1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2007,2008 Oracle. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/slab.h> 8 #include <linux/rbtree.h> 9 #include <linux/mm.h> 10 #include <linux/error-injection.h> 11 #include "messages.h" 12 #include "ctree.h" 13 #include "disk-io.h" 14 #include "transaction.h" 15 #include "print-tree.h" 16 #include "locking.h" 17 #include "volumes.h" 18 #include "qgroup.h" 19 #include "tree-mod-log.h" 20 #include "tree-checker.h" 21 #include "fs.h" 22 #include "accessors.h" 23 #include "extent-tree.h" 24 #include "relocation.h" 25 #include "file-item.h" 26 27 static struct kmem_cache *btrfs_path_cachep; 28 29 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root 30 *root, struct btrfs_path *path, int level); 31 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, 32 const struct btrfs_key *ins_key, struct btrfs_path *path, 33 int data_size, int extend); 34 static int push_node_left(struct btrfs_trans_handle *trans, 35 struct extent_buffer *dst, 36 struct extent_buffer *src, int empty); 37 static int balance_node_right(struct btrfs_trans_handle *trans, 38 struct extent_buffer *dst_buf, 39 struct extent_buffer *src_buf); 40 41 static const struct btrfs_csums { 42 u16 size; 43 const char name[10]; 44 const char driver[12]; 45 } btrfs_csums[] = { 46 [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, 47 [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, 48 [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, 49 [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", 50 .driver = "blake2b-256" }, 51 }; 52 53 /* 54 * The leaf data grows from end-to-front in the node. this returns the address 55 * of the start of the last item, which is the stop of the leaf data stack. 56 */ 57 static unsigned int leaf_data_end(const struct extent_buffer *leaf) 58 { 59 u32 nr = btrfs_header_nritems(leaf); 60 61 if (nr == 0) 62 return BTRFS_LEAF_DATA_SIZE(leaf->fs_info); 63 return btrfs_item_offset(leaf, nr - 1); 64 } 65 66 /* 67 * Move data in a @leaf (using memmove, safe for overlapping ranges). 68 * 69 * @leaf: leaf that we're doing a memmove on 70 * @dst_offset: item data offset we're moving to 71 * @src_offset: item data offset were' moving from 72 * @len: length of the data we're moving 73 * 74 * Wrapper around memmove_extent_buffer() that takes into account the header on 75 * the leaf. The btrfs_item offset's start directly after the header, so we 76 * have to adjust any offsets to account for the header in the leaf. This 77 * handles that math to simplify the callers. 78 */ 79 static inline void memmove_leaf_data(const struct extent_buffer *leaf, 80 unsigned long dst_offset, 81 unsigned long src_offset, 82 unsigned long len) 83 { 84 memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, 0) + dst_offset, 85 btrfs_item_nr_offset(leaf, 0) + src_offset, len); 86 } 87 88 /* 89 * Copy item data from @src into @dst at the given @offset. 90 * 91 * @dst: destination leaf that we're copying into 92 * @src: source leaf that we're copying from 93 * @dst_offset: item data offset we're copying to 94 * @src_offset: item data offset were' copying from 95 * @len: length of the data we're copying 96 * 97 * Wrapper around copy_extent_buffer() that takes into account the header on 98 * the leaf. The btrfs_item offset's start directly after the header, so we 99 * have to adjust any offsets to account for the header in the leaf. This 100 * handles that math to simplify the callers. 101 */ 102 static inline void copy_leaf_data(const struct extent_buffer *dst, 103 const struct extent_buffer *src, 104 unsigned long dst_offset, 105 unsigned long src_offset, unsigned long len) 106 { 107 copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, 0) + dst_offset, 108 btrfs_item_nr_offset(src, 0) + src_offset, len); 109 } 110 111 /* 112 * Move items in a @leaf (using memmove). 113 * 114 * @dst: destination leaf for the items 115 * @dst_item: the item nr we're copying into 116 * @src_item: the item nr we're copying from 117 * @nr_items: the number of items to copy 118 * 119 * Wrapper around memmove_extent_buffer() that does the math to get the 120 * appropriate offsets into the leaf from the item numbers. 121 */ 122 static inline void memmove_leaf_items(const struct extent_buffer *leaf, 123 int dst_item, int src_item, int nr_items) 124 { 125 memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, dst_item), 126 btrfs_item_nr_offset(leaf, src_item), 127 nr_items * sizeof(struct btrfs_item)); 128 } 129 130 /* 131 * Copy items from @src into @dst at the given @offset. 132 * 133 * @dst: destination leaf for the items 134 * @src: source leaf for the items 135 * @dst_item: the item nr we're copying into 136 * @src_item: the item nr we're copying from 137 * @nr_items: the number of items to copy 138 * 139 * Wrapper around copy_extent_buffer() that does the math to get the 140 * appropriate offsets into the leaf from the item numbers. 141 */ 142 static inline void copy_leaf_items(const struct extent_buffer *dst, 143 const struct extent_buffer *src, 144 int dst_item, int src_item, int nr_items) 145 { 146 copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, dst_item), 147 btrfs_item_nr_offset(src, src_item), 148 nr_items * sizeof(struct btrfs_item)); 149 } 150 151 /* This exists for btrfs-progs usages. */ 152 u16 btrfs_csum_type_size(u16 type) 153 { 154 return btrfs_csums[type].size; 155 } 156 157 int btrfs_super_csum_size(const struct btrfs_super_block *s) 158 { 159 u16 t = btrfs_super_csum_type(s); 160 /* 161 * csum type is validated at mount time 162 */ 163 return btrfs_csum_type_size(t); 164 } 165 166 const char *btrfs_super_csum_name(u16 csum_type) 167 { 168 /* csum type is validated at mount time */ 169 return btrfs_csums[csum_type].name; 170 } 171 172 /* 173 * Return driver name if defined, otherwise the name that's also a valid driver 174 * name 175 */ 176 const char *btrfs_super_csum_driver(u16 csum_type) 177 { 178 /* csum type is validated at mount time */ 179 return btrfs_csums[csum_type].driver[0] ? 180 btrfs_csums[csum_type].driver : 181 btrfs_csums[csum_type].name; 182 } 183 184 size_t __attribute_const__ btrfs_get_num_csums(void) 185 { 186 return ARRAY_SIZE(btrfs_csums); 187 } 188 189 struct btrfs_path *btrfs_alloc_path(void) 190 { 191 might_sleep(); 192 193 return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); 194 } 195 196 /* this also releases the path */ 197 void btrfs_free_path(struct btrfs_path *p) 198 { 199 if (!p) 200 return; 201 btrfs_release_path(p); 202 kmem_cache_free(btrfs_path_cachep, p); 203 } 204 205 /* 206 * path release drops references on the extent buffers in the path 207 * and it drops any locks held by this path 208 * 209 * It is safe to call this on paths that no locks or extent buffers held. 210 */ 211 noinline void btrfs_release_path(struct btrfs_path *p) 212 { 213 int i; 214 215 for (i = 0; i < BTRFS_MAX_LEVEL; i++) { 216 p->slots[i] = 0; 217 if (!p->nodes[i]) 218 continue; 219 if (p->locks[i]) { 220 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); 221 p->locks[i] = 0; 222 } 223 free_extent_buffer(p->nodes[i]); 224 p->nodes[i] = NULL; 225 } 226 } 227 228 /* 229 * We want the transaction abort to print stack trace only for errors where the 230 * cause could be a bug, eg. due to ENOSPC, and not for common errors that are 231 * caused by external factors. 232 */ 233 bool __cold abort_should_print_stack(int errno) 234 { 235 switch (errno) { 236 case -EIO: 237 case -EROFS: 238 case -ENOMEM: 239 return false; 240 } 241 return true; 242 } 243 244 /* 245 * safely gets a reference on the root node of a tree. A lock 246 * is not taken, so a concurrent writer may put a different node 247 * at the root of the tree. See btrfs_lock_root_node for the 248 * looping required. 249 * 250 * The extent buffer returned by this has a reference taken, so 251 * it won't disappear. It may stop being the root of the tree 252 * at any time because there are no locks held. 253 */ 254 struct extent_buffer *btrfs_root_node(struct btrfs_root *root) 255 { 256 struct extent_buffer *eb; 257 258 while (1) { 259 rcu_read_lock(); 260 eb = rcu_dereference(root->node); 261 262 /* 263 * RCU really hurts here, we could free up the root node because 264 * it was COWed but we may not get the new root node yet so do 265 * the inc_not_zero dance and if it doesn't work then 266 * synchronize_rcu and try again. 267 */ 268 if (atomic_inc_not_zero(&eb->refs)) { 269 rcu_read_unlock(); 270 break; 271 } 272 rcu_read_unlock(); 273 synchronize_rcu(); 274 } 275 return eb; 276 } 277 278 /* 279 * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), 280 * just get put onto a simple dirty list. Transaction walks this list to make 281 * sure they get properly updated on disk. 282 */ 283 static void add_root_to_dirty_list(struct btrfs_root *root) 284 { 285 struct btrfs_fs_info *fs_info = root->fs_info; 286 287 if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || 288 !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) 289 return; 290 291 spin_lock(&fs_info->trans_lock); 292 if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { 293 /* Want the extent tree to be the last on the list */ 294 if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) 295 list_move_tail(&root->dirty_list, 296 &fs_info->dirty_cowonly_roots); 297 else 298 list_move(&root->dirty_list, 299 &fs_info->dirty_cowonly_roots); 300 } 301 spin_unlock(&fs_info->trans_lock); 302 } 303 304 /* 305 * used by snapshot creation to make a copy of a root for a tree with 306 * a given objectid. The buffer with the new root node is returned in 307 * cow_ret, and this func returns zero on success or a negative error code. 308 */ 309 int btrfs_copy_root(struct btrfs_trans_handle *trans, 310 struct btrfs_root *root, 311 struct extent_buffer *buf, 312 struct extent_buffer **cow_ret, u64 new_root_objectid) 313 { 314 struct btrfs_fs_info *fs_info = root->fs_info; 315 struct extent_buffer *cow; 316 int ret = 0; 317 int level; 318 struct btrfs_disk_key disk_key; 319 320 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 321 trans->transid != fs_info->running_transaction->transid); 322 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 323 trans->transid != root->last_trans); 324 325 level = btrfs_header_level(buf); 326 if (level == 0) 327 btrfs_item_key(buf, &disk_key, 0); 328 else 329 btrfs_node_key(buf, &disk_key, 0); 330 331 cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, 332 &disk_key, level, buf->start, 0, 333 BTRFS_NESTING_NEW_ROOT); 334 if (IS_ERR(cow)) 335 return PTR_ERR(cow); 336 337 copy_extent_buffer_full(cow, buf); 338 btrfs_set_header_bytenr(cow, cow->start); 339 btrfs_set_header_generation(cow, trans->transid); 340 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 341 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 342 BTRFS_HEADER_FLAG_RELOC); 343 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 344 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 345 else 346 btrfs_set_header_owner(cow, new_root_objectid); 347 348 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 349 350 WARN_ON(btrfs_header_generation(buf) > trans->transid); 351 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) 352 ret = btrfs_inc_ref(trans, root, cow, 1); 353 else 354 ret = btrfs_inc_ref(trans, root, cow, 0); 355 if (ret) { 356 btrfs_tree_unlock(cow); 357 free_extent_buffer(cow); 358 btrfs_abort_transaction(trans, ret); 359 return ret; 360 } 361 362 btrfs_mark_buffer_dirty(cow); 363 *cow_ret = cow; 364 return 0; 365 } 366 367 /* 368 * check if the tree block can be shared by multiple trees 369 */ 370 int btrfs_block_can_be_shared(struct btrfs_root *root, 371 struct extent_buffer *buf) 372 { 373 /* 374 * Tree blocks not in shareable trees and tree roots are never shared. 375 * If a block was allocated after the last snapshot and the block was 376 * not allocated by tree relocation, we know the block is not shared. 377 */ 378 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 379 buf != root->node && buf != root->commit_root && 380 (btrfs_header_generation(buf) <= 381 btrfs_root_last_snapshot(&root->root_item) || 382 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) 383 return 1; 384 385 return 0; 386 } 387 388 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, 389 struct btrfs_root *root, 390 struct extent_buffer *buf, 391 struct extent_buffer *cow, 392 int *last_ref) 393 { 394 struct btrfs_fs_info *fs_info = root->fs_info; 395 u64 refs; 396 u64 owner; 397 u64 flags; 398 u64 new_flags = 0; 399 int ret; 400 401 /* 402 * Backrefs update rules: 403 * 404 * Always use full backrefs for extent pointers in tree block 405 * allocated by tree relocation. 406 * 407 * If a shared tree block is no longer referenced by its owner 408 * tree (btrfs_header_owner(buf) == root->root_key.objectid), 409 * use full backrefs for extent pointers in tree block. 410 * 411 * If a tree block is been relocating 412 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), 413 * use full backrefs for extent pointers in tree block. 414 * The reason for this is some operations (such as drop tree) 415 * are only allowed for blocks use full backrefs. 416 */ 417 418 if (btrfs_block_can_be_shared(root, buf)) { 419 ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, 420 btrfs_header_level(buf), 1, 421 &refs, &flags); 422 if (ret) 423 return ret; 424 if (unlikely(refs == 0)) { 425 btrfs_crit(fs_info, 426 "found 0 references for tree block at bytenr %llu level %d root %llu", 427 buf->start, btrfs_header_level(buf), 428 btrfs_root_id(root)); 429 ret = -EUCLEAN; 430 btrfs_abort_transaction(trans, ret); 431 return ret; 432 } 433 } else { 434 refs = 1; 435 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 436 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 437 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; 438 else 439 flags = 0; 440 } 441 442 owner = btrfs_header_owner(buf); 443 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && 444 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); 445 446 if (refs > 1) { 447 if ((owner == root->root_key.objectid || 448 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && 449 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { 450 ret = btrfs_inc_ref(trans, root, buf, 1); 451 if (ret) 452 return ret; 453 454 if (root->root_key.objectid == 455 BTRFS_TREE_RELOC_OBJECTID) { 456 ret = btrfs_dec_ref(trans, root, buf, 0); 457 if (ret) 458 return ret; 459 ret = btrfs_inc_ref(trans, root, cow, 1); 460 if (ret) 461 return ret; 462 } 463 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; 464 } else { 465 466 if (root->root_key.objectid == 467 BTRFS_TREE_RELOC_OBJECTID) 468 ret = btrfs_inc_ref(trans, root, cow, 1); 469 else 470 ret = btrfs_inc_ref(trans, root, cow, 0); 471 if (ret) 472 return ret; 473 } 474 if (new_flags != 0) { 475 ret = btrfs_set_disk_extent_flags(trans, buf, new_flags); 476 if (ret) 477 return ret; 478 } 479 } else { 480 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { 481 if (root->root_key.objectid == 482 BTRFS_TREE_RELOC_OBJECTID) 483 ret = btrfs_inc_ref(trans, root, cow, 1); 484 else 485 ret = btrfs_inc_ref(trans, root, cow, 0); 486 if (ret) 487 return ret; 488 ret = btrfs_dec_ref(trans, root, buf, 1); 489 if (ret) 490 return ret; 491 } 492 btrfs_clear_buffer_dirty(trans, buf); 493 *last_ref = 1; 494 } 495 return 0; 496 } 497 498 /* 499 * does the dirty work in cow of a single block. The parent block (if 500 * supplied) is updated to point to the new cow copy. The new buffer is marked 501 * dirty and returned locked. If you modify the block it needs to be marked 502 * dirty again. 503 * 504 * search_start -- an allocation hint for the new block 505 * 506 * empty_size -- a hint that you plan on doing more cow. This is the size in 507 * bytes the allocator should try to find free next to the block it returns. 508 * This is just a hint and may be ignored by the allocator. 509 */ 510 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, 511 struct btrfs_root *root, 512 struct extent_buffer *buf, 513 struct extent_buffer *parent, int parent_slot, 514 struct extent_buffer **cow_ret, 515 u64 search_start, u64 empty_size, 516 enum btrfs_lock_nesting nest) 517 { 518 struct btrfs_fs_info *fs_info = root->fs_info; 519 struct btrfs_disk_key disk_key; 520 struct extent_buffer *cow; 521 int level, ret; 522 int last_ref = 0; 523 int unlock_orig = 0; 524 u64 parent_start = 0; 525 526 if (*cow_ret == buf) 527 unlock_orig = 1; 528 529 btrfs_assert_tree_write_locked(buf); 530 531 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 532 trans->transid != fs_info->running_transaction->transid); 533 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 534 trans->transid != root->last_trans); 535 536 level = btrfs_header_level(buf); 537 538 if (level == 0) 539 btrfs_item_key(buf, &disk_key, 0); 540 else 541 btrfs_node_key(buf, &disk_key, 0); 542 543 if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) 544 parent_start = parent->start; 545 546 cow = btrfs_alloc_tree_block(trans, root, parent_start, 547 root->root_key.objectid, &disk_key, level, 548 search_start, empty_size, nest); 549 if (IS_ERR(cow)) 550 return PTR_ERR(cow); 551 552 /* cow is set to blocking by btrfs_init_new_buffer */ 553 554 copy_extent_buffer_full(cow, buf); 555 btrfs_set_header_bytenr(cow, cow->start); 556 btrfs_set_header_generation(cow, trans->transid); 557 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); 558 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | 559 BTRFS_HEADER_FLAG_RELOC); 560 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 561 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); 562 else 563 btrfs_set_header_owner(cow, root->root_key.objectid); 564 565 write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); 566 567 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); 568 if (ret) { 569 btrfs_tree_unlock(cow); 570 free_extent_buffer(cow); 571 btrfs_abort_transaction(trans, ret); 572 return ret; 573 } 574 575 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { 576 ret = btrfs_reloc_cow_block(trans, root, buf, cow); 577 if (ret) { 578 btrfs_tree_unlock(cow); 579 free_extent_buffer(cow); 580 btrfs_abort_transaction(trans, ret); 581 return ret; 582 } 583 } 584 585 if (buf == root->node) { 586 WARN_ON(parent && parent != buf); 587 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 588 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) 589 parent_start = buf->start; 590 591 ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); 592 if (ret < 0) { 593 btrfs_tree_unlock(cow); 594 free_extent_buffer(cow); 595 btrfs_abort_transaction(trans, ret); 596 return ret; 597 } 598 atomic_inc(&cow->refs); 599 rcu_assign_pointer(root->node, cow); 600 601 btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 602 parent_start, last_ref); 603 free_extent_buffer(buf); 604 add_root_to_dirty_list(root); 605 } else { 606 WARN_ON(trans->transid != btrfs_header_generation(parent)); 607 ret = btrfs_tree_mod_log_insert_key(parent, parent_slot, 608 BTRFS_MOD_LOG_KEY_REPLACE); 609 if (ret) { 610 btrfs_tree_unlock(cow); 611 free_extent_buffer(cow); 612 btrfs_abort_transaction(trans, ret); 613 return ret; 614 } 615 btrfs_set_node_blockptr(parent, parent_slot, 616 cow->start); 617 btrfs_set_node_ptr_generation(parent, parent_slot, 618 trans->transid); 619 btrfs_mark_buffer_dirty(parent); 620 if (last_ref) { 621 ret = btrfs_tree_mod_log_free_eb(buf); 622 if (ret) { 623 btrfs_tree_unlock(cow); 624 free_extent_buffer(cow); 625 btrfs_abort_transaction(trans, ret); 626 return ret; 627 } 628 } 629 btrfs_free_tree_block(trans, btrfs_root_id(root), buf, 630 parent_start, last_ref); 631 } 632 if (unlock_orig) 633 btrfs_tree_unlock(buf); 634 free_extent_buffer_stale(buf); 635 btrfs_mark_buffer_dirty(cow); 636 *cow_ret = cow; 637 return 0; 638 } 639 640 static inline int should_cow_block(struct btrfs_trans_handle *trans, 641 struct btrfs_root *root, 642 struct extent_buffer *buf) 643 { 644 if (btrfs_is_testing(root->fs_info)) 645 return 0; 646 647 /* Ensure we can see the FORCE_COW bit */ 648 smp_mb__before_atomic(); 649 650 /* 651 * We do not need to cow a block if 652 * 1) this block is not created or changed in this transaction; 653 * 2) this block does not belong to TREE_RELOC tree; 654 * 3) the root is not forced COW. 655 * 656 * What is forced COW: 657 * when we create snapshot during committing the transaction, 658 * after we've finished copying src root, we must COW the shared 659 * block to ensure the metadata consistency. 