb; 594 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 595 int ret = 0; 596 int reads_done; 597 598 if (!page->private) 599 goto out; 600 601 eb = (struct extent_buffer *)page->private; 602 603 /* the pending IO might have been the only thing that kept this buffer 604 * in memory. Make sure we have a ref for all this other checks 605 */ 606 extent_buffer_get(eb); 607 608 reads_done = atomic_dec_and_test(&eb->io_pages); 609 if (!reads_done) 610 goto err; 611 612 eb->read_mirror = mirror; 613 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) { 614 ret = -EIO; 615 goto err; 616 } 617 618 found_start = btrfs_header_bytenr(eb); 619 if (found_start != eb->start) { 620 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start " 621 "%llu %llu\n", 622 found_start, eb->start); 623 ret = -EIO; 624 goto err; 625 } 626 if (check_tree_block_fsid(root, eb)) { 627 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n", 628 eb->start); 629 ret = -EIO; 630 goto err; 631 } 632 found_level = btrfs_header_level(eb); 633 if (found_level >= BTRFS_MAX_LEVEL) { 634 btrfs_info(root->fs_info, "bad tree block level %d", 635 (int)btrfs_header_level(eb)); 636 ret = -EIO; 637 goto err; 638 } 639 640 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb), 641 eb, found_level); 642 643 ret = csum_tree_block(root, eb, 1); 644 if (ret) { 645 ret = -EIO; 646 goto err; 647 } 648 649 /* 650 * If this is a leaf block and it is corrupt, set the corrupt bit so 651 * that we don't try and read the other copies of this block, just 652 * return -EIO. 653 */ 654 if (found_level == 0 && check_leaf(root, eb)) { 655 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags); 656 ret = -EIO; 657 } 658 659 if (!ret) 660 set_extent_buffer_uptodate(eb); 661 err: 662 if (reads_done && 663 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 664 btree_readahead_hook(root, eb, eb->start, ret); 665 666 if (ret) { 667 /* 668 * our io error hook is going to dec the io pages 669 * again, we have to make sure it has something 670 * to decrement 671 */ 672 atomic_inc(&eb->io_pages); 673 clear_extent_buffer_uptodate(eb); 674 } 675 free_extent_buffer(eb); 676 out: 677 return ret; 678 } 679 680 static int btree_io_failed_hook(struct page *page, int failed_mirror) 681 { 682 struct extent_buffer *eb; 683 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root; 684 685 eb = (struct extent_buffer *)page->private; 686 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags); 687 eb->read_mirror = failed_mirror; 688 atomic_dec(&eb->io_pages); 689 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) 690 btree_readahead_hook(root, eb, eb->start, -EIO); 691 return -EIO; /* we fixed nothing */ 692 } 693 694 static void end_workqueue_bio(struct bio *bio, int err) 695 { 696 struct end_io_wq *end_io_wq = bio->bi_private; 697 struct btrfs_fs_info *fs_info; 698 699 fs_info = end_io_wq->info; 700 end_io_wq->error = err; 701 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL); 702 703 if (bio->bi_rw & REQ_WRITE) { 704 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) 705 btrfs_queue_work(fs_info->endio_meta_write_workers, 706 &end_io_wq->work); 707 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) 708 btrfs_queue_work(fs_info->endio_freespace_worker, 709 &end_io_wq->work); 710 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 711 btrfs_queue_work(fs_info->endio_raid56_workers, 712 &end_io_wq->work); 713 else 714 btrfs_queue_work(fs_info->endio_write_workers, 715 &end_io_wq->work); 716 } else { 717 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) 718 btrfs_queue_work(fs_info->endio_raid56_workers, 719 &end_io_wq->work); 720 else if (end_io_wq->metadata) 721 btrfs_queue_work(fs_info->endio_meta_workers, 722 &end_io_wq->work); 723 else 724 btrfs_queue_work(fs_info->endio_workers, 725 &end_io_wq->work); 726 } 727 } 728 729 /* 730 * For the metadata arg you want 731 * 732 * 0 - if data 733 * 1 - if normal metadta 734 * 2 - if writing to the free space cache area 735 * 3 - raid parity work 736 */ 737 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio, 738 int metadata) 739 { 740 struct end_io_wq *end_io_wq; 741 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS); 742 if (!end_io_wq) 743 return -ENOMEM; 744 745 end_io_wq->private = bio->bi_private; 746 end_io_wq->end_io = bio->bi_end_io; 747 end_io_wq->info = info; 748 end_io_wq->error = 0; 749 end_io_wq->bio = bio; 750 end_io_wq->metadata = metadata; 751 752 bio->bi_private = end_io_wq; 753 bio->bi_end_io = end_workqueue_bio; 754 return 0; 755 } 756 757 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info) 758 { 759 unsigned long limit = min_t(unsigned long, 760 info->thread_pool_size, 761 info->fs_devices->open_devices); 762 return 256 * limit; 763 } 764 765 static void run_one_async_start(struct btrfs_work *work) 766 { 767 struct async_submit_bio *async; 768 int ret; 769 770 async = container_of(work, struct async_submit_bio, work); 771 ret = async->submit_bio_start(async->inode, async->rw, async->bio, 772 async->mirror_num, async->bio_flags, 773 async->bio_offset); 774 if (ret) 775 async->error = ret; 776 } 777 778 static void run_one_async_done(struct btrfs_work *work) 779 { 780 struct btrfs_fs_info *fs_info; 781 struct async_submit_bio *async; 782 int limit; 783 784 async = container_of(work, struct async_submit_bio, work); 785 fs_info = BTRFS_I(async->inode)->root->fs_info; 786 787 limit = btrfs_async_submit_limit(fs_info); 788 limit = limit * 2 / 3; 789 790 if (atomic_dec_return(&fs_info->nr_async_submits) < limit && 791 waitqueue_active(&fs_info->async_submit_wait)) 792 wake_up(&fs_info->async_submit_wait); 793 794 /* If an error occured we just want to clean up the bio and move on */ 795 if (async->error) { 796 bio_endio(async->bio, async->error); 797 return; 798 } 799 800 async->submit_bio_done(async->inode, async->rw, async->bio, 801 async->mirror_num, async->bio_flags, 802 async->bio_offset); 803 } 804 805 static void run_one_async_free(struct btrfs_work *work) 806 { 807 struct async_submit_bio *async; 808 809 async = container_of(work, struct async_submit_bio, work); 810 kfree(async); 811 } 812 813 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode, 814 int rw, struct bio *bio, int mirror_num, 815 unsigned long bio_flags, 816 u64 bio_offset, 817 extent_submit_bio_hook_t *submit_bio_start, 818 extent_submit_bio_hook_t *submit_bio_done) 819 { 820 struct async_submit_bio *async; 821 822 async = kmalloc(sizeof(*async), GFP_NOFS); 823 if (!async) 824 return -ENOMEM; 825 826 async->inode = inode; 827 async->rw = rw; 828 async->bio = bio; 829 async->mirror_num = mirror_num; 830 async->submit_bio_start = submit_bio_start; 831 async->submit_bio_done = submit_bio_done; 832 833 btrfs_init_work(&async->work, run_one_async_start, 834 run_one_async_done, run_one_async_free); 835 836 async->bio_flags = bio_flags; 837 async->bio_offset = bio_offset; 838 839 async->error = 0; 840 841 atomic_inc(&fs_info->nr_async_submits); 842 843 if (rw & REQ_SYNC) 844 btrfs_set_work_high_priority(&async->work); 845 846 btrfs_queue_work(fs_info->workers, &async->work); 847 848 while (atomic_read(&fs_info->async_submit_draining) && 849 atomic_read(&fs_info->nr_async_submits)) { 850 wait_event(fs_info->async_submit_wait, 851 (atomic_read(&fs_info->nr_async_submits) == 0)); 852 } 853 854 return 0; 855 } 856 857 static int btree_csum_one_bio(struct bio *bio) 858 { 859 struct bio_vec *bvec; 860 struct btrfs_root *root; 861 int i, ret = 0; 862 863 bio_for_each_segment_all(bvec, bio, i) { 864 root = BTRFS_I(bvec->bv_page->mapping->host)->root; 865 ret = csum_dirty_buffer(root, bvec->bv_page); 866 if (ret) 867 break; 868 } 869 870 return ret; 871 } 872 873 static int __btree_submit_bio_start(struct inode *inode, int rw, 874 struct bio *bio, int mirror_num, 875 unsigned long bio_flags, 876 u64 bio_offset) 877 { 878 /* 879 * when we're called for a write, we're already in the async 880 * submission context. Just jump into btrfs_map_bio 881 */ 882 return btree_csum_one_bio(bio); 883 } 884 885 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio, 886 int mirror_num, unsigned long bio_flags, 887 u64 bio_offset) 888 { 889 int ret; 890 891 /* 892 * when we're called for a write, we're already in the async 893 * submission context. Just jump into btrfs_map_bio 894 */ 895 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1); 896 if (ret) 897 bio_endio(bio, ret); 898 return ret; 899 } 900 901 static int check_async_write(struct inode *inode, unsigned long bio_flags) 902 { 903 if (bio_flags & EXTENT_BIO_TREE_LOG) 904 return 0; 905 #ifdef CONFIG_X86 906 if (cpu_has_xmm4_2) 907 return 0; 908 #endif 909 return 1; 910 } 911 912 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio, 913 int mirror_num, unsigned long bio_flags, 914 u64 bio_offset) 915 { 916 int async = check_async_write(inode, bio_flags); 917 int ret; 918 919 if (!(rw & REQ_WRITE)) { 920 /* 921 * called for a read, do the setup so that checksum validation 922 * can happen in the async kernel threads 923 */ 924 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info, 925 bio, 1); 926 if (ret) 927 goto out_w_error; 928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 929 mirror_num, 0); 930 } else if (!async) { 931 ret = btree_csum_one_bio(bio); 932 if (ret) 933 goto out_w_error; 934 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, 935 mirror_num, 0); 936 } else { 937 /* 938 * kthread helpers are used to submit writes so that 939 * checksumming can happen in parallel across all CPUs 940 */ 941 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info, 942 inode, rw, bio, mirror_num, 0, 943 bio_offset, 944 __btree_submit_bio_start, 945 __btree_submit_bio_done); 946 } 947 948 if (ret) { 949 out_w_error: 950 bio_endio(bio, ret); 951 } 952 return ret; 953 } 954 955 #ifdef CONFIG_MIGRATION 956 static int btree_migratepage(struct address_space *mapping, 957 struct page *newpage, struct page *page, 958 enum migrate_mode mode) 959 { 960 /* 961 * we can't safely write a btree page from here, 962 * we haven't done the locking hook 963 */ 964 if (PageDirty(page)) 965 return -EAGAIN; 966 /* 967 * Buffers may be managed in a filesystem specific way. 968 * We must have no buffers or drop them. 969 */ 970 if (page_has_private(page) && 971 !try_to_release_page(page, GFP_KERNEL)) 972 return -EAGAIN; 973 return migrate_page(mapping, newpage, page, mode); 974 } 975 #endif 976 977 978 static int btree_writepages(struct address_space *mapping, 979 struct writeback_control *wbc) 980 { 981 struct btrfs_fs_info *fs_info; 982 int ret; 983 984 if (wbc->sync_mode == WB_SYNC_NONE) { 985 986 if (wbc->for_kupdate) 987 return 0; 988 989 fs_info = BTRFS_I(mapping->host)->root->fs_info; 990 /* this is a bit racy, but that's ok */ 991 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes, 992 BTRFS_DIRTY_METADATA_THRESH); 993 if (ret < 0) 994 return 0; 995 } 996 return btree_write_cache_pages(mapping, wbc); 997 } 998 999 static int btree_readpage(struct file *file, struct page *page) 1000 { 1001 struct extent_io_tree *tree; 1002 tree = &BTRFS_I(page->mapping->host)->io_tree; 1003 return extent_read_full_page(tree, page, btree_get_extent, 0); 1004 } 1005 1006 static int btree_releasepage(struct page *page, gfp_t gfp_flags) 1007 { 1008 if (PageWriteback(page) || PageDirty(page)) 1009 return 0; 1010 1011 return try_release_extent_buffer(page); 1012 } 1013 1014 static void btree_invalidatepage(struct page *page, unsigned int offset, 1015 unsigned int length) 1016 { 1017 struct extent_io_tree *tree; 1018 tree = &BTRFS_I(page->mapping->host)->io_tree; 1019 extent_invalidatepage(tree, page, offset); 1020 btree_releasepage(page, GFP_NOFS); 1021 if (PagePrivate(page)) { 1022 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info, 1023 "page private not zero on page %llu", 1024 (unsigned long long)page_offset(page)); 1025 ClearPagePrivate(page); 1026 set_page_private(page, 0); 1027 page_cache_release(page); 1028 } 1029 } 1030 1031 static int btree_set_page_dirty(struct page *page) 1032 { 1033 #ifdef DEBUG 1034 struct extent_buffer *eb; 1035 1036 BUG_ON(!PagePrivate(page)); 1037 eb = (struct extent_buffer *)page->private; 1038 BUG_ON(!eb); 1039 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)); 1040 BUG_ON(!atomic_read(&eb->refs)); 1041 btrfs_assert_tree_locked(eb); 1042 #endif 1043 return __set_page_dirty_nobuffers(page); 1044 } 1045 1046 static const struct address_space_operations btree_aops = { 1047 .readpage = btree_readpage, 1048 .writepages = btree_writepages, 1049 .releasepage = btree_releasepage, 1050 .invalidatepage = btree_invalidatepage, 1051 #ifdef CONFIG_MIGRATION 1052 .migratepage = btree_migratepage, 1053 #endif 1054 .set_page_dirty = btree_set_page_dirty, 1055 }; 1056 1057 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1058 u64 parent_transid) 1059 { 1060 struct extent_buffer *buf = NULL; 1061 struct inode *btree_inode = root->fs_info->btree_inode; 1062 int ret = 0; 1063 1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1065 if (!buf) 1066 return 0; 1067 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree, 1068 buf, 0, WAIT_NONE, btree_get_extent, 0); 1069 free_extent_buffer(buf); 1070 return ret; 1071 } 1072 1073 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize, 1074 int mirror_num, struct extent_buffer **eb) 1075 { 1076 struct extent_buffer *buf = NULL; 1077 struct inode *btree_inode = root->fs_info->btree_inode; 1078 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree; 1079 int ret; 1080 1081 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1082 if (!buf) 1083 return 0; 1084 1085 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags); 