]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 | 2 | /* |
d352ac68 | 3 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
4 | */ |
5 | ||
a6b6e75e | 6 | #include <linux/sched.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
bd989ba3 | 8 | #include <linux/rbtree.h> |
adf02123 | 9 | #include <linux/mm.h> |
e41d12f5 | 10 | #include <linux/error-injection.h> |
9b569ea0 | 11 | #include "messages.h" |
eb60ceac CM |
12 | #include "ctree.h" |
13 | #include "disk-io.h" | |
7f5c1516 | 14 | #include "transaction.h" |
5f39d397 | 15 | #include "print-tree.h" |
925baedd | 16 | #include "locking.h" |
de37aa51 | 17 | #include "volumes.h" |
f616f5cd | 18 | #include "qgroup.h" |
f3a84ccd | 19 | #include "tree-mod-log.h" |
88c602ab | 20 | #include "tree-checker.h" |
ec8eb376 | 21 | #include "fs.h" |
ad1ac501 | 22 | #include "accessors.h" |
a0231804 | 23 | #include "extent-tree.h" |
67707479 | 24 | #include "relocation.h" |
6bfd0ffa | 25 | #include "file-item.h" |
9a8dd150 | 26 | |
226463d7 JB |
27 | static struct kmem_cache *btrfs_path_cachep; |
28 | ||
e089f05c CM |
29 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
30 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
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); | |
5f39d397 | 34 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 35 | struct extent_buffer *dst, |
971a1f66 | 36 | struct extent_buffer *src, int empty); |
5f39d397 | 37 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
38 | struct extent_buffer *dst_buf, |
39 | struct extent_buffer *src_buf); | |
d97e63b6 | 40 | |
af024ed2 JT |
41 | static const struct btrfs_csums { |
42 | u16 size; | |
59a0fcdb DS |
43 | const char name[10]; |
44 | const char driver[12]; | |
af024ed2 JT |
45 | } btrfs_csums[] = { |
46 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
3951e7f0 | 47 | [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, |
3831bf00 | 48 | [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, |
352ae07b DS |
49 | [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", |
50 | .driver = "blake2b-256" }, | |
af024ed2 JT |
51 | }; |
52 | ||
3a3178c7 JB |
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 | ||
637e3b48 JB |
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 | { | |
8009adf3 JB |
84 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(leaf, 0) + dst_offset, |
85 | btrfs_item_nr_offset(leaf, 0) + src_offset, len); | |
637e3b48 JB |
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 | { | |
8009adf3 JB |
107 | copy_extent_buffer(dst, src, btrfs_item_nr_offset(dst, 0) + dst_offset, |
108 | btrfs_item_nr_offset(src, 0) + src_offset, len); | |
637e3b48 JB |
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 | ||
b3cbfb0d JB |
151 | /* This exists for btrfs-progs usages. */ |
152 | u16 btrfs_csum_type_size(u16 type) | |
153 | { | |
154 | return btrfs_csums[type].size; | |
155 | } | |
156 | ||
af024ed2 JT |
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 | */ | |
b3cbfb0d | 163 | return btrfs_csum_type_size(t); |
af024ed2 JT |
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 | ||
b4e967be DS |
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 */ | |
59a0fcdb DS |
179 | return btrfs_csums[csum_type].driver[0] ? |
180 | btrfs_csums[csum_type].driver : | |
b4e967be DS |
181 | btrfs_csums[csum_type].name; |
182 | } | |
183 | ||
604997b4 | 184 | size_t __attribute_const__ btrfs_get_num_csums(void) |
f7cea56c DS |
185 | { |
186 | return ARRAY_SIZE(btrfs_csums); | |
187 | } | |
188 | ||
df24a2b9 | 189 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 190 | { |
a4c853af C |
191 | might_sleep(); |
192 | ||
e2c89907 | 193 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
194 | } |
195 | ||
d352ac68 | 196 | /* this also releases the path */ |
df24a2b9 | 197 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 198 | { |
ff175d57 JJ |
199 | if (!p) |
200 | return; | |
b3b4aa74 | 201 | btrfs_release_path(p); |
df24a2b9 | 202 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
203 | } |
204 | ||
d352ac68 CM |
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 | */ | |
b3b4aa74 | 211 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
212 | { |
213 | int i; | |
a2135011 | 214 | |
234b63a0 | 215 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 216 | p->slots[i] = 0; |
eb60ceac | 217 | if (!p->nodes[i]) |
925baedd CM |
218 | continue; |
219 | if (p->locks[i]) { | |
bd681513 | 220 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
221 | p->locks[i] = 0; |
222 | } | |
5f39d397 | 223 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 224 | p->nodes[i] = NULL; |
eb60ceac CM |
225 | } |
226 | } | |
227 | ||
8bb808c6 DS |
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 | */ | |
ed164802 | 233 | bool __cold abort_should_print_stack(int error) |
8bb808c6 | 234 | { |
ed164802 | 235 | switch (error) { |
8bb808c6 DS |
236 | case -EIO: |
237 | case -EROFS: | |
238 | case -ENOMEM: | |
239 | return false; | |
240 | } | |
241 | return true; | |
242 | } | |
243 | ||
d352ac68 CM |
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 | */ | |
925baedd CM |
254 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
255 | { | |
256 | struct extent_buffer *eb; | |
240f62c8 | 257 | |
3083ee2e JB |
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 | |
01327610 | 264 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
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 | } | |
925baedd CM |
275 | return eb; |
276 | } | |
277 | ||
92a7cc42 QW |
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. | |
d352ac68 | 282 | */ |
0b86a832 CM |
283 | static void add_root_to_dirty_list(struct btrfs_root *root) |
284 | { | |
0b246afa JM |
285 | struct btrfs_fs_info *fs_info = root->fs_info; |
286 | ||
e7070be1 JB |
287 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
288 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
289 | return; | |
290 | ||
0b246afa | 291 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
292 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
293 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 294 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 295 | list_move_tail(&root->dirty_list, |
0b246afa | 296 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
297 | else |
298 | list_move(&root->dirty_list, | |
0b246afa | 299 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 300 | } |
0b246afa | 301 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
302 | } |
303 | ||
d352ac68 CM |
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 | */ | |
be20aa9d CM |
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 | { | |
0b246afa | 314 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 315 | struct extent_buffer *cow; |
be20aa9d CM |
316 | int ret = 0; |
317 | int level; | |
5d4f98a2 | 318 | struct btrfs_disk_key disk_key; |
60ea105a | 319 | u64 reloc_src_root = 0; |
be20aa9d | 320 | |
92a7cc42 | 321 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 322 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 323 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 324 | trans->transid != root->last_trans); |
be20aa9d CM |
325 | |
326 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
327 | if (level == 0) |
328 | btrfs_item_key(buf, &disk_key, 0); | |
329 | else | |
330 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 331 | |
60ea105a BB |
332 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
333 | reloc_src_root = btrfs_header_owner(buf); | |
4d75f8a9 | 334 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
cf6f34aa | 335 | &disk_key, level, buf->start, 0, |
60ea105a | 336 | reloc_src_root, BTRFS_NESTING_NEW_ROOT); |
5d4f98a2 | 337 | if (IS_ERR(cow)) |
be20aa9d CM |
338 | return PTR_ERR(cow); |
339 | ||
58e8012c | 340 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
341 | btrfs_set_header_bytenr(cow, cow->start); |
342 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
343 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
344 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
345 | BTRFS_HEADER_FLAG_RELOC); | |
346 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
347 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
348 | else | |
349 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 350 | |
de37aa51 | 351 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 352 | |
be20aa9d | 353 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 354 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 355 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 356 | else |
e339a6b0 | 357 | ret = btrfs_inc_ref(trans, root, cow, 0); |
867ed321 | 358 | if (ret) { |
72c9925f FM |
359 | btrfs_tree_unlock(cow); |
360 | free_extent_buffer(cow); | |
867ed321 | 361 | btrfs_abort_transaction(trans, ret); |
be20aa9d | 362 | return ret; |
867ed321 | 363 | } |
be20aa9d | 364 | |
50564b65 | 365 | btrfs_mark_buffer_dirty(trans, cow); |
be20aa9d CM |
366 | *cow_ret = cow; |
367 | return 0; | |
368 | } | |
369 | ||
5d4f98a2 YZ |
370 | /* |
371 | * check if the tree block can be shared by multiple trees | |
372 | */ | |
373 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
374 | struct extent_buffer *buf) | |
375 | { | |
376 | /* | |
92a7cc42 QW |
377 | * Tree blocks not in shareable trees and tree roots are never shared. |
378 | * If a block was allocated after the last snapshot and the block was | |
379 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 380 | */ |
92a7cc42 | 381 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
382 | buf != root->node && buf != root->commit_root && |
383 | (btrfs_header_generation(buf) <= | |
384 | btrfs_root_last_snapshot(&root->root_item) || | |
385 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
386 | return 1; | |
a79865c6 | 387 | |
5d4f98a2 YZ |
388 | return 0; |
389 | } | |
390 | ||
391 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
392 | struct btrfs_root *root, | |
393 | struct extent_buffer *buf, | |
f0486c68 YZ |
394 | struct extent_buffer *cow, |
395 | int *last_ref) | |
5d4f98a2 | 396 | { |
0b246afa | 397 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
398 | u64 refs; |
399 | u64 owner; | |
400 | u64 flags; | |
401 | u64 new_flags = 0; | |
402 | int ret; | |
403 | ||
404 | /* | |
405 | * Backrefs update rules: | |
406 | * | |
407 | * Always use full backrefs for extent pointers in tree block | |
408 | * allocated by tree relocation. | |
409 | * | |
410 | * If a shared tree block is no longer referenced by its owner | |
411 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
412 | * use full backrefs for extent pointers in tree block. | |
413 | * | |
414 | * If a tree block is been relocating | |
415 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
416 | * use full backrefs for extent pointers in tree block. | |