660 */ 661 if (btrfs_header_generation(buf) == trans->transid && 662 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && 663 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && 664 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && 665 !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) 666 return 0; 667 return 1; 668 } 669 670 /* 671 * cows a single block, see __btrfs_cow_block for the real work. 672 * This version of it has extra checks so that a block isn't COWed more than 673 * once per transaction, as long as it hasn't been written yet 674 */ 675 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, 676 struct btrfs_root *root, struct extent_buffer *buf, 677 struct extent_buffer *parent, int parent_slot, 678 struct extent_buffer **cow_ret, 679 enum btrfs_lock_nesting nest) 680 { 681 struct btrfs_fs_info *fs_info = root->fs_info; 682 u64 search_start; 683 int ret; 684 685 if (test_bit(BTRFS_ROOT_DELETING, &root->state)) 686 btrfs_err(fs_info, 687 "COW'ing blocks on a fs root that's being dropped"); 688 689 if (trans->transaction != fs_info->running_transaction) 690 WARN(1, KERN_CRIT "trans %llu running %llu\n", 691 trans->transid, 692 fs_info->running_transaction->transid); 693 694 if (trans->transid != fs_info->generation) 695 WARN(1, KERN_CRIT "trans %llu running %llu\n", 696 trans->transid, fs_info->generation); 697 698 if (!should_cow_block(trans, root, buf)) { 699 *cow_ret = buf; 700 return 0; 701 } 702 703 search_start = buf->start & ~((u64)SZ_1G - 1); 704 705 /* 706 * Before CoWing this block for later modification, check if it's 707 * the subtree root and do the delayed subtree trace if needed. 708 * 709 * Also We don't care about the error, as it's handled internally. 710 */ 711 btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); 712 ret = __btrfs_cow_block(trans, root, buf, parent, 713 parent_slot, cow_ret, search_start, 0, nest); 714 715 trace_btrfs_cow_block(root, buf, *cow_ret); 716 717 return ret; 718 } 719 ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); 720 721 /* 722 * helper function for defrag to decide if two blocks pointed to by a 723 * node are actually close by 724 */ 725 static int close_blocks(u64 blocknr, u64 other, u32 blocksize) 726 { 727 if (blocknr < other && other - (blocknr + blocksize) < 32768) 728 return 1; 729 if (blocknr > other && blocknr - (other + blocksize) < 32768) 730 return 1; 731 return 0; 732 } 733 734 #ifdef __LITTLE_ENDIAN 735 736 /* 737 * Compare two keys, on little-endian the disk order is same as CPU order and 738 * we can avoid the conversion. 739 */ 740 static int comp_keys(const struct btrfs_disk_key *disk_key, 741 const struct btrfs_key *k2) 742 { 743 const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; 744 745 return btrfs_comp_cpu_keys(k1, k2); 746 } 747 748 #else 749 750 /* 751 * compare two keys in a memcmp fashion 752 */ 753 static int comp_keys(const struct btrfs_disk_key *disk, 754 const struct btrfs_key *k2) 755 { 756 struct btrfs_key k1; 757 758 btrfs_disk_key_to_cpu(&k1, disk); 759 760 return btrfs_comp_cpu_keys(&k1, k2); 761 } 762 #endif 763 764 /* 765 * same as comp_keys only with two btrfs_key's 766 */ 767 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) 768 { 769 if (k1->objectid > k2->objectid) 770 return 1; 771 if (k1->objectid < k2->objectid) 772 return -1; 773 if (k1->type > k2->type) 774 return 1; 775 if (k1->type < k2->type) 776 return -1; 777 if (k1->offset > k2->offset) 778 return 1; 779 if (k1->offset < k2->offset) 780 return -1; 781 return 0; 782 } 783 784 /* 785 * this is used by the defrag code to go through all the 786 * leaves pointed to by a node and reallocate them so that 787 * disk order is close to key order 788 */ 789 int btrfs_realloc_node(struct btrfs_trans_handle *trans, 790 struct btrfs_root *root, struct extent_buffer *parent, 791 int start_slot, u64 *last_ret, 792 struct btrfs_key *progress) 793 { 794 struct btrfs_fs_info *fs_info = root->fs_info; 795 struct extent_buffer *cur; 796 u64 blocknr; 797 u64 search_start = *last_ret; 798 u64 last_block = 0; 799 u64 other; 800 u32 parent_nritems; 801 int end_slot; 802 int i; 803 int err = 0; 804 u32 blocksize; 805 int progress_passed = 0; 806 struct btrfs_disk_key disk_key; 807 808 WARN_ON(trans->transaction != fs_info->running_transaction); 809 WARN_ON(trans->transid != fs_info->generation); 810 811 parent_nritems = btrfs_header_nritems(parent); 812 blocksize = fs_info->nodesize; 813 end_slot = parent_nritems - 1; 814 815 if (parent_nritems <= 1) 816 return 0; 817 818 for (i = start_slot; i <= end_slot; i++) { 819 int close = 1; 820 821 btrfs_node_key(parent, &disk_key, i); 822 if (!progress_passed && comp_keys(&disk_key, progress) < 0) 823 continue; 824 825 progress_passed = 1; 826 blocknr = btrfs_node_blockptr(parent, i); 827 if (last_block == 0) 828 last_block = blocknr; 829 830 if (i > 0) { 831 other = btrfs_node_blockptr(parent, i - 1); 832 close = close_blocks(blocknr, other, blocksize); 833 } 834 if (!close && i < end_slot) { 835 other = btrfs_node_blockptr(parent, i + 1); 836 close = close_blocks(blocknr, other, blocksize); 837 } 838 if (close) { 839 last_block = blocknr; 840 continue; 841 } 842 843 cur = btrfs_read_node_slot(parent, i); 844 if (IS_ERR(cur)) 845 return PTR_ERR(cur); 846 if (search_start == 0) 847 search_start = last_block; 848 849 btrfs_tree_lock(cur); 850 err = __btrfs_cow_block(trans, root, cur, parent, i, 851 &cur, search_start, 852 min(16 * blocksize, 853 (end_slot - i) * blocksize), 854 BTRFS_NESTING_COW); 855 if (err) { 856 btrfs_tree_unlock(cur); 857 free_extent_buffer(cur); 858 break; 859 } 860 search_start = cur->start; 861 last_block = cur->start; 862 *last_ret = search_start; 863 btrfs_tree_unlock(cur); 864 free_extent_buffer(cur); 865 } 866 return err; 867 } 868 869 /* 870 * Search for a key in the given extent_buffer. 871 * 872 * The lower boundary for the search is specified by the slot number @first_slot. 873 * Use a value of 0 to search over the whole extent buffer. Works for both 874 * leaves and nodes. 875 * 876 * The slot in the extent buffer is returned via @slot. If the key exists in the 877 * extent buffer, then @slot will point to the slot where the key is, otherwise 878 * it points to the slot where you would insert the key. 879 * 880 * Slot may point to the total number of items (i.e. one position beyond the last 881 * key) if the key is bigger than the last key in the extent buffer. 882 */ 883 int btrfs_bin_search(struct extent_buffer *eb, int first_slot, 884 const struct btrfs_key *key, int *slot) 885 { 886 unsigned long p; 887 int item_size; 888 /* 889 * Use unsigned types for the low and high slots, so that we get a more 890 * efficient division in the search loop below. 891 */ 892 u32 low = first_slot; 893 u32 high = btrfs_header_nritems(eb); 894 int ret; 895 const int key_size = sizeof(struct btrfs_disk_key); 896 897 if (unlikely(low > high)) { 898 btrfs_err(eb->fs_info, 899 "%s: low (%u) > high (%u) eb %llu owner %llu level %d", 900 __func__, low, high, eb->start, 901 btrfs_header_owner(eb), btrfs_header_level(eb)); 902 return -EINVAL; 903 } 904 905 if (btrfs_header_level(eb) == 0) { 906 p = offsetof(struct btrfs_leaf, items); 907 item_size = sizeof(struct btrfs_item); 908 } else { 909 p = offsetof(struct btrfs_node, ptrs); 910 item_size = sizeof(struct btrfs_key_ptr); 911 } 912 913 while (low < high) { 914 unsigned long oip; 915 unsigned long offset; 916 struct btrfs_disk_key *tmp; 917 struct btrfs_disk_key unaligned; 918 int mid; 919 920 mid = (low + high) / 2; 921 offset = p + mid * item_size; 922 oip = offset_in_page(offset); 923 924 if (oip + key_size <= PAGE_SIZE) { 925 const unsigned long idx = get_eb_page_index(offset); 926 char *kaddr = page_address(eb->pages[idx]); 927 928 oip = get_eb_offset_in_page(eb, offset); 929 tmp = (struct btrfs_disk_key *)(kaddr + oip); 930 } else { 931 read_extent_buffer(eb, &unaligned, offset, key_size); 932 tmp = &unaligned; 933 } 934 935 ret = comp_keys(tmp, key); 936 937 if (ret < 0) 938 low = mid + 1; 939 else if (ret > 0) 940 high = mid; 941 else { 942 *slot = mid; 943 return 0; 944 } 945 } 946 *slot = low; 947 return 1; 948 } 949 950 static void root_add_used(struct btrfs_root *root, u32 size) 951 { 952 spin_lock(&root->accounting_lock); 953 btrfs_set_root_used(&root->root_item, 954 btrfs_root_used(&root->root_item) + size); 955 spin_unlock(&root->accounting_lock); 956 } 957 958 static void root_sub_used(struct btrfs_root *root, u32 size) 959 { 960 spin_lock(&root->accounting_lock); 961 btrfs_set_root_used(&root->root_item, 962 btrfs_root_used(&root->root_item) - size); 963 spin_unlock(&root->accounting_lock); 964 } 965 966 /* given a node and slot number, this reads the blocks it points to. The 967 * extent buffer is returned with a reference taken (but unlocked). 968 */ 969 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, 970 int slot) 971 { 972 int level = btrfs_header_level(parent); 973 struct btrfs_tree_parent_check check = { 0 }; 974 struct extent_buffer *eb; 975 976 if (slot < 0 || slot >= btrfs_header_nritems(parent)) 977 return ERR_PTR(-ENOENT); 978 979 ASSERT(level); 980 981 check.level = level - 1; 982 check.transid = btrfs_node_ptr_generation(parent, slot); 983 check.owner_root = btrfs_header_owner(parent); 984 check.has_first_key = true; 985 btrfs_node_key_to_cpu(parent, &check.first_key, slot); 986 987 eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), 988 &check); 989 if (IS_ERR(eb)) 990 return eb; 991 if (!extent_buffer_uptodate(eb)) { 992 free_extent_buffer(eb); 993 return ERR_PTR(-EIO); 994 } 995 996 return eb; 997 } 998 999 /* 1000 * node level balancing, used to make sure nodes are in proper order for 1001 * item deletion. We balance from the top down, so we have to make sure 1002 * that a deletion won't leave an node completely empty later on. 1003 */ 1004 static noinline int balance_level(struct btrfs_trans_handle *trans, 1005 struct btrfs_root *root, 1006 struct btrfs_path *path, int level) 1007 { 1008 struct btrfs_fs_info *fs_info = root->fs_info; 1009 struct extent_buffer *right = NULL; 1010 struct extent_buffer *mid; 1011 struct extent_buffer *left = NULL; 1012 struct extent_buffer *parent = NULL; 1013 int ret = 0; 1014 int wret; 1015 int pslot; 1016 int orig_slot = path->slots[level]; 1017 u64 orig_ptr; 1018 1019 ASSERT(level > 0); 1020 1021 mid = path->nodes[level]; 1022 1023 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); 1024 WARN_ON(btrfs_header_generation(mid) != trans->transid); 1025 1026 orig_ptr = btrfs_node_blockptr(mid, orig_slot); 1027 1028 if (level < BTRFS_MAX_LEVEL - 1) { 1029 parent = path->nodes[level + 1]; 1030 pslot = path->slots[level + 1]; 1031 } 1032 1033 /* 1034 * deal with the case where there is only one pointer in the root 1035 * by promoting the node below to a root 1036 */ 1037 if (!parent) { 1038 struct extent_buffer *child; 1039 1040 if (btrfs_header_nritems(mid) != 1) 1041 return 0; 1042 1043 /* promote the child to a root */ 1044 child = btrfs_read_node_slot(mid, 0); 1045 if (IS_ERR(child)) { 1046 ret = PTR_ERR(child); 1047 goto out; 1048 } 1049 1050 btrfs_tree_lock(child); 1051 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 1052 BTRFS_NESTING_COW); 1053 if (ret) { 1054 btrfs_tree_unlock(child); 1055 free_extent_buffer(child); 1056 goto out; 1057 } 1058 1059 ret = btrfs_tree_mod_log_insert_root(root->node, child, true); 1060 if (ret < 0) { 1061 btrfs_tree_unlock(child); 1062 free_extent_buffer(child); 1063 btrfs_abort_transaction(trans, ret); 1064 goto out; 1065 } 1066 rcu_assign_pointer(root->node, child); 1067 1068 add_root_to_dirty_list(root); 1069 btrfs_tree_unlock(child); 1070 1071 path->locks[level] = 0; 1072 path->nodes[level] = NULL; 1073 btrfs_clear_buffer_dirty(trans, mid); 1074 btrfs_tree_unlock(mid); 1075 /* once for the path */ 1076 free_extent_buffer(mid); 1077 1078 root_sub_used(root, mid->len); 1079 btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 1080 /* once for the root ptr */ 1081 free_extent_buffer_stale(mid); 1082 return 0; 1083 } 1084 if (btrfs_header_nritems(mid) > 1085 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) 1086 return 0; 1087 1088 if (pslot) { 1089 left = btrfs_read_node_slot(parent, pslot - 1); 1090 if (IS_ERR(left)) { 1091 ret = PTR_ERR(left); 1092 left = NULL; 1093 goto out; 1094 } 1095 1096 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 1097 wret = btrfs_cow_block(trans, root, left, 1098 parent, pslot - 1, &left, 1099 BTRFS_NESTING_LEFT_COW); 1100 if (wret) { 1101 ret = wret; 1102 goto out; 1103 } 1104 } 1105 1106 if (pslot + 1 < btrfs_header_nritems(parent)) { 1107 right = btrfs_read_node_slot(parent, pslot + 1); 1108 if (IS_ERR(right)) { 1109 ret = PTR_ERR(right); 1110 right = NULL; 1111 goto out; 1112 } 1113 1114 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 1115 wret = btrfs_cow_block(trans, root, right, 1116 parent, pslot + 1, &right, 1117 BTRFS_NESTING_RIGHT_COW); 1118 if (wret) { 1119 ret = wret; 1120 goto out; 1121 } 1122 } 1123 1124 /* first, try to make some room in the middle buffer */ 1125 if (left) { 1126 orig_slot += btrfs_header_nritems(left); 1127 wret = push_node_left(trans, left, mid, 1); 1128 if (wret < 0) 1129 ret = wret; 1130 } 1131 1132 /* 1133 * then try to empty the right most buffer into the middle 1134 */ 1135 if (right) { 1136 wret = push_node_left(trans, mid, right, 1); 1137 if (wret < 0 && wret != -ENOSPC) 1138 ret = wret; 1139 if (btrfs_header_nritems(right) == 0) { 1140 btrfs_clear_buffer_dirty(trans, right); 1141 btrfs_tree_unlock(right); 1142 btrfs_del_ptr(root, path, level + 1, pslot + 1); 1143 root_sub_used(root, right->len); 1144 btrfs_free_tree_block(trans, btrfs_root_id(root), right, 1145 0, 1); 1146 free_extent_buffer_stale(right); 1147 right = NULL; 1148 } else { 1149 struct btrfs_disk_key right_key; 1150 btrfs_node_key(right, &right_key, 0); 1151 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 1152 BTRFS_MOD_LOG_KEY_REPLACE); 1153 if (ret < 0) { 1154 btrfs_abort_transaction(trans, ret); 1155 goto out; 1156 } 1157 btrfs_set_node_key(parent, &right_key, pslot + 1); 1158 btrfs_mark_buffer_dirty(parent); 1159 } 1160 } 1161 if (btrfs_header_nritems(mid) == 1) { 1162 /* 1163 * we're not allowed to leave a node with one item in the 1164 * tree during a delete. A deletion from lower in the tree 1165 * could try to delete the only pointer in this node. 1166 * So, pull some keys from the left. 1167 * There has to be a left pointer at this point because 1168 * otherwise we would have pulled some pointers from the 1169 * right 1170 */ 1171 if (unlikely(!left)) { 1172 btrfs_crit(fs_info, 1173 "missing left child when middle child only has 1 item, parent bytenr %llu level %d mid bytenr %llu root %llu", 1174 parent->start, btrfs_header_level(parent), 1175 mid->start, btrfs_root_id(root)); 1176 ret = -EUCLEAN; 1177 btrfs_abort_transaction(trans, ret); 1178 goto out; 1179 } 1180 wret = balance_node_right(trans, mid, left); 1181 if (wret < 0) { 1182 ret = wret; 1183 goto out; 1184 } 1185 if (wret == 1) { 1186 wret = push_node_left(trans, left, mid, 1); 1187 if (wret < 0) 1188 ret = wret; 1189 } 1190 BUG_ON(wret == 1); 1191 } 1192 if (btrfs_header_nritems(mid) == 0) { 1193 btrfs_clear_buffer_dirty(trans, mid); 1194 btrfs_tree_unlock(mid); 1195 btrfs_del_ptr(root, path, level + 1, pslot); 1196 root_sub_used(root, mid->len); 1197 btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); 1198 free_extent_buffer_stale(mid); 1199 mid = NULL; 1200 } else { 1201 /* update the parent key to reflect our changes */ 1202 struct btrfs_disk_key mid_key; 1203 btrfs_node_key(mid, &mid_key, 0); 1204 ret = btrfs_tree_mod_log_insert_key(parent, pslot, 1205 BTRFS_MOD_LOG_KEY_REPLACE); 1206 if (ret < 0) { 1207 btrfs_abort_transaction(trans, ret); 1208 goto out; 1209 } 1210 btrfs_set_node_key(parent, &mid_key, pslot); 1211 btrfs_mark_buffer_dirty(parent); 1212 } 1213 1214 /* update the path */ 1215 if (left) { 1216 if (btrfs_header_nritems(left) > orig_slot) { 1217 atomic_inc(&left->refs); 1218 /* left was locked after cow */ 1219 path->nodes[level] = left; 1220 path->slots[level + 1] -= 1; 1221 path->slots[level] = orig_slot; 1222 if (mid) { 1223 btrfs_tree_unlock(mid); 1224 free_extent_buffer(mid); 1225 } 1226 } else { 1227 orig_slot -= btrfs_header_nritems(left); 1228 path->slots[level] = orig_slot; 1229 } 1230 } 1231 /* double check we haven't messed things up */ 1232 if (orig_ptr != 1233 btrfs_node_blockptr(path->nodes[level], path->slots[level])) 1234 BUG(); 1235 out: 1236 if (right) { 1237 btrfs_tree_unlock(right); 1238 free_extent_buffer(right); 1239 } 1240 if (left) { 1241 if (path->nodes[level] != left) 1242 btrfs_tree_unlock(left); 1243 free_extent_buffer(left); 1244 } 1245 return ret; 1246 } 1247 1248 /* Node balancing for insertion. Here we only split or push nodes around 1249 * when they are completely full. This is also done top down, so we 1250 * have to be pessimistic. 