1086 1087 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK, 1088 btree_get_extent, mirror_num); 1089 if (ret) { 1090 free_extent_buffer(buf); 1091 return ret; 1092 } 1093 1094 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) { 1095 free_extent_buffer(buf); 1096 return -EIO; 1097 } else if (extent_buffer_uptodate(buf)) { 1098 *eb = buf; 1099 } else { 1100 free_extent_buffer(buf); 1101 } 1102 return 0; 1103 } 1104 1105 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root, 1106 u64 bytenr, u32 blocksize) 1107 { 1108 return find_extent_buffer(root->fs_info, bytenr); 1109 } 1110 1111 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root, 1112 u64 bytenr, u32 blocksize) 1113 { 1114 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1115 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) 1116 return alloc_test_extent_buffer(root->fs_info, bytenr, 1117 blocksize); 1118 #endif 1119 return alloc_extent_buffer(root->fs_info, bytenr, blocksize); 1120 } 1121 1122 1123 int btrfs_write_tree_block(struct extent_buffer *buf) 1124 { 1125 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start, 1126 buf->start + buf->len - 1); 1127 } 1128 1129 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf) 1130 { 1131 return filemap_fdatawait_range(buf->pages[0]->mapping, 1132 buf->start, buf->start + buf->len - 1); 1133 } 1134 1135 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr, 1136 u32 blocksize, u64 parent_transid) 1137 { 1138 struct extent_buffer *buf = NULL; 1139 int ret; 1140 1141 buf = btrfs_find_create_tree_block(root, bytenr, blocksize); 1142 if (!buf) 1143 return NULL; 1144 1145 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 1146 if (ret) { 1147 free_extent_buffer(buf); 1148 return NULL; 1149 } 1150 return buf; 1151 1152 } 1153 1154 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root, 1155 struct extent_buffer *buf) 1156 { 1157 struct btrfs_fs_info *fs_info = root->fs_info; 1158 1159 if (btrfs_header_generation(buf) == 1160 fs_info->running_transaction->transid) { 1161 btrfs_assert_tree_locked(buf); 1162 1163 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) { 1164 __percpu_counter_add(&fs_info->dirty_metadata_bytes, 1165 -buf->len, 1166 fs_info->dirty_metadata_batch); 1167 /* ugh, clear_extent_buffer_dirty needs to lock the page */ 1168 btrfs_set_lock_blocking(buf); 1169 clear_extent_buffer_dirty(buf); 1170 } 1171 } 1172 } 1173 1174 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void) 1175 { 1176 struct btrfs_subvolume_writers *writers; 1177 int ret; 1178 1179 writers = kmalloc(sizeof(*writers), GFP_NOFS); 1180 if (!writers) 1181 return ERR_PTR(-ENOMEM); 1182 1183 ret = percpu_counter_init(&writers->counter, 0); 1184 if (ret < 0) { 1185 kfree(writers); 1186 return ERR_PTR(ret); 1187 } 1188 1189 init_waitqueue_head(&writers->wait); 1190 return writers; 1191 } 1192 1193 static void 1194 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers) 1195 { 1196 percpu_counter_destroy(&writers->counter); 1197 kfree(writers); 1198 } 1199 1200 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize, 1201 u32 stripesize, struct btrfs_root *root, 1202 struct btrfs_fs_info *fs_info, 1203 u64 objectid) 1204 { 1205 root->node = NULL; 1206 root->commit_root = NULL; 1207 root->sectorsize = sectorsize; 1208 root->nodesize = nodesize; 1209 root->leafsize = leafsize; 1210 root->stripesize = stripesize; 1211 root->state = 0; 1212 root->orphan_cleanup_state = 0; 1213 1214 root->objectid = objectid; 1215 root->last_trans = 0; 1216 root->highest_objectid = 0; 1217 root->nr_delalloc_inodes = 0; 1218 root->nr_ordered_extents = 0; 1219 root->name = NULL; 1220 root->inode_tree = RB_ROOT; 1221 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC); 1222 root->block_rsv = NULL; 1223 root->orphan_block_rsv = NULL; 1224 1225 INIT_LIST_HEAD(&root->dirty_list); 1226 INIT_LIST_HEAD(&root->root_list); 1227 INIT_LIST_HEAD(&root->delalloc_inodes); 1228 INIT_LIST_HEAD(&root->delalloc_root); 1229 INIT_LIST_HEAD(&root->ordered_extents); 1230 INIT_LIST_HEAD(&root->ordered_root); 1231 INIT_LIST_HEAD(&root->logged_list[0]); 1232 INIT_LIST_HEAD(&root->logged_list[1]); 1233 spin_lock_init(&root->orphan_lock); 1234 spin_lock_init(&root->inode_lock); 1235 spin_lock_init(&root->delalloc_lock); 1236 spin_lock_init(&root->ordered_extent_lock); 1237 spin_lock_init(&root->accounting_lock); 1238 spin_lock_init(&root->log_extents_lock[0]); 1239 spin_lock_init(&root->log_extents_lock[1]); 1240 mutex_init(&root->objectid_mutex); 1241 mutex_init(&root->log_mutex); 1242 mutex_init(&root->ordered_extent_mutex); 1243 mutex_init(&root->delalloc_mutex); 1244 init_waitqueue_head(&root->log_writer_wait); 1245 init_waitqueue_head(&root->log_commit_wait[0]); 1246 init_waitqueue_head(&root->log_commit_wait[1]); 1247 INIT_LIST_HEAD(&root->log_ctxs[0]); 1248 INIT_LIST_HEAD(&root->log_ctxs[1]); 1249 atomic_set(&root->log_commit[0], 0); 1250 atomic_set(&root->log_commit[1], 0); 1251 atomic_set(&root->log_writers, 0); 1252 atomic_set(&root->log_batch, 0); 1253 atomic_set(&root->orphan_inodes, 0); 1254 atomic_set(&root->refs, 1); 1255 atomic_set(&root->will_be_snapshoted, 0); 1256 root->log_transid = 0; 1257 root->log_transid_committed = -1; 1258 root->last_log_commit = 0; 1259 if (fs_info) 1260 extent_io_tree_init(&root->dirty_log_pages, 1261 fs_info->btree_inode->i_mapping); 1262 1263 memset(&root->root_key, 0, sizeof(root->root_key)); 1264 memset(&root->root_item, 0, sizeof(root->root_item)); 1265 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress)); 1266 memset(&root->root_kobj, 0, sizeof(root->root_kobj)); 1267 if (fs_info) 1268 root->defrag_trans_start = fs_info->generation; 1269 else 1270 root->defrag_trans_start = 0; 1271 init_completion(&root->kobj_unregister); 1272 root->root_key.objectid = objectid; 1273 root->anon_dev = 0; 1274 1275 spin_lock_init(&root->root_item_lock); 1276 } 1277 1278 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info) 1279 { 1280 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS); 1281 if (root) 1282 root->fs_info = fs_info; 1283 return root; 1284 } 1285 1286 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 1287 /* Should only be used by the testing infrastructure */ 1288 struct btrfs_root *btrfs_alloc_dummy_root(void) 1289 { 1290 struct btrfs_root *root; 1291 1292 root = btrfs_alloc_root(NULL); 1293 if (!root) 1294 return ERR_PTR(-ENOMEM); 1295 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1); 1296 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state); 1297 root->alloc_bytenr = 0; 1298 1299 return root; 1300 } 1301 #endif 1302 1303 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans, 1304 struct btrfs_fs_info *fs_info, 1305 u64 objectid) 1306 { 1307 struct extent_buffer *leaf; 1308 struct btrfs_root *tree_root = fs_info->tree_root; 1309 struct btrfs_root *root; 1310 struct btrfs_key key; 1311 int ret = 0; 1312 uuid_le uuid; 1313 1314 root = btrfs_alloc_root(fs_info); 1315 if (!root) 1316 return ERR_PTR(-ENOMEM); 1317 1318 __setup_root(tree_root->nodesize, tree_root->leafsize, 1319 tree_root->sectorsize, tree_root->stripesize, 1320 root, fs_info, objectid); 1321 root->root_key.objectid = objectid; 1322 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1323 root->root_key.offset = 0; 1324 1325 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 1326 0, objectid, NULL, 0, 0, 0); 1327 if (IS_ERR(leaf)) { 1328 ret = PTR_ERR(leaf); 1329 leaf = NULL; 1330 goto fail; 1331 } 1332 1333 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 1334 btrfs_set_header_bytenr(leaf, leaf->start); 1335 btrfs_set_header_generation(leaf, trans->transid); 1336 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1337 btrfs_set_header_owner(leaf, objectid); 1338 root->node = leaf; 1339 1340 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), 1341 BTRFS_FSID_SIZE); 1342 write_extent_buffer(leaf, fs_info->chunk_tree_uuid, 1343 btrfs_header_chunk_tree_uuid(leaf), 1344 BTRFS_UUID_SIZE); 1345 btrfs_mark_buffer_dirty(leaf); 1346 1347 root->commit_root = btrfs_root_node(root); 1348 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state); 1349 1350 root->root_item.flags = 0; 1351 root->root_item.byte_limit = 0; 1352 btrfs_set_root_bytenr(&root->root_item, leaf->start); 1353 btrfs_set_root_generation(&root->root_item, trans->transid); 1354 btrfs_set_root_level(&root->root_item, 0); 1355 btrfs_set_root_refs(&root->root_item, 1); 1356 btrfs_set_root_used(&root->root_item, leaf->len); 1357 btrfs_set_root_last_snapshot(&root->root_item, 0); 1358 btrfs_set_root_dirid(&root->root_item, 0); 1359 uuid_le_gen(&uuid); 1360 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE); 1361 root->root_item.drop_level = 0; 1362 1363 key.objectid = objectid; 1364 key.type = BTRFS_ROOT_ITEM_KEY; 1365 key.offset = 0; 1366 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item); 1367 if (ret) 1368 goto fail; 1369 1370 btrfs_tree_unlock(leaf); 1371 1372 return root; 1373 1374 fail: 1375 if (leaf) { 1376 btrfs_tree_unlock(leaf); 1377 free_extent_buffer(root->commit_root); 1378 free_extent_buffer(leaf); 1379 } 1380 kfree(root); 1381 1382 return ERR_PTR(ret); 1383 } 1384 1385 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans, 1386 struct btrfs_fs_info *fs_info) 1387 { 1388 struct btrfs_root *root; 1389 struct btrfs_root *tree_root = fs_info->tree_root; 1390 struct extent_buffer *leaf; 1391 1392 root = btrfs_alloc_root(fs_info); 1393 if (!root) 1394 return ERR_PTR(-ENOMEM); 1395 1396 __setup_root(tree_root->nodesize, tree_root->leafsize, 1397 tree_root->sectorsize, tree_root->stripesize, 1398 root, fs_info, BTRFS_TREE_LOG_OBJECTID); 1399 1400 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID; 1401 root->root_key.type = BTRFS_ROOT_ITEM_KEY; 1402 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID; 1403 1404 /* 1405 * DON'T set REF_COWS for log trees 1406 * 1407 * log trees do not get reference counted because they go away 1408 * before a real commit is actually done. They do store pointers 1409 * to file data extents, and those reference counts still get 1410 * updated (along with back refs to the log tree). 1411 */ 1412 1413 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 1414 BTRFS_TREE_LOG_OBJECTID, NULL, 1415 0, 0, 0); 1416 if (IS_ERR(leaf)) { 1417 kfree(root); 1418 return ERR_CAST(leaf); 1419 } 1420 1421 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header)); 1422 btrfs_set_header_bytenr(leaf, leaf->start); 1423 btrfs_set_header_generation(leaf, trans->transid); 1424 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV); 1425 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID); 1426 root->node = leaf; 1427 1428 write_extent_buffer(root->node, root->fs_info->fsid, 1429 btrfs_header_fsid(), BTRFS_FSID_SIZE); 1430 btrfs_mark_buffer_dirty(root->node); 1431 btrfs_tree_unlock(root->node); 1432 return root; 1433 } 1434 1435 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans, 1436 struct btrfs_fs_info *fs_info) 1437 { 1438 struct btrfs_root *log_root; 1439 1440 log_root = alloc_log_tree(trans, fs_info); 1441 if (IS_ERR(log_root)) 1442 return PTR_ERR(log_root); 1443 WARN_ON(fs_info->log_root_tree); 1444 fs_info->log_root_tree = log_root; 1445 return 0; 1446 } 1447 1448 int btrfs_add_log_tree(struct btrfs_trans_handle *trans, 1449 struct btrfs_root *root) 1450 { 1451 struct btrfs_root *log_root; 1452 struct btrfs_inode_item *inode_item; 1453 1454 log_root = alloc_log_tree(trans, root->fs_info); 1455 if (IS_ERR(log_root)) 1456 return PTR_ERR(log_root); 1457 1458 log_root->last_trans = trans->transid; 1459 log_root->root_key.offset = root->root_key.objectid; 1460 1461 inode_item = &log_root->root_item.inode; 1462 btrfs_set_stack_inode_generation(inode_item, 1); 1463 btrfs_set_stack_inode_size(inode_item, 3); 1464 btrfs_set_stack_inode_nlink(inode_item, 1); 1465 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize); 1466 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755); 1467 1468 btrfs_set_root_node(&log_root->root_item, log_root->node); 1469 1470 WARN_ON(root->log_root); 1471 root->log_root = log_root; 1472 root->log_transid = 0; 1473 root->log_transid_committed = -1; 1474 root->last_log_commit = 0; 1475 return 0; 1476 } 1477 1478 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root, 1479 struct btrfs_key *key) 1480 { 1481 struct btrfs_root *root; 1482 struct btrfs_fs_info *fs_info = tree_root->fs_info; 1483 struct btrfs_path *path; 1484 u64 generation; 1485 u32 blocksize; 1486 int ret; 1487 1488 path = btrfs_alloc_path(); 1489 if (!path) 1490 return ERR_PTR(-ENOMEM); 1491 1492 root = btrfs_alloc_root(fs_info); 1493 if (!root) { 1494 ret = -ENOMEM; 1495 goto alloc_fail; 1496 } 1497 1498 __setup_root(tree_root->nodesize, tree_root->leafsize, 1499 tree_root->sectorsize, tree_root->stripesize, 1500 root, fs_info, key->objectid); 1501 1502 ret = btrfs_find_root(tree_root, key, path, 1503 &root->root_item, &root->root_key); 1504 if (ret) { 1505 if (ret > 0) 1506 ret = -ENOENT; 1507 goto find_fail; 1508 } 1509 1510 generation = btrfs_root_generation(&root->root_item); 1511 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item)); 1512 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item), 1513 blocksize, generation); 1514 if (!root->node) { 1515 ret = -ENOMEM; 1516 goto find_fail; 1517 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) { 1518 ret = -EIO; 1519 goto read_fail; 1520 } 1521 root->commit_root = btrfs_root_node(root); 1522 out: 1523 btrfs_free_path(path); 1524 return root; 1525 1526 read_fail: 1527 free_extent_buffer(root->node); 1528 find_fail: 1529 kfree(root); 1530 alloc_fail: 1531 root = ERR_PTR(ret); 