Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#ifndef BTRFS_CTREE_H
7#define BTRFS_CTREE_H
8
9#include <linux/mm.h>
10#include <linux/sched/signal.h>
11#include <linux/highmem.h>
12#include <linux/fs.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"
37#include "locking.h"
38#include "misc.h"
39#include "fs.h"
40
41struct btrfs_trans_handle;
42struct btrfs_transaction;
43struct btrfs_pending_snapshot;
44struct btrfs_delayed_ref_root;
45struct btrfs_space_info;
46struct btrfs_block_group;
47struct btrfs_ordered_sum;
48struct btrfs_ref;
49struct btrfs_bio;
50struct btrfs_ioctl_encoded_io_args;
51struct btrfs_device;
52struct btrfs_fs_devices;
53struct btrfs_balance_control;
54struct btrfs_delayed_root;
55struct reloc_control;
56
57/* Read ahead values for struct btrfs_path.reada */
58enum {
59 READA_NONE,
60 READA_BACK,
61 READA_FORWARD,
62 /*
63 * Similar to READA_FORWARD but unlike it:
64 *
65 * 1) It will trigger readahead even for leaves that are not close to
66 * each other on disk;
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
70 * it.
71 *
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.
74 */
75 READA_FORWARD_ALWAYS,
76};
77
78/*
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.
82 *
83 * The slots array records the index of the item or block pointer
84 * used while walking the tree.
85 */
86struct btrfs_path {
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];
91 u8 reada;
92 /* keep some upper locks as we walk down */
93 u8 lowest_level;
94
95 /*
96 * set by btrfs_split_item, tells search_slot to keep all locks
97 * and to force calls to keep space in the nodes
98 */
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;
105 /*
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).
109 */
110 unsigned int search_for_extension:1;
111 /* Stop search if any locks need to be taken (for read) */
112 unsigned int nowait:1;
113};
114
115/*
116 * The state of btrfs root
117 */
118enum {
119 /*
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
124 */
125 BTRFS_ROOT_IN_TRANS_SETUP,
126
127 /*
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.
131 *
132 * This affects two things:
133 *
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.
138 *
139 * While for non-shareable trees, we just simply do a tree search
140 * with COW.
141 *
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.
146 */
147 BTRFS_ROOT_SHAREABLE,
148 BTRFS_ROOT_TRACK_DIRTY,
149 BTRFS_ROOT_IN_RADIX,
150 BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
151 BTRFS_ROOT_DEFRAG_RUNNING,
152 BTRFS_ROOT_FORCE_COW,
153 BTRFS_ROOT_MULTI_LOG_TASKS,
154 BTRFS_ROOT_DIRTY,
155 BTRFS_ROOT_DELETING,
156
157 /*
158 * Reloc tree is orphan, only kept here for qgroup delayed subtree scan
159 *
160 * Set for the subvolume tree owning the reloc tree.
161 */
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,
175};
176
177/*
178 * Record swapped tree blocks of a subvolume tree for delayed subtree trace
179 * code. For detail check comment in fs/btrfs/qgroup.c.
180 */
181struct btrfs_qgroup_swapped_blocks {
182 spinlock_t lock;
183 /* RM_EMPTY_ROOT() of above blocks[] */
184 bool swapped;
185 struct rb_root blocks[BTRFS_MAX_LEVEL];
186};
187
188/*
189 * in ram representation of the tree. extent_root is used for all allocations
190 * and for the extent tree extent_root root.
191 */
192struct btrfs_root {
193 struct rb_node rb_node;
194
195 struct extent_buffer *node;
196
197 struct extent_buffer *commit_root;
198 struct btrfs_root *log_root;
199 struct btrfs_root *reloc_root;
200
201 unsigned long state;
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;
206
207 struct mutex objectid_mutex;
208
209 spinlock_t accounting_lock;
210 struct btrfs_block_rsv *block_rsv;
211
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 */
220 atomic_t log_batch;
221 int log_transid;
222 /* No matter the commit succeeds or not*/
223 int log_transid_committed;
224 /* Just be updated when the commit succeeds. */
225 int last_log_commit;
226 pid_t log_start_pid;
227
228 u64 last_trans;
229
230 u32 type;
231
232 u64 free_objectid;
233
234 struct btrfs_key defrag_progress;
235 struct btrfs_key defrag_max;
236
237 /* The dirty list is only used by non-shareable roots */
238 struct list_head dirty_list;
239
240 struct list_head root_list;
241
242 spinlock_t log_extents_lock[2];
243 struct list_head logged_list[2];
244
245 spinlock_t inode_lock;
246 /* red-black tree that keeps track of in-memory inodes */
247 struct rb_root inode_tree;
248
249 /*
250 * radix tree that keeps track of delayed nodes of every inode,
251 * protected by inode_lock
252 */
253 struct radix_tree_root delayed_nodes_tree;
254 /*
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
257 */
258 dev_t anon_dev;
259
260 spinlock_t root_item_lock;
261 refcount_t refs;
262
263 struct mutex delalloc_mutex;
264 spinlock_t delalloc_lock;
265 /*
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.
269 */
270 struct list_head delalloc_inodes;
271 struct list_head delalloc_root;
272 u64 nr_delalloc_inodes;
273
274 struct mutex ordered_extent_mutex;
275 /*
276 * this is used by the balancing code to wait for all the pending
277 * ordered extents
278 */
279 spinlock_t ordered_extent_lock;
280
281 /*
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
285 */
286 struct list_head ordered_extents;
287 struct list_head ordered_root;
288 u64 nr_ordered_extents;
289
290 /*
291 * Not empty if this subvolume root has gone through tree block swap
292 * (relocation)
293 *
294 * Will be used by reloc_control::dirty_subvol_roots.