1251 */ 1252 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, 1253 struct btrfs_root *root, 1254 struct btrfs_path *path, int level) 1255 { 1256 struct btrfs_fs_info *fs_info = root->fs_info; 1257 struct extent_buffer *right = NULL; 1258 struct extent_buffer *mid; 1259 struct extent_buffer *left = NULL; 1260 struct extent_buffer *parent = NULL; 1261 int ret = 0; 1262 int wret; 1263 int pslot; 1264 int orig_slot = path->slots[level]; 1265 1266 if (level == 0) 1267 return 1; 1268 1269 mid = path->nodes[level]; 1270 WARN_ON(btrfs_header_generation(mid) != trans->transid); 1271 1272 if (level < BTRFS_MAX_LEVEL - 1) { 1273 parent = path->nodes[level + 1]; 1274 pslot = path->slots[level + 1]; 1275 } 1276 1277 if (!parent) 1278 return 1; 1279 1280 /* first, try to make some room in the middle buffer */ 1281 if (pslot) { 1282 u32 left_nr; 1283 1284 left = btrfs_read_node_slot(parent, pslot - 1); 1285 if (IS_ERR(left)) 1286 return PTR_ERR(left); 1287 1288 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 1289 1290 left_nr = btrfs_header_nritems(left); 1291 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 1292 wret = 1; 1293 } else { 1294 ret = btrfs_cow_block(trans, root, left, parent, 1295 pslot - 1, &left, 1296 BTRFS_NESTING_LEFT_COW); 1297 if (ret) 1298 wret = 1; 1299 else { 1300 wret = push_node_left(trans, left, mid, 0); 1301 } 1302 } 1303 if (wret < 0) 1304 ret = wret; 1305 if (wret == 0) { 1306 struct btrfs_disk_key disk_key; 1307 orig_slot += left_nr; 1308 btrfs_node_key(mid, &disk_key, 0); 1309 ret = btrfs_tree_mod_log_insert_key(parent, pslot, 1310 BTRFS_MOD_LOG_KEY_REPLACE); 1311 BUG_ON(ret < 0); 1312 btrfs_set_node_key(parent, &disk_key, pslot); 1313 btrfs_mark_buffer_dirty(parent); 1314 if (btrfs_header_nritems(left) > orig_slot) { 1315 path->nodes[level] = left; 1316 path->slots[level + 1] -= 1; 1317 path->slots[level] = orig_slot; 1318 btrfs_tree_unlock(mid); 1319 free_extent_buffer(mid); 1320 } else { 1321 orig_slot -= 1322 btrfs_header_nritems(left); 1323 path->slots[level] = orig_slot; 1324 btrfs_tree_unlock(left); 1325 free_extent_buffer(left); 1326 } 1327 return 0; 1328 } 1329 btrfs_tree_unlock(left); 1330 free_extent_buffer(left); 1331 } 1332 1333 /* 1334 * then try to empty the right most buffer into the middle 1335 */ 1336 if (pslot + 1 < btrfs_header_nritems(parent)) { 1337 u32 right_nr; 1338 1339 right = btrfs_read_node_slot(parent, pslot + 1); 1340 if (IS_ERR(right)) 1341 return PTR_ERR(right); 1342 1343 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 1344 1345 right_nr = btrfs_header_nritems(right); 1346 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { 1347 wret = 1; 1348 } else { 1349 ret = btrfs_cow_block(trans, root, right, 1350 parent, pslot + 1, 1351 &right, BTRFS_NESTING_RIGHT_COW); 1352 if (ret) 1353 wret = 1; 1354 else { 1355 wret = balance_node_right(trans, right, mid); 1356 } 1357 } 1358 if (wret < 0) 1359 ret = wret; 1360 if (wret == 0) { 1361 struct btrfs_disk_key disk_key; 1362 1363 btrfs_node_key(right, &disk_key, 0); 1364 ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, 1365 BTRFS_MOD_LOG_KEY_REPLACE); 1366 BUG_ON(ret < 0); 1367 btrfs_set_node_key(parent, &disk_key, pslot + 1); 1368 btrfs_mark_buffer_dirty(parent); 1369 1370 if (btrfs_header_nritems(mid) <= orig_slot) { 1371 path->nodes[level] = right; 1372 path->slots[level + 1] += 1; 1373 path->slots[level] = orig_slot - 1374 btrfs_header_nritems(mid); 1375 btrfs_tree_unlock(mid); 1376 free_extent_buffer(mid); 1377 } else { 1378 btrfs_tree_unlock(right); 1379 free_extent_buffer(right); 1380 } 1381 return 0; 1382 } 1383 btrfs_tree_unlock(right); 1384 free_extent_buffer(right); 1385 } 1386 return 1; 1387 } 1388 1389 /* 1390 * readahead one full node of leaves, finding things that are close 1391 * to the block in 'slot', and triggering ra on them. 1392 */ 1393 static void reada_for_search(struct btrfs_fs_info *fs_info, 1394 struct btrfs_path *path, 1395 int level, int slot, u64 objectid) 1396 { 1397 struct extent_buffer *node; 1398 struct btrfs_disk_key disk_key; 1399 u32 nritems; 1400 u64 search; 1401 u64 target; 1402 u64 nread = 0; 1403 u64 nread_max; 1404 u32 nr; 1405 u32 blocksize; 1406 u32 nscan = 0; 1407 1408 if (level != 1 && path->reada != READA_FORWARD_ALWAYS) 1409 return; 1410 1411 if (!path->nodes[level]) 1412 return; 1413 1414 node = path->nodes[level]; 1415 1416 /* 1417 * Since the time between visiting leaves is much shorter than the time 1418 * between visiting nodes, limit read ahead of nodes to 1, to avoid too 1419 * much IO at once (possibly random). 1420 */ 1421 if (path->reada == READA_FORWARD_ALWAYS) { 1422 if (level > 1) 1423 nread_max = node->fs_info->nodesize; 1424 else 1425 nread_max = SZ_128K; 1426 } else { 1427 nread_max = SZ_64K; 1428 } 1429 1430 search = btrfs_node_blockptr(node, slot); 1431 blocksize = fs_info->nodesize; 1432 if (path->reada != READA_FORWARD_ALWAYS) { 1433 struct extent_buffer *eb; 1434 1435 eb = find_extent_buffer(fs_info, search); 1436 if (eb) { 1437 free_extent_buffer(eb); 1438 return; 1439 } 1440 } 1441 1442 target = search; 1443 1444 nritems = btrfs_header_nritems(node); 1445 nr = slot; 1446 1447 while (1) { 1448 if (path->reada == READA_BACK) { 1449 if (nr == 0) 1450 break; 1451 nr--; 1452 } else if (path->reada == READA_FORWARD || 1453 path->reada == READA_FORWARD_ALWAYS) { 1454 nr++; 1455 if (nr >= nritems) 1456 break; 1457 } 1458 if (path->reada == READA_BACK && objectid) { 1459 btrfs_node_key(node, &disk_key, nr); 1460 if (btrfs_disk_key_objectid(&disk_key) != objectid) 1461 break; 1462 } 1463 search = btrfs_node_blockptr(node, nr); 1464 if (path->reada == READA_FORWARD_ALWAYS || 1465 (search <= target && target - search <= 65536) || 1466 (search > target && search - target <= 65536)) { 1467 btrfs_readahead_node_child(node, nr); 1468 nread += blocksize; 1469 } 1470 nscan++; 1471 if (nread > nread_max || nscan > 32) 1472 break; 1473 } 1474 } 1475 1476 static noinline void reada_for_balance(struct btrfs_path *path, int level) 1477 { 1478 struct extent_buffer *parent; 1479 int slot; 1480 int nritems; 1481 1482 parent = path->nodes[level + 1]; 1483 if (!parent) 1484 return; 1485 1486 nritems = btrfs_header_nritems(parent); 1487 slot = path->slots[level + 1]; 1488 1489 if (slot > 0) 1490 btrfs_readahead_node_child(parent, slot - 1); 1491 if (slot + 1 < nritems) 1492 btrfs_readahead_node_child(parent, slot + 1); 1493 } 1494 1495 1496 /* 1497 * when we walk down the tree, it is usually safe to unlock the higher layers 1498 * in the tree. The exceptions are when our path goes through slot 0, because 1499 * operations on the tree might require changing key pointers higher up in the 1500 * tree. 1501 * 1502 * callers might also have set path->keep_locks, which tells this code to keep 1503 * the lock if the path points to the last slot in the block. This is part of 1504 * walking through the tree, and selecting the next slot in the higher block. 1505 * 1506 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so 1507 * if lowest_unlock is 1, level 0 won't be unlocked 1508 */ 1509 static noinline void unlock_up(struct btrfs_path *path, int level, 1510 int lowest_unlock, int min_write_lock_level, 1511 int *write_lock_level) 1512 { 1513 int i; 1514 int skip_level = level; 1515 bool check_skip = true; 1516 1517 for (i = level; i < BTRFS_MAX_LEVEL; i++) { 1518 if (!path->nodes[i]) 1519 break; 1520 if (!path->locks[i]) 1521 break; 1522 1523 if (check_skip) { 1524 if (path->slots[i] == 0) { 1525 skip_level = i + 1; 1526 continue; 1527 } 1528 1529 if (path->keep_locks) { 1530 u32 nritems; 1531 1532 nritems = btrfs_header_nritems(path->nodes[i]); 1533 if (nritems < 1 || path->slots[i] >= nritems - 1) { 1534 skip_level = i + 1; 1535 continue; 1536 } 1537 } 1538 } 1539 1540 if (i >= lowest_unlock && i > skip_level) { 1541 check_skip = false; 1542 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); 1543 path->locks[i] = 0; 1544 if (write_lock_level && 1545 i > min_write_lock_level && 1546 i <= *write_lock_level) { 1547 *write_lock_level = i - 1; 1548 } 1549 } 1550 } 1551 } 1552 1553 /* 1554 * Helper function for btrfs_search_slot() and other functions that do a search 1555 * on a btree. The goal is to find a tree block in the cache (the radix tree at 1556 * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read 1557 * its pages from disk. 1558 * 1559 * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the 1560 * whole btree search, starting again from the current root node. 1561 */ 1562 static int 1563 read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, 1564 struct extent_buffer **eb_ret, int level, int slot, 1565 const struct btrfs_key *key) 1566 { 1567 struct btrfs_fs_info *fs_info = root->fs_info; 1568 struct btrfs_tree_parent_check check = { 0 }; 1569 u64 blocknr; 1570 u64 gen; 1571 struct extent_buffer *tmp; 1572 int ret; 1573 int parent_level; 1574 bool unlock_up; 1575 1576 unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); 1577 blocknr = btrfs_node_blockptr(*eb_ret, slot); 1578 gen = btrfs_node_ptr_generation(*eb_ret, slot); 1579 parent_level = btrfs_header_level(*eb_ret); 1580 btrfs_node_key_to_cpu(*eb_ret, &check.first_key, slot); 1581 check.has_first_key = true; 1582 check.level = parent_level - 1; 1583 check.transid = gen; 1584 check.owner_root = root->root_key.objectid; 1585 1586 /* 1587 * If we need to read an extent buffer from disk and we are holding locks 1588 * on upper level nodes, we unlock all the upper nodes before reading the 1589 * extent buffer, and then return -EAGAIN to the caller as it needs to 1590 * restart the search. We don't release the lock on the current level 1591 * because we need to walk this node to figure out which blocks to read. 1592 */ 1593 tmp = find_extent_buffer(fs_info, blocknr); 1594 if (tmp) { 1595 if (p->reada == READA_FORWARD_ALWAYS) 1596 reada_for_search(fs_info, p, level, slot, key->objectid); 1597 1598 /* first we do an atomic uptodate check */ 1599 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { 1600 /* 1601 * Do extra check for first_key, eb can be stale due to 1602 * being cached, read from scrub, or have multiple 1603 * parents (shared tree blocks). 1604 */ 1605 if (btrfs_verify_level_key(tmp, 1606 parent_level - 1, &check.first_key, gen)) { 1607 free_extent_buffer(tmp); 1608 return -EUCLEAN; 1609 } 1610 *eb_ret = tmp; 1611 return 0; 1612 } 1613 1614 if (p->nowait) { 1615 free_extent_buffer(tmp); 1616 return -EAGAIN; 1617 } 1618 1619 if (unlock_up) 1620 btrfs_unlock_up_safe(p, level + 1); 1621 1622 /* now we're allowed to do a blocking uptodate check */ 1623 ret = btrfs_read_extent_buffer(tmp, &check); 1624 if (ret) { 1625 free_extent_buffer(tmp); 1626 btrfs_release_path(p); 1627 return -EIO; 1628 } 1629 if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { 1630 free_extent_buffer(tmp); 1631 btrfs_release_path(p); 1632 return -EUCLEAN; 1633 } 1634 1635 if (unlock_up) 1636 ret = -EAGAIN; 1637 1638 goto out; 1639 } else if (p->nowait) { 1640 return -EAGAIN; 1641 } 1642 1643 if (unlock_up) { 1644 btrfs_unlock_up_safe(p, level + 1); 1645 ret = -EAGAIN; 1646 } else { 1647 ret = 0; 1648 } 1649 1650 if (p->reada != READA_NONE) 1651 reada_for_search(fs_info, p, level, slot, key->objectid); 1652 1653 tmp = read_tree_block(fs_info, blocknr, &check); 1654 if (IS_ERR(tmp)) { 1655 btrfs_release_path(p); 1656 return PTR_ERR(tmp); 1657 } 1658 /* 1659 * If the read above didn't mark this buffer up to date, 1660 * it will never end up being up to date. Set ret to EIO now 1661 * and give up so that our caller doesn't loop forever 1662 * on our EAGAINs. 1663 */ 1664 if (!extent_buffer_uptodate(tmp)) 1665 ret = -EIO; 1666 1667 out: 1668 if (ret == 0) { 1669 *eb_ret = tmp; 1670 } else { 1671 free_extent_buffer(tmp); 1672 btrfs_release_path(p); 1673 } 1674 1675 return ret; 1676 } 1677 1678 /* 1679 * helper function for btrfs_search_slot. This does all of the checks 1680 * for node-level blocks and does any balancing required based on 1681 * the ins_len. 1682 * 1683 * If no extra work was required, zero is returned. If we had to 1684 * drop the path, -EAGAIN is returned and btrfs_search_slot must 1685 * start over 1686 */ 1687 static int 1688 setup_nodes_for_search(struct btrfs_trans_handle *trans, 1689 struct btrfs_root *root, struct btrfs_path *p, 1690 struct extent_buffer *b, int level, int ins_len, 1691 int *write_lock_level) 1692 { 1693 struct btrfs_fs_info *fs_info = root->fs_info; 1694 int ret = 0; 1695 1696 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= 1697 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { 1698 1699 if (*write_lock_level < level + 1) { 1700 *write_lock_level = level + 1; 1701 btrfs_release_path(p); 1702 return -EAGAIN; 1703 } 1704 1705 reada_for_balance(p, level); 1706 ret = split_node(trans, root, p, level); 1707 1708 b = p->nodes[level]; 1709 } else if (ins_len < 0 && btrfs_header_nritems(b) < 1710 BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { 1711 1712 if (*write_lock_level < level + 1) { 1713 *write_lock_level = level + 1; 1714 btrfs_release_path(p); 1715 return -EAGAIN; 1716 } 1717 1718 reada_for_balance(p, level); 1719 ret = balance_level(trans, root, p, level); 1720 if (ret) 1721 return ret; 1722 1723 b = p->nodes[level]; 1724 if (!b) { 1725 btrfs_release_path(p); 1726 return -EAGAIN; 1727 } 1728 BUG_ON(btrfs_header_nritems(b) == 1); 1729 } 1730 return ret; 1731 } 1732 1733 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, 1734 u64 iobjectid, u64 ioff, u8 key_type, 1735 struct btrfs_key *found_key) 1736 { 1737 int ret; 1738 struct btrfs_key key; 1739 struct extent_buffer *eb; 1740 1741 ASSERT(path); 1742 ASSERT(found_key); 1743 1744 key.type = key_type; 1745 key.objectid = iobjectid; 1746 key.offset = ioff; 1747 1748 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); 1749 if (ret < 0) 1750 return ret; 1751 1752 eb = path->nodes[0]; 1753 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { 1754 ret = btrfs_next_leaf(fs_root, path); 1755 if (ret) 1756 return ret; 1757 eb = path->nodes[0]; 1758 } 1759 1760 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); 1761 if (found_key->type != key.type || 1762 found_key->objectid != key.objectid) 1763 return 1; 1764 1765 return 0; 1766 } 1767 1768 static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, 1769 struct btrfs_path *p, 1770 int write_lock_level) 1771 { 1772 struct extent_buffer *b; 1773 int root_lock = 0; 1774 int level = 0; 1775 1776 if (p->search_commit_root) { 1777 b = root->commit_root; 1778 atomic_inc(&b->refs); 1779 level = btrfs_header_level(b); 1780 /* 1781 * Ensure that all callers have set skip_locking when 1782 * p->search_commit_root = 1. 1783 */ 1784 ASSERT(p->skip_locking == 1); 1785 1786 goto out; 1787 } 1788 1789 if (p->skip_locking) { 1790 b = btrfs_root_node(root); 1791 level = btrfs_header_level(b); 1792 goto out; 1793 } 1794 1795 /* We try very hard to do read locks on the root */ 1796 root_lock = BTRFS_READ_LOCK; 1797 1798 /* 1799 * If the level is set to maximum, we can skip trying to get the read 1800 * lock. 1801 */ 1802 if (write_lock_level < BTRFS_MAX_LEVEL) { 1803 /* 1804 * We don't know the level of the root node until we actually 1805 * have it read locked 1806 */ 1807 if (p->nowait) { 1808 b = btrfs_try_read_lock_root_node(root); 1809 if (IS_ERR(b)) 1810 return b; 1811 } else { 1812 b = btrfs_read_lock_root_node(root); 1813 } 1814 level = btrfs_header_level(b); 1815 if (level > write_lock_level) 1816 goto out; 1817 1818 /* Whoops, must trade for write lock */ 1819 btrfs_tree_read_unlock(b); 1820 free_extent_buffer(b); 1821 } 1822 1823 b = btrfs_lock_root_node(root); 1824 root_lock = BTRFS_WRITE_LOCK; 1825 1826 /* The level might have changed, check again */ 1827 level = btrfs_header_level(b); 1828 1829 out: 1830 /* 1831 * The root may have failed to write out at some point, and thus is no 1832 * longer valid, return an error in this case. 1833 */ 1834 if (!extent_buffer_uptodate(b)) { 1835 if (root_lock) 1836 btrfs_tree_unlock_rw(b, root_lock); 1837 free_extent_buffer(b); 1838 return ERR_PTR(-EIO); 1839 } 1840 1841 p->nodes[level] = b; 1842 if (!p->skip_locking) 1843 p->locks[level] = root_lock; 1844 /* 1845 * Callers are responsible for dropping b's references. 1846 */ 1847 return b; 1848 } 1849 1850 /* 1851 * Replace the extent buffer at the lowest level of the path with a cloned 1852 * version. The purpose is to be able to use it safely, after releasing the 1853 * commit root semaphore, even if relocation is happening in parallel, the 1854 * transaction used for relocation is committed and the extent buffer is 1855 * reallocated in the next transaction. 1856 * 1857 * This is used in a context where the caller does not prevent transaction 1858 * commits from happening, either by holding a transaction handle or holding 1859 * some lock, while it's doing searches through a commit root. 1860 * At the moment it's only used for send operations. 1861 */ 1862 static int finish_need_commit_sem_search(struct btrfs_path *path) 1863 { 1864 const int i = path->lowest_level; 1865 const int slot = path->slots[i]; 1866 struct extent_buffer *lowest = path->nodes[i]; 1867 struct extent_buffer *clone; 1868 1869 ASSERT(path->need_commit_sem); 1870 1871 if (!lowest) 1872 return 0; 1873 1874 lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); 1875 1876 clone = btrfs_clone_extent_buffer(lowest); 1877 if (!clone) 1878 return -ENOMEM; 1879 1880 btrfs_release_path(path); 1881 path->nodes[i] = clone; 1882 path->slots[i] = slot; 1883 1884 return 0; 1885 } 1886 1887 static inline int search_for_key_slot(struct extent_buffer *eb, 1888 int search_low_slot, 1889 const struct btrfs_key *key, 1890 int prev_cmp, 1891 int *slot) 1892 { 1893 /* 1894 * If a previous call to btrfs_bin_search() on a parent node returned an 1895 * exact match (prev_cmp == 0), we can safely assume the target key will 1896 * always be at slot 0 on lower levels, since each key pointer 1897 * (struct btrfs_key_ptr) refers to the lowest key accessible from the 1898 * subtree it points to. Thus we can skip searching lower levels. 