417 | * The reason for this is some operations (such as drop tree) | |
418 | * are only allowed for blocks use full backrefs. | |
419 | */ | |
420 | ||
421 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 422 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
423 | btrfs_header_level(buf), 1, |
424 | &refs, &flags); | |
be1a5564 MF |
425 | if (ret) |
426 | return ret; | |
eced687e FM |
427 | if (unlikely(refs == 0)) { |
428 | btrfs_crit(fs_info, | |
429 | "found 0 references for tree block at bytenr %llu level %d root %llu", | |
430 | buf->start, btrfs_header_level(buf), | |
431 | btrfs_root_id(root)); | |
432 | ret = -EUCLEAN; | |
433 | btrfs_abort_transaction(trans, ret); | |
e5df9573 MF |
434 | return ret; |
435 | } | |
5d4f98a2 YZ |
436 | } else { |
437 | refs = 1; | |
438 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
439 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
440 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
441 | else | |
442 | flags = 0; | |
443 | } | |
444 | ||
445 | owner = btrfs_header_owner(buf); | |
446 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
447 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
448 | ||
449 | if (refs > 1) { | |
450 | if ((owner == root->root_key.objectid || | |
451 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
452 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 453 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
454 | if (ret) |
455 | return ret; | |
5d4f98a2 YZ |
456 | |
457 | if (root->root_key.objectid == | |
458 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 459 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
460 | if (ret) |
461 | return ret; | |
e339a6b0 | 462 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
463 | if (ret) |
464 | return ret; | |
5d4f98a2 YZ |
465 | } |
466 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
467 | } else { | |
468 | ||
469 | if (root->root_key.objectid == | |
470 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 471 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 472 | else |
e339a6b0 | 473 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
474 | if (ret) |
475 | return ret; | |
5d4f98a2 YZ |
476 | } |
477 | if (new_flags != 0) { | |
4aec05fa | 478 | ret = btrfs_set_disk_extent_flags(trans, buf, new_flags); |
be1a5564 MF |
479 | if (ret) |
480 | return ret; | |
5d4f98a2 YZ |
481 | } |
482 | } else { | |
483 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
484 | if (root->root_key.objectid == | |
485 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 486 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 487 | else |
e339a6b0 | 488 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
489 | if (ret) |
490 | return ret; | |
e339a6b0 | 491 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
492 | if (ret) |
493 | return ret; | |
5d4f98a2 | 494 | } |
190a8339 | 495 | btrfs_clear_buffer_dirty(trans, buf); |
f0486c68 | 496 | *last_ref = 1; |
5d4f98a2 YZ |
497 | } |
498 | return 0; | |
499 | } | |
500 | ||
d352ac68 | 501 | /* |
d397712b CM |
502 | * does the dirty work in cow of a single block. The parent block (if |
503 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
504 | * dirty and returned locked. If you modify the block it needs to be marked | |
505 | * dirty again. | |
d352ac68 CM |
506 | * |
507 | * search_start -- an allocation hint for the new block | |
508 | * | |
d397712b CM |
509 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
510 | * bytes the allocator should try to find free next to the block it returns. | |
511 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 512 | */ |
95f93bc4 FM |
513 | int btrfs_force_cow_block(struct btrfs_trans_handle *trans, |
514 | struct btrfs_root *root, | |
515 | struct extent_buffer *buf, | |
516 | struct extent_buffer *parent, int parent_slot, | |
517 | struct extent_buffer **cow_ret, | |
518 | u64 search_start, u64 empty_size, | |
519 | enum btrfs_lock_nesting nest) | |
02217ed2 | 520 | { |
0b246afa | 521 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 522 | struct btrfs_disk_key disk_key; |
5f39d397 | 523 | struct extent_buffer *cow; |
be1a5564 | 524 | int level, ret; |
f0486c68 | 525 | int last_ref = 0; |
925baedd | 526 | int unlock_orig = 0; |
0f5053eb | 527 | u64 parent_start = 0; |
60ea105a | 528 | u64 reloc_src_root = 0; |
7bb86316 | 529 | |
925baedd CM |
530 | if (*cow_ret == buf) |
531 | unlock_orig = 1; | |
532 | ||
49d0c642 | 533 | btrfs_assert_tree_write_locked(buf); |
925baedd | 534 | |
92a7cc42 | 535 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 536 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 537 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 538 | trans->transid != root->last_trans); |
5f39d397 | 539 | |
7bb86316 | 540 | level = btrfs_header_level(buf); |
31840ae1 | 541 | |
5d4f98a2 YZ |
542 | if (level == 0) |
543 | btrfs_item_key(buf, &disk_key, 0); | |
544 | else | |
545 | btrfs_node_key(buf, &disk_key, 0); | |
546 | ||
60ea105a BB |
547 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { |
548 | if (parent) | |
549 | parent_start = parent->start; | |
550 | reloc_src_root = btrfs_header_owner(buf); | |
551 | } | |
79bd3712 FM |
552 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
553 | root->root_key.objectid, &disk_key, level, | |
60ea105a | 554 | search_start, empty_size, reloc_src_root, nest); |
54aa1f4d CM |
555 | if (IS_ERR(cow)) |
556 | return PTR_ERR(cow); | |
6702ed49 | 557 | |
b4ce94de CM |
558 | /* cow is set to blocking by btrfs_init_new_buffer */ |
559 | ||
58e8012c | 560 | copy_extent_buffer_full(cow, buf); |
db94535d | 561 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 562 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
563 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
564 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
565 | BTRFS_HEADER_FLAG_RELOC); | |
566 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
567 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
568 | else | |
569 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 570 | |
de37aa51 | 571 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 572 | |
be1a5564 | 573 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 574 | if (ret) { |
572c83ac JB |
575 | btrfs_tree_unlock(cow); |
576 | free_extent_buffer(cow); | |
66642832 | 577 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
578 | return ret; |
579 | } | |
1a40e23b | 580 | |
92a7cc42 | 581 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 582 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 583 | if (ret) { |
572c83ac JB |
584 | btrfs_tree_unlock(cow); |
585 | free_extent_buffer(cow); | |
66642832 | 586 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 587 | return ret; |
93314e3b | 588 | } |
83d4cfd4 | 589 | } |
3fd0a558 | 590 | |
02217ed2 | 591 | if (buf == root->node) { |
925baedd | 592 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
593 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
594 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
595 | parent_start = buf->start; | |
925baedd | 596 | |
406808ab | 597 | ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); |
40b0a749 FM |
598 | if (ret < 0) { |
599 | btrfs_tree_unlock(cow); | |
600 | free_extent_buffer(cow); | |
601 | btrfs_abort_transaction(trans, ret); | |
602 | return ret; | |
603 | } | |
604 | atomic_inc(&cow->refs); | |
240f62c8 | 605 | rcu_assign_pointer(root->node, cow); |
925baedd | 606 | |
7a163608 FM |
607 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
608 | parent_start, last_ref); | |
5f39d397 | 609 | free_extent_buffer(buf); |
0b86a832 | 610 | add_root_to_dirty_list(root); |
02217ed2 | 611 | } else { |
5d4f98a2 | 612 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
d09c5152 FM |
613 | ret = btrfs_tree_mod_log_insert_key(parent, parent_slot, |
614 | BTRFS_MOD_LOG_KEY_REPLACE); | |
615 | if (ret) { | |
616 | btrfs_tree_unlock(cow); | |
617 | free_extent_buffer(cow); | |
618 | btrfs_abort_transaction(trans, ret); | |
619 | return ret; | |
620 | } | |
5f39d397 | 621 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 622 | cow->start); |
74493f7a CM |
623 | btrfs_set_node_ptr_generation(parent, parent_slot, |
624 | trans->transid); | |
50564b65 | 625 | btrfs_mark_buffer_dirty(trans, parent); |
5de865ee | 626 | if (last_ref) { |
f3a84ccd | 627 | ret = btrfs_tree_mod_log_free_eb(buf); |
5de865ee | 628 | if (ret) { |
572c83ac JB |
629 | btrfs_tree_unlock(cow); |
630 | free_extent_buffer(cow); | |
66642832 | 631 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
632 | return ret; |
633 | } | |
634 | } | |
7a163608 FM |
635 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
636 | parent_start, last_ref); | |
02217ed2 | 637 | } |
925baedd CM |
638 | if (unlock_orig) |
639 | btrfs_tree_unlock(buf); | |
3083ee2e | 640 | free_extent_buffer_stale(buf); |
50564b65 | 641 | btrfs_mark_buffer_dirty(trans, cow); |
2c90e5d6 | 642 | *cow_ret = cow; |
02217ed2 CM |
643 | return 0; |
644 | } | |
645 | ||
5d4f98a2 YZ |
646 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
647 | struct btrfs_root *root, | |
648 | struct extent_buffer *buf) | |
649 | { | |
f5ee5c9a | 650 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 651 | return 0; |
fccb84c9 | 652 | |
d1980131 DS |
653 | /* Ensure we can see the FORCE_COW bit */ |
654 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
655 | |
656 | /* | |
657 | * We do not need to cow a block if | |
658 | * 1) this block is not created or changed in this transaction; | |
659 | * 2) this block does not belong to TREE_RELOC tree; | |
660 | * 3) the root is not forced COW. | |
661 | * | |
662 | * What is forced COW: | |
01327610 | 663 | * when we create snapshot during committing the transaction, |
52042d8e | 664 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
665 | * block to ensure the metadata consistency. |
666 | */ | |
5d4f98a2 YZ |
667 | if (btrfs_header_generation(buf) == trans->transid && |
668 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
669 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 670 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 671 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
672 | return 0; |
673 | return 1; | |
674 | } | |
675 | ||
d352ac68 | 676 | /* |
95f93bc4 | 677 | * COWs a single block, see btrfs_force_cow_block() for the real work. |
01327610 | 678 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
679 | * once per transaction, as long as it hasn't been written yet |
680 | */ | |
7bff16e3 | 681 | int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
682 | struct btrfs_root *root, struct extent_buffer *buf, |
683 | struct extent_buffer *parent, int parent_slot, | |
9631e4cc JB |
684 | struct extent_buffer **cow_ret, |