295 */
296 struct list_head reloc_dirty_list;
297
298 /*
299 * Number of currently running SEND ioctls to prevent
300 * manipulation with the read-only status via SUBVOL_SETFLAGS
301 */
302 int send_in_progress;
303 /*
304 * Number of currently running deduplication operations that have a
305 * destination inode belonging to this root. Protected by the lock
306 * root_item_lock.
307 */
308 int dedupe_in_progress;
309 /* For exclusion of snapshot creation and nocow writes */
310 struct btrfs_drew_lock snapshot_lock;
311
312 atomic_t snapshot_force_cow;
313
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;
319
320 /* Number of active swapfiles */
321 atomic_t nr_swapfiles;
322
323 /* Record pairs of swapped blocks for qgroup */
324 struct btrfs_qgroup_swapped_blocks swapped_blocks;
325
326 /* Used only by log trees, when logging csum items */
327 struct extent_io_tree log_csum_range;
328
329#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
330 u64 alloc_bytenr;
331#endif
332
333#ifdef CONFIG_BTRFS_DEBUG
334 struct list_head leak_list;
335#endif
336};
337
338static inline bool btrfs_root_readonly(const struct btrfs_root *root)
339{
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;
342}
343
344static inline bool btrfs_root_dead(const struct btrfs_root *root)
345{
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;
348}
349
350static inline u64 btrfs_root_id(const struct btrfs_root *root)
351{
352 return root->root_key.objectid;
353}
354
355/*
356 * Structure that conveys information about an extent that is going to replace
357 * all the extents in a file range.
358 */
359struct btrfs_replace_extent_info {
360 u64 disk_offset;
361 u64 disk_len;
362 u64 data_offset;
363 u64 data_len;
364 u64 file_offset;
365 /* Pointer to a file extent item of type regular or prealloc. */
366 char *extent_buf;
367 /*
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.
371 */
372 bool is_new_extent;
373 /* Indicate if we should update the inode's mtime and ctime. */
374 bool update_times;
375 /* Meaningful only if is_new_extent is true. */
376 int qgroup_reserved;
377 /*
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
382 * one.
383 */
384 int insertions;
385};
386
387/* Arguments for btrfs_drop_extents() */
388struct btrfs_drop_extents_args {
389 /* Input parameters */
390
391 /*
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.
397 */
398 struct btrfs_path *path;
399 /* Start offset of the range to drop extents from */
400 u64 start;
401 /* End (exclusive, last byte + 1) of the range to drop extents from */
402 u64 end;
403 /* If true drop all the extent maps in the range */
404 bool drop_cache;
405 /*
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
410 * extent.
411 * Note: when this is set to true the path must not be NULL.
412 */
413 bool replace_extent;
414 /*
415 * Used if 'replace_extent' is true. Size of the file extent item to
416 * insert after dropping all existing extents in the range
417 */
418 u32 extent_item_size;
419
420 /* Output parameters */
421
422 /*
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.
426 */
427 u64 drop_end;
428 /*
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.
431 */
432 u64 bytes_found;
433 /*
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.
440 */
441 bool extent_inserted;
442};
443
444struct btrfs_file_private {
445 void *filldir_buf;
446 struct extent_state *llseek_cached_state;
447};
448
449static inline u32 BTRFS_LEAF_DATA_SIZE(const struct btrfs_fs_info *info)
450{
451 return info->nodesize - sizeof(struct btrfs_header);
452}
453
454static inline u32 BTRFS_MAX_ITEM_SIZE(const struct btrfs_fs_info *info)
455{
456 return BTRFS_LEAF_DATA_SIZE(info) - sizeof(struct btrfs_item);
457}
458
459static inline u32 BTRFS_NODEPTRS_PER_BLOCK(const struct btrfs_fs_info *info)
460{
461 return BTRFS_LEAF_DATA_SIZE(info) / sizeof(struct btrfs_key_ptr);
462}
463
464static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)
465{
466 return BTRFS_MAX_ITEM_SIZE(info) - sizeof(struct btrfs_dir_item);
467}
468
469#define BTRFS_BYTES_TO_BLKS(fs_info, bytes) \
470 ((bytes) >> (fs_info)->sectorsize_bits)
471
472static inline u32 btrfs_crc32c(u32 crc, const void *address, unsigned length)
473{
474 return crc32c(crc, address, length);
475}
476
477static inline void btrfs_crc32c_final(u32 crc, u8 *result)
478{
479 put_unaligned_le32(~crc, result);
480}
481
482static inline u64 btrfs_name_hash(const char *name, int len)
483{
484 return crc32c((u32)~1, name, len);
485}
486
487/*
488 * Figure the key offset of an extended inode ref
489 */
490static inline u64 btrfs_extref_hash(u64 parent_objectid, const char *name,
491 int len)
492{
493 return (u64) crc32c(parent_objectid, name, len);
494}
495
496static inline gfp_t btrfs_alloc_write_mask(struct address_space *mapping)
497{
498 return mapping_gfp_constraint(mapping, ~__GFP_FS);
499}
500
501int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
502 u64 start, u64 end);
503int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
504 u64 num_bytes, u64 *actual_bytes);
505int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range);
506
507/* ctree.c */
508int __init btrfs_ctree_init(void);
509void __cold btrfs_ctree_exit(void);
510int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key,
511 int *slot);
512int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2);
513int btrfs_previous_item(struct btrfs_root *root,
514 struct btrfs_path *path, u64 min_objectid,
515 int type);
516int btrfs_previous_extent_item(struct btrfs_root *root,
517 struct btrfs_path *path, u64 min_objectid);
518void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info,
519 struct btrfs_path *path,
520 const struct btrfs_key *new_key);
521struct extent_buffer *btrfs_root_node(struct btrfs_root *root);
522int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
523 struct btrfs_key *key, int lowest_level,
524 u64 min_trans);