1899 */ 1900 if (prev_cmp == 0) { 1901 *slot = 0; 1902 return 0; 1903 } 1904 1905 return btrfs_bin_search(eb, search_low_slot, key, slot); 1906 } 1907 1908 static int search_leaf(struct btrfs_trans_handle *trans, 1909 struct btrfs_root *root, 1910 const struct btrfs_key *key, 1911 struct btrfs_path *path, 1912 int ins_len, 1913 int prev_cmp) 1914 { 1915 struct extent_buffer *leaf = path->nodes[0]; 1916 int leaf_free_space = -1; 1917 int search_low_slot = 0; 1918 int ret; 1919 bool do_bin_search = true; 1920 1921 /* 1922 * If we are doing an insertion, the leaf has enough free space and the 1923 * destination slot for the key is not slot 0, then we can unlock our 1924 * write lock on the parent, and any other upper nodes, before doing the 1925 * binary search on the leaf (with search_for_key_slot()), allowing other 1926 * tasks to lock the parent and any other upper nodes. 1927 */ 1928 if (ins_len > 0) { 1929 /* 1930 * Cache the leaf free space, since we will need it later and it 1931 * will not change until then. 1932 */ 1933 leaf_free_space = btrfs_leaf_free_space(leaf); 1934 1935 /* 1936 * !path->locks[1] means we have a single node tree, the leaf is 1937 * the root of the tree. 1938 */ 1939 if (path->locks[1] && leaf_free_space >= ins_len) { 1940 struct btrfs_disk_key first_key; 1941 1942 ASSERT(btrfs_header_nritems(leaf) > 0); 1943 btrfs_item_key(leaf, &first_key, 0); 1944 1945 /* 1946 * Doing the extra comparison with the first key is cheap, 1947 * taking into account that the first key is very likely 1948 * already in a cache line because it immediately follows 1949 * the extent buffer's header and we have recently accessed 1950 * the header's level field. 1951 */ 1952 ret = comp_keys(&first_key, key); 1953 if (ret < 0) { 1954 /* 1955 * The first key is smaller than the key we want 1956 * to insert, so we are safe to unlock all upper 1957 * nodes and we have to do the binary search. 1958 * 1959 * We do use btrfs_unlock_up_safe() and not 1960 * unlock_up() because the later does not unlock 1961 * nodes with a slot of 0 - we can safely unlock 1962 * any node even if its slot is 0 since in this 1963 * case the key does not end up at slot 0 of the 1964 * leaf and there's no need to split the leaf. 1965 */ 1966 btrfs_unlock_up_safe(path, 1); 1967 search_low_slot = 1; 1968 } else { 1969 /* 1970 * The first key is >= then the key we want to 1971 * insert, so we can skip the binary search as 1972 * the target key will be at slot 0. 1973 * 1974 * We can not unlock upper nodes when the key is 1975 * less than the first key, because we will need 1976 * to update the key at slot 0 of the parent node 1977 * and possibly of other upper nodes too. 1978 * If the key matches the first key, then we can 1979 * unlock all the upper nodes, using 1980 * btrfs_unlock_up_safe() instead of unlock_up() 1981 * as stated above. 1982 */ 1983 if (ret == 0) 1984 btrfs_unlock_up_safe(path, 1); 1985 /* 1986 * ret is already 0 or 1, matching the result of 1987 * a btrfs_bin_search() call, so there is no need 1988 * to adjust it. 1989 */ 1990 do_bin_search = false; 1991 path->slots[0] = 0; 1992 } 1993 } 1994 } 1995 1996 if (do_bin_search) { 1997 ret = search_for_key_slot(leaf, search_low_slot, key, 1998 prev_cmp, &path->slots[0]); 1999 if (ret < 0) 2000 return ret; 2001 } 2002 2003 if (ins_len > 0) { 2004 /* 2005 * Item key already exists. In this case, if we are allowed to 2006 * insert the item (for example, in dir_item case, item key 2007 * collision is allowed), it will be merged with the original 2008 * item. Only the item size grows, no new btrfs item will be 2009 * added. If search_for_extension is not set, ins_len already 2010 * accounts the size btrfs_item, deduct it here so leaf space 2011 * check will be correct. 2012 */ 2013 if (ret == 0 && !path->search_for_extension) { 2014 ASSERT(ins_len >= sizeof(struct btrfs_item)); 2015 ins_len -= sizeof(struct btrfs_item); 2016 } 2017 2018 ASSERT(leaf_free_space >= 0); 2019 2020 if (leaf_free_space < ins_len) { 2021 int err; 2022 2023 err = split_leaf(trans, root, key, path, ins_len, 2024 (ret == 0)); 2025 ASSERT(err <= 0); 2026 if (WARN_ON(err > 0)) 2027 err = -EUCLEAN; 2028 if (err) 2029 ret = err; 2030 } 2031 } 2032 2033 return ret; 2034 } 2035 2036 /* 2037 * btrfs_search_slot - look for a key in a tree and perform necessary 2038 * modifications to preserve tree invariants. 2039 * 2040 * @trans: Handle of transaction, used when modifying the tree 2041 * @p: Holds all btree nodes along the search path 2042 * @root: The root node of the tree 2043 * @key: The key we are looking for 2044 * @ins_len: Indicates purpose of search: 2045 * >0 for inserts it's size of item inserted (*) 2046 * <0 for deletions 2047 * 0 for plain searches, not modifying the tree 2048 * 2049 * (*) If size of item inserted doesn't include 2050 * sizeof(struct btrfs_item), then p->search_for_extension must 2051 * be set. 2052 * @cow: boolean should CoW operations be performed. Must always be 1 2053 * when modifying the tree. 2054 * 2055 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. 2056 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) 2057 * 2058 * If @key is found, 0 is returned and you can find the item in the leaf level 2059 * of the path (level 0) 2060 * 2061 * If @key isn't found, 1 is returned and the leaf level of the path (level 0) 2062 * points to the slot where it should be inserted 2063 * 2064 * If an error is encountered while searching the tree a negative error number 2065 * is returned 2066 */ 2067 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, 2068 const struct btrfs_key *key, struct btrfs_path *p, 2069 int ins_len, int cow) 2070 { 2071 struct btrfs_fs_info *fs_info = root->fs_info; 2072 struct extent_buffer *b; 2073 int slot; 2074 int ret; 2075 int err; 2076 int level; 2077 int lowest_unlock = 1; 2078 /* everything at write_lock_level or lower must be write locked */ 2079 int write_lock_level = 0; 2080 u8 lowest_level = 0; 2081 int min_write_lock_level; 2082 int prev_cmp; 2083 2084 might_sleep(); 2085 2086 lowest_level = p->lowest_level; 2087 WARN_ON(lowest_level && ins_len > 0); 2088 WARN_ON(p->nodes[0] != NULL); 2089 BUG_ON(!cow && ins_len); 2090 2091 /* 2092 * For now only allow nowait for read only operations. There's no 2093 * strict reason why we can't, we just only need it for reads so it's 2094 * only implemented for reads. 2095 */ 2096 ASSERT(!p->nowait || !cow); 2097 2098 if (ins_len < 0) { 2099 lowest_unlock = 2; 2100 2101 /* when we are removing items, we might have to go up to level 2102 * two as we update tree pointers Make sure we keep write 2103 * for those levels as well 2104 */ 2105 write_lock_level = 2; 2106 } else if (ins_len > 0) { 2107 /* 2108 * for inserting items, make sure we have a write lock on 2109 * level 1 so we can update keys 2110 */ 2111 write_lock_level = 1; 2112 } 2113 2114 if (!cow) 2115 write_lock_level = -1; 2116 2117 if (cow && (p->keep_locks || p->lowest_level)) 2118 write_lock_level = BTRFS_MAX_LEVEL; 2119 2120 min_write_lock_level = write_lock_level; 2121 2122 if (p->need_commit_sem) { 2123 ASSERT(p->search_commit_root); 2124 if (p->nowait) { 2125 if (!down_read_trylock(&fs_info->commit_root_sem)) 2126 return -EAGAIN; 2127 } else { 2128 down_read(&fs_info->commit_root_sem); 2129 } 2130 } 2131 2132 again: 2133 prev_cmp = -1; 2134 b = btrfs_search_slot_get_root(root, p, write_lock_level); 2135 if (IS_ERR(b)) { 2136 ret = PTR_ERR(b); 2137 goto done; 2138 } 2139 2140 while (b) { 2141 int dec = 0; 2142 2143 level = btrfs_header_level(b); 2144 2145 if (cow) { 2146 bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); 2147 2148 /* 2149 * if we don't really need to cow this block 2150 * then we don't want to set the path blocking, 2151 * so we test it here 2152 */ 2153 if (!should_cow_block(trans, root, b)) 2154 goto cow_done; 2155 2156 /* 2157 * must have write locks on this node and the 2158 * parent 2159 */ 2160 if (level > write_lock_level || 2161 (level + 1 > write_lock_level && 2162 level + 1 < BTRFS_MAX_LEVEL && 2163 p->nodes[level + 1])) { 2164 write_lock_level = level + 1; 2165 btrfs_release_path(p); 2166 goto again; 2167 } 2168 2169 if (last_level) 2170 err = btrfs_cow_block(trans, root, b, NULL, 0, 2171 &b, 2172 BTRFS_NESTING_COW); 2173 else 2174 err = btrfs_cow_block(trans, root, b, 2175 p->nodes[level + 1], 2176 p->slots[level + 1], &b, 2177 BTRFS_NESTING_COW); 2178 if (err) { 2179 ret = err; 2180 goto done; 2181 } 2182 } 2183 cow_done: 2184 p->nodes[level] = b; 2185 2186 /* 2187 * we have a lock on b and as long as we aren't changing 2188 * the tree, there is no way to for the items in b to change. 2189 * It is safe to drop the lock on our parent before we 2190 * go through the expensive btree search on b. 2191 * 2192 * If we're inserting or deleting (ins_len != 0), then we might 2193 * be changing slot zero, which may require changing the parent. 2194 * So, we can't drop the lock until after we know which slot 2195 * we're operating on. 2196 */ 2197 if (!ins_len && !p->keep_locks) { 2198 int u = level + 1; 2199 2200 if (u < BTRFS_MAX_LEVEL && p->locks[u]) { 2201 btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); 2202 p->locks[u] = 0; 2203 } 2204 } 2205 2206 if (level == 0) { 2207 if (ins_len > 0) 2208 ASSERT(write_lock_level >= 1); 2209 2210 ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); 2211 if (!p->search_for_split) 2212 unlock_up(p, level, lowest_unlock, 2213 min_write_lock_level, NULL); 2214 goto done; 2215 } 2216 2217 ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); 2218 if (ret < 0) 2219 goto done; 2220 prev_cmp = ret; 2221 2222 if (ret && slot > 0) { 2223 dec = 1; 2224 slot--; 2225 } 2226 p->slots[level] = slot; 2227 err = setup_nodes_for_search(trans, root, p, b, level, ins_len, 2228 &write_lock_level); 2229 if (err == -EAGAIN) 2230 goto again; 2231 if (err) { 2232 ret = err; 2233 goto done; 2234 } 2235 b = p->nodes[level]; 2236 slot = p->slots[level]; 2237 2238 /* 2239 * Slot 0 is special, if we change the key we have to update 2240 * the parent pointer which means we must have a write lock on 2241 * the parent 2242 */ 2243 if (slot == 0 && ins_len && write_lock_level < level + 1) { 2244 write_lock_level = level + 1; 2245 btrfs_release_path(p); 2246 goto again; 2247 } 2248 2249 unlock_up(p, level, lowest_unlock, min_write_lock_level, 2250 &write_lock_level); 2251 2252 if (level == lowest_level) { 2253 if (dec) 2254 p->slots[level]++; 2255 goto done; 2256 } 2257 2258 err = read_block_for_search(root, p, &b, level, slot, key); 2259 if (err == -EAGAIN) 2260 goto again; 2261 if (err) { 2262 ret = err; 2263 goto done; 2264 } 2265 2266 if (!p->skip_locking) { 2267 level = btrfs_header_level(b); 2268 2269 btrfs_maybe_reset_lockdep_class(root, b); 2270 2271 if (level <= write_lock_level) { 2272 btrfs_tree_lock(b); 2273 p->locks[level] = BTRFS_WRITE_LOCK; 2274 } else { 2275 if (p->nowait) { 2276 if (!btrfs_try_tree_read_lock(b)) { 2277 free_extent_buffer(b); 2278 ret = -EAGAIN; 2279 goto done; 2280 } 2281 } else { 2282 btrfs_tree_read_lock(b); 2283 } 2284 p->locks[level] = BTRFS_READ_LOCK; 2285 } 2286 p->nodes[level] = b; 2287 } 2288 } 2289 ret = 1; 2290 done: 2291 if (ret < 0 && !p->skip_release_on_error) 2292 btrfs_release_path(p); 2293 2294 if (p->need_commit_sem) { 2295 int ret2; 2296 2297 ret2 = finish_need_commit_sem_search(p); 2298 up_read(&fs_info->commit_root_sem); 2299 if (ret2) 2300 ret = ret2; 2301 } 2302 2303 return ret; 2304 } 2305 ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); 2306 2307 /* 2308 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the 2309 * current state of the tree together with the operations recorded in the tree 2310 * modification log to search for the key in a previous version of this tree, as 2311 * denoted by the time_seq parameter. 2312 * 2313 * Naturally, there is no support for insert, delete or cow operations. 2314 * 2315 * The resulting path and return value will be set up as if we called 2316 * btrfs_search_slot at that point in time with ins_len and cow both set to 0. 2317 */ 2318 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, 2319 struct btrfs_path *p, u64 time_seq) 2320 { 2321 struct btrfs_fs_info *fs_info = root->fs_info; 2322 struct extent_buffer *b; 2323 int slot; 2324 int ret; 2325 int err; 2326 int level; 2327 int lowest_unlock = 1; 2328 u8 lowest_level = 0; 2329 2330 lowest_level = p->lowest_level; 2331 WARN_ON(p->nodes[0] != NULL); 2332 ASSERT(!p->nowait); 2333 2334 if (p->search_commit_root) { 2335 BUG_ON(time_seq); 2336 return btrfs_search_slot(NULL, root, key, p, 0, 0); 2337 } 2338 2339 again: 2340 b = btrfs_get_old_root(root, time_seq); 2341 if (!b) { 2342 ret = -EIO; 2343 goto done; 2344 } 2345 level = btrfs_header_level(b); 2346 p->locks[level] = BTRFS_READ_LOCK; 2347 2348 while (b) { 2349 int dec = 0; 2350 2351 level = btrfs_header_level(b); 2352 p->nodes[level] = b; 2353 2354 /* 2355 * we have a lock on b and as long as we aren't changing 2356 * the tree, there is no way to for the items in b to change. 2357 * It is safe to drop the lock on our parent before we 2358 * go through the expensive btree search on b. 2359 */ 2360 btrfs_unlock_up_safe(p, level + 1); 2361 2362 ret = btrfs_bin_search(b, 0, key, &slot); 2363 if (ret < 0) 2364 goto done; 2365 2366 if (level == 0) { 2367 p->slots[level] = slot; 2368 unlock_up(p, level, lowest_unlock, 0, NULL); 2369 goto done; 2370 } 2371 2372 if (ret && slot > 0) { 2373 dec = 1; 2374 slot--; 2375 } 2376 p->slots[level] = slot; 2377 unlock_up(p, level, lowest_unlock, 0, NULL); 2378 2379 if (level == lowest_level) { 2380 if (dec) 2381 p->slots[level]++; 2382 goto done; 2383 } 2384 2385 err = read_block_for_search(root, p, &b, level, slot, key); 2386 if (err == -EAGAIN) 2387 goto again; 2388 if (err) { 2389 ret = err; 2390 goto done; 2391 } 2392 2393 level = btrfs_header_level(b); 2394 btrfs_tree_read_lock(b); 2395 b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); 2396 if (!b) { 2397 ret = -ENOMEM; 2398 goto done; 2399 } 2400 p->locks[level] = BTRFS_READ_LOCK; 2401 p->nodes[level] = b; 2402 } 2403 ret = 1; 2404 done: 2405 if (ret < 0) 2406 btrfs_release_path(p); 2407 2408 return ret; 2409 } 2410 2411 /* 2412 * Search the tree again to find a leaf with smaller keys. 2413 * Returns 0 if it found something. 2414 * Returns 1 if there are no smaller keys. 2415 * Returns < 0 on error. 2416 * 2417 * This may release the path, and so you may lose any locks held at the 2418 * time you call it. 2419 */ 2420 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) 2421 { 2422 struct btrfs_key key; 2423 struct btrfs_key orig_key; 2424 struct btrfs_disk_key found_key; 2425 int ret; 2426 2427 btrfs_item_key_to_cpu(path->nodes[0], &key, 0); 2428 orig_key = key; 2429 2430 if (key.offset > 0) { 2431 key.offset--; 2432 } else if (key.type > 0) { 2433 key.type--; 2434 key.offset = (u64)-1; 2435 } else if (key.objectid > 0) { 2436 key.objectid--; 2437 key.type = (u8)-1; 2438 key.offset = (u64)-1; 2439 } else { 2440 return 1; 2441 } 2442 2443 btrfs_release_path(path); 2444 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 2445 if (ret <= 0) 2446 return ret; 2447 2448 /* 2449 * Previous key not found. Even if we were at slot 0 of the leaf we had 2450 * before releasing the path and calling btrfs_search_slot(), we now may 2451 * be in a slot pointing to the same original key - this can happen if 2452 * after we released the path, one of more items were moved from a 2453 * sibling leaf into the front of the leaf we had due to an insertion 2454 * (see push_leaf_right()). 2455 * If we hit this case and our slot is > 0 and just decrement the slot 2456 * so that the caller does not process the same key again, which may or 2457 * may not break the caller, depending on its logic. 2458 */ 2459 if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { 2460 btrfs_item_key(path->nodes[0], &found_key, path->slots[0]); 2461 ret = comp_keys(&found_key, &orig_key); 2462 if (ret == 0) { 2463 if (path->slots[0] > 0) { 2464 path->slots[0]--; 2465 return 0; 2466 } 2467 /* 2468 * At slot 0, same key as before, it means orig_key is 2469 * the lowest, leftmost, key in the tree. We're done. 2470 */ 2471 return 1; 2472 } 2473 } 2474 2475 btrfs_item_key(path->nodes[0], &found_key, 0); 2476 ret = comp_keys(&found_key, &key); 2477 /* 2478 * We might have had an item with the previous key in the tree right 2479 * before we released our path. And after we released our path, that 2480 * item might have been pushed to the first slot (0) of the leaf we 2481 * were holding due to a tree balance. Alternatively, an item with the 2482 * previous key can exist as the only element of a leaf (big fat item). 2483 * Therefore account for these 2 cases, so that our callers (like 2484 * btrfs_previous_item) don't miss an existing item with a key matching 2485 * the previous key we computed above. 2486 */ 2487 if (ret <= 0) 2488 return 0; 2489 return 1; 2490 } 2491 2492 /* 2493 * helper to use instead of search slot if no exact match is needed but 2494 * instead the next or previous item should be returned. 2495 * When find_higher is true, the next higher item is returned, the next lower 2496 * otherwise. 2497 * When return_any and find_higher are both true, and no higher item is found, 2498 * return the next lower instead. 2499 * When return_any is true and find_higher is false, and no lower item is found, 2500 * return the next higher instead. 2501 * It returns 0 if any item is found, 1 if none is found (tree empty), and 2502 * < 0 on error 2503 */ 2504 int btrfs_search_slot_for_read(struct btrfs_root *root, 2505 const struct btrfs_key *key, 2506 struct btrfs_path *p, int find_higher, 2507 int return_any) 2508 { 2509 int ret; 2510 struct extent_buffer *leaf; 2511 2512 again: 2513 ret = btrfs_search_slot(NULL, root, key, p, 0, 0); 2514 if (ret <= 0) 2515 return ret; 2516 /* 2517 * a return value of 1 means the path is at the position where the 2518 * item should be inserted. Normally this is the next bigger item, 2519 * but in case the previous item is the last in a leaf, path points 2520 * to the first free slot in the previous leaf, i.e. at an invalid 2521 * item. 2522 */ 2523 leaf = p->nodes[0]; 2524 2525 if (find_higher) { 2526 if (p->slots[0] >= btrfs_header_nritems(leaf)) { 2527 ret = btrfs_next_leaf(root, p); 2528 if (ret <= 0) 2529 return ret; 2530 if (!return_any) 2531 return 1; 2532 /* 2533 * no higher item found, return the next 2534 * lower instead 2535 */ 2536 return_any = 0; 2537 find_higher = 0; 2538 btrfs_release_path(p); 2539 goto again; 2540 } 2541 } else { 2542 if (p->slots[0] == 0) { 2543 ret = btrfs_prev_leaf(root, p); 2544 if (ret < 0) 2545 return ret; 2546 if (!ret) { 2547 leaf = p->nodes[0]; 2548 if (p->slots[0] == btrfs_header_nritems(leaf)) 2549 p->slots[0]--; 2550 return 0; 2551 } 2552 if (!return_any) 2553 return 1; 2554 /* 2555 * no lower item found, return the next 2556 * higher instead 2557 */ 2558 return_any = 0; 2559 find_higher = 1; 2560 btrfs_release_path(p); 2561 goto again; 2562 } else { 2563 --p->slots[0]; 2564 } 2565 } 2566 return 0; 2567 } 2568 2569 /* 2570 * Execute search and call btrfs_previous_item to traverse backwards if the item 2571 * was not found. 