1532 goto out; 1533 } 1534 1535 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root, 1536 struct btrfs_key *location) 1537 { 1538 struct btrfs_root *root; 1539 1540 root = btrfs_read_tree_root(tree_root, location); 1541 if (IS_ERR(root)) 1542 return root; 1543 1544 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { 1545 set_bit(BTRFS_ROOT_REF_COWS, &root->state); 1546 btrfs_check_and_init_root_item(&root->root_item); 1547 } 1548 1549 return root; 1550 } 1551 1552 int btrfs_init_fs_root(struct btrfs_root *root) 1553 { 1554 int ret; 1555 struct btrfs_subvolume_writers *writers; 1556 1557 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS); 1558 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned), 1559 GFP_NOFS); 1560 if (!root->free_ino_pinned || !root->free_ino_ctl) { 1561 ret = -ENOMEM; 1562 goto fail; 1563 } 1564 1565 writers = btrfs_alloc_subvolume_writers(); 1566 if (IS_ERR(writers)) { 1567 ret = PTR_ERR(writers); 1568 goto fail; 1569 } 1570 root->subv_writers = writers; 1571 1572 btrfs_init_free_ino_ctl(root); 1573 spin_lock_init(&root->cache_lock); 1574 init_waitqueue_head(&root->cache_wait); 1575 1576 ret = get_anon_bdev(&root->anon_dev); 1577 if (ret) 1578 goto free_writers; 1579 return 0; 1580 1581 free_writers: 1582 btrfs_free_subvolume_writers(root->subv_writers); 1583 fail: 1584 kfree(root->free_ino_ctl); 1585 kfree(root->free_ino_pinned); 1586 return ret; 1587 } 1588 1589 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info, 1590 u64 root_id) 1591 { 1592 struct btrfs_root *root; 1593 1594 spin_lock(&fs_info->fs_roots_radix_lock); 1595 root = radix_tree_lookup(&fs_info->fs_roots_radix, 1596 (unsigned long)root_id); 1597 spin_unlock(&fs_info->fs_roots_radix_lock); 1598 return root; 1599 } 1600 1601 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info, 1602 struct btrfs_root *root) 1603 { 1604 int ret; 1605 1606 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM); 1607 if (ret) 1608 return ret; 1609 1610 spin_lock(&fs_info->fs_roots_radix_lock); 1611 ret = radix_tree_insert(&fs_info->fs_roots_radix, 1612 (unsigned long)root->root_key.objectid, 1613 root); 1614 if (ret == 0) 1615 set_bit(BTRFS_ROOT_IN_RADIX, &root->state); 1616 spin_unlock(&fs_info->fs_roots_radix_lock); 1617 radix_tree_preload_end(); 1618 1619 return ret; 1620 } 1621 1622 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info, 1623 struct btrfs_key *location, 1624 bool check_ref) 1625 { 1626 struct btrfs_root *root; 1627 int ret; 1628 1629 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID) 1630 return fs_info->tree_root; 1631 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID) 1632 return fs_info->extent_root; 1633 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID) 1634 return fs_info->chunk_root; 1635 if (location->objectid == BTRFS_DEV_TREE_OBJECTID) 1636 return fs_info->dev_root; 1637 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID) 1638 return fs_info->csum_root; 1639 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID) 1640 return fs_info->quota_root ? fs_info->quota_root : 1641 ERR_PTR(-ENOENT); 1642 if (location->objectid == BTRFS_UUID_TREE_OBJECTID) 1643 return fs_info->uuid_root ? fs_info->uuid_root : 1644 ERR_PTR(-ENOENT); 1645 again: 1646 root = btrfs_lookup_fs_root(fs_info, location->objectid); 1647 if (root) { 1648 if (check_ref && btrfs_root_refs(&root->root_item) == 0) 1649 return ERR_PTR(-ENOENT); 1650 return root; 1651 } 1652 1653 root = btrfs_read_fs_root(fs_info->tree_root, location); 1654 if (IS_ERR(root)) 1655 return root; 1656 1657 if (check_ref && btrfs_root_refs(&root->root_item) == 0) { 1658 ret = -ENOENT; 1659 goto fail; 1660 } 1661 1662 ret = btrfs_init_fs_root(root); 1663 if (ret) 1664 goto fail; 1665 1666 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID, 1667 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL); 1668 if (ret < 0) 1669 goto fail; 1670 if (ret == 0) 1671 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state); 1672 1673 ret = btrfs_insert_fs_root(fs_info, root); 1674 if (ret) { 1675 if (ret == -EEXIST) { 1676 free_fs_root(root); 1677 goto again; 1678 } 1679 goto fail; 1680 } 1681 return root; 1682 fail: 1683 free_fs_root(root); 1684 return ERR_PTR(ret); 1685 } 1686 1687 static int btrfs_congested_fn(void *congested_data, int bdi_bits) 1688 { 1689 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data; 1690 int ret = 0; 1691 struct btrfs_device *device; 1692 struct backing_dev_info *bdi; 1693 1694 rcu_read_lock(); 1695 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) { 1696 if (!device->bdev) 1697 continue; 1698 bdi = blk_get_backing_dev_info(device->bdev); 1699 if (bdi && bdi_congested(bdi, bdi_bits)) { 1700 ret = 1; 1701 break; 1702 } 1703 } 1704 rcu_read_unlock(); 1705 return ret; 1706 } 1707 1708 /* 1709 * If this fails, caller must call bdi_destroy() to get rid of the 1710 * bdi again. 1711 */ 1712 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi) 1713 { 1714 int err; 1715 1716 bdi->capabilities = BDI_CAP_MAP_COPY; 1717 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY); 1718 if (err) 1719 return err; 1720 1721 bdi->ra_pages = default_backing_dev_info.ra_pages; 1722 bdi->congested_fn = btrfs_congested_fn; 1723 bdi->congested_data = info; 1724 return 0; 1725 } 1726 1727 /* 1728 * called by the kthread helper functions to finally call the bio end_io 1729 * functions. This is where read checksum verification actually happens 1730 */ 1731 static void end_workqueue_fn(struct btrfs_work *work) 1732 { 1733 struct bio *bio; 1734 struct end_io_wq *end_io_wq; 1735 int error; 1736 1737 end_io_wq = container_of(work, struct end_io_wq, work); 1738 bio = end_io_wq->bio; 1739 1740 error = end_io_wq->error; 1741 bio->bi_private = end_io_wq->private; 1742 bio->bi_end_io = end_io_wq->end_io; 1743 kfree(end_io_wq); 1744 bio_endio_nodec(bio, error); 1745 } 1746 1747 static int cleaner_kthread(void *arg) 1748 { 1749 struct btrfs_root *root = arg; 1750 int again; 1751 1752 do { 1753 again = 0; 1754 1755 /* Make the cleaner go to sleep early. */ 1756 if (btrfs_need_cleaner_sleep(root)) 1757 goto sleep; 1758 1759 if (!mutex_trylock(&root->fs_info->cleaner_mutex)) 1760 goto sleep; 1761 1762 /* 1763 * Avoid the problem that we change the status of the fs 1764 * during the above check and trylock. 1765 */ 1766 if (btrfs_need_cleaner_sleep(root)) { 1767 mutex_unlock(&root->fs_info->cleaner_mutex); 1768 goto sleep; 1769 } 1770 1771 btrfs_run_delayed_iputs(root); 1772 again = btrfs_clean_one_deleted_snapshot(root); 1773 mutex_unlock(&root->fs_info->cleaner_mutex); 1774 1775 /* 1776 * The defragger has dealt with the R/O remount and umount, 1777 * needn't do anything special here. 1778 */ 1779 btrfs_run_defrag_inodes(root->fs_info); 1780 sleep: 1781 if (!try_to_freeze() && !again) { 1782 set_current_state(TASK_INTERRUPTIBLE); 1783 if (!kthread_should_stop()) 1784 schedule(); 1785 __set_current_state(TASK_RUNNING); 1786 } 1787 } while (!kthread_should_stop()); 1788 return 0; 1789 } 1790 1791 static int transaction_kthread(void *arg) 1792 { 1793 struct btrfs_root *root = arg; 1794 struct btrfs_trans_handle *trans; 1795 struct btrfs_transaction *cur; 1796 u64 transid; 1797 unsigned long now; 1798 unsigned long delay; 1799 bool cannot_commit; 1800 1801 do { 1802 cannot_commit = false; 1803 delay = HZ * root->fs_info->commit_interval; 1804 mutex_lock(&root->fs_info->transaction_kthread_mutex); 1805 1806 spin_lock(&root->fs_info->trans_lock); 1807 cur = root->fs_info->running_transaction; 1808 if (!cur) { 1809 spin_unlock(&root->fs_info->trans_lock); 1810 goto sleep; 1811 } 1812 1813 now = get_seconds(); 1814 if (cur->state < TRANS_STATE_BLOCKED && 1815 (now < cur->start_time || 1816 now - cur->start_time < root->fs_info->commit_interval)) { 1817 spin_unlock(&root->fs_info->trans_lock); 1818 delay = HZ * 5; 1819 goto sleep; 1820 } 1821 transid = cur->transid; 1822 spin_unlock(&root->fs_info->trans_lock); 1823 1824 /* If the file system is aborted, this will always fail. */ 1825 trans = btrfs_attach_transaction(root); 1826 if (IS_ERR(trans)) { 1827 if (PTR_ERR(trans) != -ENOENT) 1828 cannot_commit = true; 1829 goto sleep; 1830 } 1831 if (transid == trans->transid) { 1832 btrfs_commit_transaction(trans, root); 1833 } else { 1834 btrfs_end_transaction(trans, root); 1835 } 1836 sleep: 1837 wake_up_process(root->fs_info->cleaner_kthread); 1838 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 1839 1840 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR, 1841 &root->fs_info->fs_state))) 1842 btrfs_cleanup_transaction(root); 1843 if (!try_to_freeze()) { 1844 set_current_state(TASK_INTERRUPTIBLE); 1845 if (!kthread_should_stop() && 1846 (!btrfs_transaction_blocked(root->fs_info) || 1847 cannot_commit)) 1848 schedule_timeout(delay); 1849 __set_current_state(TASK_RUNNING); 1850 } 1851 } while (!kthread_should_stop()); 1852 return 0; 1853 } 1854 1855 /* 1856 * this will find the highest generation in the array of 1857 * root backups. The index of the highest array is returned, 1858 * or -1 if we can't find anything. 1859 * 1860 * We check to make sure the array is valid by comparing the 1861 * generation of the latest root in the array with the generation 1862 * in the super block. If they don't match we pitch it. 1863 */ 1864 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen) 1865 { 1866 u64 cur; 1867 int newest_index = -1; 1868 struct btrfs_root_backup *root_backup; 1869 int i; 1870 1871 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) { 1872 root_backup = info->super_copy->super_roots + i; 1873 cur = btrfs_backup_tree_root_gen(root_backup); 1874 if (cur == newest_gen) 1875 newest_index = i; 1876 } 1877 1878 /* check to see if we actually wrapped around */ 1879 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) { 1880 root_backup = info->super_copy->super_roots; 1881 cur = btrfs_backup_tree_root_gen(root_backup); 1882 if (cur == newest_gen) 1883 newest_index = 0; 1884 } 1885 return newest_index; 1886 } 1887 1888 1889 /* 1890 * find the oldest backup so we know where to store new entries 1891 * in the backup array. This will set the backup_root_index 1892 * field in the fs_info struct 1893 */ 1894 static void find_oldest_super_backup(struct btrfs_fs_info *info, 1895 u64 newest_gen) 1896 { 1897 int newest_index = -1; 1898 1899 newest_index = find_newest_super_backup(info, newest_gen); 1900 /* if there was garbage in there, just move along */ 1901 if (newest_index == -1) { 1902 info->backup_root_index = 0; 1903 } else { 1904 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS; 1905 } 1906 } 1907 1908 /* 1909 * copy all the root pointers into the super backup array. 1910 * this will bump the backup pointer by one when it is 1911 * done 1912 */ 1913 static void backup_super_roots(struct btrfs_fs_info *info) 1914 { 1915 int next_backup; 1916 struct btrfs_root_backup *root_backup; 1917 int last_backup; 1918 1919 next_backup = info->backup_root_index; 1920 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) % 1921 BTRFS_NUM_BACKUP_ROOTS; 1922 1923 /* 1924 * just overwrite the last backup if we're at the same generation 1925 * this happens only at umount 1926 */ 1927 root_backup = info->super_for_commit->super_roots + last_backup; 1928 if (btrfs_backup_tree_root_gen(root_backup) == 1929 btrfs_header_generation(info->tree_root->node)) 1930 next_backup = last_backup; 1931 1932 root_backup = info->super_for_commit->super_roots + next_backup; 1933 1934 /* 1935 * make sure all of our padding and empty slots get zero filled 1936 * regardless of which ones we use today 1937 */ 1938 memset(root_backup, 0, sizeof(*root_backup)); 1939 1940 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS; 1941 1942 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start); 1943 btrfs_set_backup_tree_root_gen(root_backup, 1944 btrfs_header_generation(info->tree_root->node)); 1945 1946 btrfs_set_backup_tree_root_level(root_backup, 1947 btrfs_header_level(info->tree_root->node)); 1948 1949 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start); 1950 btrfs_set_backup_chunk_root_gen(root_backup, 1951 btrfs_header_generation(info->chunk_root->node)); 1952 btrfs_set_backup_chunk_root_level(root_backup, 1953 btrfs_header_level(info->chunk_root->node)); 1954 1955 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start); 1956 btrfs_set_backup_extent_root_gen(root_backup, 1957 btrfs_header_generation(info->extent_root->node)); 1958 btrfs_set_backup_extent_root_level(root_backup, 1959 btrfs_header_level(info->extent_root->node)); 1960 1961 /* 1962 * we might commit during log recovery, which happens before we set 1963 * the fs_root. Make sure it is valid before we fill it in. 1964 */ 1965 if (info->fs_root && info->fs_root->node) { 1966 btrfs_set_backup_fs_root(root_backup, 1967 info->fs_root->node->start); 1968 btrfs_set_backup_fs_root_gen(root_backup, 1969 btrfs_header_generation(info->fs_root->node)); 1970 btrfs_set_backup_fs_root_level(root_backup, 1971 btrfs_header_level(info->fs_root->node)); 1972 } 1973 1974 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start); 1975 btrfs_set_backup_dev_root_gen(root_backup, 1976 btrfs_header_generation(info->dev_root->node)); 1977 btrfs_set_backup_dev_root_level(root_backup, 1978 btrfs_header_level(info->dev_root->node)); 1979 1980 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start); 1981 btrfs_set_backup_csum_root_gen(root_backup, 1982 btrfs_header_generation(info->csum_root->node)); 1983 btrfs_set_backup_csum_root_level(root_backup, 1984 btrfs_header_level(info->csum_root->node)); 1985 1986 btrfs_set_backup_total_bytes(root_backup, 1987 btrfs_super_total_bytes(info->super_copy)); 