685 | enum btrfs_lock_nesting nest) | |
6702ed49 | 686 | { |
0b246afa | 687 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 688 | u64 search_start; |
f510cfec | 689 | int ret; |
dc17ff8f | 690 | |
a2caab29 FM |
691 | if (unlikely(test_bit(BTRFS_ROOT_DELETING, &root->state))) { |
692 | btrfs_abort_transaction(trans, -EUCLEAN); | |
693 | btrfs_crit(fs_info, | |
694 | "attempt to COW block %llu on root %llu that is being deleted", | |
695 | buf->start, btrfs_root_id(root)); | |
696 | return -EUCLEAN; | |
697 | } | |
83354f07 | 698 | |
48774f3b FM |
699 | /* |
700 | * COWing must happen through a running transaction, which always | |
701 | * matches the current fs generation (it's a transaction with a state | |
702 | * less than TRANS_STATE_UNBLOCKED). If it doesn't, then turn the fs | |
703 | * into error state to prevent the commit of any transaction. | |
704 | */ | |
705 | if (unlikely(trans->transaction != fs_info->running_transaction || | |
706 | trans->transid != fs_info->generation)) { | |
707 | btrfs_abort_transaction(trans, -EUCLEAN); | |
708 | btrfs_crit(fs_info, | |
709 | "unexpected transaction when attempting to COW block %llu on root %llu, transaction %llu running transaction %llu fs generation %llu", | |
710 | buf->start, btrfs_root_id(root), trans->transid, | |
711 | fs_info->running_transaction->transid, | |
712 | fs_info->generation); | |
713 | return -EUCLEAN; | |
714 | } | |
dc17ff8f | 715 | |
5d4f98a2 | 716 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
717 | *cow_ret = buf; |
718 | return 0; | |
719 | } | |
c487685d | 720 | |
b8bf4e4d | 721 | search_start = round_down(buf->start, SZ_1G); |
b4ce94de | 722 | |
f616f5cd QW |
723 | /* |
724 | * Before CoWing this block for later modification, check if it's | |
725 | * the subtree root and do the delayed subtree trace if needed. | |
726 | * | |
727 | * Also We don't care about the error, as it's handled internally. | |
728 | */ | |
729 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
95f93bc4 FM |
730 | ret = btrfs_force_cow_block(trans, root, buf, parent, parent_slot, |
731 | cow_ret, search_start, 0, nest); | |
1abe9b8a | 732 | |
733 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
734 | ||
f510cfec | 735 | return ret; |
6702ed49 | 736 | } |
f75e2b79 | 737 | ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); |
6702ed49 | 738 | |
f3465ca4 JB |
739 | /* |
740 | * same as comp_keys only with two btrfs_key's | |
741 | */ | |
e1f60a65 | 742 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
743 | { |
744 | if (k1->objectid > k2->objectid) | |
745 | return 1; | |
746 | if (k1->objectid < k2->objectid) | |
747 | return -1; | |
748 | if (k1->type > k2->type) | |
749 | return 1; | |
750 | if (k1->type < k2->type) | |
751 | return -1; | |
752 | if (k1->offset > k2->offset) | |
753 | return 1; | |
754 | if (k1->offset < k2->offset) | |
755 | return -1; | |
756 | return 0; | |
757 | } | |
081e9573 | 758 | |
74123bd7 | 759 | /* |
fb81212c | 760 | * Search for a key in the given extent_buffer. |
5f39d397 | 761 | * |
a724f313 | 762 | * The lower boundary for the search is specified by the slot number @first_slot. |
fdf8d595 AJ |
763 | * Use a value of 0 to search over the whole extent buffer. Works for both |
764 | * leaves and nodes. | |
74123bd7 | 765 | * |
fb81212c FM |
766 | * The slot in the extent buffer is returned via @slot. If the key exists in the |
767 | * extent buffer, then @slot will point to the slot where the key is, otherwise | |
768 | * it points to the slot where you would insert the key. | |
769 | * | |
770 | * Slot may point to the total number of items (i.e. one position beyond the last | |
771 | * key) if the key is bigger than the last key in the extent buffer. | |
74123bd7 | 772 | */ |
fdf8d595 AJ |
773 | int btrfs_bin_search(struct extent_buffer *eb, int first_slot, |
774 | const struct btrfs_key *key, int *slot) | |
be0e5c09 | 775 | { |
fb81212c FM |
776 | unsigned long p; |
777 | int item_size; | |
a724f313 FM |
778 | /* |
779 | * Use unsigned types for the low and high slots, so that we get a more | |
780 | * efficient division in the search loop below. | |
781 | */ | |
782 | u32 low = first_slot; | |
783 | u32 high = btrfs_header_nritems(eb); | |
be0e5c09 | 784 | int ret; |
5cd17f34 | 785 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 786 | |
a724f313 | 787 | if (unlikely(low > high)) { |
5e24e9af | 788 | btrfs_err(eb->fs_info, |
a724f313 | 789 | "%s: low (%u) > high (%u) eb %llu owner %llu level %d", |
5e24e9af LB |
790 | __func__, low, high, eb->start, |
791 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
792 | return -EINVAL; | |
793 | } | |
794 | ||
fb81212c FM |
795 | if (btrfs_header_level(eb) == 0) { |
796 | p = offsetof(struct btrfs_leaf, items); | |
797 | item_size = sizeof(struct btrfs_item); | |
798 | } else { | |
799 | p = offsetof(struct btrfs_node, ptrs); | |
800 | item_size = sizeof(struct btrfs_key_ptr); | |
801 | } | |
802 | ||
d397712b | 803 | while (low < high) { |
5cd17f34 DS |
804 | unsigned long oip; |
805 | unsigned long offset; | |
806 | struct btrfs_disk_key *tmp; | |
807 | struct btrfs_disk_key unaligned; | |
808 | int mid; | |
809 | ||
be0e5c09 | 810 | mid = (low + high) / 2; |
5f39d397 | 811 | offset = p + mid * item_size; |
5cd17f34 | 812 | oip = offset_in_page(offset); |
5f39d397 | 813 | |
5cd17f34 | 814 | if (oip + key_size <= PAGE_SIZE) { |
884b07d0 | 815 | const unsigned long idx = get_eb_page_index(offset); |
5cd17f34 | 816 | char *kaddr = page_address(eb->pages[idx]); |
5f39d397 | 817 | |
884b07d0 | 818 | oip = get_eb_offset_in_page(eb, offset); |
5cd17f34 | 819 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 820 | } else { |
5cd17f34 DS |
821 | read_extent_buffer(eb, &unaligned, offset, key_size); |
822 | tmp = &unaligned; | |
5f39d397 | 823 | } |
5cd17f34 | 824 | |
79d25df0 | 825 | ret = btrfs_comp_keys(tmp, key); |
be0e5c09 CM |
826 | |
827 | if (ret < 0) | |
828 | low = mid + 1; | |
829 | else if (ret > 0) | |
830 | high = mid; | |
831 | else { | |
832 | *slot = mid; | |
833 | return 0; | |
834 | } | |
835 | } | |
836 | *slot = low; | |
837 | return 1; | |
838 | } | |
839 | ||
02cd00fa | 840 | static void root_add_used_bytes(struct btrfs_root *root) |
f0486c68 YZ |
841 | { |
842 | spin_lock(&root->accounting_lock); | |
843 | btrfs_set_root_used(&root->root_item, | |
02cd00fa | 844 | btrfs_root_used(&root->root_item) + root->fs_info->nodesize); |
f0486c68 YZ |
845 | spin_unlock(&root->accounting_lock); |
846 | } | |
847 | ||
02cd00fa | 848 | static void root_sub_used_bytes(struct btrfs_root *root) |
f0486c68 YZ |
849 | { |
850 | spin_lock(&root->accounting_lock); | |
851 | btrfs_set_root_used(&root->root_item, | |
02cd00fa | 852 | btrfs_root_used(&root->root_item) - root->fs_info->nodesize); |
f0486c68 YZ |
853 | spin_unlock(&root->accounting_lock); |
854 | } | |
855 | ||
d352ac68 CM |
856 | /* given a node and slot number, this reads the blocks it points to. The |
857 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 858 | */ |
4b231ae4 DS |
859 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
860 | int slot) | |
bb803951 | 861 | { |
ca7a79ad | 862 | int level = btrfs_header_level(parent); |
789d6a3a | 863 | struct btrfs_tree_parent_check check = { 0 }; |
416bc658 JB |
864 | struct extent_buffer *eb; |
865 | ||
fb770ae4 LB |
866 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
867 | return ERR_PTR(-ENOENT); | |
ca7a79ad | 868 | |
d4694728 | 869 | ASSERT(level); |
ca7a79ad | 870 | |
789d6a3a QW |
871 | check.level = level - 1; |
872 | check.transid = btrfs_node_ptr_generation(parent, slot); | |
873 | check.owner_root = btrfs_header_owner(parent); | |
874 | check.has_first_key = true; | |
875 | btrfs_node_key_to_cpu(parent, &check.first_key, slot); | |
876 | ||
d0d20b0f | 877 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
789d6a3a | 878 | &check); |
4eb150d6 QW |
879 | if (IS_ERR(eb)) |
880 | return eb; | |
881 | if (!extent_buffer_uptodate(eb)) { | |
fb770ae4 | 882 | free_extent_buffer(eb); |
4eb150d6 | 883 | return ERR_PTR(-EIO); |
416bc658 JB |
884 | } |
885 | ||
886 | return eb; | |
bb803951 CM |
887 | } |
888 | ||
d352ac68 CM |
889 | /* |
890 | * node level balancing, used to make sure nodes are in proper order for | |
891 | * item deletion. We balance from the top down, so we have to make sure | |
892 | * that a deletion won't leave an node completely empty later on. | |
893 | */ | |
e02119d5 | 894 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
895 | struct btrfs_root *root, |
896 | struct btrfs_path *path, int level) | |
bb803951 | 897 | { |
0b246afa | 898 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
899 | struct extent_buffer *right = NULL; |
900 | struct extent_buffer *mid; | |
901 | struct extent_buffer *left = NULL; | |
902 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
903 | int ret = 0; |
904 | int wret; | |
905 | int pslot; | |
bb803951 | 906 | int orig_slot = path->slots[level]; |
79f95c82 | 907 | u64 orig_ptr; |
bb803951 | 908 | |
98e6b1eb | 909 | ASSERT(level > 0); |
bb803951 | 910 | |
5f39d397 | 911 | mid = path->nodes[level]; |
b4ce94de | 912 | |
ac5887c8 | 913 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); |
7bb86316 CM |
914 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
915 | ||
1d4f8a0c | 916 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 917 | |
a05a9bb1 | 918 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 919 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
920 | pslot = path->slots[level + 1]; |
921 | } | |
bb803951 | 922 | |
40689478 CM |
923 | /* |
924 | * deal with the case where there is only one pointer in the root | |
925 | * by promoting the node below to a root | |
926 | */ | |
5f39d397 CM |
927 | if (!parent) { |
928 | struct extent_buffer *child; | |
bb803951 | 929 | |
5f39d397 | 930 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
931 | return 0; |
932 | ||
933 | /* promote the child to a root */ | |
4b231ae4 | 934 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
935 | if (IS_ERR(child)) { |
936 | ret = PTR_ERR(child); | |
daefe4d4 | 937 | goto out; |
305a26af MF |
938 | } |
939 | ||
925baedd | 940 | btrfs_tree_lock(child); |
9631e4cc JB |
941 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, |
942 | BTRFS_NESTING_COW); | |
f0486c68 YZ |
943 | if (ret) { |
944 | btrfs_tree_unlock(child); | |
945 | free_extent_buffer(child); | |
daefe4d4 | 946 | goto out; |
f0486c68 | 947 | } |
2f375ab9 | 948 | |
406808ab | 949 | ret = btrfs_tree_mod_log_insert_root(root->node, child, true); |
39020d8a FM |
950 | if (ret < 0) { |
951 | btrfs_tree_unlock(child); | |
952 | free_extent_buffer(child); | |
953 | btrfs_abort_transaction(trans, ret); | |
daefe4d4 | 954 | goto out; |
39020d8a | 955 | } |