2572 * 2573 * Return 0 if found, 1 if not found and < 0 if error. 2574 */ 2575 int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, 2576 struct btrfs_path *path) 2577 { 2578 int ret; 2579 2580 ret = btrfs_search_slot(NULL, root, key, path, 0, 0); 2581 if (ret > 0) 2582 ret = btrfs_previous_item(root, path, key->objectid, key->type); 2583 2584 if (ret == 0) 2585 btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 2586 2587 return ret; 2588 } 2589 2590 /* 2591 * Search for a valid slot for the given path. 2592 * 2593 * @root: The root node of the tree. 2594 * @key: Will contain a valid item if found. 2595 * @path: The starting point to validate the slot. 2596 * 2597 * Return: 0 if the item is valid 2598 * 1 if not found 2599 * <0 if error. 2600 */ 2601 int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, 2602 struct btrfs_path *path) 2603 { 2604 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 2605 int ret; 2606 2607 ret = btrfs_next_leaf(root, path); 2608 if (ret) 2609 return ret; 2610 } 2611 2612 btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); 2613 return 0; 2614 } 2615 2616 /* 2617 * adjust the pointers going up the tree, starting at level 2618 * making sure the right key of each node is points to 'key'. 2619 * This is used after shifting pointers to the left, so it stops 2620 * fixing up pointers when a given leaf/node is not in slot 0 of the 2621 * higher levels 2622 * 2623 */ 2624 static void fixup_low_keys(struct btrfs_path *path, 2625 struct btrfs_disk_key *key, int level) 2626 { 2627 int i; 2628 struct extent_buffer *t; 2629 int ret; 2630 2631 for (i = level; i < BTRFS_MAX_LEVEL; i++) { 2632 int tslot = path->slots[i]; 2633 2634 if (!path->nodes[i]) 2635 break; 2636 t = path->nodes[i]; 2637 ret = btrfs_tree_mod_log_insert_key(t, tslot, 2638 BTRFS_MOD_LOG_KEY_REPLACE); 2639 BUG_ON(ret < 0); 2640 btrfs_set_node_key(t, key, tslot); 2641 btrfs_mark_buffer_dirty(path->nodes[i]); 2642 if (tslot != 0) 2643 break; 2644 } 2645 } 2646 2647 /* 2648 * update item key. 2649 * 2650 * This function isn't completely safe. It's the caller's responsibility 2651 * that the new key won't break the order 2652 */ 2653 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, 2654 struct btrfs_path *path, 2655 const struct btrfs_key *new_key) 2656 { 2657 struct btrfs_disk_key disk_key; 2658 struct extent_buffer *eb; 2659 int slot; 2660 2661 eb = path->nodes[0]; 2662 slot = path->slots[0]; 2663 if (slot > 0) { 2664 btrfs_item_key(eb, &disk_key, slot - 1); 2665 if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { 2666 btrfs_print_leaf(eb); 2667 btrfs_crit(fs_info, 2668 "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 2669 slot, btrfs_disk_key_objectid(&disk_key), 2670 btrfs_disk_key_type(&disk_key), 2671 btrfs_disk_key_offset(&disk_key), 2672 new_key->objectid, new_key->type, 2673 new_key->offset); 2674 BUG(); 2675 } 2676 } 2677 if (slot < btrfs_header_nritems(eb) - 1) { 2678 btrfs_item_key(eb, &disk_key, slot + 1); 2679 if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { 2680 btrfs_print_leaf(eb); 2681 btrfs_crit(fs_info, 2682 "slot %u key (%llu %u %llu) new key (%llu %u %llu)", 2683 slot, btrfs_disk_key_objectid(&disk_key), 2684 btrfs_disk_key_type(&disk_key), 2685 btrfs_disk_key_offset(&disk_key), 2686 new_key->objectid, new_key->type, 2687 new_key->offset); 2688 BUG(); 2689 } 2690 } 2691 2692 btrfs_cpu_key_to_disk(&disk_key, new_key); 2693 btrfs_set_item_key(eb, &disk_key, slot); 2694 btrfs_mark_buffer_dirty(eb); 2695 if (slot == 0) 2696 fixup_low_keys(path, &disk_key, 1); 2697 } 2698 2699 /* 2700 * Check key order of two sibling extent buffers. 2701 * 2702 * Return true if something is wrong. 2703 * Return false if everything is fine. 2704 * 2705 * Tree-checker only works inside one tree block, thus the following 2706 * corruption can not be detected by tree-checker: 2707 * 2708 * Leaf @left | Leaf @right 2709 * -------------------------------------------------------------- 2710 * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | 2711 * 2712 * Key f6 in leaf @left itself is valid, but not valid when the next 2713 * key in leaf @right is 7. 2714 * This can only be checked at tree block merge time. 2715 * And since tree checker has ensured all key order in each tree block 2716 * is correct, we only need to bother the last key of @left and the first 2717 * key of @right. 2718 */ 2719 static bool check_sibling_keys(struct extent_buffer *left, 2720 struct extent_buffer *right) 2721 { 2722 struct btrfs_key left_last; 2723 struct btrfs_key right_first; 2724 int level = btrfs_header_level(left); 2725 int nr_left = btrfs_header_nritems(left); 2726 int nr_right = btrfs_header_nritems(right); 2727 2728 /* No key to check in one of the tree blocks */ 2729 if (!nr_left || !nr_right) 2730 return false; 2731 2732 if (level) { 2733 btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); 2734 btrfs_node_key_to_cpu(right, &right_first, 0); 2735 } else { 2736 btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); 2737 btrfs_item_key_to_cpu(right, &right_first, 0); 2738 } 2739 2740 if (unlikely(btrfs_comp_cpu_keys(&left_last, &right_first) >= 0)) { 2741 btrfs_crit(left->fs_info, "left extent buffer:"); 2742 btrfs_print_tree(left, false); 2743 btrfs_crit(left->fs_info, "right extent buffer:"); 2744 btrfs_print_tree(right, false); 2745 btrfs_crit(left->fs_info, 2746 "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", 2747 left_last.objectid, left_last.type, 2748 left_last.offset, right_first.objectid, 2749 right_first.type, right_first.offset); 2750 return true; 2751 } 2752 return false; 2753 } 2754 2755 /* 2756 * try to push data from one node into the next node left in the 2757 * tree. 2758 * 2759 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible 2760 * error, and > 0 if there was no room in the left hand block. 2761 */ 2762 static int push_node_left(struct btrfs_trans_handle *trans, 2763 struct extent_buffer *dst, 2764 struct extent_buffer *src, int empty) 2765 { 2766 struct btrfs_fs_info *fs_info = trans->fs_info; 2767 int push_items = 0; 2768 int src_nritems; 2769 int dst_nritems; 2770 int ret = 0; 2771 2772 src_nritems = btrfs_header_nritems(src); 2773 dst_nritems = btrfs_header_nritems(dst); 2774 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 2775 WARN_ON(btrfs_header_generation(src) != trans->transid); 2776 WARN_ON(btrfs_header_generation(dst) != trans->transid); 2777 2778 if (!empty && src_nritems <= 8) 2779 return 1; 2780 2781 if (push_items <= 0) 2782 return 1; 2783 2784 if (empty) { 2785 push_items = min(src_nritems, push_items); 2786 if (push_items < src_nritems) { 2787 /* leave at least 8 pointers in the node if 2788 * we aren't going to empty it 2789 */ 2790 if (src_nritems - push_items < 8) { 2791 if (push_items <= 8) 2792 return 1; 2793 push_items -= 8; 2794 } 2795 } 2796 } else 2797 push_items = min(src_nritems - 8, push_items); 2798 2799 /* dst is the left eb, src is the middle eb */ 2800 if (check_sibling_keys(dst, src)) { 2801 ret = -EUCLEAN; 2802 btrfs_abort_transaction(trans, ret); 2803 return ret; 2804 } 2805 ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); 2806 if (ret) { 2807 btrfs_abort_transaction(trans, ret); 2808 return ret; 2809 } 2810 copy_extent_buffer(dst, src, 2811 btrfs_node_key_ptr_offset(dst, dst_nritems), 2812 btrfs_node_key_ptr_offset(src, 0), 2813 push_items * sizeof(struct btrfs_key_ptr)); 2814 2815 if (push_items < src_nritems) { 2816 /* 2817 * btrfs_tree_mod_log_eb_copy handles logging the move, so we 2818 * don't need to do an explicit tree mod log operation for it. 2819 */ 2820 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(src, 0), 2821 btrfs_node_key_ptr_offset(src, push_items), 2822 (src_nritems - push_items) * 2823 sizeof(struct btrfs_key_ptr)); 2824 } 2825 btrfs_set_header_nritems(src, src_nritems - push_items); 2826 btrfs_set_header_nritems(dst, dst_nritems + push_items); 2827 btrfs_mark_buffer_dirty(src); 2828 btrfs_mark_buffer_dirty(dst); 2829 2830 return ret; 2831 } 2832 2833 /* 2834 * try to push data from one node into the next node right in the 2835 * tree. 2836 * 2837 * returns 0 if some ptrs were pushed, < 0 if there was some horrible 2838 * error, and > 0 if there was no room in the right hand block. 2839 * 2840 * this will only push up to 1/2 the contents of the left node over 2841 */ 2842 static int balance_node_right(struct btrfs_trans_handle *trans, 2843 struct extent_buffer *dst, 2844 struct extent_buffer *src) 2845 { 2846 struct btrfs_fs_info *fs_info = trans->fs_info; 2847 int push_items = 0; 2848 int max_push; 2849 int src_nritems; 2850 int dst_nritems; 2851 int ret = 0; 2852 2853 WARN_ON(btrfs_header_generation(src) != trans->transid); 2854 WARN_ON(btrfs_header_generation(dst) != trans->transid); 2855 2856 src_nritems = btrfs_header_nritems(src); 2857 dst_nritems = btrfs_header_nritems(dst); 2858 push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; 2859 if (push_items <= 0) 2860 return 1; 2861 2862 if (src_nritems < 4) 2863 return 1; 2864 2865 max_push = src_nritems / 2 + 1; 2866 /* don't try to empty the node */ 2867 if (max_push >= src_nritems) 2868 return 1; 2869 2870 if (max_push < push_items) 2871 push_items = max_push; 2872 2873 /* dst is the right eb, src is the middle eb */ 2874 if (check_sibling_keys(src, dst)) { 2875 ret = -EUCLEAN; 2876 btrfs_abort_transaction(trans, ret); 2877 return ret; 2878 } 2879 2880 /* 2881 * btrfs_tree_mod_log_eb_copy handles logging the move, so we don't 2882 * need to do an explicit tree mod log operation for it. 2883 */ 2884 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(dst, push_items), 2885 btrfs_node_key_ptr_offset(dst, 0), 2886 (dst_nritems) * 2887 sizeof(struct btrfs_key_ptr)); 2888 2889 ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, 2890 push_items); 2891 if (ret) { 2892 btrfs_abort_transaction(trans, ret); 2893 return ret; 2894 } 2895 copy_extent_buffer(dst, src, 2896 btrfs_node_key_ptr_offset(dst, 0), 2897 btrfs_node_key_ptr_offset(src, src_nritems - push_items), 2898 push_items * sizeof(struct btrfs_key_ptr)); 2899 2900 btrfs_set_header_nritems(src, src_nritems - push_items); 2901 btrfs_set_header_nritems(dst, dst_nritems + push_items); 2902 2903 btrfs_mark_buffer_dirty(src); 2904 btrfs_mark_buffer_dirty(dst); 2905 2906 return ret; 2907 } 2908 2909 /* 2910 * helper function to insert a new root level in the tree. 2911 * A new node is allocated, and a single item is inserted to 2912 * point to the existing root 2913 * 2914 * returns zero on success or < 0 on failure. 2915 */ 2916 static noinline int insert_new_root(struct btrfs_trans_handle *trans, 2917 struct btrfs_root *root, 2918 struct btrfs_path *path, int level) 2919 { 2920 struct btrfs_fs_info *fs_info = root->fs_info; 2921 u64 lower_gen; 2922 struct extent_buffer *lower; 2923 struct extent_buffer *c; 2924 struct extent_buffer *old; 2925 struct btrfs_disk_key lower_key; 2926 int ret; 2927 2928 BUG_ON(path->nodes[level]); 2929 BUG_ON(path->nodes[level-1] != root->node); 2930 2931 lower = path->nodes[level-1]; 2932 if (level == 1) 2933 btrfs_item_key(lower, &lower_key, 0); 2934 else 2935 btrfs_node_key(lower, &lower_key, 0); 2936 2937 c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 2938 &lower_key, level, root->node->start, 0, 2939 BTRFS_NESTING_NEW_ROOT); 2940 if (IS_ERR(c)) 2941 return PTR_ERR(c); 2942 2943 root_add_used(root, fs_info->nodesize); 2944 2945 btrfs_set_header_nritems(c, 1); 2946 btrfs_set_node_key(c, &lower_key, 0); 2947 btrfs_set_node_blockptr(c, 0, lower->start); 2948 lower_gen = btrfs_header_generation(lower); 2949 WARN_ON(lower_gen != trans->transid); 2950 2951 btrfs_set_node_ptr_generation(c, 0, lower_gen); 2952 2953 btrfs_mark_buffer_dirty(c); 2954 2955 old = root->node; 2956 ret = btrfs_tree_mod_log_insert_root(root->node, c, false); 2957 BUG_ON(ret < 0); 2958 rcu_assign_pointer(root->node, c); 2959 2960 /* the super has an extra ref to root->node */ 2961 free_extent_buffer(old); 2962 2963 add_root_to_dirty_list(root); 2964 atomic_inc(&c->refs); 2965 path->nodes[level] = c; 2966 path->locks[level] = BTRFS_WRITE_LOCK; 2967 path->slots[level] = 0; 2968 return 0; 2969 } 2970 2971 /* 2972 * worker function to insert a single pointer in a node. 2973 * the node should have enough room for the pointer already 2974 * 2975 * slot and level indicate where you want the key to go, and 2976 * blocknr is the block the key points to. 2977 */ 2978 static void insert_ptr(struct btrfs_trans_handle *trans, 2979 struct btrfs_path *path, 2980 struct btrfs_disk_key *key, u64 bytenr, 2981 int slot, int level) 2982 { 2983 struct extent_buffer *lower; 2984 int nritems; 2985 int ret; 2986 2987 BUG_ON(!path->nodes[level]); 2988 btrfs_assert_tree_write_locked(path->nodes[level]); 2989 lower = path->nodes[level]; 2990 nritems = btrfs_header_nritems(lower); 2991 BUG_ON(slot > nritems); 2992 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); 2993 if (slot != nritems) { 2994 if (level) { 2995 ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, 2996 slot, nritems - slot); 2997 BUG_ON(ret < 0); 2998 } 2999 memmove_extent_buffer(lower, 3000 btrfs_node_key_ptr_offset(lower, slot + 1), 3001 btrfs_node_key_ptr_offset(lower, slot), 3002 (nritems - slot) * sizeof(struct btrfs_key_ptr)); 3003 } 3004 if (level) { 3005 ret = btrfs_tree_mod_log_insert_key(lower, slot, 3006 BTRFS_MOD_LOG_KEY_ADD); 3007 BUG_ON(ret < 0); 3008 } 3009 btrfs_set_node_key(lower, key, slot); 3010 btrfs_set_node_blockptr(lower, slot, bytenr); 3011 WARN_ON(trans->transid == 0); 3012 btrfs_set_node_ptr_generation(lower, slot, trans->transid); 3013 btrfs_set_header_nritems(lower, nritems + 1); 3014 btrfs_mark_buffer_dirty(lower); 3015 } 3016 3017 /* 3018 * split the node at the specified level in path in two. 3019 * The path is corrected to point to the appropriate node after the split 3020 * 3021 * Before splitting this tries to make some room in the node by pushing 3022 * left and right, if either one works, it returns right away. 3023 * 3024 * returns 0 on success and < 0 on failure 3025 */ 3026 static noinline int split_node(struct btrfs_trans_handle *trans, 3027 struct btrfs_root *root, 3028 struct btrfs_path *path, int level) 3029 { 3030 struct btrfs_fs_info *fs_info = root->fs_info; 3031 struct extent_buffer *c; 3032 struct extent_buffer *split; 3033 struct btrfs_disk_key disk_key; 3034 int mid; 3035 int ret; 3036 u32 c_nritems; 3037 3038 c = path->nodes[level]; 3039 WARN_ON(btrfs_header_generation(c) != trans->transid); 3040 if (c == root->node) { 3041 /* 3042 * trying to split the root, lets make a new one 3043 * 3044 * tree mod log: We don't log_removal old root in 3045 * insert_new_root, because that root buffer will be kept as a 3046 * normal node. We are going to log removal of half of the 3047 * elements below with btrfs_tree_mod_log_eb_copy(). We're 3048 * holding a tree lock on the buffer, which is why we cannot 3049 * race with other tree_mod_log users. 3050 */ 3051 ret = insert_new_root(trans, root, path, level + 1); 3052 if (ret) 3053 return ret; 3054 } else { 3055 ret = push_nodes_for_insert(trans, root, path, level); 3056 c = path->nodes[level]; 3057 if (!ret && btrfs_header_nritems(c) < 3058 BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) 3059 return 0; 3060 if (ret < 0) 3061 return ret; 3062 } 3063 3064 c_nritems = btrfs_header_nritems(c); 3065 mid = (c_nritems + 1) / 2; 3066 btrfs_node_key(c, &disk_key, mid); 3067 3068 split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 3069 &disk_key, level, c->start, 0, 3070 BTRFS_NESTING_SPLIT); 3071 if (IS_ERR(split)) 3072 return PTR_ERR(split); 3073 3074 root_add_used(root, fs_info->nodesize); 3075 ASSERT(btrfs_header_level(c) == level); 3076 3077 ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); 3078 if (ret) { 3079 btrfs_tree_unlock(split); 3080 free_extent_buffer(split); 3081 btrfs_abort_transaction(trans, ret); 3082 return ret; 3083 } 3084 copy_extent_buffer(split, c, 3085 btrfs_node_key_ptr_offset(split, 0), 3086 btrfs_node_key_ptr_offset(c, mid), 3087 (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); 3088 btrfs_set_header_nritems(split, c_nritems - mid); 3089 btrfs_set_header_nritems(c, mid); 3090 3091 btrfs_mark_buffer_dirty(c); 3092 btrfs_mark_buffer_dirty(split); 3093 3094 insert_ptr(trans, path, &disk_key, split->start, 3095 path->slots[level + 1] + 1, level + 1); 3096 3097 if (path->slots[level] >= mid) { 3098 path->slots[level] -= mid; 3099 btrfs_tree_unlock(c); 3100 free_extent_buffer(c); 3101 path->nodes[level] = split; 3102 path->slots[level + 1] += 1; 3103 } else { 3104 btrfs_tree_unlock(split); 3105 free_extent_buffer(split); 3106 } 3107 return 0; 3108 } 3109 3110 /* 3111 * how many bytes are required to store the items in a leaf. start 3112 * and nr indicate which items in the leaf to check. This totals up the 3113 * space used both by the item structs and the item data 3114 */ 3115 static int leaf_space_used(const struct extent_buffer *l, int start, int nr) 3116 { 3117 int data_len; 3118 int nritems = btrfs_header_nritems(l); 3119 int end = min(nritems, start + nr) - 1; 3120 3121 if (!nr) 3122 return 0; 3123 data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); 3124 data_len = data_len - btrfs_item_offset(l, end); 3125 data_len += sizeof(struct btrfs_item) * nr; 3126 WARN_ON(data_len < 0); 3127 return data_len; 3128 } 3129 3130 /* 3131 * The space between the end of the leaf items and 3132 * the start of the leaf data. IOW, how much room 3133 * the leaf has left for both items and data 3134 */ 3135 int btrfs_leaf_free_space(const struct extent_buffer *leaf) 3136 { 3137 struct btrfs_fs_info *fs_info = leaf->fs_info; 3138 int nritems = btrfs_header_nritems(leaf); 3139 int ret; 3140 3141 ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); 3142 if (ret < 0) { 3143 btrfs_crit(fs_info, 3144 "leaf free space ret %d, leaf data size %lu, used %d nritems %d", 3145 ret, 3146 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), 3147 leaf_space_used(leaf, 0, nritems), nritems); 3148 } 3149 return ret; 3150 } 3151 3152 /* 3153 * min slot controls the lowest index we're willing to push to the 3154 * right. We'll push up to and including min_slot, but no lower 3155 */ 3156 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, 3157 struct btrfs_path *path, 3158 int data_size, int empty, 3159 struct extent_buffer *right, 3160 int free_space, u32 left_nritems, 3161 u32 min_slot) 3162 { 3163 struct btrfs_fs_info *fs_info = right->fs_info; 3164 struct extent_buffer *left = path->nodes[0]; 3165 struct extent_buffer *upper = path->nodes[1]; 3166 struct btrfs_map_token token; 3167 struct btrfs_disk_key disk_key; 3168 int slot; 3169 u32 i; 3170 int push_space = 0; 3171 int push_items = 0; 3172 u32 nr; 3173 u32 right_nritems; 3174 u32 data_end; 3175 u32 this_item_size; 3176 3177 if (empty) 3178 nr = 0; 3179 else 3180 nr = max_t(u32, 1, min_slot); 3181 3182 if (path->slots[0] >= left_nritems) 3183 push_space += data_size; 3184 3185 slot = path->slots[1]; 3186 i = left_nritems - 1; 3187 while (i >= nr) { 3188 if (!empty && push_items > 0) { 3189 if (path->slots[0] > i) 3190 break; 3191 if (path->slots[0] == i) { 3192 int space = btrfs_leaf_free_space(left); 3193 3194 if (space + push_space * 2 > free_space) 3195 break; 3196 } 3197 } 3198 3199 if (path->slots[0] == i) 3200 push_space += data_size; 3201 3202 this_item_size = btrfs_item_size(left, i); 3203 if (this_item_size + sizeof(struct btrfs_item) + 3204 push_space > free_space) 3205 break; 3206 3207 push_items++; 3208 push_space += this_item_size + sizeof(struct btrfs_item); 3209 if (i == 0) 3210 break; 3211 i--; 3212 } 3213 3214 if (push_items == 0) 3215 goto out_unlock; 3216 3217 WARN_ON(!empty && push_items == left_nritems); 3218 3219 /* push left to right */ 3220 right_nritems = btrfs_header_nritems(right); 3221 3222 push_space = btrfs_item_data_end(left, left_nritems - push_items); 3223 push_space -= leaf_data_end(left); 3224 3225 /* make room in the right data area */ 3226 data_end = leaf_data_end(right); 3227 memmove_leaf_data(right, data_end - push_space, data_end, 3228 BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); 3229 3230 /* copy from the left data area */ 3231 copy_leaf_data(right, left, BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 3232 leaf_data_end(left), push_space); 3233 3234 memmove_leaf_items(right, push_items, 0, right_nritems); 3235 3236 /* copy the items from left to right */ 3237 copy_leaf_items(right, left, 0, left_nritems - push_items, push_items); 3238 3239 /* update the item pointers */ 3240 btrfs_init_map_token(&token, right); 3241 right_nritems += push_items; 3242 btrfs_set_header_nritems(right, right_nritems); 3243 push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 3244 for (i = 0; i < right_nritems; i++) { 3245 push_space -= btrfs_token_item_size(&token, i); 3246 btrfs_set_token_item_offset(&token, i, push_space); 3247 } 3248 3249 left_nritems -= push_items; 3250 btrfs_set_header_nritems(left, left_nritems); 3251 3252 if (left_nritems) 3253 btrfs_mark_buffer_dirty(left); 3254 else 3255 btrfs_clear_buffer_dirty(trans, left); 3256 3257 btrfs_mark_buffer_dirty(right); 3258 3259 btrfs_item_key(right, &disk_key, 0); 3260 btrfs_set_node_key(upper, &disk_key, slot + 1); 3261 btrfs_mark_buffer_dirty(upper); 3262 3263 /* then fixup the leaf pointer in the path */ 3264 if (path->slots[0] >= left_nritems) { 3265 path->slots[0] -= left_nritems; 3266 if (btrfs_header_nritems(path->nodes[0]) == 0) 3267 btrfs_clear_buffer_dirty(trans, path->nodes[0]); 3268 btrfs_tree_unlock(path->nodes[0]); 3269 free_extent_buffer(path->nodes[0]); 3270 path->nodes[0] = right; 3271 path->slots[1] += 1; 3272 } else { 3273 btrfs_tree_unlock(right); 3274 free_extent_buffer(right); 3275 } 3276 return 0; 3277 3278 out_unlock: 3279 btrfs_tree_unlock(right); 3280 free_extent_buffer(right); 3281 return 1; 3282 } 3283 3284 /* 3285 * push some data in the path leaf to the right, trying to free up at 3286 * least data_size bytes. returns zero if the push worked, nonzero otherwise 3287 * 3288 * returns 1 if the push failed because the other node didn't have enough 3289 * room, 0 if everything worked out and < 0 if there were major errors. 3290 * 3291 * this will push starting from min_slot to the end of the leaf. It won't 3292 * push any slot lower than min_slot 3293 */ 3294 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root 3295 *root, struct btrfs_path *path, 3296 int min_data_size, int data_size, 3297 int empty, u32 min_slot) 3298 { 3299 struct extent_buffer *left = path->nodes[0]; 3300 struct extent_buffer *right; 3301 struct extent_buffer *upper; 3302 int slot; 3303 int free_space; 3304 u32 left_nritems; 3305 int ret; 3306 3307 if (!path->nodes[1]) 3308 return 1; 3309 3310 slot = path->slots[1]; 3311 upper = path->nodes[1]; 3312 if (slot >= btrfs_header_nritems(upper) - 1) 3313 return 1; 3314 3315 btrfs_assert_tree_write_locked(path->nodes[1]); 3316 3317 right = btrfs_read_node_slot(upper, slot + 1); 3318 if (IS_ERR(right)) 3319 return PTR_ERR(right); 3320 3321 __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); 3322 3323 free_space = btrfs_leaf_free_space(right); 3324 if (free_space < data_size) 3325 goto out_unlock; 3326 3327 ret = btrfs_cow_block(trans, root, right, upper, 3328 slot + 1, &right, BTRFS_NESTING_RIGHT_COW); 3329 if (ret) 3330 goto out_unlock; 3331 3332 left_nritems = btrfs_header_nritems(left); 3333 if (left_nritems == 0) 3334 goto out_unlock; 3335 3336 if (check_sibling_keys(left, right)) { 3337 ret = -EUCLEAN; 3338 btrfs_abort_transaction(trans, ret); 3339 btrfs_tree_unlock(right); 3340 free_extent_buffer(right); 3341 return ret; 3342 } 3343 if (path->slots[0] == left_nritems && !empty) { 3344 /* Key greater than all keys in the leaf, right neighbor has 3345 * enough room for it and we're not emptying our leaf to delete 3346 * it, therefore use right neighbor to insert the new item and 3347 * no need to touch/dirty our left leaf. */ 3348 btrfs_tree_unlock(left); 3349 free_extent_buffer(left); 3350 path->nodes[0] = right; 3351 path->slots[0] = 0; 3352 path->slots[1]++; 3353 return 0; 3354 } 3355 3356 return __push_leaf_right(trans, path, min_data_size, empty, right, 3357 free_space, left_nritems, min_slot); 3358 out_unlock: 3359 btrfs_tree_unlock(right); 3360 free_extent_buffer(right); 3361 return 1; 3362 } 3363 3364 /* 3365 * push some data in the path leaf to the left, trying to free up at 3366 * least data_size bytes. returns zero if the push worked, nonzero otherwise 3367 * 3368 * max_slot can put a limit on how far into the leaf we'll push items. The 3369 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the 3370 * items 3371 */ 3372 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, 3373 struct btrfs_path *path, int data_size, 3374 int empty, struct extent_buffer *left, 3375 int free_space, u32 right_nritems, 3376 u32 max_slot) 3377 { 3378 struct btrfs_fs_info *fs_info = left->fs_info; 3379 struct btrfs_disk_key disk_key; 3380 struct extent_buffer *right = path->nodes[0]; 3381 int i; 3382 int push_space = 0; 3383 int push_items = 0; 3384 u32 old_left_nritems; 3385 u32 nr; 3386 int ret = 0; 3387 u32 this_item_size; 3388 u32 old_left_item_size; 3389 struct btrfs_map_token token; 3390 3391 if (empty) 3392 nr = min(right_nritems, max_slot); 3393 else 3394 nr = min(right_nritems - 1, max_slot); 3395 3396 for (i = 0; i < nr; i++) { 3397 if (!empty && push_items > 0) { 3398 if (path->slots[0] < i) 3399 break; 3400 if (path->slots[0] == i) { 3401 int space = btrfs_leaf_free_space(right); 3402 3403 if (space + push_space * 2 > free_space) 3404 break; 3405 } 3406 } 3407 3408 if (path->slots[0] == i) 3409 push_space += data_size; 3410 3411 this_item_size = btrfs_item_size(right, i); 3412 if (this_item_size + sizeof(struct btrfs_item) + push_space > 3413 free_space) 3414 break; 3415 3416 push_items++; 3417 push_space += this_item_size + sizeof(struct btrfs_item); 3418 } 3419 3420 if (push_items == 0) { 3421 ret = 1; 3422 goto out; 3423 } 3424 WARN_ON(!empty && push_items == btrfs_header_nritems(right)); 3425 3426 /* push data from right to left */ 3427 copy_leaf_items(left, right, btrfs_header_nritems(left), 0, push_items); 3428 3429 push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - 3430 btrfs_item_offset(right, push_items - 1); 3431 3432 copy_leaf_data(left, right, leaf_data_end(left) - push_space, 3433 btrfs_item_offset(right, push_items - 1), push_space); 3434 old_left_nritems = btrfs_header_nritems(left); 3435 BUG_ON(old_left_nritems <= 0); 3436 3437 btrfs_init_map_token(&token, left); 3438 old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); 3439 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { 3440 u32 ioff; 3441 3442 ioff = btrfs_token_item_offset(&token, i); 3443 btrfs_set_token_item_offset(&token, i, 3444 ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); 3445 } 3446 btrfs_set_header_nritems(left, old_left_nritems + push_items); 3447 3448 /* fixup right node */ 3449 if (push_items > right_nritems) 3450 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, 3451 right_nritems); 3452 3453 if (push_items < right_nritems) { 3454 push_space = btrfs_item_offset(right, push_items - 1) - 3455 leaf_data_end(right); 3456 memmove_leaf_data(right, 3457 BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, 3458 leaf_data_end(right), push_space); 3459 3460 memmove_leaf_items(right, 0, push_items, 3461 btrfs_header_nritems(right) - push_items); 3462 } 3463 3464 btrfs_init_map_token(&token, right); 3465 right_nritems -= push_items; 3466 btrfs_set_header_nritems(right, right_nritems); 3467 push_space = BTRFS_LEAF_DATA_SIZE(fs_info); 3468 for (i = 0; i < right_nritems; i++) { 3469 push_space = push_space - btrfs_token_item_size(&token, i); 3470 btrfs_set_token_item_offset(&token, i, push_space); 3471 } 3472 3473 btrfs_mark_buffer_dirty(left); 3474 if (right_nritems) 3475 btrfs_mark_buffer_dirty(right); 3476 else 3477 btrfs_clear_buffer_dirty(trans, right); 3478 3479 btrfs_item_key(right, &disk_key, 0); 3480 fixup_low_keys(path, &disk_key, 1); 3481 3482 /* then fixup the leaf pointer in the path */ 3483 if (path->slots[0] < push_items) { 3484 path->slots[0] += old_left_nritems; 3485 btrfs_tree_unlock(path->nodes[0]); 3486 free_extent_buffer(path->nodes[0]); 3487 path->nodes[0] = left; 3488 path->slots[1] -= 1; 3489 } else { 3490 btrfs_tree_unlock(left); 3491 free_extent_buffer(left); 3492 path->slots[0] -= push_items; 3493 } 3494 BUG_ON(path->slots[0] < 0); 3495 return ret; 3496 out: 3497 btrfs_tree_unlock(left); 3498 free_extent_buffer(left); 3499 return ret; 3500 } 3501 3502 /* 3503 * push some data in the path leaf to the left, trying to free up at 3504 * least data_size bytes. returns zero if the push worked, nonzero otherwise 3505 * 3506 * max_slot can put a limit on how far into the leaf we'll push items. The 3507 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the 3508 * items 3509 */ 3510 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root 3511 *root, struct btrfs_path *path, int min_data_size, 3512 int data_size, int empty, u32 max_slot) 3513 { 3514 struct extent_buffer *right = path->nodes[0]; 3515 struct extent_buffer *left; 3516 int slot; 3517 int free_space; 3518 u32 right_nritems; 3519 int ret = 0; 3520 3521 slot = path->slots[1]; 3522 if (slot == 0) 3523 return 1; 3524 if (!path->nodes[1]) 3525 return 1; 3526 3527 right_nritems = btrfs_header_nritems(right); 3528 if (right_nritems == 0) 3529 return 1; 3530 3531 btrfs_assert_tree_write_locked(path->nodes[1]); 3532 3533 left = btrfs_read_node_slot(path->nodes[1], slot - 1); 3534 if (IS_ERR(left)) 3535 return PTR_ERR(left); 3536 3537 __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); 3538 3539 free_space = btrfs_leaf_free_space(left); 3540 if (free_space < data_size) { 3541 ret = 1; 3542 goto out; 3543 } 3544 3545 ret = btrfs_cow_block(trans, root, left, 3546 path->nodes[1], slot - 1, &left, 3547 BTRFS_NESTING_LEFT_COW); 3548 if (ret) { 3549 /* we hit -ENOSPC, but it isn't fatal here */ 3550 if (ret == -ENOSPC) 3551 ret = 1; 3552 goto out; 3553 } 3554 3555 if (check_sibling_keys(left, right)) { 3556 ret = -EUCLEAN; 3557 btrfs_abort_transaction(trans, ret); 3558 goto out; 3559 } 3560 return __push_leaf_left(trans, path, min_data_size, empty, left, 3561 free_space, right_nritems, max_slot); 3562 out: 3563 btrfs_tree_unlock(left); 3564 free_extent_buffer(left); 3565 return ret; 3566 } 3567 3568 /* 3569 * split the path's leaf in two, making sure there is at least data_size 3570 * available for the resulting leaf level of the path. 3571 */ 3572 static noinline void copy_for_split(struct btrfs_trans_handle *trans, 3573 struct btrfs_path *path, 3574 struct extent_buffer *l, 3575 struct extent_buffer *right, 3576 int slot, int mid, int nritems) 3577 { 3578 struct btrfs_fs_info *fs_info = trans->fs_info; 3579 int data_copy_size; 3580 int rt_data_off; 3581 int i; 3582 struct btrfs_disk_key disk_key; 3583 struct btrfs_map_token token; 3584 3585 nritems = nritems - mid; 3586 btrfs_set_header_nritems(right, nritems); 3587 data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); 3588 3589 copy_leaf_items(right, l, 0, mid, nritems); 3590 3591 copy_leaf_data(right, l, BTRFS_LEAF_DATA_SIZE(fs_info) - data_copy_size, 3592 leaf_data_end(l), data_copy_size); 3593 3594 rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); 3595 3596 btrfs_init_map_token(&token, right); 3597 for (i = 0; i < nritems; i++) { 3598 u32 ioff; 3599 3600 ioff = btrfs_token_item_offset(&token, i); 3601 btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); 3602 } 3603 3604 btrfs_set_header_nritems(l, mid); 3605 btrfs_item_key(right, &disk_key, 0); 3606 insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); 3607 3608 btrfs_mark_buffer_dirty(right); 3609 btrfs_mark_buffer_dirty(l); 3610 BUG_ON(path->slots[0] != slot); 3611 3612 if (mid <= slot) { 3613 btrfs_tree_unlock(path->nodes[0]); 3614 free_extent_buffer(path->nodes[0]); 3615 path->nodes[0] = right; 3616 path->slots[0] -= mid; 3617 path->slots[1] += 1; 3618 } else { 3619 btrfs_tree_unlock(right); 3620 free_extent_buffer(right); 3621 } 3622 3623 BUG_ON(path->slots[0] < 0); 3624 } 3625 3626 /* 3627 * double splits happen when we need to insert a big item in the middle 3628 * of a leaf. A double split can leave us with 3 mostly empty leaves: 3629 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] 3630 * A B C 3631 * 3632 * We avoid this by trying to push the items on either side of our target 3633 * into the adjacent leaves. If all goes well we can avoid the double split 3634 * completely. 3635 */ 3636 static noinline int push_for_double_split(struct btrfs_trans_handle *trans, 3637 struct btrfs_root *root, 3638 struct btrfs_path *path, 3639 int data_size) 3640 { 3641 int ret; 3642 int progress = 0; 3643 int slot; 3644 u32 nritems; 3645 int space_needed = data_size; 3646 3647 slot = path->slots[0]; 3648 if (slot < btrfs_header_nritems(path->nodes[0])) 3649 space_needed -= btrfs_leaf_free_space(path->nodes[0]); 3650 3651 /* 3652 * try to push all the items after our slot into the 3653 * right leaf 3654 */ 3655 ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); 3656 if (ret < 0) 3657 return ret; 3658 3659 if (ret == 0) 3660 progress++; 3661 3662 nritems = btrfs_header_nritems(path->nodes[0]); 3663 /* 3664 * our goal is to get our slot at the start or end of a leaf. If 3665 * we've done so we're done 3666 */ 3667 if (path->slots[0] == 0 || path->slots[0] == nritems) 3668 return 0; 3669 3670 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 3671 return 0; 3672 3673 /* try to push all the items before our slot into the next leaf */ 3674 slot = path->slots[0]; 3675 space_needed = data_size; 3676 if (slot > 0) 3677 space_needed -= btrfs_leaf_free_space(path->nodes[0]); 3678 ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); 3679 if (ret < 0) 3680 return ret; 3681 3682 if (ret == 0) 3683 progress++; 3684 3685 if (progress) 3686 return 0; 3687 return 1; 3688 } 3689 3690 /* 3691 * split the path's leaf in two, making sure there is at least data_size 3692 * available for the resulting leaf level of the path. 3693 * 3694 * returns 0 if all went well and < 0 on failure. 3695 */ 3696 static noinline int split_leaf(struct btrfs_trans_handle *trans, 3697 struct btrfs_root *root, 3698 const struct btrfs_key *ins_key, 3699 struct btrfs_path *path, int data_size, 3700 int extend) 3701 { 3702 struct btrfs_disk_key disk_key; 3703 struct extent_buffer *l; 3704 u32 nritems; 3705 int mid; 3706 int slot; 3707 struct extent_buffer *right; 3708 struct btrfs_fs_info *fs_info = root->fs_info; 3709 int ret = 0; 3710 int wret; 3711 int split; 3712 int num_doubles = 0; 3713 int tried_avoid_double = 0; 3714 3715 l = path->nodes[0]; 3716 slot = path->slots[0]; 3717 if (extend && data_size + btrfs_item_size(l, slot) + 3718 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) 3719 return -EOVERFLOW; 3720 3721 /* first try to make some room by pushing left and right */ 3722 if (data_size && path->nodes[1]) { 3723 int space_needed = data_size; 3724 3725 if (slot < btrfs_header_nritems(l)) 3726 space_needed -= btrfs_leaf_free_space(l); 3727 3728 wret = push_leaf_right(trans, root, path, space_needed, 3729 space_needed, 0, 0); 3730 if (wret < 0) 3731 return wret; 3732 if (wret) { 3733 space_needed = data_size; 3734 if (slot > 0) 3735 space_needed -= btrfs_leaf_free_space(l); 3736 wret = push_leaf_left(trans, root, path, space_needed, 3737 space_needed, 0, (u32)-1); 3738 if (wret < 0) 3739 return wret; 3740 } 3741 l = path->nodes[0]; 3742 3743 /* did the pushes work? */ 3744 if (btrfs_leaf_free_space(l) >= data_size) 3745 return 0; 3746 } 3747 3748 if (!path->nodes[1]) { 3749 ret = insert_new_root(trans, root, path, 1); 3750 if (ret) 3751 return ret; 3752 } 3753 again: 3754 split = 1; 3755 l = path->nodes[0]; 3756 slot = path->slots[0]; 3757 nritems = btrfs_header_nritems(l); 3758 mid = (nritems + 1) / 2; 3759 3760 if (mid <= slot) { 3761 if (nritems == 1 || 3762 leaf_space_used(l, mid, nritems - mid) + data_size > 3763 BTRFS_LEAF_DATA_SIZE(fs_info)) { 3764 if (slot >= nritems) { 3765 split = 0; 3766 } else { 3767 mid = slot; 3768 if (mid != nritems && 3769 leaf_space_used(l, mid, nritems - mid) + 3770 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 3771 if (data_size && !tried_avoid_double) 3772 goto push_for_double; 3773 split = 2; 3774 } 3775 } 3776 } 3777 } else { 3778 if (leaf_space_used(l, 0, mid) + data_size > 3779 BTRFS_LEAF_DATA_SIZE(fs_info)) { 3780 if (!extend && data_size && slot == 0) { 3781 split = 0; 3782 } else if ((extend || !data_size) && slot == 0) { 3783 mid = 1; 3784 } else { 3785 mid = slot; 3786 if (mid != nritems && 3787 leaf_space_used(l, mid, nritems - mid) + 3788 data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { 3789 if (data_size && !tried_avoid_double) 3790 goto push_for_double; 3791 split = 2; 3792 } 3793 } 3794 } 3795 } 3796 3797 if (split == 0) 3798 btrfs_cpu_key_to_disk(&disk_key, ins_key); 3799 else 3800 btrfs_item_key(l, &disk_key, mid); 3801 3802 /* 3803 * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double 3804 * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES 3805 * subclasses, which is 8 at the time of this patch, and we've maxed it 3806 * out. In the future we could add a 3807 * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just 3808 * use BTRFS_NESTING_NEW_ROOT. 3809 */ 3810 right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, 3811 &disk_key, 0, l->start, 0, 3812 num_doubles ? BTRFS_NESTING_NEW_ROOT : 3813 BTRFS_NESTING_SPLIT); 3814 if (IS_ERR(right)) 3815 return PTR_ERR(right); 3816 3817 root_add_used(root, fs_info->nodesize); 3818 3819 if (split == 0) { 3820 if (mid <= slot) { 3821 btrfs_set_header_nritems(right, 0); 3822 insert_ptr(trans, path, &disk_key, 3823 right->start, path->slots[1] + 1, 1); 3824 btrfs_tree_unlock(path->nodes[0]); 3825 free_extent_buffer(path->nodes[0]); 3826 path->nodes[0] = right; 3827 path->slots[0] = 0; 3828 path->slots[1] += 1; 3829 } else { 3830 btrfs_set_header_nritems(right, 0); 3831 insert_ptr(trans, path, &disk_key, 3832 right->start, path->slots[1], 1); 3833 btrfs_tree_unlock(path->nodes[0]); 3834 free_extent_buffer(path->nodes[0]); 3835 path->nodes[0] = right; 3836 path->slots[0] = 0; 3837 if (path->slots[1] == 0) 3838 fixup_low_keys(path, &disk_key, 1); 3839 } 3840 /* 3841 * We create a new leaf 'right' for the required ins_len and 3842 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying 3843 * the content of ins_len to 'right'. 