1988 btrfs_set_backup_bytes_used(root_backup, 1989 btrfs_super_bytes_used(info->super_copy)); 1990 btrfs_set_backup_num_devices(root_backup, 1991 btrfs_super_num_devices(info->super_copy)); 1992 1993 /* 1994 * if we don't copy this out to the super_copy, it won't get remembered 1995 * for the next commit 1996 */ 1997 memcpy(&info->super_copy->super_roots, 1998 &info->super_for_commit->super_roots, 1999 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS); 2000 } 2001 2002 /* 2003 * this copies info out of the root backup array and back into 2004 * the in-memory super block. It is meant to help iterate through 2005 * the array, so you send it the number of backups you've already 2006 * tried and the last backup index you used. 2007 * 2008 * this returns -1 when it has tried all the backups 2009 */ 2010 static noinline int next_root_backup(struct btrfs_fs_info *info, 2011 struct btrfs_super_block *super, 2012 int *num_backups_tried, int *backup_index) 2013 { 2014 struct btrfs_root_backup *root_backup; 2015 int newest = *backup_index; 2016 2017 if (*num_backups_tried == 0) { 2018 u64 gen = btrfs_super_generation(super); 2019 2020 newest = find_newest_super_backup(info, gen); 2021 if (newest == -1) 2022 return -1; 2023 2024 *backup_index = newest; 2025 *num_backups_tried = 1; 2026 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) { 2027 /* we've tried all the backups, all done */ 2028 return -1; 2029 } else { 2030 /* jump to the next oldest backup */ 2031 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) % 2032 BTRFS_NUM_BACKUP_ROOTS; 2033 *backup_index = newest; 2034 *num_backups_tried += 1; 2035 } 2036 root_backup = super->super_roots + newest; 2037 2038 btrfs_set_super_generation(super, 2039 btrfs_backup_tree_root_gen(root_backup)); 2040 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup)); 2041 btrfs_set_super_root_level(super, 2042 btrfs_backup_tree_root_level(root_backup)); 2043 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup)); 2044 2045 /* 2046 * fixme: the total bytes and num_devices need to match or we should 2047 * need a fsck 2048 */ 2049 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup)); 2050 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup)); 2051 return 0; 2052 } 2053 2054 /* helper to cleanup workers */ 2055 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info) 2056 { 2057 btrfs_destroy_workqueue(fs_info->fixup_workers); 2058 btrfs_destroy_workqueue(fs_info->delalloc_workers); 2059 btrfs_destroy_workqueue(fs_info->workers); 2060 btrfs_destroy_workqueue(fs_info->endio_workers); 2061 btrfs_destroy_workqueue(fs_info->endio_meta_workers); 2062 btrfs_destroy_workqueue(fs_info->endio_raid56_workers); 2063 btrfs_destroy_workqueue(fs_info->rmw_workers); 2064 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers); 2065 btrfs_destroy_workqueue(fs_info->endio_write_workers); 2066 btrfs_destroy_workqueue(fs_info->endio_freespace_worker); 2067 btrfs_destroy_workqueue(fs_info->submit_workers); 2068 btrfs_destroy_workqueue(fs_info->delayed_workers); 2069 btrfs_destroy_workqueue(fs_info->caching_workers); 2070 btrfs_destroy_workqueue(fs_info->readahead_workers); 2071 btrfs_destroy_workqueue(fs_info->flush_workers); 2072 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers); 2073 btrfs_destroy_workqueue(fs_info->extent_workers); 2074 } 2075 2076 static void free_root_extent_buffers(struct btrfs_root *root) 2077 { 2078 if (root) { 2079 free_extent_buffer(root->node); 2080 free_extent_buffer(root->commit_root); 2081 root->node = NULL; 2082 root->commit_root = NULL; 2083 } 2084 } 2085 2086 /* helper to cleanup tree roots */ 2087 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root) 2088 { 2089 free_root_extent_buffers(info->tree_root); 2090 2091 free_root_extent_buffers(info->dev_root); 2092 free_root_extent_buffers(info->extent_root); 2093 free_root_extent_buffers(info->csum_root); 2094 free_root_extent_buffers(info->quota_root); 2095 free_root_extent_buffers(info->uuid_root); 2096 if (chunk_root) 2097 free_root_extent_buffers(info->chunk_root); 2098 } 2099 2100 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info) 2101 { 2102 int ret; 2103 struct btrfs_root *gang[8]; 2104 int i; 2105 2106 while (!list_empty(&fs_info->dead_roots)) { 2107 gang[0] = list_entry(fs_info->dead_roots.next, 2108 struct btrfs_root, root_list); 2109 list_del(&gang[0]->root_list); 2110 2111 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) { 2112 btrfs_drop_and_free_fs_root(fs_info, gang[0]); 2113 } else { 2114 free_extent_buffer(gang[0]->node); 2115 free_extent_buffer(gang[0]->commit_root); 2116 btrfs_put_fs_root(gang[0]); 2117 } 2118 } 2119 2120 while (1) { 2121 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 2122 (void **)gang, 0, 2123 ARRAY_SIZE(gang)); 2124 if (!ret) 2125 break; 2126 for (i = 0; i < ret; i++) 2127 btrfs_drop_and_free_fs_root(fs_info, gang[i]); 2128 } 2129 2130 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 2131 btrfs_free_log_root_tree(NULL, fs_info); 2132 btrfs_destroy_pinned_extent(fs_info->tree_root, 2133 fs_info->pinned_extents); 2134 } 2135 } 2136 2137 int open_ctree(struct super_block *sb, 2138 struct btrfs_fs_devices *fs_devices, 2139 char *options) 2140 { 2141 u32 sectorsize; 2142 u32 nodesize; 2143 u32 leafsize; 2144 u32 blocksize; 2145 u32 stripesize; 2146 u64 generation; 2147 u64 features; 2148 struct btrfs_key location; 2149 struct buffer_head *bh; 2150 struct btrfs_super_block *disk_super; 2151 struct btrfs_fs_info *fs_info = btrfs_sb(sb); 2152 struct btrfs_root *tree_root; 2153 struct btrfs_root *extent_root; 2154 struct btrfs_root *csum_root; 2155 struct btrfs_root *chunk_root; 2156 struct btrfs_root *dev_root; 2157 struct btrfs_root *quota_root; 2158 struct btrfs_root *uuid_root; 2159 struct btrfs_root *log_tree_root; 2160 int ret; 2161 int err = -EINVAL; 2162 int num_backups_tried = 0; 2163 int backup_index = 0; 2164 int max_active; 2165 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND; 2166 bool create_uuid_tree; 2167 bool check_uuid_tree; 2168 2169 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info); 2170 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info); 2171 if (!tree_root || !chunk_root) { 2172 err = -ENOMEM; 2173 goto fail; 2174 } 2175 2176 ret = init_srcu_struct(&fs_info->subvol_srcu); 2177 if (ret) { 2178 err = ret; 2179 goto fail; 2180 } 2181 2182 ret = setup_bdi(fs_info, &fs_info->bdi); 2183 if (ret) { 2184 err = ret; 2185 goto fail_srcu; 2186 } 2187 2188 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0); 2189 if (ret) { 2190 err = ret; 2191 goto fail_bdi; 2192 } 2193 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE * 2194 (1 + ilog2(nr_cpu_ids)); 2195 2196 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0); 2197 if (ret) { 2198 err = ret; 2199 goto fail_dirty_metadata_bytes; 2200 } 2201 2202 ret = percpu_counter_init(&fs_info->bio_counter, 0); 2203 if (ret) { 2204 err = ret; 2205 goto fail_delalloc_bytes; 2206 } 2207 2208 fs_info->btree_inode = new_inode(sb); 2209 if (!fs_info->btree_inode) { 2210 err = -ENOMEM; 2211 goto fail_bio_counter; 2212 } 2213 2214 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS); 2215 2216 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC); 2217 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC); 2218 INIT_LIST_HEAD(&fs_info->trans_list); 2219 INIT_LIST_HEAD(&fs_info->dead_roots); 2220 INIT_LIST_HEAD(&fs_info->delayed_iputs); 2221 INIT_LIST_HEAD(&fs_info->delalloc_roots); 2222 INIT_LIST_HEAD(&fs_info->caching_block_groups); 2223 spin_lock_init(&fs_info->delalloc_root_lock); 2224 spin_lock_init(&fs_info->trans_lock); 2225 spin_lock_init(&fs_info->fs_roots_radix_lock); 2226 spin_lock_init(&fs_info->delayed_iput_lock); 2227 spin_lock_init(&fs_info->defrag_inodes_lock); 2228 spin_lock_init(&fs_info->free_chunk_lock); 2229 spin_lock_init(&fs_info->tree_mod_seq_lock); 2230 spin_lock_init(&fs_info->super_lock); 2231 spin_lock_init(&fs_info->qgroup_op_lock); 2232 spin_lock_init(&fs_info->buffer_lock); 2233 rwlock_init(&fs_info->tree_mod_log_lock); 2234 mutex_init(&fs_info->reloc_mutex); 2235 mutex_init(&fs_info->delalloc_root_mutex); 2236 seqlock_init(&fs_info->profiles_lock); 2237 2238 init_completion(&fs_info->kobj_unregister); 2239 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots); 2240 INIT_LIST_HEAD(&fs_info->space_info); 2241 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list); 2242 btrfs_mapping_init(&fs_info->mapping_tree); 2243 btrfs_init_block_rsv(&fs_info->global_block_rsv, 2244 BTRFS_BLOCK_RSV_GLOBAL); 2245 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv, 2246 BTRFS_BLOCK_RSV_DELALLOC); 2247 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS); 2248 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK); 2249 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY); 2250 btrfs_init_block_rsv(&fs_info->delayed_block_rsv, 2251 BTRFS_BLOCK_RSV_DELOPS); 2252 atomic_set(&fs_info->nr_async_submits, 0); 2253 atomic_set(&fs_info->async_delalloc_pages, 0); 2254 atomic_set(&fs_info->async_submit_draining, 0); 2255 atomic_set(&fs_info->nr_async_bios, 0); 2256 atomic_set(&fs_info->defrag_running, 0); 2257 atomic_set(&fs_info->qgroup_op_seq, 0); 2258 atomic64_set(&fs_info->tree_mod_seq, 0); 2259 fs_info->sb = sb; 2260 fs_info->max_inline = 8192 * 1024; 2261 fs_info->metadata_ratio = 0; 2262 fs_info->defrag_inodes = RB_ROOT; 2263 fs_info->free_chunk_space = 0; 2264 fs_info->tree_mod_log = RB_ROOT; 2265 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL; 2266 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64); 2267 /* readahead state */ 2268 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT); 2269 spin_lock_init(&fs_info->reada_lock); 2270 2271 fs_info->thread_pool_size = min_t(unsigned long, 2272 num_online_cpus() + 2, 8); 2273 2274 INIT_LIST_HEAD(&fs_info->ordered_roots); 2275 spin_lock_init(&fs_info->ordered_root_lock); 2276 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root), 2277 GFP_NOFS); 2278 if (!fs_info->delayed_root) { 2279 err = -ENOMEM; 2280 goto fail_iput; 2281 } 2282 btrfs_init_delayed_root(fs_info->delayed_root); 2283 2284 mutex_init(&fs_info->scrub_lock); 2285 atomic_set(&fs_info->scrubs_running, 0); 2286 atomic_set(&fs_info->scrub_pause_req, 0); 2287 atomic_set(&fs_info->scrubs_paused, 0); 2288 atomic_set(&fs_info->scrub_cancel_req, 0); 2289 init_waitqueue_head(&fs_info->replace_wait); 2290 init_waitqueue_head(&fs_info->scrub_pause_wait); 2291 fs_info->scrub_workers_refcnt = 0; 2292 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2293 fs_info->check_integrity_print_mask = 0; 2294 #endif 2295 2296 spin_lock_init(&fs_info->balance_lock); 2297 mutex_init(&fs_info->balance_mutex); 2298 atomic_set(&fs_info->balance_running, 0); 2299 atomic_set(&fs_info->balance_pause_req, 0); 2300 atomic_set(&fs_info->balance_cancel_req, 0); 2301 fs_info->balance_ctl = NULL; 2302 init_waitqueue_head(&fs_info->balance_wait_q); 2303 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work); 2304 2305 sb->s_blocksize = 4096; 2306 sb->s_blocksize_bits = blksize_bits(4096); 2307 sb->s_bdi = &fs_info->bdi; 2308 2309 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID; 2310 set_nlink(fs_info->btree_inode, 1); 2311 /* 2312 * we set the i_size on the btree inode to the max possible int. 2313 * the real end of the address space is determined by all of 2314 * the devices in the system 2315 */ 2316 fs_info->btree_inode->i_size = OFFSET_MAX; 2317 fs_info->btree_inode->i_mapping->a_ops = &btree_aops; 2318 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi; 2319 2320 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node); 2321 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree, 2322 fs_info->btree_inode->i_mapping); 2323 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0; 2324 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree); 2325 2326 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops; 2327 2328 BTRFS_I(fs_info->btree_inode)->root = tree_root; 2329 memset(&BTRFS_I(fs_info->btree_inode)->location, 0, 2330 sizeof(struct btrfs_key)); 2331 set_bit(BTRFS_INODE_DUMMY, 2332 &BTRFS_I(fs_info->btree_inode)->runtime_flags); 2333 btrfs_insert_inode_hash(fs_info->btree_inode); 2334 2335 spin_lock_init(&fs_info->block_group_cache_lock); 2336 fs_info->block_group_cache_tree = RB_ROOT; 2337 fs_info->first_logical_byte = (u64)-1; 2338 2339 extent_io_tree_init(&fs_info->freed_extents[0], 2340 fs_info->btree_inode->i_mapping); 2341 extent_io_tree_init(&fs_info->freed_extents[1], 2342 fs_info->btree_inode->i_mapping); 2343 fs_info->pinned_extents = &fs_info->freed_extents[0]; 2344 fs_info->do_barriers = 1; 2345 2346 2347 mutex_init(&fs_info->ordered_operations_mutex); 2348 mutex_init(&fs_info->ordered_extent_flush_mutex); 2349 mutex_init(&fs_info->tree_log_mutex); 2350 mutex_init(&fs_info->chunk_mutex); 2351 mutex_init(&fs_info->transaction_kthread_mutex); 2352 mutex_init(&fs_info->cleaner_mutex); 2353 mutex_init(&fs_info->volume_mutex); 2354 init_rwsem(&fs_info->commit_root_sem); 2355 init_rwsem(&fs_info->cleanup_work_sem); 2356 init_rwsem(&fs_info->subvol_sem); 2357 sema_init(&fs_info->uuid_tree_rescan_sem, 1); 2358 fs_info->dev_replace.lock_owner = 0; 2359 atomic_set(&fs_info->dev_replace.nesting_level, 0); 2360 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount); 2361 mutex_init(&fs_info->dev_replace.lock_management_lock); 2362 mutex_init(&fs_info->dev_replace.lock); 2363 2364 spin_lock_init(&fs_info->qgroup_lock); 2365 mutex_init(&fs_info->qgroup_ioctl_lock); 2366 fs_info->qgroup_tree = RB_ROOT; 2367 fs_info->qgroup_op_tree = RB_ROOT; 2368 INIT_LIST_HEAD(&fs_info->dirty_qgroups); 2369 fs_info->qgroup_seq = 1; 2370 fs_info->quota_enabled = 0; 2371 fs_info->pending_quota_state = 0; 2372 fs_info->qgroup_ulist = NULL; 2373 mutex_init(&fs_info->qgroup_rescan_lock); 2374 2375 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster); 2376 btrfs_init_free_cluster(&fs_info->data_alloc_cluster); 2377 2378 init_waitqueue_head(&fs_info->transaction_throttle); 2379 init_waitqueue_head(&fs_info->transaction_wait); 2380 init_waitqueue_head(&fs_info->transaction_blocked_wait); 2381 init_waitqueue_head(&fs_info->async_submit_wait); 2382 2383 ret = btrfs_alloc_stripe_hash_table(fs_info); 2384 if (ret) { 2385 err = ret; 2386 goto fail_alloc; 2387 } 2388 2389 __setup_root(4096, 4096, 4096, 4096, tree_root, 2390 fs_info, BTRFS_ROOT_TREE_OBJECTID); 2391 2392 invalidate_bdev(fs_devices->latest_bdev); 2393 2394 /* 2395 * Read super block and check the signature bytes only 2396 */ 2397 bh = btrfs_read_dev_super(fs_devices->latest_bdev); 2398 if (!bh) { 2399 err = -EINVAL; 2400 goto fail_alloc; 2401 } 2402 2403 /* 2404 * We want to check superblock checksum, the type is stored inside. 