1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2007 Oracle. All rights reserved.
10 #include <linux/sched/signal.h>
11 #include <linux/highmem.h>
13 #include <linux/rwsem.h>
14 #include <linux/semaphore.h>
15 #include <linux/completion.h>
16 #include <linux/backing-dev.h>
17 #include <linux/wait.h>
18 #include <linux/slab.h>
19 #include <trace/events/btrfs.h>
20 #include <asm/unaligned.h>
21 #include <linux/pagemap.h>
22 #include <linux/btrfs.h>
23 #include <linux/btrfs_tree.h>
24 #include <linux/workqueue.h>
25 #include <linux/security.h>
26 #include <linux/sizes.h>
27 #include <linux/dynamic_debug.h>
28 #include <linux/refcount.h>
29 #include <linux/crc32c.h>
30 #include <linux/iomap.h>
31 #include <linux/fscrypt.h>
32 #include "extent-io-tree.h"
33 #include "extent_io.h"
34 #include "extent_map.h"
35 #include "async-thread.h"
36 #include "block-rsv.h"
41 struct btrfs_trans_handle;
42 struct btrfs_transaction;
43 struct btrfs_pending_snapshot;
44 struct btrfs_delayed_ref_root;
45 struct btrfs_space_info;
46 struct btrfs_block_group;
47 struct btrfs_ordered_sum;
50 struct btrfs_ioctl_encoded_io_args;
52 struct btrfs_fs_devices;
53 struct btrfs_balance_control;
54 struct btrfs_delayed_root;
57 /* Read ahead values for struct btrfs_path.reada */
63 * Similar to READA_FORWARD but unlike it:
65 * 1) It will trigger readahead even for leaves that are not close to
67 * 2) It also triggers readahead for nodes;
68 * 3) During a search, even when a node or leaf is already in memory, it
69 * will still trigger readahead for other nodes and leaves that follow
72 * This is meant to be used only when we know we are iterating over the
73 * entire tree or a very large part of it.
79 * btrfs_paths remember the path taken from the root down to the leaf.
80 * level 0 is always the leaf, and nodes[1...BTRFS_MAX_LEVEL] will point
81 * to any other levels that are present.
83 * The slots array records the index of the item or block pointer
84 * used while walking the tree.
87 struct extent_buffer *nodes[BTRFS_MAX_LEVEL];
88 int slots[BTRFS_MAX_LEVEL];
89 /* if there is real range locking, this locks field will change */
90 u8 locks[BTRFS_MAX_LEVEL];
92 /* keep some upper locks as we walk down */
96 * set by btrfs_split_item, tells search_slot to keep all locks
97 * and to force calls to keep space in the nodes
99 unsigned int search_for_split:1;
100 unsigned int keep_locks:1;
101 unsigned int skip_locking:1;
102 unsigned int search_commit_root:1;
103 unsigned int need_commit_sem:1;
104 unsigned int skip_release_on_error:1;
106 * Indicate that new item (btrfs_search_slot) is extending already
107 * existing item and ins_len contains only the data size and not item
108 * header (ie. sizeof(struct btrfs_item) is not included).
110 unsigned int search_for_extension:1;
111 /* Stop search if any locks need to be taken (for read) */
112 unsigned int nowait:1;
116 * The state of btrfs root
120 * btrfs_record_root_in_trans is a multi-step process, and it can race
121 * with the balancing code. But the race is very small, and only the
122 * first time the root is added to each transaction. So IN_TRANS_SETUP
123 * is used to tell us when more checks are required
125 BTRFS_ROOT_IN_TRANS_SETUP,
128 * Set if tree blocks of this root can be shared by other roots.
129 * Only subvolume trees and their reloc trees have this bit set.
130 * Conflicts with TRACK_DIRTY bit.
132 * This affects two things:
134 * - How balance works
135 * For shareable roots, we need to use reloc tree and do path
136 * replacement for balance, and need various pre/post hooks for
137 * snapshot creation to handle them.
139 * While for non-shareable trees, we just simply do a tree search
142 * - How dirty roots are tracked
143 * For shareable roots, btrfs_record_root_in_trans() is needed to
144 * track them, while non-subvolume roots have TRACK_DIRTY bit, they
145 * don't need to set this manually.