3844 */ 3845 return ret; 3846 } 3847 3848 copy_for_split(trans, path, l, right, slot, mid, nritems); 3849 3850 if (split == 2) { 3851 BUG_ON(num_doubles != 0); 3852 num_doubles++; 3853 goto again; 3854 } 3855 3856 return 0; 3857 3858 push_for_double: 3859 push_for_double_split(trans, root, path, data_size); 3860 tried_avoid_double = 1; 3861 if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) 3862 return 0; 3863 goto again; 3864 } 3865 3866 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, 3867 struct btrfs_root *root, 3868 struct btrfs_path *path, int ins_len) 3869 { 3870 struct btrfs_key key; 3871 struct extent_buffer *leaf; 3872 struct btrfs_file_extent_item *fi; 3873 u64 extent_len = 0; 3874 u32 item_size; 3875 int ret; 3876 3877 leaf = path->nodes[0]; 3878 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3879 3880 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && 3881 key.type != BTRFS_EXTENT_CSUM_KEY); 3882 3883 if (btrfs_leaf_free_space(leaf) >= ins_len) 3884 return 0; 3885 3886 item_size = btrfs_item_size(leaf, path->slots[0]); 3887 if (key.type == BTRFS_EXTENT_DATA_KEY) { 3888 fi = btrfs_item_ptr(leaf, path->slots[0], 3889 struct btrfs_file_extent_item); 3890 extent_len = btrfs_file_extent_num_bytes(leaf, fi); 3891 } 3892 btrfs_release_path(path); 3893 3894 path->keep_locks = 1; 3895 path->search_for_split = 1; 3896 ret = btrfs_search_slot(trans, root, &key, path, 0, 1); 3897 path->search_for_split = 0; 3898 if (ret > 0) 3899 ret = -EAGAIN; 3900 if (ret < 0) 3901 goto err; 3902 3903 ret = -EAGAIN; 3904 leaf = path->nodes[0]; 3905 /* if our item isn't there, return now */ 3906 if (item_size != btrfs_item_size(leaf, path->slots[0])) 3907 goto err; 3908 3909 /* the leaf has changed, it now has room. return now */ 3910 if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) 3911 goto err; 3912 3913 if (key.type == BTRFS_EXTENT_DATA_KEY) { 3914 fi = btrfs_item_ptr(leaf, path->slots[0], 3915 struct btrfs_file_extent_item); 3916 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) 3917 goto err; 3918 } 3919 3920 ret = split_leaf(trans, root, &key, path, ins_len, 1); 3921 if (ret) 3922 goto err; 3923 3924 path->keep_locks = 0; 3925 btrfs_unlock_up_safe(path, 1); 3926 return 0; 3927 err: 3928 path->keep_locks = 0; 3929 return ret; 3930 } 3931 3932 static noinline int split_item(struct btrfs_path *path, 3933 const struct btrfs_key *new_key, 3934 unsigned long split_offset) 3935 { 3936 struct extent_buffer *leaf; 3937 int orig_slot, slot; 3938 char *buf; 3939 u32 nritems; 3940 u32 item_size; 3941 u32 orig_offset; 3942 struct btrfs_disk_key disk_key; 3943 3944 leaf = path->nodes[0]; 3945 BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); 3946 3947 orig_slot = path->slots[0]; 3948 orig_offset = btrfs_item_offset(leaf, path->slots[0]); 3949 item_size = btrfs_item_size(leaf, path->slots[0]); 3950 3951 buf = kmalloc(item_size, GFP_NOFS); 3952 if (!buf) 3953 return -ENOMEM; 3954 3955 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, 3956 path->slots[0]), item_size); 3957 3958 slot = path->slots[0] + 1; 3959 nritems = btrfs_header_nritems(leaf); 3960 if (slot != nritems) { 3961 /* shift the items */ 3962 memmove_leaf_items(leaf, slot + 1, slot, nritems - slot); 3963 } 3964 3965 btrfs_cpu_key_to_disk(&disk_key, new_key); 3966 btrfs_set_item_key(leaf, &disk_key, slot); 3967 3968 btrfs_set_item_offset(leaf, slot, orig_offset); 3969 btrfs_set_item_size(leaf, slot, item_size - split_offset); 3970 3971 btrfs_set_item_offset(leaf, orig_slot, 3972 orig_offset + item_size - split_offset); 3973 btrfs_set_item_size(leaf, orig_slot, split_offset); 3974 3975 btrfs_set_header_nritems(leaf, nritems + 1); 3976 3977 /* write the data for the start of the original item */ 3978 write_extent_buffer(leaf, buf, 3979 btrfs_item_ptr_offset(leaf, path->slots[0]), 3980 split_offset); 3981 3982 /* write the data for the new item */ 3983 write_extent_buffer(leaf, buf + split_offset, 3984 btrfs_item_ptr_offset(leaf, slot), 3985 item_size - split_offset); 3986 btrfs_mark_buffer_dirty(leaf); 3987 3988 BUG_ON(btrfs_leaf_free_space(leaf) < 0); 3989 kfree(buf); 3990 return 0; 3991 } 3992 3993 /* 3994 * This function splits a single item into two items, 3995 * giving 'new_key' to the new item and splitting the 3996 * old one at split_offset (from the start of the item). 3997 * 3998 * The path may be released by this operation. After 3999 * the split, the path is pointing to the old item. The 4000 * new item is going to be in the same node as the old one. 4001 * 4002 * Note, the item being split must be smaller enough to live alone on 4003 * a tree block with room for one extra struct btrfs_item 4004 * 4005 * This allows us to split the item in place, keeping a lock on the 4006 * leaf the entire time. 4007 */ 4008 int btrfs_split_item(struct btrfs_trans_handle *trans, 4009 struct btrfs_root *root, 4010 struct btrfs_path *path, 4011 const struct btrfs_key *new_key, 4012 unsigned long split_offset) 4013 { 4014 int ret; 4015 ret = setup_leaf_for_split(trans, root, path, 4016 sizeof(struct btrfs_item)); 4017 if (ret) 4018 return ret; 4019 4020 ret = split_item(path, new_key, split_offset); 4021 return ret; 4022 } 4023 4024 /* 4025 * make the item pointed to by the path smaller. new_size indicates 4026 * how small to make it, and from_end tells us if we just chop bytes 4027 * off the end of the item or if we shift the item to chop bytes off 4028 * the front. 4029 */ 4030 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) 4031 { 4032 int slot; 4033 struct extent_buffer *leaf; 4034 u32 nritems; 4035 unsigned int data_end; 4036 unsigned int old_data_start; 4037 unsigned int old_size; 4038 unsigned int size_diff; 4039 int i; 4040 struct btrfs_map_token token; 4041 4042 leaf = path->nodes[0]; 4043 slot = path->slots[0]; 4044 4045 old_size = btrfs_item_size(leaf, slot); 4046 if (old_size == new_size) 4047 return; 4048 4049 nritems = btrfs_header_nritems(leaf); 4050 data_end = leaf_data_end(leaf); 4051 4052 old_data_start = btrfs_item_offset(leaf, slot); 4053 4054 size_diff = old_size - new_size; 4055 4056 BUG_ON(slot < 0); 4057 BUG_ON(slot >= nritems); 4058 4059 /* 4060 * item0..itemN ... dataN.offset..dataN.size .. data0.size 4061 */ 4062 /* first correct the data pointers */ 4063 btrfs_init_map_token(&token, leaf); 4064 for (i = slot; i < nritems; i++) { 4065 u32 ioff; 4066 4067 ioff = btrfs_token_item_offset(&token, i); 4068 btrfs_set_token_item_offset(&token, i, ioff + size_diff); 4069 } 4070 4071 /* shift the data */ 4072 if (from_end) { 4073 memmove_leaf_data(leaf, data_end + size_diff, data_end, 4074 old_data_start + new_size - data_end); 4075 } else { 4076 struct btrfs_disk_key disk_key; 4077 u64 offset; 4078 4079 btrfs_item_key(leaf, &disk_key, slot); 4080 4081 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { 4082 unsigned long ptr; 4083 struct btrfs_file_extent_item *fi; 4084 4085 fi = btrfs_item_ptr(leaf, slot, 4086 struct btrfs_file_extent_item); 4087 fi = (struct btrfs_file_extent_item *)( 4088 (unsigned long)fi - size_diff); 4089 4090 if (btrfs_file_extent_type(leaf, fi) == 4091 BTRFS_FILE_EXTENT_INLINE) { 4092 ptr = btrfs_item_ptr_offset(leaf, slot); 4093 memmove_extent_buffer(leaf, ptr, 4094 (unsigned long)fi, 4095 BTRFS_FILE_EXTENT_INLINE_DATA_START); 4096 } 4097 } 4098 4099 memmove_leaf_data(leaf, data_end + size_diff, data_end, 4100 old_data_start - data_end); 4101 4102 offset = btrfs_disk_key_offset(&disk_key); 4103 btrfs_set_disk_key_offset(&disk_key, offset + size_diff); 4104 btrfs_set_item_key(leaf, &disk_key, slot); 4105 if (slot == 0) 4106 fixup_low_keys(path, &disk_key, 1); 4107 } 4108 4109 btrfs_set_item_size(leaf, slot, new_size); 4110 btrfs_mark_buffer_dirty(leaf); 4111 4112 if (btrfs_leaf_free_space(leaf) < 0) { 4113 btrfs_print_leaf(leaf); 4114 BUG(); 4115 } 4116 } 4117 4118 /* 4119 * make the item pointed to by the path bigger, data_size is the added size. 4120 */ 4121 void btrfs_extend_item(struct btrfs_path *path, u32 data_size) 4122 { 4123 int slot; 4124 struct extent_buffer *leaf; 4125 u32 nritems; 4126 unsigned int data_end; 4127 unsigned int old_data; 4128 unsigned int old_size; 4129 int i; 4130 struct btrfs_map_token token; 4131 4132 leaf = path->nodes[0]; 4133 4134 nritems = btrfs_header_nritems(leaf); 4135 data_end = leaf_data_end(leaf); 4136 4137 if (btrfs_leaf_free_space(leaf) < data_size) { 4138 btrfs_print_leaf(leaf); 4139 BUG(); 4140 } 4141 slot = path->slots[0]; 4142 old_data = btrfs_item_data_end(leaf, slot); 4143 4144 BUG_ON(slot < 0); 4145 if (slot >= nritems) { 4146 btrfs_print_leaf(leaf); 4147 btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", 4148 slot, nritems); 4149 BUG(); 4150 } 4151 4152 /* 4153 * item0..itemN ... dataN.offset..dataN.size .. data0.size 4154 */ 4155 /* first correct the data pointers */ 4156 btrfs_init_map_token(&token, leaf); 4157 for (i = slot; i < nritems; i++) { 4158 u32 ioff; 4159 4160 ioff = btrfs_token_item_offset(&token, i); 4161 btrfs_set_token_item_offset(&token, i, ioff - data_size); 4162 } 4163 4164 /* shift the data */ 4165 memmove_leaf_data(leaf, data_end - data_size, data_end, 4166 old_data - data_end); 4167 4168 data_end = old_data; 4169 old_size = btrfs_item_size(leaf, slot); 4170 btrfs_set_item_size(leaf, slot, old_size + data_size); 4171 btrfs_mark_buffer_dirty(leaf); 4172 4173 if (btrfs_leaf_free_space(leaf) < 0) { 4174 btrfs_print_leaf(leaf); 4175 BUG(); 4176 } 4177 } 4178 4179 /* 4180 * Make space in the node before inserting one or more items. 4181 * 4182 * @root: root we are inserting items to 4183 * @path: points to the leaf/slot where we are going to insert new items 4184 * @batch: information about the batch of items to insert 4185 * 4186 * Main purpose is to save stack depth by doing the bulk of the work in a 4187 * function that doesn't call btrfs_search_slot 4188 */ 4189 static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, 4190 const struct btrfs_item_batch *batch) 4191 { 4192 struct btrfs_fs_info *fs_info = root->fs_info; 4193 int i; 4194 u32 nritems; 4195 unsigned int data_end; 4196 struct btrfs_disk_key disk_key; 4197 struct extent_buffer *leaf; 4198 int slot; 4199 struct btrfs_map_token token; 4200 u32 total_size; 4201 4202 /* 4203 * Before anything else, update keys in the parent and other ancestors 4204 * if needed, then release the write locks on them, so that other tasks 4205 * can use them while we modify the leaf. 4206 */ 4207 if (path->slots[0] == 0) { 4208 btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); 4209 fixup_low_keys(path, &disk_key, 1); 4210 } 4211 btrfs_unlock_up_safe(path, 1); 4212 4213 leaf = path->nodes[0]; 4214 slot = path->slots[0]; 4215 4216 nritems = btrfs_header_nritems(leaf); 4217 data_end = leaf_data_end(leaf); 4218 total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4219 4220 if (btrfs_leaf_free_space(leaf) < total_size) { 4221 btrfs_print_leaf(leaf); 4222 btrfs_crit(fs_info, "not enough freespace need %u have %d", 4223 total_size, btrfs_leaf_free_space(leaf)); 4224 BUG(); 4225 } 4226 4227 btrfs_init_map_token(&token, leaf); 4228 if (slot != nritems) { 4229 unsigned int old_data = btrfs_item_data_end(leaf, slot); 4230 4231 if (old_data < data_end) { 4232 btrfs_print_leaf(leaf); 4233 btrfs_crit(fs_info, 4234 "item at slot %d with data offset %u beyond data end of leaf %u", 4235 slot, old_data, data_end); 4236 BUG(); 4237 } 4238 /* 4239 * item0..itemN ... dataN.offset..dataN.size .. data0.size 4240 */ 4241 /* first correct the data pointers */ 4242 for (i = slot; i < nritems; i++) { 4243 u32 ioff; 4244 4245 ioff = btrfs_token_item_offset(&token, i); 4246 btrfs_set_token_item_offset(&token, i, 4247 ioff - batch->total_data_size); 4248 } 4249 /* shift the items */ 4250 memmove_leaf_items(leaf, slot + batch->nr, slot, nritems - slot); 4251 4252 /* shift the data */ 4253 memmove_leaf_data(leaf, data_end - batch->total_data_size, 4254 data_end, old_data - data_end); 4255 data_end = old_data; 4256 } 4257 4258 /* setup the item for the new data */ 4259 for (i = 0; i < batch->nr; i++) { 4260 btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); 4261 btrfs_set_item_key(leaf, &disk_key, slot + i); 4262 data_end -= batch->data_sizes[i]; 4263 btrfs_set_token_item_offset(&token, slot + i, data_end); 4264 btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); 4265 } 4266 4267 btrfs_set_header_nritems(leaf, nritems + batch->nr); 4268 btrfs_mark_buffer_dirty(leaf); 4269 4270 if (btrfs_leaf_free_space(leaf) < 0) { 4271 btrfs_print_leaf(leaf); 4272 BUG(); 4273 } 4274 } 4275 4276 /* 4277 * Insert a new item into a leaf. 4278 * 4279 * @root: The root of the btree. 4280 * @path: A path pointing to the target leaf and slot. 4281 * @key: The key of the new item. 4282 * @data_size: The size of the data associated with the new key. 4283 */ 4284 void btrfs_setup_item_for_insert(struct btrfs_root *root, 4285 struct btrfs_path *path, 4286 const struct btrfs_key *key, 4287 u32 data_size) 4288 { 4289 struct btrfs_item_batch batch; 4290 4291 batch.keys = key; 4292 batch.data_sizes = &data_size; 4293 batch.total_data_size = data_size; 4294 batch.nr = 1; 4295 4296 setup_items_for_insert(root, path, &batch); 4297 } 4298 4299 /* 4300 * Given a key and some data, insert items into the tree. 4301 * This does all the path init required, making room in the tree if needed. 4302 */ 4303 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, 4304 struct btrfs_root *root, 4305 struct btrfs_path *path, 4306 const struct btrfs_item_batch *batch) 4307 { 4308 int ret = 0; 4309 int slot; 4310 u32 total_size; 4311 4312 total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); 4313 ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); 4314 if (ret == 0) 4315 return -EEXIST; 4316 if (ret < 0) 4317 return ret; 4318 4319 slot = path->slots[0]; 4320 BUG_ON(slot < 0); 4321 4322 setup_items_for_insert(root, path, batch); 4323 return 0; 4324 } 4325 4326 /* 4327 * Given a key and some data, insert an item into the tree. 4328 * This does all the path init required, making room in the tree if needed. 4329 */ 4330 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4331 const struct btrfs_key *cpu_key, void *data, 4332 u32 data_size) 4333 { 4334 int ret = 0; 4335 struct btrfs_path *path; 4336 struct extent_buffer *leaf; 4337 unsigned long ptr; 4338 4339 path = btrfs_alloc_path(); 4340 if (!path) 4341 return -ENOMEM; 4342 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); 4343 if (!ret) { 4344 leaf = path->nodes[0]; 4345 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); 4346 write_extent_buffer(leaf, data, ptr, data_size); 4347 btrfs_mark_buffer_dirty(leaf); 4348 } 4349 btrfs_free_path(path); 4350 return ret; 4351 } 4352 4353 /* 4354 * This function duplicates an item, giving 'new_key' to the new item. 4355 * It guarantees both items live in the same tree leaf and the new item is 4356 * contiguous with the original item. 4357 * 4358 * This allows us to split a file extent in place, keeping a lock on the leaf 4359 * the entire time. 4360 */ 4361 int btrfs_duplicate_item(struct btrfs_trans_handle *trans, 4362 struct btrfs_root *root, 4363 struct btrfs_path *path, 4364 const struct btrfs_key *new_key) 4365 { 4366 struct extent_buffer *leaf; 4367 int ret; 4368 u32 item_size; 4369 4370 leaf = path->nodes[0]; 4371 item_size = btrfs_item_size(leaf, path->slots[0]); 4372 ret = setup_leaf_for_split(trans, root, path, 4373 item_size + sizeof(struct btrfs_item)); 4374 if (ret) 4375 return ret; 4376 4377 path->slots[0]++; 4378 btrfs_setup_item_for_insert(root, path, new_key, item_size); 4379 leaf = path->nodes[0]; 4380 memcpy_extent_buffer(leaf, 4381 btrfs_item_ptr_offset(leaf, path->slots[0]), 4382 btrfs_item_ptr_offset(leaf, path->slots[0] - 1), 4383 item_size); 4384 return 0; 4385 } 4386 4387 /* 4388 * delete the pointer from a given node. 4389 * 4390 * the tree should have been previously balanced so the deletion does not 4391 * empty a node. 4392 * 4393 * This is exported for use inside btrfs-progs, don't un-export it. 4394 */ 4395 void btrfs_del_ptr(struct btrfs_root *root, struct btrfs_path *path, int level, 4396 int slot) 4397 { 4398 struct extent_buffer *parent = path->nodes[level]; 4399 u32 nritems; 4400 int ret; 4401 4402 nritems = btrfs_header_nritems(parent); 4403 if (slot != nritems - 1) { 4404 if (level) { 4405 ret = btrfs_tree_mod_log_insert_move(parent, slot, 4406 slot + 1, nritems - slot - 1); 4407 BUG_ON(ret < 0); 4408 } 4409 memmove_extent_buffer(parent, 4410 btrfs_node_key_ptr_offset(parent, slot), 4411 btrfs_node_key_ptr_offset(parent, slot + 1), 4412 sizeof(struct btrfs_key_ptr) * 4413 (nritems - slot - 1)); 4414 } else if (level) { 4415 ret = btrfs_tree_mod_log_insert_key(parent, slot, 4416 BTRFS_MOD_LOG_KEY_REMOVE); 4417 BUG_ON(ret < 0); 4418 } 4419 4420 nritems--; 4421 btrfs_set_header_nritems(parent, nritems); 4422 if (nritems == 0 && parent == root->node) { 4423 BUG_ON(btrfs_header_level(root->node) != 1); 4424 /* just turn the root into a leaf and break */ 4425 btrfs_set_header_level(root->node, 0); 4426 } else if (slot == 0) { 4427 struct btrfs_disk_key disk_key; 4428 4429 btrfs_node_key(parent, &disk_key, 0); 4430 fixup_low_keys(path, &disk_key, level + 1); 4431 } 4432 btrfs_mark_buffer_dirty(parent); 4433 } 4434 4435 /* 4436 * a helper function to delete the leaf pointed to by path->slots[1] and 4437 * path->nodes[1]. 