2405 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k). 2406 */ 2407 if (btrfs_check_super_csum(bh->b_data)) { 2408 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n"); 2409 err = -EINVAL; 2410 goto fail_alloc; 2411 } 2412 2413 /* 2414 * super_copy is zeroed at allocation time and we never touch the 2415 * following bytes up to INFO_SIZE, the checksum is calculated from 2416 * the whole block of INFO_SIZE 2417 */ 2418 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy)); 2419 memcpy(fs_info->super_for_commit, fs_info->super_copy, 2420 sizeof(*fs_info->super_for_commit)); 2421 brelse(bh); 2422 2423 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE); 2424 2425 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY); 2426 if (ret) { 2427 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n"); 2428 err = -EINVAL; 2429 goto fail_alloc; 2430 } 2431 2432 disk_super = fs_info->super_copy; 2433 if (!btrfs_super_root(disk_super)) 2434 goto fail_alloc; 2435 2436 /* check FS state, whether FS is broken. */ 2437 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR) 2438 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state); 2439 2440 /* 2441 * run through our array of backup supers and setup 2442 * our ring pointer to the oldest one 2443 */ 2444 generation = btrfs_super_generation(disk_super); 2445 find_oldest_super_backup(fs_info, generation); 2446 2447 /* 2448 * In the long term, we'll store the compression type in the super 2449 * block, and it'll be used for per file compression control. 2450 */ 2451 fs_info->compress_type = BTRFS_COMPRESS_ZLIB; 2452 2453 ret = btrfs_parse_options(tree_root, options); 2454 if (ret) { 2455 err = ret; 2456 goto fail_alloc; 2457 } 2458 2459 features = btrfs_super_incompat_flags(disk_super) & 2460 ~BTRFS_FEATURE_INCOMPAT_SUPP; 2461 if (features) { 2462 printk(KERN_ERR "BTRFS: couldn't mount because of " 2463 "unsupported optional features (%Lx).\n", 2464 features); 2465 err = -EINVAL; 2466 goto fail_alloc; 2467 } 2468 2469 if (btrfs_super_leafsize(disk_super) != 2470 btrfs_super_nodesize(disk_super)) { 2471 printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2472 "blocksizes don't match. node %d leaf %d\n", 2473 btrfs_super_nodesize(disk_super), 2474 btrfs_super_leafsize(disk_super)); 2475 err = -EINVAL; 2476 goto fail_alloc; 2477 } 2478 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) { 2479 printk(KERN_ERR "BTRFS: couldn't mount because metadata " 2480 "blocksize (%d) was too large\n", 2481 btrfs_super_leafsize(disk_super)); 2482 err = -EINVAL; 2483 goto fail_alloc; 2484 } 2485 2486 features = btrfs_super_incompat_flags(disk_super); 2487 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF; 2488 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO) 2489 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO; 2490 2491 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA) 2492 printk(KERN_ERR "BTRFS: has skinny extents\n"); 2493 2494 /* 2495 * flag our filesystem as having big metadata blocks if 2496 * they are bigger than the page size 2497 */ 2498 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) { 2499 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA)) 2500 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n"); 2501 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA; 2502 } 2503 2504 nodesize = btrfs_super_nodesize(disk_super); 2505 leafsize = btrfs_super_leafsize(disk_super); 2506 sectorsize = btrfs_super_sectorsize(disk_super); 2507 stripesize = btrfs_super_stripesize(disk_super); 2508 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids)); 2509 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids)); 2510 2511 /* 2512 * mixed block groups end up with duplicate but slightly offset 2513 * extent buffers for the same range. It leads to corruptions 2514 */ 2515 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && 2516 (sectorsize != leafsize)) { 2517 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes " 2518 "are not allowed for mixed block groups on %s\n", 2519 sb->s_id); 2520 goto fail_alloc; 2521 } 2522 2523 /* 2524 * Needn't use the lock because there is no other task which will 2525 * update the flag. 2526 */ 2527 btrfs_set_super_incompat_flags(disk_super, features); 2528 2529 features = btrfs_super_compat_ro_flags(disk_super) & 2530 ~BTRFS_FEATURE_COMPAT_RO_SUPP; 2531 if (!(sb->s_flags & MS_RDONLY) && features) { 2532 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of " 2533 "unsupported option features (%Lx).\n", 2534 features); 2535 err = -EINVAL; 2536 goto fail_alloc; 2537 } 2538 2539 max_active = fs_info->thread_pool_size; 2540 2541 fs_info->workers = 2542 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI, 2543 max_active, 16); 2544 2545 fs_info->delalloc_workers = 2546 btrfs_alloc_workqueue("delalloc", flags, max_active, 2); 2547 2548 fs_info->flush_workers = 2549 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0); 2550 2551 fs_info->caching_workers = 2552 btrfs_alloc_workqueue("cache", flags, max_active, 0); 2553 2554 /* 2555 * a higher idle thresh on the submit workers makes it much more 2556 * likely that bios will be send down in a sane order to the 2557 * devices 2558 */ 2559 fs_info->submit_workers = 2560 btrfs_alloc_workqueue("submit", flags, 2561 min_t(u64, fs_devices->num_devices, 2562 max_active), 64); 2563 2564 fs_info->fixup_workers = 2565 btrfs_alloc_workqueue("fixup", flags, 1, 0); 2566 2567 /* 2568 * endios are largely parallel and should have a very 2569 * low idle thresh 2570 */ 2571 fs_info->endio_workers = 2572 btrfs_alloc_workqueue("endio", flags, max_active, 4); 2573 fs_info->endio_meta_workers = 2574 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4); 2575 fs_info->endio_meta_write_workers = 2576 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2); 2577 fs_info->endio_raid56_workers = 2578 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4); 2579 fs_info->rmw_workers = 2580 btrfs_alloc_workqueue("rmw", flags, max_active, 2); 2581 fs_info->endio_write_workers = 2582 btrfs_alloc_workqueue("endio-write", flags, max_active, 2); 2583 fs_info->endio_freespace_worker = 2584 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0); 2585 fs_info->delayed_workers = 2586 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0); 2587 fs_info->readahead_workers = 2588 btrfs_alloc_workqueue("readahead", flags, max_active, 2); 2589 fs_info->qgroup_rescan_workers = 2590 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0); 2591 fs_info->extent_workers = 2592 btrfs_alloc_workqueue("extent-refs", flags, 2593 min_t(u64, fs_devices->num_devices, 2594 max_active), 8); 2595 2596 if (!(fs_info->workers && fs_info->delalloc_workers && 2597 fs_info->submit_workers && fs_info->flush_workers && 2598 fs_info->endio_workers && fs_info->endio_meta_workers && 2599 fs_info->endio_meta_write_workers && 2600 fs_info->endio_write_workers && fs_info->endio_raid56_workers && 2601 fs_info->endio_freespace_worker && fs_info->rmw_workers && 2602 fs_info->caching_workers && fs_info->readahead_workers && 2603 fs_info->fixup_workers && fs_info->delayed_workers && 2604 fs_info->fixup_workers && fs_info->extent_workers && 2605 fs_info->qgroup_rescan_workers)) { 2606 err = -ENOMEM; 2607 goto fail_sb_buffer; 2608 } 2609 2610 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super); 2611 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages, 2612 4 * 1024 * 1024 / PAGE_CACHE_SIZE); 2613 2614 tree_root->nodesize = nodesize; 2615 tree_root->leafsize = leafsize; 2616 tree_root->sectorsize = sectorsize; 2617 tree_root->stripesize = stripesize; 2618 2619 sb->s_blocksize = sectorsize; 2620 sb->s_blocksize_bits = blksize_bits(sectorsize); 2621 2622 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) { 2623 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id); 2624 goto fail_sb_buffer; 2625 } 2626 2627 if (sectorsize != PAGE_SIZE) { 2628 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) " 2629 "found on %s\n", (unsigned long)sectorsize, sb->s_id); 2630 goto fail_sb_buffer; 2631 } 2632 2633 mutex_lock(&fs_info->chunk_mutex); 2634 ret = btrfs_read_sys_array(tree_root); 2635 mutex_unlock(&fs_info->chunk_mutex); 2636 if (ret) { 2637 printk(KERN_WARNING "BTRFS: failed to read the system " 2638 "array on %s\n", sb->s_id); 2639 goto fail_sb_buffer; 2640 } 2641 2642 blocksize = btrfs_level_size(tree_root, 2643 btrfs_super_chunk_root_level(disk_super)); 2644 generation = btrfs_super_chunk_root_generation(disk_super); 2645 2646 __setup_root(nodesize, leafsize, sectorsize, stripesize, 2647 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID); 2648 2649 chunk_root->node = read_tree_block(chunk_root, 2650 btrfs_super_chunk_root(disk_super), 2651 blocksize, generation); 2652 if (!chunk_root->node || 2653 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) { 2654 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n", 2655 sb->s_id); 2656 goto fail_tree_roots; 2657 } 2658 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node); 2659 chunk_root->commit_root = btrfs_root_node(chunk_root); 2660 2661 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid, 2662 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE); 2663 2664 ret = btrfs_read_chunk_tree(chunk_root); 2665 if (ret) { 2666 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n", 2667 sb->s_id); 2668 goto fail_tree_roots; 2669 } 2670 2671 /* 2672 * keep the device that is marked to be the target device for the 2673 * dev_replace procedure 2674 */ 2675 btrfs_close_extra_devices(fs_info, fs_devices, 0); 2676 2677 if (!fs_devices->latest_bdev) { 2678 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n", 2679 sb->s_id); 2680 goto fail_tree_roots; 2681 } 2682 2683 retry_root_backup: 2684 blocksize = btrfs_level_size(tree_root, 2685 btrfs_super_root_level(disk_super)); 2686 generation = btrfs_super_generation(disk_super); 2687 2688 tree_root->node = read_tree_block(tree_root, 2689 btrfs_super_root(disk_super), 2690 blocksize, generation); 2691 if (!tree_root->node || 2692 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) { 2693 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n", 2694 sb->s_id); 2695 2696 goto recovery_tree_root; 2697 } 2698 2699 btrfs_set_root_node(&tree_root->root_item, tree_root->node); 2700 tree_root->commit_root = btrfs_root_node(tree_root); 2701 btrfs_set_root_refs(&tree_root->root_item, 1); 2702 2703 location.objectid = BTRFS_EXTENT_TREE_OBJECTID; 2704 location.type = BTRFS_ROOT_ITEM_KEY; 2705 location.offset = 0; 2706 2707 extent_root = btrfs_read_tree_root(tree_root, &location); 2708 if (IS_ERR(extent_root)) { 2709 ret = PTR_ERR(extent_root); 2710 goto recovery_tree_root; 2711 } 2712 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state); 2713 fs_info->extent_root = extent_root; 2714 2715 location.objectid = BTRFS_DEV_TREE_OBJECTID; 2716 dev_root = btrfs_read_tree_root(tree_root, &location); 2717 if (IS_ERR(dev_root)) { 2718 ret = PTR_ERR(dev_root); 2719 goto recovery_tree_root; 2720 } 2721 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state); 2722 fs_info->dev_root = dev_root; 2723 btrfs_init_devices_late(fs_info); 2724 2725 location.objectid = BTRFS_CSUM_TREE_OBJECTID; 2726 csum_root = btrfs_read_tree_root(tree_root, &location); 2727 if (IS_ERR(csum_root)) { 2728 ret = PTR_ERR(csum_root); 2729 goto recovery_tree_root; 2730 } 2731 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state); 2732 fs_info->csum_root = csum_root; 2733 2734 location.objectid = BTRFS_QUOTA_TREE_OBJECTID; 2735 quota_root = btrfs_read_tree_root(tree_root, &location); 2736 if (!IS_ERR(quota_root)) { 2737 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state); 2738 fs_info->quota_enabled = 1; 2739 fs_info->pending_quota_state = 1; 2740 fs_info->quota_root = quota_root; 2741 } 2742 2743 location.objectid = BTRFS_UUID_TREE_OBJECTID; 2744 uuid_root = btrfs_read_tree_root(tree_root, &location); 2745 if (IS_ERR(uuid_root)) { 2746 ret = PTR_ERR(uuid_root); 2747 if (ret != -ENOENT) 2748 goto recovery_tree_root; 2749 create_uuid_tree = true; 2750 check_uuid_tree = false; 2751 } else { 2752 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state); 2753 fs_info->uuid_root = uuid_root; 2754 create_uuid_tree = false; 2755 check_uuid_tree = 2756 generation != btrfs_super_uuid_tree_generation(disk_super); 2757 } 2758 2759 fs_info->generation = generation; 2760 fs_info->last_trans_committed = generation; 2761 2762 ret = btrfs_recover_balance(fs_info); 2763 if (ret) { 2764 printk(KERN_WARNING "BTRFS: failed to recover balance\n"); 2765 goto fail_block_groups; 2766 } 2767 2768 ret = btrfs_init_dev_stats(fs_info); 2769 if (ret) { 2770 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n", 2771 ret); 2772 goto fail_block_groups; 2773 } 2774 2775 ret = btrfs_init_dev_replace(fs_info); 2776 if (ret) { 2777 pr_err("BTRFS: failed to init dev_replace: %d\n", ret); 2778 goto fail_block_groups; 2779 } 2780 2781 btrfs_close_extra_devices(fs_info, fs_devices, 1); 2782 2783 ret = btrfs_sysfs_add_one(fs_info); 2784 if (ret) { 2785 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret); 2786 goto fail_block_groups; 2787 } 2788 2789 ret = btrfs_init_space_info(fs_info); 2790 if (ret) { 2791 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret); 2792 goto fail_sysfs; 2793 } 2794 2795 ret = btrfs_read_block_groups(extent_root); 2796 if (ret) { 2797 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret); 2798 goto fail_sysfs; 2799 } 2800 fs_info->num_tolerated_disk_barrier_failures = 2801 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); 2802 if (fs_info->fs_devices->missing_devices > 2803 fs_info->num_tolerated_disk_barrier_failures && 2804 !