525int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
526 struct btrfs_path *path,
527 u64 min_trans);
528struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
529 int slot);
530
531int btrfs_cow_block(struct btrfs_trans_handle *trans,
532 struct btrfs_root *root, struct extent_buffer *buf,
533 struct extent_buffer *parent, int parent_slot,
534 struct extent_buffer **cow_ret,
535 enum btrfs_lock_nesting nest);
536int btrfs_copy_root(struct btrfs_trans_handle *trans,
537 struct btrfs_root *root,
538 struct extent_buffer *buf,
539 struct extent_buffer **cow_ret, u64 new_root_objectid);
540int btrfs_block_can_be_shared(struct btrfs_root *root,
541 struct extent_buffer *buf);
542void btrfs_extend_item(struct btrfs_path *path, u32 data_size);
543void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end);
544int btrfs_split_item(struct btrfs_trans_handle *trans,
545 struct btrfs_root *root,
546 struct btrfs_path *path,
547 const struct btrfs_key *new_key,
548 unsigned long split_offset);
549int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
550 struct btrfs_root *root,
551 struct btrfs_path *path,
552 const struct btrfs_key *new_key);
553int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
554 u64 inum, u64 ioff, u8 key_type, struct btrfs_key *found_key);
555int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
556 const struct btrfs_key *key, struct btrfs_path *p,
557 int ins_len, int cow);
558int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
559 struct btrfs_path *p, u64 time_seq);
560int btrfs_search_slot_for_read(struct btrfs_root *root,
561 const struct btrfs_key *key,
562 struct btrfs_path *p, int find_higher,
563 int return_any);
564int btrfs_realloc_node(struct btrfs_trans_handle *trans,
565 struct btrfs_root *root, struct extent_buffer *parent,
566 int start_slot, u64 *last_ret,
567 struct btrfs_key *progress);
568void btrfs_release_path(struct btrfs_path *p);
569struct btrfs_path *btrfs_alloc_path(void);
570void btrfs_free_path(struct btrfs_path *p);
571
572int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
573 struct btrfs_path *path, int slot, int nr);
574static inline int btrfs_del_item(struct btrfs_trans_handle *trans,
575 struct btrfs_root *root,
576 struct btrfs_path *path)
577{
578 return btrfs_del_items(trans, root, path, path->slots[0], 1);
579}
580
581/*
582 * Describes a batch of items to insert in a btree. This is used by
583 * btrfs_insert_empty_items().
584 */
585struct btrfs_item_batch {
586 /*
587 * Pointer to an array containing the keys of the items to insert (in
588 * sorted order).
589 */
590 const struct btrfs_key *keys;
591 /* Pointer to an array containing the data size for each item to insert. */
592 const u32 *data_sizes;
593 /*
594 * The sum of data sizes for all items. The caller can compute this while
595 * setting up the data_sizes array, so it ends up being more efficient
596 * than having btrfs_insert_empty_items() or setup_item_for_insert()
597 * doing it, as it would avoid an extra loop over a potentially large
598 * array, and in the case of setup_item_for_insert(), we would be doing
599 * it while holding a write lock on a leaf and often on upper level nodes
600 * too, unnecessarily increasing the size of a critical section.
601 */
602 u32 total_data_size;
603 /* Size of the keys and data_sizes arrays (number of items in the batch). */
604 int nr;
605};
606
607void btrfs_setup_item_for_insert(struct btrfs_root *root,
608 struct btrfs_path *path,
609 const struct btrfs_key *key,
610 u32 data_size);
611int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
612 const struct btrfs_key *key, void *data, u32 data_size);
613int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
614 struct btrfs_root *root,
615 struct btrfs_path *path,
616 const struct btrfs_item_batch *batch);
617
618static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
619 struct btrfs_root *root,
620 struct btrfs_path *path,
621 const struct btrfs_key *key,
622 u32 data_size)
623{
624 struct btrfs_item_batch batch;
625
626 batch.keys = key;
627 batch.data_sizes = &data_size;
628 batch.total_data_size = data_size;
629 batch.nr = 1;
630
631 return btrfs_insert_empty_items(trans, root, path, &batch);
632}
633
634int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
635int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
636 u64 time_seq);
637
638int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key,
639 struct btrfs_path *path);
640
641int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key,
642 struct btrfs_path *path);
643
644/*
645 * Search in @root for a given @key, and store the slot found in @found_key.
646 *
647 * @root: The root node of the tree.
648 * @key: The key we are looking for.
649 * @found_key: Will hold the found item.
650 * @path: Holds the current slot/leaf.
651 * @iter_ret: Contains the value returned from btrfs_search_slot or
652 * btrfs_get_next_valid_item, whichever was executed last.
653 *
654 * The @iter_ret is an output variable that will contain the return value of
655 * btrfs_search_slot, if it encountered an error, or the value returned from
656 * btrfs_get_next_valid_item otherwise. That return value can be 0, if a valid
657 * slot was found, 1 if there were no more leaves, and <0 if there was an error.
658 *
659 * It's recommended to use a separate variable for iter_ret and then use it to
660 * set the function return value so there's no confusion of the 0/1/errno
661 * values stemming from btrfs_search_slot.
662 */
663#define btrfs_for_each_slot(root, key, found_key, path, iter_ret) \
664 for (iter_ret = btrfs_search_slot(NULL, (root), (key), (path), 0, 0); \
665 (iter_ret) >= 0 && \
666 (iter_ret = btrfs_get_next_valid_item((root), (found_key), (path))) == 0; \
667 (path)->slots[0]++ \
668 )
669
670int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq);
671
672/*
673 * Search the tree again to find a leaf with greater keys.