240f62c8 | 956 | rcu_assign_pointer(root->node, child); |
925baedd | 957 | |
0b86a832 | 958 | add_root_to_dirty_list(root); |
925baedd | 959 | btrfs_tree_unlock(child); |
b4ce94de | 960 | |
925baedd | 961 | path->locks[level] = 0; |
bb803951 | 962 | path->nodes[level] = NULL; |
190a8339 | 963 | btrfs_clear_buffer_dirty(trans, mid); |
925baedd | 964 | btrfs_tree_unlock(mid); |
bb803951 | 965 | /* once for the path */ |
5f39d397 | 966 | free_extent_buffer(mid); |
f0486c68 | 967 | |
02cd00fa | 968 | root_sub_used_bytes(root); |
7a163608 | 969 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
bb803951 | 970 | /* once for the root ptr */ |
3083ee2e | 971 | free_extent_buffer_stale(mid); |
f0486c68 | 972 | return 0; |
bb803951 | 973 | } |
5f39d397 | 974 | if (btrfs_header_nritems(mid) > |
0b246afa | 975 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
976 | return 0; |
977 | ||
9cf14029 JB |
978 | if (pslot) { |
979 | left = btrfs_read_node_slot(parent, pslot - 1); | |
980 | if (IS_ERR(left)) { | |
981 | ret = PTR_ERR(left); | |
982 | left = NULL; | |
daefe4d4 | 983 | goto out; |
9cf14029 | 984 | } |
fb770ae4 | 985 | |
bf77467a | 986 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
5f39d397 | 987 | wret = btrfs_cow_block(trans, root, left, |
9631e4cc | 988 | parent, pslot - 1, &left, |
bf59a5a2 | 989 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
990 | if (wret) { |
991 | ret = wret; | |
daefe4d4 | 992 | goto out; |
54aa1f4d | 993 | } |
2cc58cf2 | 994 | } |
fb770ae4 | 995 | |
9cf14029 JB |
996 | if (pslot + 1 < btrfs_header_nritems(parent)) { |
997 | right = btrfs_read_node_slot(parent, pslot + 1); | |
998 | if (IS_ERR(right)) { | |
999 | ret = PTR_ERR(right); | |
1000 | right = NULL; | |
daefe4d4 | 1001 | goto out; |
9cf14029 | 1002 | } |
fb770ae4 | 1003 | |
bf77467a | 1004 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
5f39d397 | 1005 | wret = btrfs_cow_block(trans, root, right, |
9631e4cc | 1006 | parent, pslot + 1, &right, |
bf59a5a2 | 1007 | BTRFS_NESTING_RIGHT_COW); |
2cc58cf2 CM |
1008 | if (wret) { |
1009 | ret = wret; | |
daefe4d4 | 1010 | goto out; |
2cc58cf2 CM |
1011 | } |
1012 | } | |
1013 | ||
1014 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1015 | if (left) { |
1016 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 1017 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
1018 | if (wret < 0) |
1019 | ret = wret; | |
bb803951 | 1020 | } |
79f95c82 CM |
1021 | |
1022 | /* | |
1023 | * then try to empty the right most buffer into the middle | |
1024 | */ | |
5f39d397 | 1025 | if (right) { |
d30a668f | 1026 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1027 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1028 | ret = wret; |
5f39d397 | 1029 | if (btrfs_header_nritems(right) == 0) { |
190a8339 | 1030 | btrfs_clear_buffer_dirty(trans, right); |
925baedd | 1031 | btrfs_tree_unlock(right); |
751a2761 FM |
1032 | ret = btrfs_del_ptr(trans, root, path, level + 1, pslot + 1); |
1033 | if (ret < 0) { | |
1034 | free_extent_buffer_stale(right); | |
1035 | right = NULL; | |
1036 | goto out; | |
1037 | } | |
02cd00fa | 1038 | root_sub_used_bytes(root); |
7a163608 FM |
1039 | btrfs_free_tree_block(trans, btrfs_root_id(root), right, |
1040 | 0, 1); | |
3083ee2e | 1041 | free_extent_buffer_stale(right); |
f0486c68 | 1042 | right = NULL; |
bb803951 | 1043 | } else { |
5f39d397 CM |
1044 | struct btrfs_disk_key right_key; |
1045 | btrfs_node_key(right, &right_key, 0); | |
f3a84ccd | 1046 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
33cff222 | 1047 | BTRFS_MOD_LOG_KEY_REPLACE); |
39020d8a FM |
1048 | if (ret < 0) { |
1049 | btrfs_abort_transaction(trans, ret); | |
daefe4d4 | 1050 | goto out; |
39020d8a | 1051 | } |
5f39d397 | 1052 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
50564b65 | 1053 | btrfs_mark_buffer_dirty(trans, parent); |
bb803951 CM |
1054 | } |
1055 | } | |
5f39d397 | 1056 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1057 | /* |
1058 | * we're not allowed to leave a node with one item in the | |
1059 | * tree during a delete. A deletion from lower in the tree | |
1060 | * could try to delete the only pointer in this node. | |
1061 | * So, pull some keys from the left. | |
1062 | * There has to be a left pointer at this point because | |
1063 | * otherwise we would have pulled some pointers from the | |
1064 | * right | |
1065 | */ | |
725026ed FM |
1066 | if (unlikely(!left)) { |
1067 | btrfs_crit(fs_info, | |
1068 | "missing left child when middle child only has 1 item, parent bytenr %llu level %d mid bytenr %llu root %llu", | |
1069 | parent->start, btrfs_header_level(parent), | |
1070 | mid->start, btrfs_root_id(root)); | |
1071 | ret = -EUCLEAN; | |
1072 | btrfs_abort_transaction(trans, ret); | |
daefe4d4 | 1073 | goto out; |
305a26af | 1074 | } |
55d32ed8 | 1075 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1076 | if (wret < 0) { |
79f95c82 | 1077 | ret = wret; |
daefe4d4 | 1078 | goto out; |
54aa1f4d | 1079 | } |
bce4eae9 | 1080 | if (wret == 1) { |
d30a668f | 1081 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1082 | if (wret < 0) |
1083 | ret = wret; | |
1084 | } | |
79f95c82 CM |
1085 | BUG_ON(wret == 1); |
1086 | } | |
5f39d397 | 1087 | if (btrfs_header_nritems(mid) == 0) { |
190a8339 | 1088 | btrfs_clear_buffer_dirty(trans, mid); |
925baedd | 1089 | btrfs_tree_unlock(mid); |
751a2761 FM |
1090 | ret = btrfs_del_ptr(trans, root, path, level + 1, pslot); |
1091 | if (ret < 0) { | |
1092 | free_extent_buffer_stale(mid); | |
1093 | mid = NULL; | |
1094 | goto out; | |
1095 | } | |
02cd00fa | 1096 | root_sub_used_bytes(root); |
7a163608 | 1097 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
3083ee2e | 1098 | free_extent_buffer_stale(mid); |
f0486c68 | 1099 | mid = NULL; |
79f95c82 CM |
1100 | } else { |
1101 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1102 | struct btrfs_disk_key mid_key; |
1103 | btrfs_node_key(mid, &mid_key, 0); | |
f3a84ccd | 1104 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
33cff222 | 1105 | BTRFS_MOD_LOG_KEY_REPLACE); |
39020d8a FM |
1106 | if (ret < 0) { |
1107 | btrfs_abort_transaction(trans, ret); | |
daefe4d4 | 1108 | goto out; |
39020d8a | 1109 | } |
5f39d397 | 1110 | btrfs_set_node_key(parent, &mid_key, pslot); |
50564b65 | 1111 | btrfs_mark_buffer_dirty(trans, parent); |
79f95c82 | 1112 | } |
bb803951 | 1113 | |
79f95c82 | 1114 | /* update the path */ |
5f39d397 CM |
1115 | if (left) { |
1116 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1117 | atomic_inc(&left->refs); |
925baedd | 1118 | /* left was locked after cow */ |
5f39d397 | 1119 | path->nodes[level] = left; |
bb803951 CM |
1120 | path->slots[level + 1] -= 1; |
1121 | path->slots[level] = orig_slot; | |
925baedd CM |
1122 | if (mid) { |
1123 | btrfs_tree_unlock(mid); | |
5f39d397 | 1124 | free_extent_buffer(mid); |
925baedd | 1125 | } |
bb803951 | 1126 | } else { |
5f39d397 | 1127 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1128 | path->slots[level] = orig_slot; |
1129 | } | |
1130 | } | |
79f95c82 | 1131 | /* double check we haven't messed things up */ |
e20d96d6 | 1132 | if (orig_ptr != |
5f39d397 | 1133 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1134 | BUG(); |
daefe4d4 | 1135 | out: |
925baedd CM |
1136 | if (right) { |
1137 | btrfs_tree_unlock(right); | |
5f39d397 | 1138 | free_extent_buffer(right); |
925baedd CM |
1139 | } |
1140 | if (left) { | |
1141 | if (path->nodes[level] != left) | |
1142 | btrfs_tree_unlock(left); | |
5f39d397 | 1143 | free_extent_buffer(left); |
925baedd | 1144 | } |
bb803951 CM |
1145 | return ret; |
1146 | } | |
1147 | ||
d352ac68 CM |
1148 | /* Node balancing for insertion. Here we only split or push nodes around |
1149 | * when they are completely full. This is also done top down, so we | |
1150 | * have to be pessimistic. | |
1151 | */ | |
d397712b | 1152 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1153 | struct btrfs_root *root, |
1154 | struct btrfs_path *path, int level) | |
e66f709b | 1155 | { |
0b246afa | 1156 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1157 | struct extent_buffer *right = NULL; |
1158 | struct extent_buffer *mid; | |
1159 | struct extent_buffer *left = NULL; | |
1160 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1161 | int ret = 0; |
1162 | int wret; | |
1163 | int pslot; | |
1164 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1165 | |
1166 | if (level == 0) | |
1167 | return 1; | |
1168 | ||
5f39d397 | 1169 | mid = path->nodes[level]; |
7bb86316 | 1170 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1171 | |
a05a9bb1 | 1172 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1173 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1174 | pslot = path->slots[level + 1]; |
1175 | } | |
e66f709b | 1176 | |
5f39d397 | 1177 | if (!parent) |
e66f709b | 1178 | return 1; |
e66f709b | 1179 | |
e66f709b | 1180 | /* first, try to make some room in the middle buffer */ |
9cf14029 | 1181 | if (pslot) { |
e66f709b | 1182 | u32 left_nr; |
925baedd | 1183 | |
9cf14029 JB |
1184 | left = btrfs_read_node_slot(parent, pslot - 1); |
1185 | if (IS_ERR(left)) | |
1186 | return PTR_ERR(left); | |
1187 | ||
bf77467a | 1188 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
b4ce94de | 1189 | |
5f39d397 | 1190 | left_nr = btrfs_header_nritems(left); |
0b246afa | 1191 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1192 | wret = 1; |
1193 | } else { | |
5f39d397 | 1194 | ret = btrfs_cow_block(trans, root, left, parent, |
9631e4cc | 1195 | pslot - 1, &left, |
bf59a5a2 | 1196 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
1197 | if (ret) |
1198 | wret = 1; | |
1199 | else { | |
d30a668f | 1200 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 1201 | } |
33ade1f8 | 1202 | } |
e66f709b CM |
1203 | if (wret < 0) |
1204 | ret = wret; | |
1205 | if (wret == 0) { | |
5f39d397 | 1206 | struct btrfs_disk_key disk_key; |
e66f709b | 1207 | orig_slot += left_nr; |
5f39d397 | 1208 | btrfs_node_key(mid, &disk_key, 0); |
f3a84ccd | 1209 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
33cff222 | 1210 | BTRFS_MOD_LOG_KEY_REPLACE); |
11d6ae03 FM |
1211 | if (ret < 0) { |
1212 | btrfs_tree_unlock(left); | |
1213 | free_extent_buffer(left); | |
1214 | btrfs_abort_transaction(trans, ret); | |
1215 | return ret; | |
1216 | } | |
5f39d397 | 1217 | btrfs_set_node_key(parent, &disk_key, pslot); |
50564b65 | 1218 | btrfs_mark_buffer_dirty(trans, parent); |
5f39d397 CM |
1219 | if (btrfs_header_nritems(left) > orig_slot) { |
1220 | path->nodes[level] = left; | |
e66f709b CM |
1221 | path->slots[level + 1] -= 1; |
1222 | path->slots[level] = orig_slot; | |
925baedd | 1223 | btrfs_tree_unlock(mid); |
5f39d397 | 1224 | free_extent_buffer(mid); |
e66f709b CM |
1225 | } else { |
1226 | orig_slot -= | |