147 BTRFS_ROOT_SHAREABLE,
148 BTRFS_ROOT_TRACK_DIRTY,
150 BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
151 BTRFS_ROOT_DEFRAG_RUNNING,
152 BTRFS_ROOT_FORCE_COW,
153 BTRFS_ROOT_MULTI_LOG_TASKS,
158 * Reloc tree is orphan, only kept here for qgroup delayed subtree scan
160 * Set for the subvolume tree owning the reloc tree.
162 BTRFS_ROOT_DEAD_RELOC_TREE,
163 /* Mark dead root stored on device whose cleanup needs to be resumed */
164 BTRFS_ROOT_DEAD_TREE,
165 /* The root has a log tree. Used for subvolume roots and the tree root. */
166 BTRFS_ROOT_HAS_LOG_TREE,
167 /* Qgroup flushing is in progress */
168 BTRFS_ROOT_QGROUP_FLUSHING,
169 /* We started the orphan cleanup for this root. */
170 BTRFS_ROOT_ORPHAN_CLEANUP,
171 /* This root has a drop operation that was started previously. */
172 BTRFS_ROOT_UNFINISHED_DROP,
173 /* This reloc root needs to have its buffers lockdep class reset. */
174 BTRFS_ROOT_RESET_LOCKDEP_CLASS,
178 * Record swapped tree blocks of a subvolume tree for delayed subtree trace
179 * code. For detail check comment in fs/btrfs/qgroup.c.
181 struct btrfs_qgroup_swapped_blocks {
183 /* RM_EMPTY_ROOT() of above blocks[] */
185 struct rb_root blocks[BTRFS_MAX_LEVEL];
189 * in ram representation of the tree. extent_root is used for all allocations
190 * and for the extent tree extent_root root.
193 struct rb_node rb_node;
195 struct extent_buffer *node;
197 struct extent_buffer *commit_root;
198 struct btrfs_root *log_root;
199 struct btrfs_root *reloc_root;
202 struct btrfs_root_item root_item;
203 struct btrfs_key root_key;
204 struct btrfs_fs_info *fs_info;
205 struct extent_io_tree dirty_log_pages;
207 struct mutex objectid_mutex;
209 spinlock_t accounting_lock;
210 struct btrfs_block_rsv *block_rsv;
212 struct mutex log_mutex;
213 wait_queue_head_t log_writer_wait;
214 wait_queue_head_t log_commit_wait[2];
215 struct list_head log_ctxs[2];
216 /* Used only for log trees of subvolumes, not for the log root tree */
217 atomic_t log_writers;
218 atomic_t log_commit[2];
219 /* Used only for log trees of subvolumes, not for the log root tree */
222 /* No matter the commit succeeds or not*/
223 int log_transid_committed;
224 /* Just be updated when the commit succeeds. */
234 struct btrfs_key defrag_progress;
235 struct btrfs_key defrag_max;
237 /* The dirty list is only used by non-shareable roots */
238 struct list_head dirty_list;
240 struct list_head root_list;
242 spinlock_t log_extents_lock[2];
243 struct list_head logged_list[2];
245 spinlock_t inode_lock;
246 /* red-black tree that keeps track of in-memory inodes */
247 struct rb_root inode_tree;
250 * radix tree that keeps track of delayed nodes of every inode,
251 * protected by inode_lock
253 struct radix_tree_root delayed_nodes_tree;
255 * right now this just gets used so that a root has its own devid
256 * for stat. It may be used for more later
260 spinlock_t root_item_lock;
263 struct mutex delalloc_mutex;
264 spinlock_t delalloc_lock;
266 * all of the inodes that have delalloc bytes. It is possible for
267 * this list to be empty even when there is still dirty data=ordered
268 * extents waiting to finish IO.
270 struct list_head delalloc_inodes;
271 struct list_head delalloc_root;
272 u64 nr_delalloc_inodes;
274 struct mutex ordered_extent_mutex;
276 * this is used by the balancing code to wait for all the pending
279 spinlock_t ordered_extent_lock;
282 * all of the data=ordered extents pending writeback
283 * these can span multiple transactions and basically include
284 * every dirty data page that isn't from nodatacow
286 struct list_head ordered_extents;
287 struct list_head ordered_root;
288 u64 nr_ordered_extents;
291 * Not empty if this subvolume root has gone through tree block swap
294 * Will be used by reloc_control::dirty_subvol_roots.