(sb->s_flags & MS_RDONLY)) { 2805 printk(KERN_WARNING "BTRFS: " 2806 "too many missing devices, writeable mount is not allowed\n"); 2807 goto fail_sysfs; 2808 } 2809 2810 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root, 2811 "btrfs-cleaner"); 2812 if (IS_ERR(fs_info->cleaner_kthread)) 2813 goto fail_sysfs; 2814 2815 fs_info->transaction_kthread = kthread_run(transaction_kthread, 2816 tree_root, 2817 "btrfs-transaction"); 2818 if (IS_ERR(fs_info->transaction_kthread)) 2819 goto fail_cleaner; 2820 2821 if (!btrfs_test_opt(tree_root, SSD) && 2822 !btrfs_test_opt(tree_root, NOSSD) && 2823 !fs_info->fs_devices->rotating) { 2824 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD " 2825 "mode\n"); 2826 btrfs_set_opt(fs_info->mount_opt, SSD); 2827 } 2828 2829 /* Set the real inode map cache flag */ 2830 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE)) 2831 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE); 2832 2833 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 2834 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) { 2835 ret = btrfsic_mount(tree_root, fs_devices, 2836 btrfs_test_opt(tree_root, 2837 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ? 2838 1 : 0, 2839 fs_info->check_integrity_print_mask); 2840 if (ret) 2841 printk(KERN_WARNING "BTRFS: failed to initialize" 2842 " integrity check module %s\n", sb->s_id); 2843 } 2844 #endif 2845 ret = btrfs_read_qgroup_config(fs_info); 2846 if (ret) 2847 goto fail_trans_kthread; 2848 2849 /* do not make disk changes in broken FS */ 2850 if (btrfs_super_log_root(disk_super) != 0) { 2851 u64 bytenr = btrfs_super_log_root(disk_super); 2852 2853 if (fs_devices->rw_devices == 0) { 2854 printk(KERN_WARNING "BTRFS: log replay required " 2855 "on RO media\n"); 2856 err = -EIO; 2857 goto fail_qgroup; 2858 } 2859 blocksize = 2860 btrfs_level_size(tree_root, 2861 btrfs_super_log_root_level(disk_super)); 2862 2863 log_tree_root = btrfs_alloc_root(fs_info); 2864 if (!log_tree_root) { 2865 err = -ENOMEM; 2866 goto fail_qgroup; 2867 } 2868 2869 __setup_root(nodesize, leafsize, sectorsize, stripesize, 2870 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID); 2871 2872 log_tree_root->node = read_tree_block(tree_root, bytenr, 2873 blocksize, 2874 generation + 1); 2875 if (!log_tree_root->node || 2876 !extent_buffer_uptodate(log_tree_root->node)) { 2877 printk(KERN_ERR "BTRFS: failed to read log tree\n"); 2878 free_extent_buffer(log_tree_root->node); 2879 kfree(log_tree_root); 2880 goto fail_qgroup; 2881 } 2882 /* returns with log_tree_root freed on success */ 2883 ret = btrfs_recover_log_trees(log_tree_root); 2884 if (ret) { 2885 btrfs_error(tree_root->fs_info, ret, 2886 "Failed to recover log tree"); 2887 free_extent_buffer(log_tree_root->node); 2888 kfree(log_tree_root); 2889 goto fail_qgroup; 2890 } 2891 2892 if (sb->s_flags & MS_RDONLY) { 2893 ret = btrfs_commit_super(tree_root); 2894 if (ret) 2895 goto fail_qgroup; 2896 } 2897 } 2898 2899 ret = btrfs_find_orphan_roots(tree_root); 2900 if (ret) 2901 goto fail_qgroup; 2902 2903 if (!(sb->s_flags & MS_RDONLY)) { 2904 ret = btrfs_cleanup_fs_roots(fs_info); 2905 if (ret) 2906 goto fail_qgroup; 2907 2908 mutex_lock(&fs_info->cleaner_mutex); 2909 ret = btrfs_recover_relocation(tree_root); 2910 mutex_unlock(&fs_info->cleaner_mutex); 2911 if (ret < 0) { 2912 printk(KERN_WARNING 2913 "BTRFS: failed to recover relocation\n"); 2914 err = -EINVAL; 2915 goto fail_qgroup; 2916 } 2917 } 2918 2919 location.objectid = BTRFS_FS_TREE_OBJECTID; 2920 location.type = BTRFS_ROOT_ITEM_KEY; 2921 location.offset = 0; 2922 2923 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location); 2924 if (IS_ERR(fs_info->fs_root)) { 2925 err = PTR_ERR(fs_info->fs_root); 2926 goto fail_qgroup; 2927 } 2928 2929 if (sb->s_flags & MS_RDONLY) 2930 return 0; 2931 2932 down_read(&fs_info->cleanup_work_sem); 2933 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) || 2934 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) { 2935 up_read(&fs_info->cleanup_work_sem); 2936 close_ctree(tree_root); 2937 return ret; 2938 } 2939 up_read(&fs_info->cleanup_work_sem); 2940 2941 ret = btrfs_resume_balance_async(fs_info); 2942 if (ret) { 2943 printk(KERN_WARNING "BTRFS: failed to resume balance\n"); 2944 close_ctree(tree_root); 2945 return ret; 2946 } 2947 2948 ret = btrfs_resume_dev_replace_async(fs_info); 2949 if (ret) { 2950 pr_warn("BTRFS: failed to resume dev_replace\n"); 2951 close_ctree(tree_root); 2952 return ret; 2953 } 2954 2955 btrfs_qgroup_rescan_resume(fs_info); 2956 2957 if (create_uuid_tree) { 2958 pr_info("BTRFS: creating UUID tree\n"); 2959 ret = btrfs_create_uuid_tree(fs_info); 2960 if (ret) { 2961 pr_warn("BTRFS: failed to create the UUID tree %d\n", 2962 ret); 2963 close_ctree(tree_root); 2964 return ret; 2965 } 2966 } else if (check_uuid_tree || 2967 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) { 2968 pr_info("BTRFS: checking UUID tree\n"); 2969 ret = btrfs_check_uuid_tree(fs_info); 2970 if (ret) { 2971 pr_warn("BTRFS: failed to check the UUID tree %d\n", 2972 ret); 2973 close_ctree(tree_root); 2974 return ret; 2975 } 2976 } else { 2977 fs_info->update_uuid_tree_gen = 1; 2978 } 2979 2980 return 0; 2981 2982 fail_qgroup: 2983 btrfs_free_qgroup_config(fs_info); 2984 fail_trans_kthread: 2985 kthread_stop(fs_info->transaction_kthread); 2986 btrfs_cleanup_transaction(fs_info->tree_root); 2987 btrfs_free_fs_roots(fs_info); 2988 fail_cleaner: 2989 kthread_stop(fs_info->cleaner_kthread); 2990 2991 /* 2992 * make sure we're done with the btree inode before we stop our 2993 * kthreads 2994 */ 2995 filemap_write_and_wait(fs_info->btree_inode->i_mapping); 2996 2997 fail_sysfs: 2998 btrfs_sysfs_remove_one(fs_info); 2999 3000 fail_block_groups: 3001 btrfs_put_block_group_cache(fs_info); 3002 btrfs_free_block_groups(fs_info); 3003 3004 fail_tree_roots: 3005 free_root_pointers(fs_info, 1); 3006 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3007 3008 fail_sb_buffer: 3009 btrfs_stop_all_workers(fs_info); 3010 fail_alloc: 3011 fail_iput: 3012 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3013 3014 iput(fs_info->btree_inode); 3015 fail_bio_counter: 3016 percpu_counter_destroy(&fs_info->bio_counter); 3017 fail_delalloc_bytes: 3018 percpu_counter_destroy(&fs_info->delalloc_bytes); 3019 fail_dirty_metadata_bytes: 3020 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3021 fail_bdi: 3022 bdi_destroy(&fs_info->bdi); 3023 fail_srcu: 3024 cleanup_srcu_struct(&fs_info->subvol_srcu); 3025 fail: 3026 btrfs_free_stripe_hash_table(fs_info); 3027 btrfs_close_devices(fs_info->fs_devices); 3028 return err; 3029 3030 recovery_tree_root: 3031 if (!btrfs_test_opt(tree_root, RECOVERY)) 3032 goto fail_tree_roots; 3033 3034 free_root_pointers(fs_info, 0); 3035 3036 /* don't use the log in recovery mode, it won't be valid */ 3037 btrfs_set_super_log_root(disk_super, 0); 3038 3039 /* we can't trust the free space cache either */ 3040 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE); 3041 3042 ret = next_root_backup(fs_info, fs_info->super_copy, 3043 &num_backups_tried, &backup_index); 3044 if (ret == -1) 3045 goto fail_block_groups; 3046 goto retry_root_backup; 3047 } 3048 3049 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate) 3050 { 3051 if (uptodate) { 3052 set_buffer_uptodate(bh); 3053 } else { 3054 struct btrfs_device *device = (struct btrfs_device *) 3055 bh->b_private; 3056 3057 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to " 3058 "I/O error on %s\n", 3059 rcu_str_deref(device->name)); 3060 /* note, we dont' set_buffer_write_io_error because we have 3061 * our own ways of dealing with the IO errors 3062 */ 3063 clear_buffer_uptodate(bh); 3064 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS); 3065 } 3066 unlock_buffer(bh); 3067 put_bh(bh); 3068 } 3069 3070 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev) 3071 { 3072 struct buffer_head *bh; 3073 struct buffer_head *latest = NULL; 3074 struct btrfs_super_block *super; 3075 int i; 3076 u64 transid = 0; 3077 u64 bytenr; 3078 3079 /* we would like to check all the supers, but that would make 3080 * a btrfs mount succeed after a mkfs from a different FS. 3081 * So, we need to add a special mount option to scan for 3082 * later supers, using BTRFS_SUPER_MIRROR_MAX instead 3083 */ 3084 for (i = 0; i < 1; i++) { 3085 bytenr = btrfs_sb_offset(i); 3086 if (bytenr + BTRFS_SUPER_INFO_SIZE >= 3087 i_size_read(bdev->bd_inode)) 3088 break; 3089 bh = __bread(bdev, bytenr / 4096, 3090 BTRFS_SUPER_INFO_SIZE); 3091 if (!bh) 3092 continue; 3093 3094 super = (struct btrfs_super_block *)bh->b_data; 3095 if (btrfs_super_bytenr(super) != bytenr || 3096 btrfs_super_magic(super) != BTRFS_MAGIC) { 3097 brelse(bh); 3098 continue; 3099 } 3100 3101 if (!latest || btrfs_super_generation(super) > transid) { 3102 brelse(latest); 3103 latest = bh; 3104 transid = btrfs_super_generation(super); 3105 } else { 3106 brelse(bh); 3107 } 3108 } 3109 return latest; 3110 } 3111 3112 /* 3113 * this should be called twice, once with wait == 0 and 3114 * once with wait == 1. When wait == 0 is done, all the buffer heads 3115 * we write are pinned. 3116 * 3117 * They are released when wait == 1 is done. 3118 * max_mirrors must be the same for both runs, and it indicates how 3119 * many supers on this one device should be written. 3120 * 3121 * max_mirrors == 0 means to write them all. 3122 */ 3123 static int write_dev_supers(struct btrfs_device *device, 3124 struct btrfs_super_block *sb, 3125 int do_barriers, int wait, int max_mirrors) 3126 { 3127 struct buffer_head *bh; 3128 int i; 3129 int ret; 3130 int errors = 0; 3131 u32 crc; 3132 u64 bytenr; 3133 3134 if (max_mirrors == 0) 3135 max_mirrors = BTRFS_SUPER_MIRROR_MAX; 3136 3137 for (i = 0; i < max_mirrors; i++) { 3138 bytenr = btrfs_sb_offset(i); 3139 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes) 3140 break; 3141 3142 if (wait) { 3143 bh = __find_get_block(device->bdev, bytenr / 4096, 3144 BTRFS_SUPER_INFO_SIZE); 3145 if (!bh) { 3146 errors++; 3147 continue; 3148 } 3149 wait_on_buffer(bh); 3150 if (!buffer_uptodate(bh)) 3151 errors++; 3152 3153 /* drop our reference */ 3154 brelse(bh); 3155 3156 /* drop the reference from the wait == 0 run */ 3157 brelse(bh); 3158 continue; 3159 } else { 3160 btrfs_set_super_bytenr(sb, bytenr); 3161 3162 crc = ~(u32)0; 3163 crc = btrfs_csum_data((char *)sb + 3164 BTRFS_CSUM_SIZE, crc, 3165 BTRFS_SUPER_INFO_SIZE - 3166 BTRFS_CSUM_SIZE); 3167 btrfs_csum_final(crc, sb->csum); 3168 3169 /* 3170 * one reference for us, and we leave it for the 3171 * caller 3172 */ 3173 bh = __getblk(device->bdev, bytenr / 4096, 3174 BTRFS_SUPER_INFO_SIZE); 3175 if (!bh) { 3176 printk(KERN_ERR "BTRFS: couldn't get super " 3177 "buffer head for bytenr %Lu\n", bytenr); 3178 errors++; 3179 continue; 3180 } 3181 3182 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE); 3183 3184 /* one reference for submit_bh */ 3185 get_bh(bh); 3186 3187 set_buffer_uptodate(bh); 3188 lock_buffer(bh); 3189 bh->b_end_io = btrfs_end_buffer_write_sync; 3190 bh->b_private = device; 3191 } 3192 3193 /* 3194 * we fua the first super. The others we allow 3195 * to go down lazy. 3196 */ 3197 if (i == 0) 3198 ret = btrfsic_submit_bh(WRITE_FUA, bh); 3199 else 3200 ret = btrfsic_submit_bh(WRITE_SYNC, bh); 3201 if (ret) 3202 errors++; 3203 } 3204 return errors < i ? 0 : -1; 3205 } 3206 3207 /* 3208 * endio for the write_dev_flush, this will wake anyone waiting 3209 * for the barrier when it is done 3210 */ 3211 static void btrfs_end_empty_barrier(struct bio *bio, int err) 3212 { 3213 if (err) { 3214 if (err == -EOPNOTSUPP) 3215 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 3216 clear_bit(BIO_UPTODATE, &bio->bi_flags); 3217 } 3218 if (bio->bi_private) 3219 complete(bio->bi_private); 3220 bio_put(bio); 3221 } 3222 3223 /* 3224 * trigger flushes for one the devices. If you pass wait == 0, the flushes are 3225 * sent down. With wait == 1, it waits for the previous flush. 