674 *
675 * Returns 0 if it found something or 1 if there are no greater leaves.
676 * Returns < 0 on error.
677 */
678static inline int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
679{
680 return btrfs_next_old_leaf(root, path, 0);
681}
682
683static inline int btrfs_next_item(struct btrfs_root *root, struct btrfs_path *p)
684{
685 return btrfs_next_old_item(root, p, 0);
686}
687int btrfs_leaf_free_space(struct extent_buffer *leaf);
688
689static inline int is_fstree(u64 rootid)
690{
691 if (rootid == BTRFS_FS_TREE_OBJECTID ||
692 ((s64)rootid >= (s64)BTRFS_FIRST_FREE_OBJECTID &&
693 !btrfs_qgroup_level(rootid)))
694 return 1;
695 return 0;
696}
697
698static inline bool btrfs_is_data_reloc_root(const struct btrfs_root *root)
699{
700 return root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID;
701}
702
703int btrfs_super_csum_size(const struct btrfs_super_block *s);
704const char *btrfs_super_csum_name(u16 csum_type);
705const char *btrfs_super_csum_driver(u16 csum_type);
706size_t __attribute_const__ btrfs_get_num_csums(void);
707
708/*
709 * We use page status Private2 to indicate there is an ordered extent with
710 * unfinished IO.
711 *
712 * Rename the Private2 accessors to Ordered, to improve readability.
713 */
714#define PageOrdered(page) PagePrivate2(page)
715#define SetPageOrdered(page) SetPagePrivate2(page)
716#define ClearPageOrdered(page) ClearPagePrivate2(page)
717#define folio_test_ordered(folio) folio_test_private_2(folio)
718#define folio_set_ordered(folio) folio_set_private_2(folio)
719#define folio_clear_ordered(folio) folio_clear_private_2(folio)
720
721#endif
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#ifndef BTRFS_CTREE_H
7#define BTRFS_CTREE_H
8
9#include <linux/pagemap.h>
10#include <linux/spinlock.h>
11#include <linux/rbtree.h>
12#include <linux/mutex.h>
13#include <linux/wait.h>
14#include <linux/list.h>
15#include <linux/atomic.h>
16#include <linux/xarray.h>
17#include <linux/refcount.h>
18#include <uapi/linux/btrfs_tree.h>
19#include "locking.h"
20#include "fs.h"
21#include "accessors.h"
22#include "extent-io-tree.h"
23
24struct extent_buffer;
25struct btrfs_block_rsv;
26struct btrfs_trans_handle;
27struct btrfs_block_group;
28
29/* Read ahead values for struct btrfs_path.reada */
30enum {
31 READA_NONE,
32 READA_BACK,
33 READA_FORWARD,
34 /*
35 * Similar to READA_FORWARD but unlike it:
36 *
37 * 1) It will trigger readahead even for leaves that are not close to
38 * each other on disk;
39 * 2) It also triggers readahead for nodes;
40 * 3) During a search, even when a node or leaf is already in memory, it
41 * will still trigger readahead for other nodes and leaves that follow
42 * it.
43 *
44 * This is meant to be used only when we know we are iterating over the
45 * entire tree or a very large part of it.
46 */
47 READA_FORWARD_ALWAYS,
48};
49
50/*
51 * btrfs_paths remember the path taken from the root down to the leaf.
52 * level 0 is always the leaf, and nodes[1...BTRFS_MAX_LEVEL] will point
53 * to any other levels that are present.
54 *
55 * The slots array records the index of the item or block pointer
56 * used while walking the tree.
57 */
58struct btrfs_path {
59 struct extent_buffer *nodes[BTRFS_MAX_LEVEL];
60 int slots[BTRFS_MAX_LEVEL];
61 /* if there is real range locking, this locks field will change */
62 u8 locks[BTRFS_MAX_LEVEL];
63 u8 reada;
64 /* keep some upper locks as we walk down */
65 u8 lowest_level;
66
67 /*
68 * set by btrfs_split_item, tells search_slot to keep all locks
69 * and to force calls to keep space in the nodes
70 */
71 unsigned int search_for_split:1;
72 unsigned int keep_locks:1;
73 unsigned int skip_locking:1;
74 unsigned int search_commit_root:1;
75 unsigned int need_commit_sem:1;
76 unsigned int skip_release_on_error:1;
77 /*
78 * Indicate that new item (btrfs_search_slot) is extending already
79 * existing item and ins_len contains only the data size and not item
80 * header (ie. sizeof(struct btrfs_item) is not included).
81 */
82 unsigned int search_for_extension:1;
83 /* Stop search if any locks need to be taken (for read) */
84 unsigned int nowait:1;
85};
86
87/*
88 * The state of btrfs root
89 */
90enum {
91 /*
92 * btrfs_record_root_in_trans is a multi-step process, and it can race
93 * with the balancing code. But the race is very small, and only the
94 * first time the root is added to each transaction. So IN_TRANS_SETUP
95 * is used to tell us when more checks are required
96 */
97 BTRFS_ROOT_IN_TRANS_SETUP,
98
99 /*
100 * Set if tree blocks of this root can be shared by other roots.
101 * Only subvolume trees and their reloc trees have this bit set.
102 * Conflicts with TRACK_DIRTY bit.
103 *
104 * This affects two things:
105 *
106 * - How balance works
107 * For shareable roots, we need to use reloc tree and do path
108 * replacement for balance, and need various pre/post hooks for
109 * snapshot creation to handle them.
110 *
111 * While for non-shareable trees, we just simply do a tree search
112 * with COW.