5f39d397 | 1227 | btrfs_header_nritems(left); |
e66f709b | 1228 | path->slots[level] = orig_slot; |
925baedd | 1229 | btrfs_tree_unlock(left); |
5f39d397 | 1230 | free_extent_buffer(left); |
e66f709b | 1231 | } |
e66f709b CM |
1232 | return 0; |
1233 | } | |
925baedd | 1234 | btrfs_tree_unlock(left); |
5f39d397 | 1235 | free_extent_buffer(left); |
e66f709b | 1236 | } |
e66f709b CM |
1237 | |
1238 | /* | |
1239 | * then try to empty the right most buffer into the middle | |
1240 | */ | |
9cf14029 | 1241 | if (pslot + 1 < btrfs_header_nritems(parent)) { |
33ade1f8 | 1242 | u32 right_nr; |
b4ce94de | 1243 | |
9cf14029 JB |
1244 | right = btrfs_read_node_slot(parent, pslot + 1); |
1245 | if (IS_ERR(right)) | |
1246 | return PTR_ERR(right); | |
1247 | ||
bf77467a | 1248 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
b4ce94de | 1249 | |
5f39d397 | 1250 | right_nr = btrfs_header_nritems(right); |
0b246afa | 1251 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1252 | wret = 1; |
1253 | } else { | |
5f39d397 CM |
1254 | ret = btrfs_cow_block(trans, root, right, |
1255 | parent, pslot + 1, | |
bf59a5a2 | 1256 | &right, BTRFS_NESTING_RIGHT_COW); |
54aa1f4d CM |
1257 | if (ret) |
1258 | wret = 1; | |
1259 | else { | |
55d32ed8 | 1260 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 1261 | } |
33ade1f8 | 1262 | } |
e66f709b CM |
1263 | if (wret < 0) |
1264 | ret = wret; | |
1265 | if (wret == 0) { | |
5f39d397 CM |
1266 | struct btrfs_disk_key disk_key; |
1267 | ||
1268 | btrfs_node_key(right, &disk_key, 0); | |
f3a84ccd | 1269 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
33cff222 | 1270 | BTRFS_MOD_LOG_KEY_REPLACE); |
11d6ae03 FM |
1271 | if (ret < 0) { |
1272 | btrfs_tree_unlock(right); | |
1273 | free_extent_buffer(right); | |
1274 | btrfs_abort_transaction(trans, ret); | |
1275 | return ret; | |
1276 | } | |
5f39d397 | 1277 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
50564b65 | 1278 | btrfs_mark_buffer_dirty(trans, parent); |
5f39d397 CM |
1279 | |
1280 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1281 | path->nodes[level] = right; | |
e66f709b CM |
1282 | path->slots[level + 1] += 1; |
1283 | path->slots[level] = orig_slot - | |
5f39d397 | 1284 | btrfs_header_nritems(mid); |
925baedd | 1285 | btrfs_tree_unlock(mid); |
5f39d397 | 1286 | free_extent_buffer(mid); |
e66f709b | 1287 | } else { |
925baedd | 1288 | btrfs_tree_unlock(right); |
5f39d397 | 1289 | free_extent_buffer(right); |
e66f709b | 1290 | } |
e66f709b CM |
1291 | return 0; |
1292 | } | |
925baedd | 1293 | btrfs_tree_unlock(right); |
5f39d397 | 1294 | free_extent_buffer(right); |
e66f709b | 1295 | } |
e66f709b CM |
1296 | return 1; |
1297 | } | |
1298 | ||
3c69faec | 1299 | /* |
d352ac68 CM |
1300 | * readahead one full node of leaves, finding things that are close |
1301 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1302 | */ |
2ff7e61e | 1303 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
1304 | struct btrfs_path *path, |
1305 | int level, int slot, u64 objectid) | |
3c69faec | 1306 | { |
5f39d397 | 1307 | struct extent_buffer *node; |
01f46658 | 1308 | struct btrfs_disk_key disk_key; |
3c69faec | 1309 | u32 nritems; |
3c69faec | 1310 | u64 search; |
a7175319 | 1311 | u64 target; |
6b80053d | 1312 | u64 nread = 0; |
ace75066 | 1313 | u64 nread_max; |
6b80053d CM |
1314 | u32 nr; |
1315 | u32 blocksize; | |
1316 | u32 nscan = 0; | |
db94535d | 1317 | |
ace75066 | 1318 | if (level != 1 && path->reada != READA_FORWARD_ALWAYS) |
6702ed49 CM |
1319 | return; |
1320 | ||
1321 | if (!path->nodes[level]) | |
3c69faec CM |
1322 | return; |
1323 | ||
5f39d397 | 1324 | node = path->nodes[level]; |
925baedd | 1325 | |
ace75066 FM |
1326 | /* |
1327 | * Since the time between visiting leaves is much shorter than the time | |
1328 | * between visiting nodes, limit read ahead of nodes to 1, to avoid too | |
1329 | * much IO at once (possibly random). | |
1330 | */ | |
1331 | if (path->reada == READA_FORWARD_ALWAYS) { | |
1332 | if (level > 1) | |
1333 | nread_max = node->fs_info->nodesize; | |
1334 | else | |
1335 | nread_max = SZ_128K; | |
1336 | } else { | |
1337 | nread_max = SZ_64K; | |
1338 | } | |
1339 | ||
3c69faec | 1340 | search = btrfs_node_blockptr(node, slot); |
0b246afa | 1341 | blocksize = fs_info->nodesize; |
069a2e37 FM |
1342 | if (path->reada != READA_FORWARD_ALWAYS) { |
1343 | struct extent_buffer *eb; | |
1344 | ||
1345 | eb = find_extent_buffer(fs_info, search); | |
1346 | if (eb) { | |
1347 | free_extent_buffer(eb); | |
1348 | return; | |
1349 | } | |
3c69faec CM |
1350 | } |
1351 | ||
a7175319 | 1352 | target = search; |
6b80053d | 1353 | |
5f39d397 | 1354 | nritems = btrfs_header_nritems(node); |
6b80053d | 1355 | nr = slot; |
25b8b936 | 1356 | |
d397712b | 1357 | while (1) { |
e4058b54 | 1358 | if (path->reada == READA_BACK) { |
6b80053d CM |
1359 | if (nr == 0) |
1360 | break; | |
1361 | nr--; | |
ace75066 FM |
1362 | } else if (path->reada == READA_FORWARD || |
1363 | path->reada == READA_FORWARD_ALWAYS) { | |
6b80053d CM |
1364 | nr++; |
1365 | if (nr >= nritems) | |
1366 | break; | |
3c69faec | 1367 | } |
e4058b54 | 1368 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
1369 | btrfs_node_key(node, &disk_key, nr); |
1370 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1371 | break; | |
1372 | } | |
6b80053d | 1373 | search = btrfs_node_blockptr(node, nr); |
ace75066 FM |
1374 | if (path->reada == READA_FORWARD_ALWAYS || |
1375 | (search <= target && target - search <= 65536) || | |
a7175319 | 1376 | (search > target && search - target <= 65536)) { |
bfb484d9 | 1377 | btrfs_readahead_node_child(node, nr); |
6b80053d CM |
1378 | nread += blocksize; |
1379 | } | |
1380 | nscan++; | |
ace75066 | 1381 | if (nread > nread_max || nscan > 32) |
6b80053d | 1382 | break; |
3c69faec CM |
1383 | } |
1384 | } | |
925baedd | 1385 | |
bfb484d9 | 1386 | static noinline void reada_for_balance(struct btrfs_path *path, int level) |
b4ce94de | 1387 | { |
bfb484d9 | 1388 | struct extent_buffer *parent; |
b4ce94de CM |
1389 | int slot; |
1390 | int nritems; | |
b4ce94de | 1391 | |
8c594ea8 | 1392 | parent = path->nodes[level + 1]; |
b4ce94de | 1393 | if (!parent) |
0b08851f | 1394 | return; |
b4ce94de CM |
1395 | |
1396 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1397 | slot = path->slots[level + 1]; |
b4ce94de | 1398 | |
bfb484d9 JB |
1399 | if (slot > 0) |
1400 | btrfs_readahead_node_child(parent, slot - 1); | |
1401 | if (slot + 1 < nritems) | |
1402 | btrfs_readahead_node_child(parent, slot + 1); | |
b4ce94de CM |
1403 | } |
1404 | ||
1405 | ||
d352ac68 | 1406 | /* |
d397712b CM |
1407 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1408 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1409 | * operations on the tree might require changing key pointers higher up in the | |
1410 | * tree. | |
d352ac68 | 1411 | * |
d397712b CM |
1412 | * callers might also have set path->keep_locks, which tells this code to keep |
1413 | * the lock if the path points to the last slot in the block. This is part of | |
1414 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1415 | * |
d397712b CM |
1416 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1417 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1418 | */ |
e02119d5 | 1419 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1420 | int lowest_unlock, int min_write_lock_level, |
1421 | int *write_lock_level) | |
925baedd CM |
1422 | { |
1423 | int i; | |
1424 | int skip_level = level; | |
c1227996 | 1425 | bool check_skip = true; |
925baedd CM |
1426 | |
1427 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1428 | if (!path->nodes[i]) | |
1429 | break; | |
1430 | if (!path->locks[i]) | |
1431 | break; | |
c1227996 NB |
1432 | |
1433 | if (check_skip) { | |
1434 | if (path->slots[i] == 0) { | |
925baedd CM |
1435 | skip_level = i + 1; |
1436 | continue; | |
1437 | } | |
c1227996 NB |
1438 | |
1439 | if (path->keep_locks) { | |
1440 | u32 nritems; | |
1441 | ||
1442 | nritems = btrfs_header_nritems(path->nodes[i]); | |
1443 | if (nritems < 1 || path->slots[i] >= nritems - 1) { | |
1444 | skip_level = i + 1; | |
1445 | continue; | |
1446 | } | |
1447 | } | |
925baedd | 1448 | } |
051e1b9f | 1449 | |
d80bb3f9 | 1450 | if (i >= lowest_unlock && i > skip_level) { |
c1227996 NB |
1451 | check_skip = false; |
1452 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); | |
925baedd | 1453 | path->locks[i] = 0; |
f7c79f30 CM |
1454 | if (write_lock_level && |
1455 | i > min_write_lock_level && | |
1456 | i <= *write_lock_level) { | |
1457 | *write_lock_level = i - 1; | |
1458 | } | |
925baedd CM |
1459 | } |
1460 | } | |
1461 | } | |
1462 | ||
c8c42864 | 1463 | /* |
376a21d7 FM |
1464 | * Helper function for btrfs_search_slot() and other functions that do a search |
1465 | * on a btree. The goal is to find a tree block in the cache (the radix tree at | |
1466 | * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read | |
1467 | * its pages from disk. | |
c8c42864 | 1468 | * |
376a21d7 FM |
1469 | * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the |
1470 | * whole btree search, starting again from the current root node. | |
c8c42864 CM |
1471 | */ |
1472 | static int | |
d07b8528 LB |
1473 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
1474 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 1475 | const struct btrfs_key *key) |
c8c42864 | 1476 | { |
0b246afa | 1477 | struct btrfs_fs_info *fs_info = root->fs_info; |
789d6a3a | 1478 | struct btrfs_tree_parent_check check = { 0 }; |
c8c42864 CM |
1479 | u64 blocknr; |
1480 | u64 gen; | |
c8c42864 | 1481 | struct extent_buffer *tmp; |
76a05b35 | 1482 | int ret; |
581c1760 | 1483 | int parent_level; |
b246666e | 1484 | bool unlock_up; |
c8c42864 | 1485 | |
b246666e | 1486 | unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); |
213ff4b7 NB |
1487 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
1488 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
1489 | parent_level = btrfs_header_level(*eb_ret); | |
789d6a3a QW |
1490 | btrfs_node_key_to_cpu(*eb_ret, &check.first_key, slot); |
1491 | check.has_first_key = true; | |
1492 | check.level = parent_level - 1; | |
1493 | check.transid = gen; | |
1494 | check.owner_root = root->root_key.objectid; | |
c8c42864 | 1495 | |
b246666e FM |
1496 | /* |
1497 | * If we need to read an extent buffer from disk and we are holding locks | |
1498 | * on upper level nodes, we unlock all the upper nodes before reading the | |
1499 | * extent buffer, and then return -EAGAIN to the caller as it needs to | |
1500 | * restart the search. We don't release the lock on the current level | |
1501 | * because we need to walk this node to figure out which blocks to read. | |
1502 | */ | |
0b246afa | 1503 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 1504 | if (tmp) { |