3226 * 3227 * any device where the flush fails with eopnotsupp are flagged as not-barrier 3228 * capable 3229 */ 3230 static int write_dev_flush(struct btrfs_device *device, int wait) 3231 { 3232 struct bio *bio; 3233 int ret = 0; 3234 3235 if (device->nobarriers) 3236 return 0; 3237 3238 if (wait) { 3239 bio = device->flush_bio; 3240 if (!bio) 3241 return 0; 3242 3243 wait_for_completion(&device->flush_wait); 3244 3245 if (bio_flagged(bio, BIO_EOPNOTSUPP)) { 3246 printk_in_rcu("BTRFS: disabling barriers on dev %s\n", 3247 rcu_str_deref(device->name)); 3248 device->nobarriers = 1; 3249 } else if (!bio_flagged(bio, BIO_UPTODATE)) { 3250 ret = -EIO; 3251 btrfs_dev_stat_inc_and_print(device, 3252 BTRFS_DEV_STAT_FLUSH_ERRS); 3253 } 3254 3255 /* drop the reference from the wait == 0 run */ 3256 bio_put(bio); 3257 device->flush_bio = NULL; 3258 3259 return ret; 3260 } 3261 3262 /* 3263 * one reference for us, and we leave it for the 3264 * caller 3265 */ 3266 device->flush_bio = NULL; 3267 bio = btrfs_io_bio_alloc(GFP_NOFS, 0); 3268 if (!bio) 3269 return -ENOMEM; 3270 3271 bio->bi_end_io = btrfs_end_empty_barrier; 3272 bio->bi_bdev = device->bdev; 3273 init_completion(&device->flush_wait); 3274 bio->bi_private = &device->flush_wait; 3275 device->flush_bio = bio; 3276 3277 bio_get(bio); 3278 btrfsic_submit_bio(WRITE_FLUSH, bio); 3279 3280 return 0; 3281 } 3282 3283 /* 3284 * send an empty flush down to each device in parallel, 3285 * then wait for them 3286 */ 3287 static int barrier_all_devices(struct btrfs_fs_info *info) 3288 { 3289 struct list_head *head; 3290 struct btrfs_device *dev; 3291 int errors_send = 0; 3292 int errors_wait = 0; 3293 int ret; 3294 3295 /* send down all the barriers */ 3296 head = &info->fs_devices->devices; 3297 list_for_each_entry_rcu(dev, head, dev_list) { 3298 if (dev->missing) 3299 continue; 3300 if (!dev->bdev) { 3301 errors_send++; 3302 continue; 3303 } 3304 if (!dev->in_fs_metadata || !dev->writeable) 3305 continue; 3306 3307 ret = write_dev_flush(dev, 0); 3308 if (ret) 3309 errors_send++; 3310 } 3311 3312 /* wait for all the barriers */ 3313 list_for_each_entry_rcu(dev, head, dev_list) { 3314 if (dev->missing) 3315 continue; 3316 if (!dev->bdev) { 3317 errors_wait++; 3318 continue; 3319 } 3320 if (!dev->in_fs_metadata || !dev->writeable) 3321 continue; 3322 3323 ret = write_dev_flush(dev, 1); 3324 if (ret) 3325 errors_wait++; 3326 } 3327 if (errors_send > info->num_tolerated_disk_barrier_failures || 3328 errors_wait > info->num_tolerated_disk_barrier_failures) 3329 return -EIO; 3330 return 0; 3331 } 3332 3333 int btrfs_calc_num_tolerated_disk_barrier_failures( 3334 struct btrfs_fs_info *fs_info) 3335 { 3336 struct btrfs_ioctl_space_info space; 3337 struct btrfs_space_info *sinfo; 3338 u64 types[] = {BTRFS_BLOCK_GROUP_DATA, 3339 BTRFS_BLOCK_GROUP_SYSTEM, 3340 BTRFS_BLOCK_GROUP_METADATA, 3341 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA}; 3342 int num_types = 4; 3343 int i; 3344 int c; 3345 int num_tolerated_disk_barrier_failures = 3346 (int)fs_info->fs_devices->num_devices; 3347 3348 for (i = 0; i < num_types; i++) { 3349 struct btrfs_space_info *tmp; 3350 3351 sinfo = NULL; 3352 rcu_read_lock(); 3353 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) { 3354 if (tmp->flags == types[i]) { 3355 sinfo = tmp; 3356 break; 3357 } 3358 } 3359 rcu_read_unlock(); 3360 3361 if (!sinfo) 3362 continue; 3363 3364 down_read(&sinfo->groups_sem); 3365 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) { 3366 if (!list_empty(&sinfo->block_groups[c])) { 3367 u64 flags; 3368 3369 btrfs_get_block_group_info( 3370 &sinfo->block_groups[c], &space); 3371 if (space.total_bytes == 0 || 3372 space.used_bytes == 0) 3373 continue; 3374 flags = space.flags; 3375 /* 3376 * return 3377 * 0: if dup, single or RAID0 is configured for 3378 * any of metadata, system or data, else 3379 * 1: if RAID5 is configured, or if RAID1 or 3380 * RAID10 is configured and only two mirrors 3381 * are used, else 3382 * 2: if RAID6 is configured, else 3383 * num_mirrors - 1: if RAID1 or RAID10 is 3384 * configured and more than 3385 * 2 mirrors are used. 3386 */ 3387 if (num_tolerated_disk_barrier_failures > 0 && 3388 ((flags & (BTRFS_BLOCK_GROUP_DUP | 3389 BTRFS_BLOCK_GROUP_RAID0)) || 3390 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) 3391 == 0))) 3392 num_tolerated_disk_barrier_failures = 0; 3393 else if (num_tolerated_disk_barrier_failures > 1) { 3394 if (flags & (BTRFS_BLOCK_GROUP_RAID1 | 3395 BTRFS_BLOCK_GROUP_RAID5 | 3396 BTRFS_BLOCK_GROUP_RAID10)) { 3397 num_tolerated_disk_barrier_failures = 1; 3398 } else if (flags & 3399 BTRFS_BLOCK_GROUP_RAID6) { 3400 num_tolerated_disk_barrier_failures = 2; 3401 } 3402 } 3403 } 3404 } 3405 up_read(&sinfo->groups_sem); 3406 } 3407 3408 return num_tolerated_disk_barrier_failures; 3409 } 3410 3411 static int write_all_supers(struct btrfs_root *root, int max_mirrors) 3412 { 3413 struct list_head *head; 3414 struct btrfs_device *dev; 3415 struct btrfs_super_block *sb; 3416 struct btrfs_dev_item *dev_item; 3417 int ret; 3418 int do_barriers; 3419 int max_errors; 3420 int total_errors = 0; 3421 u64 flags; 3422 3423 do_barriers = !btrfs_test_opt(root, NOBARRIER); 3424 backup_super_roots(root->fs_info); 3425 3426 sb = root->fs_info->super_for_commit; 3427 dev_item = &sb->dev_item; 3428 3429 mutex_lock(&root->fs_info->fs_devices->device_list_mutex); 3430 head = &root->fs_info->fs_devices->devices; 3431 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1; 3432 3433 if (do_barriers) { 3434 ret = barrier_all_devices(root->fs_info); 3435 if (ret) { 3436 mutex_unlock( 3437 &root->fs_info->fs_devices->device_list_mutex); 3438 btrfs_error(root->fs_info, ret, 3439 "errors while submitting device barriers."); 3440 return ret; 3441 } 3442 } 3443 3444 list_for_each_entry_rcu(dev, head, dev_list) { 3445 if (!dev->bdev) { 3446 total_errors++; 3447 continue; 3448 } 3449 if (!dev->in_fs_metadata || !dev->writeable) 3450 continue; 3451 3452 btrfs_set_stack_device_generation(dev_item, 0); 3453 btrfs_set_stack_device_type(dev_item, dev->type); 3454 btrfs_set_stack_device_id(dev_item, dev->devid); 3455 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes); 3456 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used); 3457 btrfs_set_stack_device_io_align(dev_item, dev->io_align); 3458 btrfs_set_stack_device_io_width(dev_item, dev->io_width); 3459 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size); 3460 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE); 3461 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE); 3462 3463 flags = btrfs_super_flags(sb); 3464 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN); 3465 3466 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors); 3467 if (ret) 3468 total_errors++; 3469 } 3470 if (total_errors > max_errors) { 3471 btrfs_err(root->fs_info, "%d errors while writing supers", 3472 total_errors); 3473 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3474 3475 /* FUA is masked off if unsupported and can't be the reason */ 3476 btrfs_error(root->fs_info, -EIO, 3477 "%d errors while writing supers", total_errors); 3478 return -EIO; 3479 } 3480 3481 total_errors = 0; 3482 list_for_each_entry_rcu(dev, head, dev_list) { 3483 if (!dev->bdev) 3484 continue; 3485 if (!dev->in_fs_metadata || !dev->writeable) 3486 continue; 3487 3488 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors); 3489 if (ret) 3490 total_errors++; 3491 } 3492 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); 3493 if (total_errors > max_errors) { 3494 btrfs_error(root->fs_info, -EIO, 3495 "%d errors while writing supers", total_errors); 3496 return -EIO; 3497 } 3498 return 0; 3499 } 3500 3501 int write_ctree_super(struct btrfs_trans_handle *trans, 3502 struct btrfs_root *root, int max_mirrors) 3503 { 3504 return write_all_supers(root, max_mirrors); 3505 } 3506 3507 /* Drop a fs root from the radix tree and free it. */ 3508 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info, 3509 struct btrfs_root *root) 3510 { 3511 spin_lock(&fs_info->fs_roots_radix_lock); 3512 radix_tree_delete(&fs_info->fs_roots_radix, 3513 (unsigned long)root->root_key.objectid); 3514 spin_unlock(&fs_info->fs_roots_radix_lock); 3515 3516 if (btrfs_root_refs(&root->root_item) == 0) 3517 synchronize_srcu(&fs_info->subvol_srcu); 3518 3519 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 3520 btrfs_free_log(NULL, root); 3521 3522 if (root->free_ino_pinned) 3523 __btrfs_remove_free_space_cache(root->free_ino_pinned); 3524 if (root->free_ino_ctl) 3525 __btrfs_remove_free_space_cache(root->free_ino_ctl); 3526 free_fs_root(root); 3527 } 3528 3529 static void free_fs_root(struct btrfs_root *root) 3530 { 3531 iput(root->cache_inode); 3532 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree)); 3533 btrfs_free_block_rsv(root, root->orphan_block_rsv); 3534 root->orphan_block_rsv = NULL; 3535 if (root->anon_dev) 3536 free_anon_bdev(root->anon_dev); 3537 if (root->subv_writers) 3538 btrfs_free_subvolume_writers(root->subv_writers); 3539 free_extent_buffer(root->node); 3540 free_extent_buffer(root->commit_root); 3541 kfree(root->free_ino_ctl); 3542 kfree(root->free_ino_pinned); 3543 kfree(root->name); 3544 btrfs_put_fs_root(root); 3545 } 3546 3547 void btrfs_free_fs_root(struct btrfs_root *root) 3548 { 3549 free_fs_root(root); 3550 } 3551 3552 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info) 3553 { 3554 u64 root_objectid = 0; 3555 struct btrfs_root *gang[8]; 3556 int i = 0; 3557 int err = 0; 3558 unsigned int ret = 0; 3559 int index; 3560 3561 while (1) { 3562 index = srcu_read_lock(&fs_info->subvol_srcu); 3563 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix, 3564 (void **)gang, root_objectid, 3565 ARRAY_SIZE(gang)); 3566 if (!ret) { 3567 srcu_read_unlock(&fs_info->subvol_srcu, index); 3568 break; 3569 } 3570 root_objectid = gang[ret - 1]->root_key.objectid + 1; 3571 3572 for (i = 0; i < ret; i++) { 3573 /* Avoid to grab roots in dead_roots */ 3574 if (btrfs_root_refs(&gang[i]->root_item) == 0) { 3575 gang[i] = NULL; 3576 continue; 3577 } 3578 /* grab all the search result for later use */ 3579 gang[i] = btrfs_grab_fs_root(gang[i]); 3580 } 3581 srcu_read_unlock(&fs_info->subvol_srcu, index); 3582 3583 for (i = 0; i < ret; i++) { 3584 if (!gang[i]) 3585 continue; 3586 root_objectid = gang[i]->root_key.objectid; 3587 err = btrfs_orphan_cleanup(gang[i]); 3588 if (err) 3589 break; 3590 btrfs_put_fs_root(gang[i]); 3591 } 3592 root_objectid++; 3593 } 3594 3595 /* release the uncleaned roots due to error */ 3596 for (; i < ret; i++) { 3597 if (gang[i]) 3598 btrfs_put_fs_root(gang[i]); 3599 } 3600 return err; 3601 } 3602 3603 int btrfs_commit_super(struct btrfs_root *root) 3604 { 3605 struct btrfs_trans_handle *trans; 3606 3607 mutex_lock(&root->fs_info->cleaner_mutex); 3608 btrfs_run_delayed_iputs(root); 3609 mutex_unlock(&root->fs_info->cleaner_mutex); 3610 wake_up_process(root->fs_info->cleaner_kthread); 3611 3612 /* wait until ongoing cleanup work done */ 3613 down_write(&root->fs_info->cleanup_work_sem); 3614 up_write(&root->fs_info->cleanup_work_sem); 3615 3616 trans = btrfs_join_transaction(root); 3617 if (IS_ERR(trans)) 3618 return PTR_ERR(trans); 3619 return btrfs_commit_transaction(trans, root); 3620 } 3621 3622 int close_ctree(struct btrfs_root *root) 3623 { 3624 struct btrfs_fs_info *fs_info = root->fs_info; 3625 int ret; 3626 3627 fs_info->closing = 1; 3628 smp_mb(); 3629 3630 /* wait for the uuid_scan task to finish */ 3631 down(&fs_info->uuid_tree_rescan_sem); 3632 /* avoid complains from lockdep et al., set sem back to initial state */ 3633 up(&fs_info->uuid_tree_rescan_sem); 3634 3635 /* pause restriper - we want to resume on mount */ 3636 btrfs_pause_balance(fs_info); 3637 3638 btrfs_dev_replace_suspend_for_unmount(fs_info); 3639 3640 btrfs_scrub_cancel(fs_info); 3641 3642 /* wait for any defraggers to finish */ 3643 wait_event(fs_info->transaction_wait, 3644 (atomic_read(&fs_info->defrag_running) == 0)); 3645 3646 /* clear out the rbtree of defraggable inodes */ 3647 btrfs_cleanup_defrag_inodes(fs_info); 3648 3649 cancel_work_sync(&fs_info->async_reclaim_work); 3650 3651 if (!(fs_info->sb->s_flags & MS_RDONLY)) { 3652 ret = btrfs_commit_super(root); 3653 if (ret) 3654 btrfs_err(root->fs_info, "commit super ret %d", ret); 3655 } 3656 3657 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 3658 btrfs_error_commit_super(root); 3659 3660 kthread_stop(fs_info->transaction_kthread); 3661 kthread_stop(fs_info->cleaner_kthread); 3662 3663 fs_info->closing = 2; 3664 smp_mb(); 3665 3666 btrfs_free_qgroup_config(root->fs_info); 3667 3668 if (percpu_counter_sum(&fs_info->delalloc_bytes)) { 3669 btrfs_info(root->fs_info, "at unmount delalloc count %lld", 3670 percpu_counter_sum(&fs_info->delalloc_bytes)); 3671 } 3672 3673 btrfs_sysfs_remove_one(fs_info); 3674 3675 btrfs_free_fs_roots(fs_info); 3676 3677 btrfs_put_block_group_cache(fs_info); 3678 3679 btrfs_free_block_groups(fs_info); 3680 3681 /* 3682 * we must make sure there is not any read request to 3683 * submit after we stopping all workers. 3684 */ 3685 invalidate_inode_pages2(fs_info->btree_inode->i_mapping); 3686 btrfs_stop_all_workers(fs_info); 3687 3688 free_root_pointers(fs_info, 1); 3689 3690 iput(fs_info->btree_inode); 3691 3692 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3693 if (btrfs_test_opt(root, CHECK_INTEGRITY)) 3694 btrfsic_unmount(root, fs_info->fs_devices); 3695 #endif 3696 3697 btrfs_close_devices(fs_info->fs_devices); 3698 btrfs_mapping_tree_free(&fs_info->mapping_tree); 3699 3700 percpu_counter_destroy(&fs_info->dirty_metadata_bytes); 3701 percpu_counter_destroy(&fs_info->delalloc_bytes); 3702 percpu_counter_destroy(&fs_info->bio_counter); 3703 bdi_destroy(&fs_info->bdi); 3704 cleanup_srcu_struct(&fs_info->subvol_srcu); 3705 3706 btrfs_free_stripe_hash_table(fs_info); 3707 3708 btrfs_free_block_rsv(root, root->orphan_block_rsv); 3709 root->orphan_block_rsv = NULL; 3710 3711 return 0; 3712 } 3713 3714 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid, 3715 int atomic) 3716 { 3717 int ret; 3718 struct inode *btree_inode = buf->pages[0]->mapping->host; 3719 3720 ret = extent_buffer_uptodate(buf); 3721 if (!ret) 3722 return ret; 3723 3724 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf, 3725 parent_transid, atomic); 3726 if (ret == -EAGAIN) 3727 return ret; 3728 return !ret; 3729 } 3730 3731 int btrfs_set_buffer_uptodate(struct extent_buffer *buf) 3732 { 3733 return set_extent_buffer_uptodate(buf); 3734 } 3735 3736 void btrfs_mark_buffer_dirty(struct extent_buffer *buf) 3737 { 3738 struct btrfs_root *root; 3739 u64 transid = btrfs_header_generation(buf); 3740 int was_dirty; 3741 3742 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS 3743 /* 3744 * This is a fast path so only do this check if we have sanity tests 3745 * enabled. Normal people shouldn't be marking dummy buffers as dirty 3746 * outside of the sanity tests. 