296 struct list_head reloc_dirty_list;
299 * Number of currently running SEND ioctls to prevent
300 * manipulation with the read-only status via SUBVOL_SETFLAGS
302 int send_in_progress;
304 * Number of currently running deduplication operations that have a
305 * destination inode belonging to this root. Protected by the lock
308 int dedupe_in_progress;
309 /* For exclusion of snapshot creation and nocow writes */
310 struct btrfs_drew_lock snapshot_lock;
312 atomic_t snapshot_force_cow;
314 /* For qgroup metadata reserved space */
315 spinlock_t qgroup_meta_rsv_lock;
316 u64 qgroup_meta_rsv_pertrans;
317 u64 qgroup_meta_rsv_prealloc;
318 wait_queue_head_t qgroup_flush_wait;
320 /* Number of active swapfiles */
321 atomic_t nr_swapfiles;
323 /* Record pairs of swapped blocks for qgroup */
324 struct btrfs_qgroup_swapped_blocks swapped_blocks;
326 /* Used only by log trees, when logging csum items */
327 struct extent_io_tree log_csum_range;
329 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
333 #ifdef CONFIG_BTRFS_DEBUG
334 struct list_head leak_list;
338 static inline bool btrfs_root_readonly(const struct btrfs_root *root)
340 /* Byte-swap the constant at compile time, root_item::flags is LE */
341 return (root->root_item.flags & cpu_to_le64(BTRFS_ROOT_SUBVOL_RDONLY)) != 0;
344 static inline bool btrfs_root_dead(const struct btrfs_root *root)
346 /* Byte-swap the constant at compile time, root_item::flags is LE */
347 return (root->root_item.flags & cpu_to_le64(BTRFS_ROOT_SUBVOL_DEAD)) != 0;
350 static inline u64 btrfs_root_id(const struct btrfs_root *root)
352 return root->root_key.objectid;
356 * Structure that conveys information about an extent that is going to replace
357 * all the extents in a file range.
359 struct btrfs_replace_extent_info {
365 /* Pointer to a file extent item of type regular or prealloc. */
368 * Set to true when attempting to replace a file range with a new extent
369 * described by this structure, set to false when attempting to clone an
370 * existing extent into a file range.
373 /* Indicate if we should update the inode's mtime and ctime. */
375 /* Meaningful only if is_new_extent is true. */
378 * Meaningful only if is_new_extent is true.
379 * Used to track how many extent items we have already inserted in a
380 * subvolume tree that refer to the extent described by this structure,
381 * so that we know when to create a new delayed ref or update an existing
387 /* Arguments for btrfs_drop_extents() */
388 struct btrfs_drop_extents_args {
389 /* Input parameters */
392 * If NULL, btrfs_drop_extents() will allocate and free its own path.
393 * If 'replace_extent' is true, this must not be NULL. Also the path
394 * is always released except if 'replace_extent' is true and
395 * btrfs_drop_extents() sets 'extent_inserted' to true, in which case
396 * the path is kept locked.
398 struct btrfs_path *path;
399 /* Start offset of the range to drop extents from */
401 /* End (exclusive, last byte + 1) of the range to drop extents from */
403 /* If true drop all the extent maps in the range */
406 * If true it means we want to insert a new extent after dropping all
407 * the extents in the range. If this is true, the 'extent_item_size'
408 * parameter must be set as well and the 'extent_inserted' field will
409 * be set to true by btrfs_drop_extents() if it could insert the new
411 * Note: when this is set to true the path must not be NULL.
415 * Used if 'replace_extent' is true. Size of the file extent item to
416 * insert after dropping all existing extents in the range
418 u32 extent_item_size;
420 /* Output parameters */
423 * Set to the minimum between the input parameter 'end' and the end
424 * (exclusive, last byte + 1) of the last dropped extent. This is always
425 * set even if btrfs_drop_extents() returns an error.
429 * The number of allocated bytes found in the range. This can be smaller
430 * than the range's length when there are holes in the range.
434 * Only set if 'replace_extent' is true. Set to true if we were able
435 * to insert a replacement extent after dropping all extents in the
436 * range, otherwise set to false by btrfs_drop_extents().
437 * Also, if btrfs_drop_extents() has set this to true it means it
438 * returned with the path locked, otherwise if it has set this to
439 * false it has returned with the path released.