113 *
114 * - How dirty roots are tracked
115 * For shareable roots, btrfs_record_root_in_trans() is needed to
116 * track them, while non-subvolume roots have TRACK_DIRTY bit, they
117 * don't need to set this manually.
118 */
119 BTRFS_ROOT_SHAREABLE,
120 BTRFS_ROOT_TRACK_DIRTY,
121 BTRFS_ROOT_IN_RADIX,
122 BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
123 BTRFS_ROOT_DEFRAG_RUNNING,
124 BTRFS_ROOT_FORCE_COW,
125 BTRFS_ROOT_MULTI_LOG_TASKS,
126 BTRFS_ROOT_DIRTY,
127 BTRFS_ROOT_DELETING,
128
129 /*
130 * Reloc tree is orphan, only kept here for qgroup delayed subtree scan
131 *
132 * Set for the subvolume tree owning the reloc tree.
133 */
134 BTRFS_ROOT_DEAD_RELOC_TREE,
135 /* Mark dead root stored on device whose cleanup needs to be resumed */
136 BTRFS_ROOT_DEAD_TREE,
137 /* The root has a log tree. Used for subvolume roots and the tree root. */
138 BTRFS_ROOT_HAS_LOG_TREE,
139 /* Qgroup flushing is in progress */
140 BTRFS_ROOT_QGROUP_FLUSHING,
141 /* We started the orphan cleanup for this root. */
142 BTRFS_ROOT_ORPHAN_CLEANUP,
143 /* This root has a drop operation that was started previously. */
144 BTRFS_ROOT_UNFINISHED_DROP,
145 /* This reloc root needs to have its buffers lockdep class reset. */
146 BTRFS_ROOT_RESET_LOCKDEP_CLASS,
147};
148
149/*
150 * Record swapped tree blocks of a subvolume tree for delayed subtree trace
151 * code. For detail check comment in fs/btrfs/qgroup.c.
152 */
153struct btrfs_qgroup_swapped_blocks {
154 spinlock_t lock;
155 /* RM_EMPTY_ROOT() of above blocks[] */
156 bool swapped;
157 struct rb_root blocks[BTRFS_MAX_LEVEL];
158};
159
160/*
161 * in ram representation of the tree. extent_root is used for all allocations
162 * and for the extent tree extent_root root.
163 */
164struct btrfs_root {
165 struct rb_node rb_node;
166
167 struct extent_buffer *node;
168
169 struct extent_buffer *commit_root;
170 struct btrfs_root *log_root;
171 struct btrfs_root *reloc_root;
172
173 unsigned long state;
174 struct btrfs_root_item root_item;
175 struct btrfs_key root_key;
176 struct btrfs_fs_info *fs_info;
177 struct extent_io_tree dirty_log_pages;
178
179 struct mutex objectid_mutex;
180
181 spinlock_t accounting_lock;
182 struct btrfs_block_rsv *block_rsv;
183
184 struct mutex log_mutex;
185 wait_queue_head_t log_writer_wait;
186 wait_queue_head_t log_commit_wait[2];
187 struct list_head log_ctxs[2];
188 /* Used only for log trees of subvolumes, not for the log root tree */
189 atomic_t log_writers;
190 atomic_t log_commit[2];
191 /* Used only for log trees of subvolumes, not for the log root tree */
192 atomic_t log_batch;
193 /*
194 * Protected by the 'log_mutex' lock but can be read without holding
195 * that lock to avoid unnecessary lock contention, in which case it
196 * should be read using btrfs_get_root_log_transid() except if it's a
197 * log tree in which case it can be directly accessed. Updates to this
198 * field should always use btrfs_set_root_log_transid(), except for log
199 * trees where the field can be updated directly.
200 */
201 int log_transid;
202 /* No matter the commit succeeds or not*/
203 int log_transid_committed;
204 /*
205 * Just be updated when the commit succeeds. Use
206 * btrfs_get_root_last_log_commit() and btrfs_set_root_last_log_commit()
207 * to access this field.
208 */
209 int last_log_commit;
210 pid_t log_start_pid;
211
212 u64 last_trans;
213
214 u64 free_objectid;
215
216 struct btrfs_key defrag_progress;
217 struct btrfs_key defrag_max;
218
219 /* The dirty list is only used by non-shareable roots */
220 struct list_head dirty_list;
221
222 struct list_head root_list;
223
224 spinlock_t inode_lock;
225 /* red-black tree that keeps track of in-memory inodes */
226 struct rb_root inode_tree;
227
228 /*
229 * Xarray that keeps track of delayed nodes of every inode, protected
230 * by @inode_lock.
231 */
232 struct xarray delayed_nodes;
233 /*
234 * right now this just gets used so that a root has its own devid
235 * for stat. It may be used for more later
236 */
237 dev_t anon_dev;
238
239 spinlock_t root_item_lock;
240 refcount_t refs;
241
242 struct mutex delalloc_mutex;
243 spinlock_t delalloc_lock;
244 /*
245 * all of the inodes that have delalloc bytes. It is possible for
246 * this list to be empty even when there is still dirty data=ordered
247 * extents waiting to finish IO.
248 */
249 struct list_head delalloc_inodes;
250 struct list_head delalloc_root;
251 u64 nr_delalloc_inodes;
252
253 struct mutex ordered_extent_mutex;
254 /*
255 * this is used by the balancing code to wait for all the pending
256 * ordered extents
257 */
258 spinlock_t ordered_extent_lock;
259
260 /*
261 * all of the data=ordered extents pending writeback
262 * these can span multiple transactions and basically include
263 * every dirty data page that isn't from nodatacow
264 */
265 struct list_head ordered_extents;
266 struct list_head ordered_root;
267 u64 nr_ordered_extents;
268
269 /*
270 * Not empty if this subvolume root has gone through tree block swap
271 * (relocation)
272 *
273 * Will be used by reloc_control::dirty_subvol_roots.