3747 */ 3748 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags))) 3749 return; 3750 #endif 3751 root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3752 btrfs_assert_tree_locked(buf); 3753 if (transid != root->fs_info->generation) 3754 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, " 3755 "found %llu running %llu\n", 3756 buf->start, transid, root->fs_info->generation); 3757 was_dirty = set_extent_buffer_dirty(buf); 3758 if (!was_dirty) 3759 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes, 3760 buf->len, 3761 root->fs_info->dirty_metadata_batch); 3762 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY 3763 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) { 3764 btrfs_print_leaf(root, buf); 3765 ASSERT(0); 3766 } 3767 #endif 3768 } 3769 3770 static void __btrfs_btree_balance_dirty(struct btrfs_root *root, 3771 int flush_delayed) 3772 { 3773 /* 3774 * looks as though older kernels can get into trouble with 3775 * this code, they end up stuck in balance_dirty_pages forever 3776 */ 3777 int ret; 3778 3779 if (current->flags & PF_MEMALLOC) 3780 return; 3781 3782 if (flush_delayed) 3783 btrfs_balance_delayed_items(root); 3784 3785 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes, 3786 BTRFS_DIRTY_METADATA_THRESH); 3787 if (ret > 0) { 3788 balance_dirty_pages_ratelimited( 3789 root->fs_info->btree_inode->i_mapping); 3790 } 3791 return; 3792 } 3793 3794 void btrfs_btree_balance_dirty(struct btrfs_root *root) 3795 { 3796 __btrfs_btree_balance_dirty(root, 1); 3797 } 3798 3799 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root) 3800 { 3801 __btrfs_btree_balance_dirty(root, 0); 3802 } 3803 3804 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid) 3805 { 3806 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root; 3807 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid); 3808 } 3809 3810 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info, 3811 int read_only) 3812 { 3813 /* 3814 * Placeholder for checks 3815 */ 3816 return 0; 3817 } 3818 3819 static void btrfs_error_commit_super(struct btrfs_root *root) 3820 { 3821 mutex_lock(&root->fs_info->cleaner_mutex); 3822 btrfs_run_delayed_iputs(root); 3823 mutex_unlock(&root->fs_info->cleaner_mutex); 3824 3825 down_write(&root->fs_info->cleanup_work_sem); 3826 up_write(&root->fs_info->cleanup_work_sem); 3827 3828 /* cleanup FS via transaction */ 3829 btrfs_cleanup_transaction(root); 3830 } 3831 3832 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t, 3833 struct btrfs_root *root) 3834 { 3835 struct btrfs_inode *btrfs_inode; 3836 struct list_head splice; 3837 3838 INIT_LIST_HEAD(&splice); 3839 3840 mutex_lock(&root->fs_info->ordered_operations_mutex); 3841 spin_lock(&root->fs_info->ordered_root_lock); 3842 3843 list_splice_init(&t->ordered_operations, &splice); 3844 while (!list_empty(&splice)) { 3845 btrfs_inode = list_entry(splice.next, struct btrfs_inode, 3846 ordered_operations); 3847 3848 list_del_init(&btrfs_inode->ordered_operations); 3849 spin_unlock(&root->fs_info->ordered_root_lock); 3850 3851 btrfs_invalidate_inodes(btrfs_inode->root); 3852 3853 spin_lock(&root->fs_info->ordered_root_lock); 3854 } 3855 3856 spin_unlock(&root->fs_info->ordered_root_lock); 3857 mutex_unlock(&root->fs_info->ordered_operations_mutex); 3858 } 3859 3860 static void btrfs_destroy_ordered_extents(struct btrfs_root *root) 3861 { 3862 struct btrfs_ordered_extent *ordered; 3863 3864 spin_lock(&root->ordered_extent_lock); 3865 /* 3866 * This will just short circuit the ordered completion stuff which will 3867 * make sure the ordered extent gets properly cleaned up. 3868 */ 3869 list_for_each_entry(ordered, &root->ordered_extents, 3870 root_extent_list) 3871 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 3872 spin_unlock(&root->ordered_extent_lock); 3873 } 3874 3875 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info) 3876 { 3877 struct btrfs_root *root; 3878 struct list_head splice; 3879 3880 INIT_LIST_HEAD(&splice); 3881 3882 spin_lock(&fs_info->ordered_root_lock); 3883 list_splice_init(&fs_info->ordered_roots, &splice); 3884 while (!list_empty(&splice)) { 3885 root = list_first_entry(&splice, struct btrfs_root, 3886 ordered_root); 3887 list_move_tail(&root->ordered_root, 3888 &fs_info->ordered_roots); 3889 3890 spin_unlock(&fs_info->ordered_root_lock); 3891 btrfs_destroy_ordered_extents(root); 3892 3893 cond_resched(); 3894 spin_lock(&fs_info->ordered_root_lock); 3895 } 3896 spin_unlock(&fs_info->ordered_root_lock); 3897 } 3898 3899 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans, 3900 struct btrfs_root *root) 3901 { 3902 struct rb_node *node; 3903 struct btrfs_delayed_ref_root *delayed_refs; 3904 struct btrfs_delayed_ref_node *ref; 3905 int ret = 0; 3906 3907 delayed_refs = &trans->delayed_refs; 3908 3909 spin_lock(&delayed_refs->lock); 3910 if (atomic_read(&delayed_refs->num_entries) == 0) { 3911 spin_unlock(&delayed_refs->lock); 3912 btrfs_info(root->fs_info, "delayed_refs has NO entry"); 3913 return ret; 3914 } 3915 3916 while ((node = rb_first(&delayed_refs->href_root)) != NULL) { 3917 struct btrfs_delayed_ref_head *head; 3918 bool pin_bytes = false; 3919 3920 head = rb_entry(node, struct btrfs_delayed_ref_head, 3921 href_node); 3922 if (!mutex_trylock(&head->mutex)) { 3923 atomic_inc(&head->node.refs); 3924 spin_unlock(&delayed_refs->lock); 3925 3926 mutex_lock(&head->mutex); 3927 mutex_unlock(&head->mutex); 3928 btrfs_put_delayed_ref(&head->node); 3929 spin_lock(&delayed_refs->lock); 3930 continue; 3931 } 3932 spin_lock(&head->lock); 3933 while ((node = rb_first(&head->ref_root)) != NULL) { 3934 ref = rb_entry(node, struct btrfs_delayed_ref_node, 3935 rb_node); 3936 ref->in_tree = 0; 3937 rb_erase(&ref->rb_node, &head->ref_root); 3938 atomic_dec(&delayed_refs->num_entries); 3939 btrfs_put_delayed_ref(ref); 3940 } 3941 if (head->must_insert_reserved) 3942 pin_bytes = true; 3943 btrfs_free_delayed_extent_op(head->extent_op); 3944 delayed_refs->num_heads--; 3945 if (head->processing == 0) 3946 delayed_refs->num_heads_ready--; 3947 atomic_dec(&delayed_refs->num_entries); 3948 head->node.in_tree = 0; 3949 rb_erase(&head->href_node, &delayed_refs->href_root); 3950 spin_unlock(&head->lock); 3951 spin_unlock(&delayed_refs->lock); 3952 mutex_unlock(&head->mutex); 3953 3954 if (pin_bytes) 3955 btrfs_pin_extent(root, head->node.bytenr, 3956 head->node.num_bytes, 1); 3957 btrfs_put_delayed_ref(&head->node); 3958 cond_resched(); 3959 spin_lock(&delayed_refs->lock); 3960 } 3961 3962 spin_unlock(&delayed_refs->lock); 3963 3964 return ret; 3965 } 3966 3967 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root) 3968 { 3969 struct btrfs_inode *btrfs_inode; 3970 struct list_head splice; 3971 3972 INIT_LIST_HEAD(&splice); 3973 3974 spin_lock(&root->delalloc_lock); 3975 list_splice_init(&root->delalloc_inodes, &splice); 3976 3977 while (!list_empty(&splice)) { 3978 btrfs_inode = list_first_entry(&splice, struct btrfs_inode, 3979 delalloc_inodes); 3980 3981 list_del_init(&btrfs_inode->delalloc_inodes); 3982 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST, 3983 &btrfs_inode->runtime_flags); 3984 spin_unlock(&root->delalloc_lock); 3985 3986 btrfs_invalidate_inodes(btrfs_inode->root); 3987 3988 spin_lock(&root->delalloc_lock); 3989 } 3990 3991 spin_unlock(&root->delalloc_lock); 3992 } 3993 3994 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info) 3995 { 3996 struct btrfs_root *root; 3997 struct list_head splice; 3998 3999 INIT_LIST_HEAD(&splice); 4000 4001 spin_lock(&fs_info->delalloc_root_lock); 4002 list_splice_init(&fs_info->delalloc_roots, &splice); 4003 while (!list_empty(&splice)) { 4004 root = list_first_entry(&splice, struct btrfs_root, 4005 delalloc_root); 4006 list_del_init(&root->delalloc_root); 4007 root = btrfs_grab_fs_root(root); 4008 BUG_ON(!root); 4009 spin_unlock(&fs_info->delalloc_root_lock); 4010 4011 btrfs_destroy_delalloc_inodes(root); 4012 btrfs_put_fs_root(root); 4013 4014 spin_lock(&fs_info->delalloc_root_lock); 4015 } 4016 spin_unlock(&fs_info->delalloc_root_lock); 4017 } 4018 4019 static int btrfs_destroy_marked_extents(struct btrfs_root *root, 4020 struct extent_io_tree *dirty_pages, 4021 int mark) 4022 { 4023 int ret; 4024 struct extent_buffer *eb; 4025 u64 start = 0; 4026 u64 end; 4027 4028 while (1) { 4029 ret = find_first_extent_bit(dirty_pages, start, &start, &end, 4030 mark, NULL); 4031 if (ret) 4032 break; 4033 4034 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS); 4035 while (start <= end) { 4036 eb = btrfs_find_tree_block(root, start, 4037 root->leafsize); 4038 start += root->leafsize; 4039 if (!eb) 4040 continue; 4041 wait_on_extent_buffer_writeback(eb); 4042 4043 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, 4044 &eb->bflags)) 4045 clear_extent_buffer_dirty(eb); 4046 free_extent_buffer_stale(eb); 4047 } 4048 } 4049 4050 return ret; 4051 } 4052 4053 static int btrfs_destroy_pinned_extent(struct btrfs_root *root, 4054 struct extent_io_tree *pinned_extents) 4055 { 4056 struct extent_io_tree *unpin; 4057 u64 start; 4058 u64 end; 4059 int ret; 4060 bool loop = true; 4061 4062 unpin = pinned_extents; 4063 again: 4064 while (1) { 4065 ret = find_first_extent_bit(unpin, 0, &start, &end, 4066 EXTENT_DIRTY, NULL); 4067 if (ret) 4068 break; 4069 4070 /* opt_discard */ 4071 if (btrfs_test_opt(root, DISCARD)) 4072 ret = btrfs_error_discard_extent(root, start, 4073 end + 1 - start, 4074 NULL); 4075 4076 clear_extent_dirty(unpin, start, end, GFP_NOFS); 4077 btrfs_error_unpin_extent_range(root, start, end); 4078 cond_resched(); 4079 } 4080 4081 if (loop) { 4082 if (unpin == &root->fs_info->freed_extents[0]) 4083 unpin = &root->fs_info->freed_extents[1]; 4084 else 4085 unpin = &root->fs_info->freed_extents[0]; 4086 loop = false; 4087 goto again; 4088 } 4089 4090 return 0; 4091 } 4092 4093 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans, 4094 struct btrfs_root *root) 4095 { 4096 btrfs_destroy_ordered_operations(cur_trans, root); 4097 4098 btrfs_destroy_delayed_refs(cur_trans, root); 4099 4100 cur_trans->state = TRANS_STATE_COMMIT_START; 4101 wake_up(&root->fs_info->transaction_blocked_wait); 4102 4103 cur_trans->state = TRANS_STATE_UNBLOCKED; 4104 wake_up(&root->fs_info->transaction_wait); 4105 4106 btrfs_destroy_delayed_inodes(root); 4107 btrfs_assert_delayed_root_empty(root); 4108 4109 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages, 4110 EXTENT_DIRTY); 4111 btrfs_destroy_pinned_extent(root, 4112 root->fs_info->pinned_extents); 4113 4114 cur_trans->state =TRANS_STATE_COMPLETED; 4115 wake_up(&cur_trans->commit_wait); 4116 4117 /* 4118 memset(cur_trans, 0, sizeof(*cur_trans)); 4119 kmem_cache_free(btrfs_transaction_cachep, cur_trans); 4120 */ 4121 } 4122 4123 static int btrfs_cleanup_transaction(struct btrfs_root *root) 4124 { 4125 struct btrfs_transaction *t; 4126 4127 mutex_lock(&root->fs_info->transaction_kthread_mutex); 4128 4129 spin_lock(&root->fs_info->trans_lock); 4130 while (!list_empty(&root->fs_info->trans_list)) { 4131 t = list_first_entry(&root->fs_info->trans_list, 4132 struct btrfs_transaction, list); 4133 if (t->state >= TRANS_STATE_COMMIT_START) { 4134 atomic_inc(&t->use_count); 4135 spin_unlock(&root->fs_info->trans_lock); 4136 btrfs_wait_for_commit(root, t->transid); 4137 btrfs_put_transaction(t); 4138 spin_lock(&root->fs_info->trans_lock); 4139 continue; 4140 } 4141 if (t == root->fs_info->running_transaction) { 4142 t->state = TRANS_STATE_COMMIT_DOING; 4143 spin_unlock(&root->fs_info->trans_lock); 4144 /* 4145 * We wait for 0 num_writers since we don't hold a trans 4146 * handle open currently for this transaction. 4147 */ 4148 wait_event(t->writer_wait, 4149 atomic_read(&t->num_writers) == 0); 4150 } else { 4151 spin_unlock(&root->fs_info->trans_lock); 4152 } 4153 btrfs_cleanup_one_transaction(t, root); 4154 4155 spin_lock(&root->fs_info->trans_lock); 4156 if (t == root->fs_info->running_transaction) 4157 root->fs_info->running_transaction = NULL; 4158 list_del_init(&t->list); 4159 spin_unlock(&root->fs_info->trans_lock); 4160 4161 btrfs_put_transaction(t); 4162 trace_btrfs_transaction_commit(root); 4163 spin_lock(&root->fs_info->trans_lock); 4164 } 4165 spin_unlock(&root->fs_info->trans_lock); 4166 btrfs_destroy_all_ordered_extents(root->fs_info); 4167 btrfs_destroy_delayed_inodes(root); 4168 btrfs_assert_delayed_root_empty(root); 4169 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents); 4170 btrfs_destroy_all_delalloc_inodes(root->fs_info); 4171 mutex_unlock(&root->fs_info->transaction_kthread_mutex); 4172 4173 return 0; 4174 } 4175 4176 static struct extent_io_ops btree_extent_io_ops = { 4177 .readpage_end_io_hook = btree_readpage_end_io_hook, 4178 .readpage_io_failed_hook = btree_io_failed_hook, 4179 .submit_bio_hook = btree_submit_bio_hook, 4180 /* note we're sharing with inode.c for the merge bio hook */ 4181 .merge_bio_hook = btrfs_merge_bio_hook, 4182 }; 4183