441 bool extent_inserted;
444 struct btrfs_file_private {
447 struct extent_state *llseek_cached_state;
450 static inline u32 BTRFS_LEAF_DATA_SIZE(const struct btrfs_fs_info *info)
452 return info->nodesize - sizeof(struct btrfs_header);
455 static inline u32 BTRFS_MAX_ITEM_SIZE(const struct btrfs_fs_info *info)
457 return BTRFS_LEAF_DATA_SIZE(info) - sizeof(struct btrfs_item);
460 static inline u32 BTRFS_NODEPTRS_PER_BLOCK(const struct btrfs_fs_info *info)
462 return BTRFS_LEAF_DATA_SIZE(info) / sizeof(struct btrfs_key_ptr);
465 static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)
467 return BTRFS_MAX_ITEM_SIZE(info) - sizeof(struct btrfs_dir_item);
470 #define BTRFS_BYTES_TO_BLKS(fs_info, bytes) \
471 ((bytes) >> (fs_info)->sectorsize_bits)
473 static inline u32 btrfs_crc32c(u32 crc, const void *address, unsigned length)
475 return crc32c(crc, address, length);
478 static inline void btrfs_crc32c_final(u32 crc, u8 *result)
480 put_unaligned_le32(~crc, result);
483 static inline u64 btrfs_name_hash(const char *name, int len)
485 return crc32c((u32)~1, name, len);
489 * Figure the key offset of an extended inode ref
491 static inline u64 btrfs_extref_hash(u64 parent_objectid, const char *name,
494 return (u64) crc32c(parent_objectid, name, len);
497 static inline gfp_t btrfs_alloc_write_mask(struct address_space *mapping)
499 return mapping_gfp_constraint(mapping, ~__GFP_FS);
502 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
504 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
505 u64 num_bytes, u64 *actual_bytes);
506 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range);
509 int __init btrfs_ctree_init(void);
510 void __cold btrfs_ctree_exit(void);
512 int btrfs_bin_search(struct extent_buffer *eb, int first_slot,
513 const struct btrfs_key *key, int *slot);
515 int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2);
516 int btrfs_previous_item(struct btrfs_root *root,
517 struct btrfs_path *path, u64 min_objectid,
519 int btrfs_previous_extent_item(struct btrfs_root *root,
520 struct btrfs_path *path, u64 min_objectid);
521 void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
522 struct btrfs_path *path,
523 const struct btrfs_key *new_key);
524 struct extent_buffer *btrfs_root_node(struct btrfs_root *root);
525 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
526 struct btrfs_key *key, int lowest_level,
528 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
529 struct btrfs_path *path,
531 struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
534 int btrfs_cow_block(struct btrfs_trans_handle *trans,
535 struct btrfs_root *root, struct extent_buffer *buf,
536 struct extent_buffer *parent, int parent_slot,
537 struct extent_buffer **cow_ret,
538 enum btrfs_lock_nesting nest);
539 int btrfs_copy_root(struct btrfs_trans_handle *trans,
540 struct btrfs_root *root,
541 struct extent_buffer *buf,
542 struct extent_buffer **cow_ret, u64 new_root_objectid);
543 int btrfs_block_can_be_shared(struct btrfs_root *root,
544 struct extent_buffer *buf);
545 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
546 struct btrfs_path *path, int level, int slot);
547 void btrfs_extend_item(struct btrfs_path *path, u32 data_size);
548 void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end);
549 int btrfs_split_item(struct btrfs_trans_handle *trans,
550 struct btrfs_root *root,
551 struct btrfs_path *path,
552 const struct btrfs_key *new_key,
553 unsigned long split_offset);
554 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
555 struct btrfs_root *root,
556 struct btrfs_path *path,
557 const struct btrfs_key *new_key);
558 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
559 u64 inum, u64 ioff, u8 key_type, struct btrfs_key *found_key);
560 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
561 const struct btrfs_key *key, struct btrfs_path *p,
562 int ins_len, int cow);
563 int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
564 struct btrfs_path *p, u64 time_seq);
565 int btrfs_search_slot_for_read(struct btrfs_root *root,
566 const struct btrfs_key *key,
567 struct btrfs_path *p, int find_higher,
569 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
570 struct btrfs_root *root, struct extent_buffer *parent,
571 int start_slot, u64 *last_ret,
572 struct btrfs_key *progress);
573 void btrfs_release_path(struct btrfs_path *p);
574 struct btrfs_path *btrfs_alloc_path(void);
575 void btrfs_free_path(struct btrfs_path *p);
577 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
578 struct btrfs_path *path, int slot, int nr);
579 static inline int btrfs_del_item(struct btrfs_trans_handle *trans,
580 struct btrfs_root *root,
581 struct btrfs_path *path)
583 return btrfs_del_items(trans, root, path, path->slots[0], 1);
587 * Describes a batch of items to insert in a btree. This is used by
588 * btrfs_insert_empty_items().