274 */
275 struct list_head reloc_dirty_list;
276
277 /*
278 * Number of currently running SEND ioctls to prevent
279 * manipulation with the read-only status via SUBVOL_SETFLAGS
280 */
281 int send_in_progress;
282 /*
283 * Number of currently running deduplication operations that have a
284 * destination inode belonging to this root. Protected by the lock
285 * root_item_lock.
286 */
287 int dedupe_in_progress;
288 /* For exclusion of snapshot creation and nocow writes */
289 struct btrfs_drew_lock snapshot_lock;
290
291 atomic_t snapshot_force_cow;
292
293 /* For qgroup metadata reserved space */
294 spinlock_t qgroup_meta_rsv_lock;
295 u64 qgroup_meta_rsv_pertrans;
296 u64 qgroup_meta_rsv_prealloc;
297 wait_queue_head_t qgroup_flush_wait;
298
299 /* Number of active swapfiles */
300 atomic_t nr_swapfiles;
301
302 /* Record pairs of swapped blocks for qgroup */
303 struct btrfs_qgroup_swapped_blocks swapped_blocks;
304
305 /* Used only by log trees, when logging csum items */
306 struct extent_io_tree log_csum_range;
307
308 /* Used in simple quotas, track root during relocation. */
309 u64 relocation_src_root;
310
311#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
312 u64 alloc_bytenr;
313#endif
314
315#ifdef CONFIG_BTRFS_DEBUG
316 struct list_head leak_list;
317#endif
318};
319
320static inline bool btrfs_root_readonly(const struct btrfs_root *root)
321{
322 /* Byte-swap the constant at compile time, root_item::flags is LE */
323 return (root->root_item.flags & cpu_to_le64(BTRFS_ROOT_SUBVOL_RDONLY)) != 0;
324}
325
326static inline bool btrfs_root_dead(const struct btrfs_root *root)
327{
328 /* Byte-swap the constant at compile time, root_item::flags is LE */
329 return (root->root_item.flags & cpu_to_le64(BTRFS_ROOT_SUBVOL_DEAD)) != 0;
330}
331
332static inline u64 btrfs_root_id(const struct btrfs_root *root)
333{
334 return root->root_key.objectid;
335}
336
337static inline int btrfs_get_root_log_transid(const struct btrfs_root *root)
338{
339 return READ_ONCE(root->log_transid);
340}
341
342static inline void btrfs_set_root_log_transid(struct btrfs_root *root, int log_transid)
343{
344 WRITE_ONCE(root->log_transid, log_transid);
345}
346
347static inline int btrfs_get_root_last_log_commit(const struct btrfs_root *root)
348{
349 return READ_ONCE(root->last_log_commit);
350}
351
352static inline void btrfs_set_root_last_log_commit(struct btrfs_root *root, int commit_id)
353{
354 WRITE_ONCE(root->last_log_commit, commit_id);
355}
356
357/*
358 * Structure that conveys information about an extent that is going to replace
359 * all the extents in a file range.
360 */
361struct btrfs_replace_extent_info {
362 u64 disk_offset;
363 u64 disk_len;
364 u64 data_offset;
365 u64 data_len;
366 u64 file_offset;
367 /* Pointer to a file extent item of type regular or prealloc. */
368 char *extent_buf;
369 /*
370 * Set to true when attempting to replace a file range with a new extent
371 * described by this structure, set to false when attempting to clone an
372 * existing extent into a file range.
373 */
374 bool is_new_extent;
375 /* Indicate if we should update the inode's mtime and ctime. */
376 bool update_times;
377 /* Meaningful only if is_new_extent is true. */
378 int qgroup_reserved;
379 /*
380 * Meaningful only if is_new_extent is true.
381 * Used to track how many extent items we have already inserted in a
382 * subvolume tree that refer to the extent described by this structure,
383 * so that we know when to create a new delayed ref or update an existing
384 * one.
385 */
386 int insertions;
387};
388
389/* Arguments for btrfs_drop_extents() */
390struct btrfs_drop_extents_args {
391 /* Input parameters */
392
393 /*
394 * If NULL, btrfs_drop_extents() will allocate and free its own path.
395 * If 'replace_extent' is true, this must not be NULL. Also the path
396 * is always released except if 'replace_extent' is true and
397 * btrfs_drop_extents() sets 'extent_inserted' to true, in which case
398 * the path is kept locked.
399 */
400 struct btrfs_path *path;
401 /* Start offset of the range to drop extents from */
402 u64 start;
403 /* End (exclusive, last byte + 1) of the range to drop extents from */
404 u64 end;
405 /* If true drop all the extent maps in the range */
406 bool drop_cache;
407 /*
408 * If true it means we want to insert a new extent after dropping all
409 * the extents in the range. If this is true, the 'extent_item_size'
410 * parameter must be set as well and the 'extent_inserted' field will
411 * be set to true by btrfs_drop_extents() if it could insert the new
412 * extent.
413 * Note: when this is set to true the path must not be NULL.
414 */
415 bool replace_extent;
416 /*
417 * Used if 'replace_extent' is true. Size of the file extent item to
418 * insert after dropping all existing extents in the range
419 */
420 u32 extent_item_size;
421
422 /* Output parameters */
423
424 /*
425 * Set to the minimum between the input parameter 'end' and the end
426 * (exclusive, last byte + 1) of the last dropped extent. This is always
427 * set even if btrfs_drop_extents() returns an error.
428 */
429 u64 drop_end;
430 /*
431 * The number of allocated bytes found in the range. This can be smaller
432 * than the range's length when there are holes in the range.