ace75066 FM |
1505 | if (p->reada == READA_FORWARD_ALWAYS) |
1506 | reada_for_search(fs_info, p, level, slot, key->objectid); | |
1507 | ||
b9fab919 | 1508 | /* first we do an atomic uptodate check */ |
bdf7c00e | 1509 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
1510 | /* |
1511 | * Do extra check for first_key, eb can be stale due to | |
1512 | * being cached, read from scrub, or have multiple | |
1513 | * parents (shared tree blocks). | |
1514 | */ | |
e064d5e9 | 1515 | if (btrfs_verify_level_key(tmp, |
789d6a3a | 1516 | parent_level - 1, &check.first_key, gen)) { |
448de471 QW |
1517 | free_extent_buffer(tmp); |
1518 | return -EUCLEAN; | |
1519 | } | |
bdf7c00e JB |
1520 | *eb_ret = tmp; |
1521 | return 0; | |
1522 | } | |
1523 | ||
857bc13f JB |
1524 | if (p->nowait) { |
1525 | free_extent_buffer(tmp); | |
1526 | return -EAGAIN; | |
1527 | } | |
1528 | ||
b246666e FM |
1529 | if (unlock_up) |
1530 | btrfs_unlock_up_safe(p, level + 1); | |
1531 | ||
bdf7c00e | 1532 | /* now we're allowed to do a blocking uptodate check */ |
789d6a3a | 1533 | ret = btrfs_read_extent_buffer(tmp, &check); |
9a4ffa1b QW |
1534 | if (ret) { |
1535 | free_extent_buffer(tmp); | |
1536 | btrfs_release_path(p); | |
1537 | return -EIO; | |
cb44921a | 1538 | } |
88c602ab QW |
1539 | if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { |
1540 | free_extent_buffer(tmp); | |
1541 | btrfs_release_path(p); | |
1542 | return -EUCLEAN; | |
1543 | } | |
b246666e FM |
1544 | |
1545 | if (unlock_up) | |
1546 | ret = -EAGAIN; | |
1547 | ||
1548 | goto out; | |
857bc13f JB |
1549 | } else if (p->nowait) { |
1550 | return -EAGAIN; | |
c8c42864 CM |
1551 | } |
1552 | ||
b246666e | 1553 | if (unlock_up) { |
4bb59055 FM |
1554 | btrfs_unlock_up_safe(p, level + 1); |
1555 | ret = -EAGAIN; | |
1556 | } else { | |
1557 | ret = 0; | |
1558 | } | |
8c594ea8 | 1559 | |
e4058b54 | 1560 | if (p->reada != READA_NONE) |
2ff7e61e | 1561 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 1562 | |
789d6a3a | 1563 | tmp = read_tree_block(fs_info, blocknr, &check); |
4eb150d6 QW |
1564 | if (IS_ERR(tmp)) { |
1565 | btrfs_release_path(p); | |
1566 | return PTR_ERR(tmp); | |
76a05b35 | 1567 | } |
4eb150d6 QW |
1568 | /* |
1569 | * If the read above didn't mark this buffer up to date, | |
1570 | * it will never end up being up to date. Set ret to EIO now | |
1571 | * and give up so that our caller doesn't loop forever | |
1572 | * on our EAGAINs. | |
1573 | */ | |
1574 | if (!extent_buffer_uptodate(tmp)) | |
1575 | ret = -EIO; | |
02a3307a | 1576 | |
b246666e | 1577 | out: |
4bb59055 FM |
1578 | if (ret == 0) { |
1579 | *eb_ret = tmp; | |
1580 | } else { | |
1581 | free_extent_buffer(tmp); | |
1582 | btrfs_release_path(p); | |
1583 | } | |
1584 | ||
76a05b35 | 1585 | return ret; |
c8c42864 CM |
1586 | } |
1587 | ||
1588 | /* | |
1589 | * helper function for btrfs_search_slot. This does all of the checks | |
1590 | * for node-level blocks and does any balancing required based on | |
1591 | * the ins_len. | |
1592 | * | |
1593 | * If no extra work was required, zero is returned. If we had to | |
1594 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1595 | * start over | |
1596 | */ | |
1597 | static int | |
1598 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1599 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1600 | struct extent_buffer *b, int level, int ins_len, |
1601 | int *write_lock_level) | |
c8c42864 | 1602 | { |
0b246afa | 1603 | struct btrfs_fs_info *fs_info = root->fs_info; |
95b982de | 1604 | int ret = 0; |
0b246afa | 1605 | |
c8c42864 | 1606 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 1607 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 | 1608 | |
bd681513 CM |
1609 | if (*write_lock_level < level + 1) { |
1610 | *write_lock_level = level + 1; | |
1611 | btrfs_release_path(p); | |
95b982de | 1612 | return -EAGAIN; |
bd681513 CM |
1613 | } |
1614 | ||
bfb484d9 | 1615 | reada_for_balance(p, level); |
95b982de | 1616 | ret = split_node(trans, root, p, level); |
c8c42864 | 1617 | |
c8c42864 CM |
1618 | b = p->nodes[level]; |
1619 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 1620 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 | 1621 | |
bd681513 CM |
1622 | if (*write_lock_level < level + 1) { |
1623 | *write_lock_level = level + 1; | |
1624 | btrfs_release_path(p); | |
95b982de | 1625 | return -EAGAIN; |
bd681513 CM |
1626 | } |
1627 | ||
bfb484d9 | 1628 | reada_for_balance(p, level); |
95b982de NB |
1629 | ret = balance_level(trans, root, p, level); |
1630 | if (ret) | |
1631 | return ret; | |
c8c42864 | 1632 | |
c8c42864 CM |
1633 | b = p->nodes[level]; |
1634 | if (!b) { | |
b3b4aa74 | 1635 | btrfs_release_path(p); |
95b982de | 1636 | return -EAGAIN; |
c8c42864 CM |
1637 | } |
1638 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1639 | } | |
c8c42864 CM |
1640 | return ret; |
1641 | } | |
1642 | ||
381cf658 | 1643 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
1644 | u64 iobjectid, u64 ioff, u8 key_type, |
1645 | struct btrfs_key *found_key) | |
1646 | { | |
1647 | int ret; | |
1648 | struct btrfs_key key; | |
1649 | struct extent_buffer *eb; | |
381cf658 DS |
1650 | |
1651 | ASSERT(path); | |
1d4c08e0 | 1652 | ASSERT(found_key); |
e33d5c3d KN |
1653 | |
1654 | key.type = key_type; | |
1655 | key.objectid = iobjectid; | |
1656 | key.offset = ioff; | |
1657 | ||
1658 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 1659 | if (ret < 0) |
e33d5c3d KN |
1660 | return ret; |
1661 | ||
1662 | eb = path->nodes[0]; | |
1663 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
1664 | ret = btrfs_next_leaf(fs_root, path); | |
1665 | if (ret) | |
1666 | return ret; | |
1667 | eb = path->nodes[0]; | |
1668 | } | |
1669 | ||
1670 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
1671 | if (found_key->type != key.type || | |
1672 | found_key->objectid != key.objectid) | |
1673 | return 1; | |
1674 | ||
1675 | return 0; | |
1676 | } | |
1677 | ||
1fc28d8e LB |
1678 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
1679 | struct btrfs_path *p, | |
1680 | int write_lock_level) | |
1681 | { | |
1fc28d8e | 1682 | struct extent_buffer *b; |
120de408 | 1683 | int root_lock = 0; |
1fc28d8e LB |
1684 | int level = 0; |
1685 | ||
1fc28d8e | 1686 | if (p->search_commit_root) { |
d96b3424 FM |
1687 | b = root->commit_root; |
1688 | atomic_inc(&b->refs); | |
be6821f8 | 1689 | level = btrfs_header_level(b); |
f9ddfd05 LB |
1690 | /* |
1691 | * Ensure that all callers have set skip_locking when | |
1692 | * p->search_commit_root = 1. | |
1693 | */ | |
1694 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
1695 | |
1696 | goto out; | |
1697 | } | |
1698 | ||
1699 | if (p->skip_locking) { | |
1700 | b = btrfs_root_node(root); | |
1701 | level = btrfs_header_level(b); | |
1702 | goto out; | |
1703 | } | |
1704 | ||
120de408 JB |
1705 | /* We try very hard to do read locks on the root */ |
1706 | root_lock = BTRFS_READ_LOCK; | |
1707 | ||
1fc28d8e | 1708 | /* |
662c653b LB |
1709 | * If the level is set to maximum, we can skip trying to get the read |
1710 | * lock. | |
1fc28d8e | 1711 | */ |
662c653b LB |
1712 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
1713 | /* | |
1714 | * We don't know the level of the root node until we actually | |
1715 | * have it read locked | |
1716 | */ | |
857bc13f JB |
1717 | if (p->nowait) { |
1718 | b = btrfs_try_read_lock_root_node(root); | |
1719 | if (IS_ERR(b)) | |
1720 | return b; | |
1721 | } else { | |
1722 | b = btrfs_read_lock_root_node(root); | |
1723 | } | |
662c653b LB |
1724 | level = btrfs_header_level(b); |
1725 | if (level > write_lock_level) | |
1726 | goto out; | |
1727 | ||
1728 | /* Whoops, must trade for write lock */ | |
1729 | btrfs_tree_read_unlock(b); | |
1730 | free_extent_buffer(b); | |
1731 | } | |
1fc28d8e | 1732 | |
1fc28d8e LB |
1733 | b = btrfs_lock_root_node(root); |
1734 | root_lock = BTRFS_WRITE_LOCK; | |
1735 | ||
1736 | /* The level might have changed, check again */ | |
1737 | level = btrfs_header_level(b); | |
1738 | ||
1739 | out: | |
120de408 JB |
1740 | /* |
1741 | * The root may have failed to write out at some point, and thus is no | |
1742 | * longer valid, return an error in this case. | |
1743 | */ | |
1744 | if (!extent_buffer_uptodate(b)) { | |
1745 | if (root_lock) | |
1746 | btrfs_tree_unlock_rw(b, root_lock); | |
1747 | free_extent_buffer(b); | |
1748 | return ERR_PTR(-EIO); | |
1749 | } | |
1750 | ||
1fc28d8e LB |
1751 | p->nodes[level] = b; |
1752 | if (!p->skip_locking) | |
1753 | p->locks[level] = root_lock; | |
1754 | /* | |
1755 | * Callers are responsible for dropping b's references. | |
1756 | */ | |
1757 | return b; | |
1758 | } | |
1759 | ||
d96b3424 FM |
1760 | /* |
1761 | * Replace the extent buffer at the lowest level of the path with a cloned | |
1762 | * version. The purpose is to be able to use it safely, after releasing the | |
1763 | * commit root semaphore, even if relocation is happening in parallel, the | |
1764 | * transaction used for relocation is committed and the extent buffer is | |
1765 | * reallocated in the next transaction. | |
1766 | * | |
1767 | * This is used in a context where the caller does not prevent transaction | |
1768 | * commits from happening, either by holding a transaction handle or holding | |
1769 | * some lock, while it's doing searches through a commit root. | |
1770 | * At the moment it's only used for send operations. | |
1771 | */ | |
1772 | static int finish_need_commit_sem_search(struct btrfs_path *path) | |
1773 | { | |
1774 | const int i = path->lowest_level; | |
1775 | const int slot = path->slots[i]; | |
1776 | struct extent_buffer *lowest = path->nodes[i]; | |
1777 | struct extent_buffer *clone; | |
1778 | ||
1779 | ASSERT(path->need_commit_sem); | |
1780 | ||
1781 | if (!lowest) | |
1782 | return 0; | |
1783 | ||
1784 | lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); | |
1785 | ||
1786 | clone = btrfs_clone_extent_buffer(lowest); | |
1787 | if (!clone) | |
1788 | return -ENOMEM; | |
1789 | ||
1790 | btrfs_release_path(path); | |
1791 | path->nodes[i] = clone; | |
1792 | path->slots[i] = slot; | |
1793 | ||
1794 | return 0; | |
1795 | } | |
1fc28d8e | 1796 | |
e2e58d0f FM |
1797 | static inline int search_for_key_slot(struct extent_buffer *eb, |
1798 | int search_low_slot, | |
1799 | const struct btrfs_key *key, | |
1800 | int prev_cmp, | |
1801 | int *slot) | |
1802 | { | |
1803 | /* | |
1804 | * If a previous call to btrfs_bin_search() on a parent node returned an | |
1805 | * exact match (prev_cmp == 0), we can safely assume the target key will | |
1806 | * always be at slot 0 on lower levels, since each key pointer | |
1807 | * (struct btrfs_key_ptr) refers to the lowest key accessible from the | |