590 struct btrfs_item_batch {
592 * Pointer to an array containing the keys of the items to insert (in
595 const struct btrfs_key *keys;
596 /* Pointer to an array containing the data size for each item to insert. */
597 const u32 *data_sizes;
599 * The sum of data sizes for all items. The caller can compute this while
600 * setting up the data_sizes array, so it ends up being more efficient
601 * than having btrfs_insert_empty_items() or setup_item_for_insert()
602 * doing it, as it would avoid an extra loop over a potentially large
603 * array, and in the case of setup_item_for_insert(), we would be doing
604 * it while holding a write lock on a leaf and often on upper level nodes
605 * too, unnecessarily increasing the size of a critical section.
608 /* Size of the keys and data_sizes arrays (number of items in the batch). */
612 void btrfs_setup_item_for_insert(struct btrfs_root *root,
613 struct btrfs_path *path,
614 const struct btrfs_key *key,
616 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
617 const struct btrfs_key *key, void *data, u32 data_size);
618 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
619 struct btrfs_root *root,
620 struct btrfs_path *path,
621 const struct btrfs_item_batch *batch);
623 static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root,
625 struct btrfs_path *path,
626 const struct btrfs_key *key,
629 struct btrfs_item_batch batch;
632 batch.data_sizes = &data_size;
633 batch.total_data_size = data_size;
636 return btrfs_insert_empty_items(trans, root, path, &batch);
639 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
642 int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key,
643 struct btrfs_path *path);
645 int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key,
646 struct btrfs_path *path);
649 * Search in @root for a given @key, and store the slot found in @found_key.
651 * @root: The root node of the tree.
652 * @key: The key we are looking for.
653 * @found_key: Will hold the found item.
654 * @path: Holds the current slot/leaf.
655 * @iter_ret: Contains the value returned from btrfs_search_slot or
656 * btrfs_get_next_valid_item, whichever was executed last.
658 * The @iter_ret is an output variable that will contain the return value of
659 * btrfs_search_slot, if it encountered an error, or the value returned from
660 * btrfs_get_next_valid_item otherwise. That return value can be 0, if a valid
661 * slot was found, 1 if there were no more leaves, and <0 if there was an error.
663 * It's recommended to use a separate variable for iter_ret and then use it to
664 * set the function return value so there's no confusion of the 0/1/errno
665 * values stemming from btrfs_search_slot.
667 #define btrfs_for_each_slot(root, key, found_key, path, iter_ret) \
668 for (iter_ret = btrfs_search_slot(NULL, (root), (key), (path), 0, 0); \
670 (iter_ret = btrfs_get_next_valid_item((root), (found_key), (path))) == 0; \
674 int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq);
677 * Search the tree again to find a leaf with greater keys.
679 * Returns 0 if it found something or 1 if there are no greater leaves.
680 * Returns < 0 on error.
682 static inline int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
684 return btrfs_next_old_leaf(root, path, 0);
687 static inline int btrfs_next_item(struct btrfs_root *root, struct btrfs_path *p)
689 return btrfs_next_old_item(root, p, 0);
691 int btrfs_leaf_free_space(const struct extent_buffer *leaf);
693 static inline int is_fstree(u64 rootid)
695 if (rootid == BTRFS_FS_TREE_OBJECTID ||
696 ((s64)rootid >= (s64)BTRFS_FIRST_FREE_OBJECTID &&
697 !btrfs_qgroup_level(rootid)))
702 static inline bool btrfs_is_data_reloc_root(const struct btrfs_root *root)
704 return root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID;
707 u16 btrfs_csum_type_size(u16 type);
708 int btrfs_super_csum_size(const struct btrfs_super_block *s);
709 const char *btrfs_super_csum_name(u16 csum_type);
710 const char *btrfs_super_csum_driver(u16 csum_type);
711 size_t __attribute_const__ btrfs_get_num_csums(void);
714 * We use page status Private2 to indicate there is an ordered extent with
717 * Rename the Private2 accessors to Ordered, to improve readability.
719 #define PageOrdered(page) PagePrivate2(page)
720 #define SetPageOrdered(page) SetPagePrivate2(page)
721 #define ClearPageOrdered(page) ClearPagePrivate2(page)
722 #define folio_test_ordered(folio) folio_test_private_2(folio)
723 #define folio_set_ordered(folio) folio_set_private_2(folio)
724 #define folio_clear_ordered(folio) folio_clear_private_2(folio)