4438 * 4439 * This deletes the pointer in path->nodes[1] and frees the leaf 4440 * block extent. zero is returned if it all worked out, < 0 otherwise. 4441 * 4442 * The path must have already been setup for deleting the leaf, including 4443 * all the proper balancing. path->nodes[1] must be locked. 4444 */ 4445 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, 4446 struct btrfs_root *root, 4447 struct btrfs_path *path, 4448 struct extent_buffer *leaf) 4449 { 4450 WARN_ON(btrfs_header_generation(leaf) != trans->transid); 4451 btrfs_del_ptr(root, path, 1, path->slots[1]); 4452 4453 /* 4454 * btrfs_free_extent is expensive, we want to make sure we 4455 * aren't holding any locks when we call it 4456 */ 4457 btrfs_unlock_up_safe(path, 0); 4458 4459 root_sub_used(root, leaf->len); 4460 4461 atomic_inc(&leaf->refs); 4462 btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); 4463 free_extent_buffer_stale(leaf); 4464 } 4465 /* 4466 * delete the item at the leaf level in path. If that empties 4467 * the leaf, remove it from the tree 4468 */ 4469 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, 4470 struct btrfs_path *path, int slot, int nr) 4471 { 4472 struct btrfs_fs_info *fs_info = root->fs_info; 4473 struct extent_buffer *leaf; 4474 int ret = 0; 4475 int wret; 4476 u32 nritems; 4477 4478 leaf = path->nodes[0]; 4479 nritems = btrfs_header_nritems(leaf); 4480 4481 if (slot + nr != nritems) { 4482 const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); 4483 const int data_end = leaf_data_end(leaf); 4484 struct btrfs_map_token token; 4485 u32 dsize = 0; 4486 int i; 4487 4488 for (i = 0; i < nr; i++) 4489 dsize += btrfs_item_size(leaf, slot + i); 4490 4491 memmove_leaf_data(leaf, data_end + dsize, data_end, 4492 last_off - data_end); 4493 4494 btrfs_init_map_token(&token, leaf); 4495 for (i = slot + nr; i < nritems; i++) { 4496 u32 ioff; 4497 4498 ioff = btrfs_token_item_offset(&token, i); 4499 btrfs_set_token_item_offset(&token, i, ioff + dsize); 4500 } 4501 4502 memmove_leaf_items(leaf, slot, slot + nr, nritems - slot - nr); 4503 } 4504 btrfs_set_header_nritems(leaf, nritems - nr); 4505 nritems -= nr; 4506 4507 /* delete the leaf if we've emptied it */ 4508 if (nritems == 0) { 4509 if (leaf == root->node) { 4510 btrfs_set_header_level(leaf, 0); 4511 } else { 4512 btrfs_clear_buffer_dirty(trans, leaf); 4513 btrfs_del_leaf(trans, root, path, leaf); 4514 } 4515 } else { 4516 int used = leaf_space_used(leaf, 0, nritems); 4517 if (slot == 0) { 4518 struct btrfs_disk_key disk_key; 4519 4520 btrfs_item_key(leaf, &disk_key, 0); 4521 fixup_low_keys(path, &disk_key, 1); 4522 } 4523 4524 /* 4525 * Try to delete the leaf if it is mostly empty. We do this by 4526 * trying to move all its items into its left and right neighbours. 4527 * If we can't move all the items, then we don't delete it - it's 4528 * not ideal, but future insertions might fill the leaf with more 4529 * items, or items from other leaves might be moved later into our 4530 * leaf due to deletions on those leaves. 4531 */ 4532 if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { 4533 u32 min_push_space; 4534 4535 /* push_leaf_left fixes the path. 4536 * make sure the path still points to our leaf 4537 * for possible call to btrfs_del_ptr below 4538 */ 4539 slot = path->slots[1]; 4540 atomic_inc(&leaf->refs); 4541 /* 4542 * We want to be able to at least push one item to the 4543 * left neighbour leaf, and that's the first item. 4544 */ 4545 min_push_space = sizeof(struct btrfs_item) + 4546 btrfs_item_size(leaf, 0); 4547 wret = push_leaf_left(trans, root, path, 0, 4548 min_push_space, 1, (u32)-1); 4549 if (wret < 0 && wret != -ENOSPC) 4550 ret = wret; 4551 4552 if (path->nodes[0] == leaf && 4553 btrfs_header_nritems(leaf)) { 4554 /* 4555 * If we were not able to push all items from our 4556 * leaf to its left neighbour, then attempt to 4557 * either push all the remaining items to the 4558 * right neighbour or none. There's no advantage 4559 * in pushing only some items, instead of all, as 4560 * it's pointless to end up with a leaf having 4561 * too few items while the neighbours can be full 4562 * or nearly full. 4563 */ 4564 nritems = btrfs_header_nritems(leaf); 4565 min_push_space = leaf_space_used(leaf, 0, nritems); 4566 wret = push_leaf_right(trans, root, path, 0, 4567 min_push_space, 1, 0); 4568 if (wret < 0 && wret != -ENOSPC) 4569 ret = wret; 4570 } 4571 4572 if (btrfs_header_nritems(leaf) == 0) { 4573 path->slots[1] = slot; 4574 btrfs_del_leaf(trans, root, path, leaf); 4575 free_extent_buffer(leaf); 4576 ret = 0; 4577 } else { 4578 /* if we're still in the path, make sure 4579 * we're dirty. Otherwise, one of the 4580 * push_leaf functions must have already 4581 * dirtied this buffer 4582 */ 4583 if (path->nodes[0] == leaf) 4584 btrfs_mark_buffer_dirty(leaf); 4585 free_extent_buffer(leaf); 4586 } 4587 } else { 4588 btrfs_mark_buffer_dirty(leaf); 4589 } 4590 } 4591 return ret; 4592 } 4593 4594 /* 4595 * A helper function to walk down the tree starting at min_key, and looking 4596 * for nodes or leaves that are have a minimum transaction id. 4597 * This is used by the btree defrag code, and tree logging 4598 * 4599 * This does not cow, but it does stuff the starting key it finds back 4600 * into min_key, so you can call btrfs_search_slot with cow=1 on the 4601 * key and get a writable path. 4602 * 4603 * This honors path->lowest_level to prevent descent past a given level 4604 * of the tree. 4605 * 4606 * min_trans indicates the oldest transaction that you are interested 4607 * in walking through. Any nodes or leaves older than min_trans are 4608 * skipped over (without reading them). 4609 * 4610 * returns zero if something useful was found, < 0 on error and 1 if there 4611 * was nothing in the tree that matched the search criteria. 4612 */ 4613 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, 4614 struct btrfs_path *path, 4615 u64 min_trans) 4616 { 4617 struct extent_buffer *cur; 4618 struct btrfs_key found_key; 4619 int slot; 4620 int sret; 4621 u32 nritems; 4622 int level; 4623 int ret = 1; 4624 int keep_locks = path->keep_locks; 4625 4626 ASSERT(!path->nowait); 4627 path->keep_locks = 1; 4628 again: 4629 cur = btrfs_read_lock_root_node(root); 4630 level = btrfs_header_level(cur); 4631 WARN_ON(path->nodes[level]); 4632 path->nodes[level] = cur; 4633 path->locks[level] = BTRFS_READ_LOCK; 4634 4635 if (btrfs_header_generation(cur) < min_trans) { 4636 ret = 1; 4637 goto out; 4638 } 4639 while (1) { 4640 nritems = btrfs_header_nritems(cur); 4641 level = btrfs_header_level(cur); 4642 sret = btrfs_bin_search(cur, 0, min_key, &slot); 4643 if (sret < 0) { 4644 ret = sret; 4645 goto out; 4646 } 4647 4648 /* at the lowest level, we're done, setup the path and exit */ 4649 if (level == path->lowest_level) { 4650 if (slot >= nritems) 4651 goto find_next_key; 4652 ret = 0; 4653 path->slots[level] = slot; 4654 btrfs_item_key_to_cpu(cur, &found_key, slot); 4655 goto out; 4656 } 4657 if (sret && slot > 0) 4658 slot--; 4659 /* 4660 * check this node pointer against the min_trans parameters. 4661 * If it is too old, skip to the next one. 4662 */ 4663 while (slot < nritems) { 4664 u64 gen; 4665 4666 gen = btrfs_node_ptr_generation(cur, slot); 4667 if (gen < min_trans) { 4668 slot++; 4669 continue; 4670 } 4671 break; 4672 } 4673 find_next_key: 4674 /* 4675 * we didn't find a candidate key in this node, walk forward 4676 * and find another one 4677 */ 4678 if (slot >= nritems) { 4679 path->slots[level] = slot; 4680 sret = btrfs_find_next_key(root, path, min_key, level, 4681 min_trans); 4682 if (sret == 0) { 4683 btrfs_release_path(path); 4684 goto again; 4685 } else { 4686 goto out; 4687 } 4688 } 4689 /* save our key for returning back */ 4690 btrfs_node_key_to_cpu(cur, &found_key, slot); 4691 path->slots[level] = slot; 4692 if (level == path->lowest_level) { 4693 ret = 0; 4694 goto out; 4695 } 4696 cur = btrfs_read_node_slot(cur, slot); 4697 if (IS_ERR(cur)) { 4698 ret = PTR_ERR(cur); 4699 goto out; 4700 } 4701 4702 btrfs_tree_read_lock(cur); 4703 4704 path->locks[level - 1] = BTRFS_READ_LOCK; 4705 path->nodes[level - 1] = cur; 4706 unlock_up(path, level, 1, 0, NULL); 4707 } 4708 out: 4709 path->keep_locks = keep_locks; 4710 if (ret == 0) { 4711 btrfs_unlock_up_safe(path, path->lowest_level + 1); 4712 memcpy(min_key, &found_key, sizeof(found_key)); 4713 } 4714 return ret; 4715 } 4716 4717 /* 4718 * this is similar to btrfs_next_leaf, but does not try to preserve 4719 * and fixup the path. It looks for and returns the next key in the 4720 * tree based on the current path and the min_trans parameters. 4721 * 4722 * 0 is returned if another key is found, < 0 if there are any errors 4723 * and 1 is returned if there are no higher keys in the tree 4724 * 4725 * path->keep_locks should be set to 1 on the search made before 4726 * calling this function. 4727 */ 4728 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, 4729 struct btrfs_key *key, int level, u64 min_trans) 4730 { 4731 int slot; 4732 struct extent_buffer *c; 4733 4734 WARN_ON(!path->keep_locks && !path->skip_locking); 4735 while (level < BTRFS_MAX_LEVEL) { 4736 if (!path->nodes[level]) 4737 return 1; 4738 4739 slot = path->slots[level] + 1; 4740 c = path->nodes[level]; 4741 next: 4742 if (slot >= btrfs_header_nritems(c)) { 4743 int ret; 4744 int orig_lowest; 4745 struct btrfs_key cur_key; 4746 if (level + 1 >= BTRFS_MAX_LEVEL || 4747 !path->nodes[level + 1]) 4748 return 1; 4749 4750 if (path->locks[level + 1] || path->skip_locking) { 4751 level++; 4752 continue; 4753 } 4754 4755 slot = btrfs_header_nritems(c) - 1; 4756 if (level == 0) 4757 btrfs_item_key_to_cpu(c, &cur_key, slot); 4758 else 4759 btrfs_node_key_to_cpu(c, &cur_key, slot); 4760 4761 orig_lowest = path->lowest_level; 4762 btrfs_release_path(path); 4763 path->lowest_level = level; 4764 ret = btrfs_search_slot(NULL, root, &cur_key, path, 4765 0, 0); 4766 path->lowest_level = orig_lowest; 4767 if (ret < 0) 4768 return ret; 4769 4770 c = path->nodes[level]; 4771 slot = path->slots[level]; 4772 if (ret == 0) 4773 slot++; 4774 goto next; 4775 } 4776 4777 if (level == 0) 4778 btrfs_item_key_to_cpu(c, key, slot); 4779 else { 4780 u64 gen = btrfs_node_ptr_generation(c, slot); 4781 4782 if (gen < min_trans) { 4783 slot++; 4784 goto next; 4785 } 4786 btrfs_node_key_to_cpu(c, key, slot); 4787 } 4788 return 0; 4789 } 4790 return 1; 4791 } 4792 4793 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, 4794 u64 time_seq) 4795 { 4796 int slot; 4797 int level; 4798 struct extent_buffer *c; 4799 struct extent_buffer *next; 4800 struct btrfs_fs_info *fs_info = root->fs_info; 4801 struct btrfs_key key; 4802 bool need_commit_sem = false; 4803 u32 nritems; 4804 int ret; 4805 int i; 4806 4807 /* 4808 * The nowait semantics are used only for write paths, where we don't 4809 * use the tree mod log and sequence numbers. 4810 */ 4811 if (time_seq) 4812 ASSERT(!path->nowait); 4813 4814 nritems = btrfs_header_nritems(path->nodes[0]); 4815 if (nritems == 0) 4816 return 1; 4817 4818 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); 4819 again: 4820 level = 1; 4821 next = NULL; 4822 btrfs_release_path(path); 4823 4824 path->keep_locks = 1; 4825 4826 if (time_seq) { 4827 ret = btrfs_search_old_slot(root, &key, path, time_seq); 4828 } else { 4829 if (path->need_commit_sem) { 4830 path->need_commit_sem = 0; 4831 need_commit_sem = true; 4832 if (path->nowait) { 4833 if (!down_read_trylock(&fs_info->commit_root_sem)) { 4834 ret = -EAGAIN; 4835 goto done; 4836 } 4837 } else { 4838 down_read(&fs_info->commit_root_sem); 4839 } 4840 } 4841 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 4842 } 4843 path->keep_locks = 0; 4844 4845 if (ret < 0) 4846 goto done; 4847 4848 nritems = btrfs_header_nritems(path->nodes[0]); 4849 /* 4850 * by releasing the path above we dropped all our locks. A balance 4851 * could have added more items next to the key that used to be 4852 * at the very end of the block. So, check again here and 4853 * advance the path if there are now more items available. 4854 */ 4855 if (nritems > 0 && path->slots[0] < nritems - 1) { 4856 if (ret == 0) 4857 path->slots[0]++; 4858 ret = 0; 4859 goto done; 4860 } 4861 /* 4862 * So the above check misses one case: 4863 * - after releasing the path above, someone has removed the item that 4864 * used to be at the very end of the block, and balance between leafs 4865 * gets another one with bigger key.offset to replace it. 4866 * 4867 * This one should be returned as well, or we can get leaf corruption 4868 * later(esp. in __btrfs_drop_extents()). 4869 * 4870 * And a bit more explanation about this check, 4871 * with ret > 0, the key isn't found, the path points to the slot 4872 * where it should be inserted, so the path->slots[0] item must be the 4873 * bigger one. 4874 */ 4875 if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { 4876 ret = 0; 4877 goto done; 4878 } 4879 4880 while (level < BTRFS_MAX_LEVEL) { 4881 if (!path->nodes[level]) { 4882 ret = 1; 4883 goto done; 4884 } 4885 4886 slot = path->slots[level] + 1; 4887 c = path->nodes[level]; 4888 if (slot >= btrfs_header_nritems(c)) { 4889 level++; 4890 if (level == BTRFS_MAX_LEVEL) { 4891 ret = 1; 4892 goto done; 4893 } 4894 continue; 4895 } 4896 4897 4898 /* 4899 * Our current level is where we're going to start from, and to 4900 * make sure lockdep doesn't complain we need to drop our locks 4901 * and nodes from 0 to our current level. 4902 */ 4903 for (i = 0; i < level; i++) { 4904 if (path->locks[level]) { 4905 btrfs_tree_read_unlock(path->nodes[i]); 4906 path->locks[i] = 0; 4907 } 4908 free_extent_buffer(path->nodes[i]); 4909 path->nodes[i] = NULL; 4910 } 4911 4912 next = c; 4913 ret = read_block_for_search(root, path, &next, level, 4914 slot, &key); 4915 if (ret == -EAGAIN && !path->nowait) 4916 goto again; 4917 4918 if (ret < 0) { 4919 btrfs_release_path(path); 4920 goto done; 4921 } 4922 4923 if (!path->skip_locking) { 4924 ret = btrfs_try_tree_read_lock(next); 4925 if (!ret && path->nowait) { 4926 ret = -EAGAIN; 4927 goto done; 4928 } 4929 if (!ret && time_seq) { 4930 /* 4931 * If we don't get the lock, we may be racing 4932 * with push_leaf_left, holding that lock while 4933 * itself waiting for the leaf we've currently 4934 * locked. To solve this situation, we give up 4935 * on our lock and cycle. 4936 */ 4937 free_extent_buffer(next); 4938 btrfs_release_path(path); 4939 cond_resched(); 4940 goto again; 4941 } 4942 if (!ret) 4943 btrfs_tree_read_lock(next); 4944 } 4945 break; 4946 } 4947 path->slots[level] = slot; 4948 while (1) { 4949 level--; 4950 path->nodes[level] = next; 4951 path->slots[level] = 0; 4952 if (!path->skip_locking) 4953 path->locks[level] = BTRFS_READ_LOCK; 4954 if (!level) 4955 break; 4956 4957 ret = read_block_for_search(root, path, &next, level, 4958 0, &key); 4959 if (ret == -EAGAIN && !path->nowait) 4960 goto again; 4961 4962 if (ret < 0) { 4963 btrfs_release_path(path); 4964 goto done; 4965 } 4966 4967 if (!path->skip_locking) { 4968 if (path->nowait) { 4969 if (!btrfs_try_tree_read_lock(next)) { 4970 ret = -EAGAIN; 4971 goto done; 4972 } 4973 } else { 4974 btrfs_tree_read_lock(next); 4975 } 4976 } 4977 } 4978 ret = 0; 4979 done: 4980 unlock_up(path, 0, 1, 0, NULL); 4981 if (need_commit_sem) { 4982 int ret2; 4983 4984 path->need_commit_sem = 1; 4985 ret2 = finish_need_commit_sem_search(path); 4986 up_read(&fs_info->commit_root_sem); 4987 if (ret2) 4988 ret = ret2; 4989 } 4990 4991 return ret; 4992 } 4993 4994 int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq) 4995 { 4996 path->slots[0]++; 4997 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) 4998 return btrfs_next_old_leaf(root, path, time_seq); 4999 return 0; 5000 } 5001 5002 /* 5003 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps 5004 * searching until it gets past min_objectid or finds an item of 'type' 5005 * 5006 * returns 0 if something is found, 1 if nothing was found and < 0 on error 5007 */ 5008 int btrfs_previous_item(struct btrfs_root *root, 5009 struct btrfs_path *path, u64 min_objectid, 5010 int type) 5011 { 5012 struct btrfs_key found_key; 5013 struct extent_buffer *leaf; 5014 u32 nritems; 5015 int ret; 5016 5017 while (1) { 5018 if (path->slots[0] == 0) { 5019 ret = btrfs_prev_leaf(root, path); 5020 if (ret != 0) 5021 return ret; 5022 } else { 5023 path->slots[0]--; 5024 } 5025 leaf = path->nodes[0]; 5026 nritems = btrfs_header_nritems(leaf); 5027 if (nritems == 0) 5028 return 1; 5029 if (path->slots[0] == nritems) 5030 path->slots[0]--; 5031 5032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5033 if (found_key.objectid < min_objectid) 5034 break; 5035 if (found_key.type == type) 5036 return 0; 5037 if (found_key.objectid == min_objectid && 5038 found_key.type < type) 5039 break; 5040 } 5041 return 1; 5042 } 5043 5044 /* 5045 * search in extent tree to find a previous Metadata/Data extent item with 5046 * min objecitd. 5047 * 5048 * returns 0 if something is found, 1 if nothing was found and < 0 on error 5049 */ 5050 int btrfs_previous_extent_item(struct btrfs_root *root, 5051 struct btrfs_path *path, u64 min_objectid) 5052 { 5053 struct btrfs_key found_key; 5054 struct extent_buffer *leaf; 5055 u32 nritems; 5056 int ret; 5057 5058 while (1) { 5059 if (path->slots[0] == 0) { 5060 ret = btrfs_prev_leaf(root, path); 5061 if (ret != 0) 5062 return ret; 5063 } else { 5064 path->slots[0]--; 5065 } 5066 leaf = path->nodes[0]; 5067 nritems = btrfs_header_nritems(leaf); 5068 if (nritems == 0) 5069 return 1; 5070 if (path->slots[0] == nritems) 5071 path->slots[0]--; 5072 5073 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); 5074 if (found_key.objectid < min_objectid) 5075 break; 5076 if (found_key.type == BTRFS_EXTENT_ITEM_KEY || 5077 found_key.type == BTRFS_METADATA_ITEM_KEY) 5078 return 0; 5079 if (found_key.objectid == min_objectid && 5080 found_key.type < BTRFS_EXTENT_ITEM_KEY) 5081 break; 5082 } 5083 return 1; 5084 } 5085 5086 int __init btrfs_ctree_init(void) 5087 { 5088 btrfs_path_cachep = kmem_cache_create("btrfs_path", 5089 sizeof(struct btrfs_path), 0, 5090 SLAB_MEM_SPREAD, NULL); 5091 if (!btrfs_path_cachep) 5092 return -ENOMEM; 5093 return 0; 5094 } 5095 5096 void __cold btrfs_ctree_exit(void) 5097 { 5098 kmem_cache_destroy(btrfs_path_cachep); 5099 } 5100