1808 | * subtree it points to. Thus we can skip searching lower levels. | |
1809 | */ | |
1810 | if (prev_cmp == 0) { | |
1811 | *slot = 0; | |
1812 | return 0; | |
1813 | } | |
1814 | ||
fdf8d595 | 1815 | return btrfs_bin_search(eb, search_low_slot, key, slot); |
e2e58d0f FM |
1816 | } |
1817 | ||
109324cf FM |
1818 | static int search_leaf(struct btrfs_trans_handle *trans, |
1819 | struct btrfs_root *root, | |
1820 | const struct btrfs_key *key, | |
1821 | struct btrfs_path *path, | |
1822 | int ins_len, | |
1823 | int prev_cmp) | |
1824 | { | |
1825 | struct extent_buffer *leaf = path->nodes[0]; | |
1826 | int leaf_free_space = -1; | |
1827 | int search_low_slot = 0; | |
1828 | int ret; | |
1829 | bool do_bin_search = true; | |
1830 | ||
1831 | /* | |
1832 | * If we are doing an insertion, the leaf has enough free space and the | |
1833 | * destination slot for the key is not slot 0, then we can unlock our | |
1834 | * write lock on the parent, and any other upper nodes, before doing the | |
1835 | * binary search on the leaf (with search_for_key_slot()), allowing other | |
1836 | * tasks to lock the parent and any other upper nodes. | |
1837 | */ | |
1838 | if (ins_len > 0) { | |
1839 | /* | |
1840 | * Cache the leaf free space, since we will need it later and it | |
1841 | * will not change until then. | |
1842 | */ | |
1843 | leaf_free_space = btrfs_leaf_free_space(leaf); | |
1844 | ||
1845 | /* | |
1846 | * !path->locks[1] means we have a single node tree, the leaf is | |
1847 | * the root of the tree. | |
1848 | */ | |
1849 | if (path->locks[1] && leaf_free_space >= ins_len) { | |
1850 | struct btrfs_disk_key first_key; | |
1851 | ||
1852 | ASSERT(btrfs_header_nritems(leaf) > 0); | |
1853 | btrfs_item_key(leaf, &first_key, 0); | |
1854 | ||
1855 | /* | |
1856 | * Doing the extra comparison with the first key is cheap, | |
1857 | * taking into account that the first key is very likely | |
1858 | * already in a cache line because it immediately follows | |
1859 | * the extent buffer's header and we have recently accessed | |
1860 | * the header's level field. | |
1861 | */ | |
79d25df0 | 1862 | ret = btrfs_comp_keys(&first_key, key); |
109324cf FM |
1863 | if (ret < 0) { |
1864 | /* | |
1865 | * The first key is smaller than the key we want | |
1866 | * to insert, so we are safe to unlock all upper | |
1867 | * nodes and we have to do the binary search. | |
1868 | * | |
1869 | * We do use btrfs_unlock_up_safe() and not | |
1870 | * unlock_up() because the later does not unlock | |
1871 | * nodes with a slot of 0 - we can safely unlock | |
1872 | * any node even if its slot is 0 since in this | |
1873 | * case the key does not end up at slot 0 of the | |
1874 | * leaf and there's no need to split the leaf. | |
1875 | */ | |
1876 | btrfs_unlock_up_safe(path, 1); | |
1877 | search_low_slot = 1; | |
1878 | } else { | |
1879 | /* | |
1880 | * The first key is >= then the key we want to | |
1881 | * insert, so we can skip the binary search as | |
1882 | * the target key will be at slot 0. | |
1883 | * | |
1884 | * We can not unlock upper nodes when the key is | |
1885 | * less than the first key, because we will need | |
1886 | * to update the key at slot 0 of the parent node | |
1887 | * and possibly of other upper nodes too. | |
1888 | * If the key matches the first key, then we can | |
1889 | * unlock all the upper nodes, using | |
1890 | * btrfs_unlock_up_safe() instead of unlock_up() | |
1891 | * as stated above. | |
1892 | */ | |
1893 | if (ret == 0) | |
1894 | btrfs_unlock_up_safe(path, 1); | |
1895 | /* | |
1896 | * ret is already 0 or 1, matching the result of | |
1897 | * a btrfs_bin_search() call, so there is no need | |
1898 | * to adjust it. | |
1899 | */ | |
1900 | do_bin_search = false; | |
1901 | path->slots[0] = 0; | |
1902 | } | |
1903 | } | |
1904 | } | |
1905 | ||
1906 | if (do_bin_search) { | |
1907 | ret = search_for_key_slot(leaf, search_low_slot, key, | |
1908 | prev_cmp, &path->slots[0]); | |
1909 | if (ret < 0) | |
1910 | return ret; | |
1911 | } | |
1912 | ||
1913 | if (ins_len > 0) { | |
1914 | /* | |
1915 | * Item key already exists. In this case, if we are allowed to | |
1916 | * insert the item (for example, in dir_item case, item key | |
1917 | * collision is allowed), it will be merged with the original | |
1918 | * item. Only the item size grows, no new btrfs item will be | |
1919 | * added. If search_for_extension is not set, ins_len already | |
1920 | * accounts the size btrfs_item, deduct it here so leaf space | |
1921 | * check will be correct. | |
1922 | */ | |
1923 | if (ret == 0 && !path->search_for_extension) { | |
1924 | ASSERT(ins_len >= sizeof(struct btrfs_item)); | |
1925 | ins_len -= sizeof(struct btrfs_item); | |
1926 | } | |
1927 | ||
1928 | ASSERT(leaf_free_space >= 0); | |
1929 | ||
1930 | if (leaf_free_space < ins_len) { | |
1931 | int err; | |
1932 | ||
1933 | err = split_leaf(trans, root, key, path, ins_len, | |
1934 | (ret == 0)); | |
bb8e9a60 FM |
1935 | ASSERT(err <= 0); |
1936 | if (WARN_ON(err > 0)) | |
1937 | err = -EUCLEAN; | |
109324cf FM |
1938 | if (err) |
1939 | ret = err; | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | return ret; | |
1944 | } | |
1945 | ||
74123bd7 | 1946 | /* |
9580503b DS |
1947 | * Look for a key in a tree and perform necessary modifications to preserve |
1948 | * tree invariants. | |
74123bd7 | 1949 | * |
4271ecea NB |
1950 | * @trans: Handle of transaction, used when modifying the tree |
1951 | * @p: Holds all btree nodes along the search path | |
1952 | * @root: The root node of the tree | |
1953 | * @key: The key we are looking for | |
9a664971 | 1954 | * @ins_len: Indicates purpose of search: |
1955 | * >0 for inserts it's size of item inserted (*) | |
1956 | * <0 for deletions | |
1957 | * 0 for plain searches, not modifying the tree | |
1958 | * | |
1959 | * (*) If size of item inserted doesn't include | |
1960 | * sizeof(struct btrfs_item), then p->search_for_extension must | |
1961 | * be set. | |
4271ecea NB |
1962 | * @cow: boolean should CoW operations be performed. Must always be 1 |
1963 | * when modifying the tree. | |
97571fd0 | 1964 | * |
4271ecea NB |
1965 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
1966 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
1967 | * | |
1968 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
1969 | * of the path (level 0) | |
1970 | * | |
1971 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
1972 | * points to the slot where it should be inserted | |
1973 | * | |
1974 | * If an error is encountered while searching the tree a negative error number | |
1975 | * is returned | |
74123bd7 | 1976 | */ |
310712b2 OS |
1977 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1978 | const struct btrfs_key *key, struct btrfs_path *p, | |
1979 | int ins_len, int cow) | |
be0e5c09 | 1980 | { |
d96b3424 | 1981 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 1982 | struct extent_buffer *b; |
be0e5c09 CM |
1983 | int slot; |
1984 | int ret; | |
33c66f43 | 1985 | int err; |
be0e5c09 | 1986 | int level; |
925baedd | 1987 | int lowest_unlock = 1; |
bd681513 CM |
1988 | /* everything at write_lock_level or lower must be write locked */ |
1989 | int write_lock_level = 0; | |
9f3a7427 | 1990 | u8 lowest_level = 0; |
f7c79f30 | 1991 | int min_write_lock_level; |
d7396f07 | 1992 | int prev_cmp; |
9f3a7427 | 1993 | |
a4c853af C |
1994 | might_sleep(); |
1995 | ||
6702ed49 | 1996 | lowest_level = p->lowest_level; |
323ac95b | 1997 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1998 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 1999 | BUG_ON(!cow && ins_len); |
25179201 | 2000 | |
857bc13f JB |
2001 | /* |
2002 | * For now only allow nowait for read only operations. There's no | |
2003 | * strict reason why we can't, we just only need it for reads so it's | |
2004 | * only implemented for reads. | |
2005 | */ | |
2006 | ASSERT(!p->nowait || !cow); | |
2007 | ||
bd681513 | 2008 | if (ins_len < 0) { |
925baedd | 2009 | lowest_unlock = 2; |
65b51a00 | 2010 | |
bd681513 CM |
2011 | /* when we are removing items, we might have to go up to level |
2012 | * two as we update tree pointers Make sure we keep write | |
2013 | * for those levels as well | |
2014 | */ | |
2015 | write_lock_level = 2; | |
2016 | } else if (ins_len > 0) { | |
2017 | /* | |
2018 | * for inserting items, make sure we have a write lock on | |
2019 | * level 1 so we can update keys | |
2020 | */ | |
2021 | write_lock_level = 1; | |
2022 | } | |
2023 | ||
2024 | if (!cow) | |
2025 | write_lock_level = -1; | |
2026 | ||
09a2a8f9 | 2027 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2028 | write_lock_level = BTRFS_MAX_LEVEL; |
2029 | ||
f7c79f30 CM |
2030 | min_write_lock_level = write_lock_level; |
2031 | ||
d96b3424 FM |
2032 | if (p->need_commit_sem) { |
2033 | ASSERT(p->search_commit_root); | |
857bc13f JB |
2034 | if (p->nowait) { |
2035 | if (!down_read_trylock(&fs_info->commit_root_sem)) | |
2036 | return -EAGAIN; | |
2037 | } else { | |
2038 | down_read(&fs_info->commit_root_sem); | |
2039 | } | |
d96b3424 FM |
2040 | } |
2041 | ||
bb803951 | 2042 | again: |
d7396f07 | 2043 | prev_cmp = -1; |
1fc28d8e | 2044 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
2045 | if (IS_ERR(b)) { |
2046 | ret = PTR_ERR(b); | |
2047 | goto done; | |
2048 | } | |
925baedd | 2049 | |
eb60ceac | 2050 | while (b) { |
f624d976 QW |
2051 | int dec = 0; |
2052 | ||
5f39d397 | 2053 | level = btrfs_header_level(b); |
65b51a00 | 2054 | |
02217ed2 | 2055 | if (cow) { |
9ea2c7c9 NB |
2056 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
2057 | ||
c8c42864 CM |
2058 | /* |
2059 | * if we don't really need to cow this block | |
2060 | * then we don't want to set the path blocking, | |
2061 | * so we test it here | |
2062 | */ | |
5963ffca | 2063 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2064 | goto cow_done; |
5d4f98a2 | 2065 | |
bd681513 CM |
2066 | /* |
2067 | * must have write locks on this node and the | |
2068 | * parent | |
2069 | */ | |
5124e00e JB |
2070 | if (level > write_lock_level || |
2071 | (level + 1 > write_lock_level && | |
2072 | level + 1 < BTRFS_MAX_LEVEL && | |
2073 | p->nodes[level + 1])) { | |
bd681513 CM |
2074 | write_lock_level = level + 1; |
2075 | btrfs_release_path(p); | |
2076 | goto again; | |
2077 | } | |
2078 | ||
9ea2c7c9 NB |
2079 | if (last_level) |
2080 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
9631e4cc JB |
2081 | &b, |
2082 | BTRFS_NESTING_COW); | |
9ea2c7c9 NB |
2083 | else |
2084 | err = btrfs_cow_block(trans, root, b, | |
2085 | p->nodes[level + 1], | |
9631e4cc JB |
2086 | p->slots[level + 1], &b, |
2087 | BTRFS_NESTING_COW); | |
33c66f43 | 2088 | if (err) { |
33c66f43 | 2089 | ret = err; |
65b51a00 | 2090 | goto done; |
54aa1f4d | 2091 | } |