433 */
434 u64 bytes_found;
435 /*
436 * Only set if 'replace_extent' is true. Set to true if we were able
437 * to insert a replacement extent after dropping all extents in the
438 * range, otherwise set to false by btrfs_drop_extents().
439 * Also, if btrfs_drop_extents() has set this to true it means it
440 * returned with the path locked, otherwise if it has set this to
441 * false it has returned with the path released.
442 */
443 bool extent_inserted;
444};
445
446struct btrfs_file_private {
447 void *filldir_buf;
448 u64 last_index;
449 struct extent_state *llseek_cached_state;
450};
451
452static inline u32 BTRFS_LEAF_DATA_SIZE(const struct btrfs_fs_info *info)
453{
454 return info->nodesize - sizeof(struct btrfs_header);
455}
456
457static inline u32 BTRFS_MAX_ITEM_SIZE(const struct btrfs_fs_info *info)
458{
459 return BTRFS_LEAF_DATA_SIZE(info) - sizeof(struct btrfs_item);
460}
461
462static inline u32 BTRFS_NODEPTRS_PER_BLOCK(const struct btrfs_fs_info *info)
463{
464 return BTRFS_LEAF_DATA_SIZE(info) / sizeof(struct btrfs_key_ptr);
465}
466
467static inline u32 BTRFS_MAX_XATTR_SIZE(const struct btrfs_fs_info *info)
468{
469 return BTRFS_MAX_ITEM_SIZE(info) - sizeof(struct btrfs_dir_item);
470}
471
472#define BTRFS_BYTES_TO_BLKS(fs_info, bytes) \
473 ((bytes) >> (fs_info)->sectorsize_bits)
474
475static inline gfp_t btrfs_alloc_write_mask(struct address_space *mapping)
476{
477 return mapping_gfp_constraint(mapping, ~__GFP_FS);
478}
479
480void btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, u64 start, u64 end);
481int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
482 u64 num_bytes, u64 *actual_bytes);
483int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range);
484
485/* ctree.c */
486int __init btrfs_ctree_init(void);
487void __cold btrfs_ctree_exit(void);
488
489int btrfs_bin_search(struct extent_buffer *eb, int first_slot,
490 const struct btrfs_key *key, int *slot);
491
492int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2);
493
494#ifdef __LITTLE_ENDIAN
495
496/*
497 * Compare two keys, on little-endian the disk order is same as CPU order and
498 * we can avoid the conversion.
499 */
500static inline int btrfs_comp_keys(const struct btrfs_disk_key *disk_key,
501 const struct btrfs_key *k2)
502{
503 const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key;
504
505 return btrfs_comp_cpu_keys(k1, k2);
506}
507
508#else
509
510/* Compare two keys in a memcmp fashion. */
511static inline int btrfs_comp_keys(const struct btrfs_disk_key *disk,
512 const struct btrfs_key *k2)
513{
514 struct btrfs_key k1;
515
516 btrfs_disk_key_to_cpu(&k1, disk);
517
518 return btrfs_comp_cpu_keys(&k1, k2);
519}
520
521#endif
522
523int btrfs_previous_item(struct btrfs_root *root,
524 struct btrfs_path *path, u64 min_objectid,
525 int type);
526int btrfs_previous_extent_item(struct btrfs_root *root,
527 struct btrfs_path *path, u64 min_objectid);
528void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
529 struct btrfs_path *path,
530 const struct btrfs_key *new_key);
531struct extent_buffer *btrfs_root_node(struct btrfs_root *root);
532int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
533 struct btrfs_key *key, int lowest_level,
534 u64 min_trans);
535int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
536 struct btrfs_path *path,
537 u64 min_trans);
538struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent,
539 int slot);
540
541int btrfs_cow_block(struct btrfs_trans_handle *trans,
542 struct btrfs_root *root, struct extent_buffer *buf,
543 struct extent_buffer *parent, int parent_slot,
544 struct extent_buffer **cow_ret,
545 enum btrfs_lock_nesting nest);
546int btrfs_force_cow_block(struct btrfs_trans_handle *trans,
547 struct btrfs_root *root,
548 struct extent_buffer *buf,
549 struct extent_buffer *parent, int parent_slot,
550 struct extent_buffer **cow_ret,
551 u64 search_start, u64 empty_size,
552 enum btrfs_lock_nesting nest);
553int btrfs_copy_root(struct btrfs_trans_handle *trans,
554 struct btrfs_root *root,
555 struct extent_buffer *buf,
556 struct extent_buffer **cow_ret, u64 new_root_objectid);
557bool btrfs_block_can_be_shared(struct btrfs_trans_handle *trans,
558 struct btrfs_root *root,
559 struct extent_buffer *buf);
560int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
561 struct btrfs_path *path, int level, int slot);
562void btrfs_extend_item(struct btrfs_trans_handle *trans,
563 struct btrfs_path *path, u32 data_size);
564void btrfs_truncate_item(struct btrfs_trans_handle *trans,
565 struct btrfs_path *path, u32 new_size, int from_end);
566int btrfs_split_item(struct btrfs_trans_handle *trans,
567 struct btrfs_root *root,
568 struct btrfs_path *path,
569 const struct btrfs_key *new_key,
570 unsigned long split_offset);
571int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
572 struct btrfs_root *root,
573 struct btrfs_path *path,
574 const struct btrfs_key *new_key);
575int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path,
576 u64 inum, u64 ioff, u8 key_type, struct btrfs_key *found_key);
577int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root,
578 const struct btrfs_key *key, struct btrfs_path *p,
579 int ins_len, int cow);
580int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key,
581 struct btrfs_path *p, u64 time_seq);
582int btrfs_search_slot_for_read(struct btrfs_root *root,
583 const struct btrfs_key *key,
584 struct btrfs_path *p, int find_higher,
585 int return_any);
586void btrfs_release_path(struct btrfs_path *p);
587struct btrfs_path *btrfs_alloc_path(void);
588void btrfs_free_path(struct btrfs_path *p);
589
590int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
591 struct btrfs_path *path, int slot, int nr);
592static inline int btrfs_del_item(struct btrfs_trans_handle *trans,
593 struct btrfs_root *root,
594 struct btrfs_path *path)
595{
596 return btrfs_del_items(trans, root, path, path->slots[0], 1);
597}
598
599/*
600 * Describes a batch of items to insert in a btree. This is used by
601 * btrfs_insert_empty_items().
602 */
603struct btrfs_item_batch {
604 /*
605 * Pointer to an array containing the keys of the items to insert (in
606 * sorted order).
607 */
608 const struct btrfs_key *keys;
609 /* Pointer to an array containing the data size for each item to insert. */
610 const u32 *data_sizes;
611 /*
612 * The sum of data sizes for all items. The caller can compute this while
613 * setting up the data_sizes array, so it ends up being more efficient
614 * than having btrfs_insert_empty_items() or setup_item_for_insert()
615 * doing it, as it would avoid an extra loop over a potentially large
616 * array, and in the case of setup_item_for_insert(), we would be doing
617 * it while holding a write lock on a leaf and often on upper level nodes
618 * too, unnecessarily increasing the size of a critical section.
619 */
620 u32 total_data_size;
621 /* Size of the keys and data_sizes arrays (number of items in the batch). */
622 int nr;
623};
624
625void btrfs_setup_item_for_insert(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root,
627 struct btrfs_path *path,
628 const struct btrfs_key *key,
629 u32 data_size);
630int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
631 const struct btrfs_key *key, void *data, u32 data_size);
632int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
633 struct btrfs_root *root,
634 struct btrfs_path *path,
635 const struct btrfs_item_batch *batch);
636
637static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root,
639 struct btrfs_path *path,
640 const struct btrfs_key *key,
641 u32 data_size)
642{
643 struct btrfs_item_batch batch;
644
645 batch.keys = key;
646 batch.data_sizes = &data_size;
647 batch.total_data_size = data_size;
648 batch.nr = 1;
649
650 return btrfs_insert_empty_items(trans, root, path, &batch);
651}
652
653int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
654 u64 time_seq);
655
656int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key,
657 struct btrfs_path *path);
658
659int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key,
660 struct btrfs_path *path);
661
662/*
663 * Search in @root for a given @key, and store the slot found in @found_key.
664 *
665 * @root: The root node of the tree.
666 * @key: The key we are looking for.
667 * @found_key: Will hold the found item.
668 * @path: Holds the current slot/leaf.
669 * @iter_ret: Contains the value returned from btrfs_search_slot or
670 * btrfs_get_next_valid_item, whichever was executed last.
671 *
672 * The @iter_ret is an output variable that will contain the return value of
673 * btrfs_search_slot, if it encountered an error, or the value returned from
674 * btrfs_get_next_valid_item otherwise. That return value can be 0, if a valid
675 * slot was found, 1 if there were no more leaves, and <0 if there was an error.
676 *
677 * It's recommended to use a separate variable for iter_ret and then use it to
678 * set the function return value so there's no confusion of the 0/1/errno
679 * values stemming from btrfs_search_slot.
680 */
681#define btrfs_for_each_slot(root, key, found_key, path, iter_ret) \
682 for (iter_ret = btrfs_search_slot(NULL, (root), (key), (path), 0, 0); \
683 (iter_ret) >= 0 && \
684 (iter_ret = btrfs_get_next_valid_item((root), (found_key), (path))) == 0; \
685 (path)->slots[0]++ \
686 )
687
688int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq);
689
690/*
691 * Search the tree again to find a leaf with greater keys.
692 *
693 * Returns 0 if it found something or 1 if there are no greater leaves.
694 * Returns < 0 on error.
695 */
696static inline int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
697{
698 return btrfs_next_old_leaf(root, path, 0);
699}
700
701static inline int btrfs_next_item(struct btrfs_root *root, struct btrfs_path *p)
702{
703 return btrfs_next_old_item(root, p, 0);
704}
705int btrfs_leaf_free_space(const struct extent_buffer *leaf);
706
707static inline int is_fstree(u64 rootid)
708{
709 if (rootid == BTRFS_FS_TREE_OBJECTID ||
710 ((s64)rootid >= (s64)BTRFS_FIRST_FREE_OBJECTID &&
711 !btrfs_qgroup_level(rootid)))
712 return 1;
713 return 0;
714}
715
716static inline bool btrfs_is_data_reloc_root(const struct btrfs_root *root)
717{
718 return root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID;
719}
720
721u16 btrfs_csum_type_size(u16 type);
722int btrfs_super_csum_size(const struct btrfs_super_block *s);
723const char *btrfs_super_csum_name(u16 csum_type);
724const char *btrfs_super_csum_driver(u16 csum_type);
725size_t __attribute_const__ btrfs_get_num_csums(void);
726
727/*
728 * We use page status Private2 to indicate there is an ordered extent with
729 * unfinished IO.
730 *
731 * Rename the Private2 accessors to Ordered, to improve readability.
732 */
733#define PageOrdered(page) PagePrivate2(page)
734#define SetPageOrdered(page) SetPagePrivate2(page)
735#define ClearPageOrdered(page) ClearPagePrivate2(page)
736#define folio_test_ordered(folio) folio_test_private_2(folio)
737#define folio_set_ordered(folio) folio_set_private_2(folio)
738#define folio_clear_ordered(folio) folio_clear_private_2(folio)
739
740#endif