02217ed2 | 2092 | } |
65b51a00 | 2093 | cow_done: |
eb60ceac | 2094 | p->nodes[level] = b; |
b4ce94de CM |
2095 | |
2096 | /* | |
2097 | * we have a lock on b and as long as we aren't changing | |
2098 | * the tree, there is no way to for the items in b to change. | |
2099 | * It is safe to drop the lock on our parent before we | |
2100 | * go through the expensive btree search on b. | |
2101 | * | |
eb653de1 FDBM |
2102 | * If we're inserting or deleting (ins_len != 0), then we might |
2103 | * be changing slot zero, which may require changing the parent. | |
2104 | * So, we can't drop the lock until after we know which slot | |
2105 | * we're operating on. | |
b4ce94de | 2106 | */ |
eb653de1 FDBM |
2107 | if (!ins_len && !p->keep_locks) { |
2108 | int u = level + 1; | |
2109 | ||
2110 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2111 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2112 | p->locks[u] = 0; | |
2113 | } | |
2114 | } | |
b4ce94de | 2115 | |
e2e58d0f | 2116 | if (level == 0) { |
109324cf | 2117 | if (ins_len > 0) |
e5e1c174 | 2118 | ASSERT(write_lock_level >= 1); |
bd681513 | 2119 | |
109324cf | 2120 | ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); |
459931ec | 2121 | if (!p->search_for_split) |
f7c79f30 | 2122 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2123 | min_write_lock_level, NULL); |
65b51a00 | 2124 | goto done; |
be0e5c09 | 2125 | } |
e2e58d0f FM |
2126 | |
2127 | ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); | |
2128 | if (ret < 0) | |
2129 | goto done; | |
2130 | prev_cmp = ret; | |
2131 | ||
f624d976 QW |
2132 | if (ret && slot > 0) { |
2133 | dec = 1; | |
2134 | slot--; | |
2135 | } | |
2136 | p->slots[level] = slot; | |
2137 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2138 | &write_lock_level); | |
2139 | if (err == -EAGAIN) | |
2140 | goto again; | |
2141 | if (err) { | |
2142 | ret = err; | |
2143 | goto done; | |
2144 | } | |
2145 | b = p->nodes[level]; | |
2146 | slot = p->slots[level]; | |
2147 | ||
2148 | /* | |
2149 | * Slot 0 is special, if we change the key we have to update | |
2150 | * the parent pointer which means we must have a write lock on | |
2151 | * the parent | |
2152 | */ | |
2153 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2154 | write_lock_level = level + 1; | |
2155 | btrfs_release_path(p); | |
2156 | goto again; | |
2157 | } | |
2158 | ||
2159 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2160 | &write_lock_level); | |
2161 | ||
2162 | if (level == lowest_level) { | |
2163 | if (dec) | |
2164 | p->slots[level]++; | |
2165 | goto done; | |
2166 | } | |
2167 | ||
2168 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2169 | if (err == -EAGAIN) | |
2170 | goto again; | |
2171 | if (err) { | |
2172 | ret = err; | |
2173 | goto done; | |
2174 | } | |
2175 | ||
2176 | if (!p->skip_locking) { | |
2177 | level = btrfs_header_level(b); | |
b40130b2 JB |
2178 | |
2179 | btrfs_maybe_reset_lockdep_class(root, b); | |
2180 | ||
f624d976 | 2181 | if (level <= write_lock_level) { |
ac5887c8 | 2182 | btrfs_tree_lock(b); |
f624d976 QW |
2183 | p->locks[level] = BTRFS_WRITE_LOCK; |
2184 | } else { | |
857bc13f JB |
2185 | if (p->nowait) { |
2186 | if (!btrfs_try_tree_read_lock(b)) { | |
2187 | free_extent_buffer(b); | |
2188 | ret = -EAGAIN; | |
2189 | goto done; | |
2190 | } | |
2191 | } else { | |
2192 | btrfs_tree_read_lock(b); | |
2193 | } | |
f624d976 QW |
2194 | p->locks[level] = BTRFS_READ_LOCK; |
2195 | } | |
2196 | p->nodes[level] = b; | |
2197 | } | |
be0e5c09 | 2198 | } |
65b51a00 CM |
2199 | ret = 1; |
2200 | done: | |
5f5bc6b1 | 2201 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2202 | btrfs_release_path(p); |
d96b3424 FM |
2203 | |
2204 | if (p->need_commit_sem) { | |
2205 | int ret2; | |
2206 | ||
2207 | ret2 = finish_need_commit_sem_search(p); | |
2208 | up_read(&fs_info->commit_root_sem); | |
2209 | if (ret2) | |
2210 | ret = ret2; | |
2211 | } | |
2212 | ||
65b51a00 | 2213 | return ret; |
be0e5c09 | 2214 | } |
f75e2b79 | 2215 | ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); |
be0e5c09 | 2216 | |
5d9e75c4 JS |
2217 | /* |
2218 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2219 | * current state of the tree together with the operations recorded in the tree | |
2220 | * modification log to search for the key in a previous version of this tree, as | |
2221 | * denoted by the time_seq parameter. | |
2222 | * | |
2223 | * Naturally, there is no support for insert, delete or cow operations. | |
2224 | * | |
2225 | * The resulting path and return value will be set up as if we called | |
2226 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2227 | */ | |
310712b2 | 2228 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2229 | struct btrfs_path *p, u64 time_seq) |
2230 | { | |
0b246afa | 2231 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2232 | struct extent_buffer *b; |
2233 | int slot; | |
2234 | int ret; | |
2235 | int err; | |
2236 | int level; | |
2237 | int lowest_unlock = 1; | |
2238 | u8 lowest_level = 0; | |
2239 | ||
2240 | lowest_level = p->lowest_level; | |
2241 | WARN_ON(p->nodes[0] != NULL); | |
c922b016 | 2242 | ASSERT(!p->nowait); |
5d9e75c4 JS |
2243 | |
2244 | if (p->search_commit_root) { | |
2245 | BUG_ON(time_seq); | |
2246 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2247 | } | |
2248 | ||
2249 | again: | |
f3a84ccd | 2250 | b = btrfs_get_old_root(root, time_seq); |
315bed43 NB |
2251 | if (!b) { |
2252 | ret = -EIO; | |
2253 | goto done; | |
2254 | } | |
5d9e75c4 | 2255 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2256 | p->locks[level] = BTRFS_READ_LOCK; |
2257 | ||
2258 | while (b) { | |
abe9339d QW |
2259 | int dec = 0; |
2260 | ||
5d9e75c4 JS |
2261 | level = btrfs_header_level(b); |
2262 | p->nodes[level] = b; | |
5d9e75c4 JS |
2263 | |
2264 | /* | |
2265 | * we have a lock on b and as long as we aren't changing | |
2266 | * the tree, there is no way to for the items in b to change. | |
2267 | * It is safe to drop the lock on our parent before we | |
2268 | * go through the expensive btree search on b. | |
2269 | */ | |
2270 | btrfs_unlock_up_safe(p, level + 1); | |
2271 | ||
fdf8d595 | 2272 | ret = btrfs_bin_search(b, 0, key, &slot); |
cbca7d59 FM |
2273 | if (ret < 0) |
2274 | goto done; | |
5d9e75c4 | 2275 | |
abe9339d | 2276 | if (level == 0) { |
5d9e75c4 JS |
2277 | p->slots[level] = slot; |
2278 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2279 | goto done; |
2280 | } | |
5d9e75c4 | 2281 | |
abe9339d QW |
2282 | if (ret && slot > 0) { |
2283 | dec = 1; | |
2284 | slot--; | |
2285 | } | |
2286 | p->slots[level] = slot; | |
2287 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2288 | |
abe9339d QW |
2289 | if (level == lowest_level) { |
2290 | if (dec) | |
2291 | p->slots[level]++; | |
2292 | goto done; | |
2293 | } | |
5d9e75c4 | 2294 | |
abe9339d QW |
2295 | err = read_block_for_search(root, p, &b, level, slot, key); |
2296 | if (err == -EAGAIN) | |
2297 | goto again; | |
2298 | if (err) { | |
2299 | ret = err; | |
5d9e75c4 JS |
2300 | goto done; |
2301 | } | |
abe9339d QW |
2302 | |
2303 | level = btrfs_header_level(b); | |
ac5887c8 | 2304 | btrfs_tree_read_lock(b); |
f3a84ccd | 2305 | b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); |
abe9339d QW |
2306 | if (!b) { |
2307 | ret = -ENOMEM; | |
2308 | goto done; | |
2309 | } | |
2310 | p->locks[level] = BTRFS_READ_LOCK; | |
2311 | p->nodes[level] = b; | |
5d9e75c4 JS |
2312 | } |
2313 | ret = 1; | |
2314 | done: | |
5d9e75c4 JS |
2315 | if (ret < 0) |
2316 | btrfs_release_path(p); | |
2317 | ||
2318 | return ret; | |
2319 | } | |
2320 | ||
f469c8bd FM |
2321 | /* |
2322 | * Search the tree again to find a leaf with smaller keys. | |
2323 | * Returns 0 if it found something. | |
2324 | * Returns 1 if there are no smaller keys. | |
2325 | * Returns < 0 on error. | |
2326 | * | |
2327 | * This may release the path, and so you may lose any locks held at the | |
2328 | * time you call it. | |
2329 | */ | |
2330 | static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
2331 | { | |
2332 | struct btrfs_key key; | |
2333 | struct btrfs_key orig_key; | |
2334 | struct btrfs_disk_key found_key; | |
2335 | int ret; | |
2336 | ||
2337 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); | |
2338 | orig_key = key; | |
2339 | ||
2340 | if (key.offset > 0) { | |
2341 | key.offset--; | |
2342 | } else if (key.type > 0) { | |
2343 | key.type--; | |
2344 | key.offset = (u64)-1; | |
2345 | } else if (key.objectid > 0) { | |
2346 | key.objectid--; | |
2347 | key.type = (u8)-1; | |
2348 | key.offset = (u64)-1; | |
2349 | } else { | |
2350 | return 1; | |
2351 | } | |
2352 | ||
2353 | btrfs_release_path(path); | |
2354 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2355 | if (ret <= 0) | |
2356 | return ret; | |
2357 | ||
2358 | /* | |
2359 | * Previous key not found. Even if we were at slot 0 of the leaf we had | |
2360 | * before releasing the path and calling btrfs_search_slot(), we now may | |
2361 | * be in a slot pointing to the same original key - this can happen if | |
2362 | * after we released the path, one of more items were moved from a | |
2363 | * sibling leaf into the front of the leaf we had due to an insertion | |
2364 | * (see push_leaf_right()). | |
2365 | * If we hit this case and our slot is > 0 and just decrement the slot | |
2366 | * so that the caller does not process the same key again, which may or | |
2367 | * may not break the caller, depending on its logic. | |
2368 | */ | |
2369 | if (path->slots[0] < btrfs_header_nritems(path->nodes[0])) { | |
2370 | btrfs_item_key(path->nodes[0], &found_key, path->slots[0]); | |
79d25df0 | 2371 | ret = btrfs_comp_keys(&found_key, &orig_key); |
f469c8bd FM |
2372 | if (ret == 0) { |
2373 | if (path->slots[0] > 0) { | |
2374 | path->slots[0]--; | |
2375 | return 0; | |
2376 | } | |
2377 | /* | |
2378 | * At slot 0, same key as before, it means orig_key is | |
2379 | * the lowest, leftmost, key in the tree. We're done. | |
2380 | */ | |
2381 | return 1; | |
2382 | } | |
2383 | } | |
2384 | ||
2385 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
79d25df0 | 2386 | ret = btrfs_comp_keys(&found_key, &key); |
f469c8bd FM |
2387 | /* |
2388 | * We might have had an item with the previous key in the tree right | |
2389 | * before we released our path. And after we released our path, that | |
2390 | * item might have been pushed to the first slot (0) of the leaf we | |
2391 | * were holding due to a tree balance. Alternatively, an item with the | |
2392 | * previous key can exist as the only element of a leaf (big fat item). | |
2393 | * Therefore account for these 2 cases, so that our callers (like | |
2394 | * btrfs_previous_item) don't miss an existing item with a key matching | |
2395 | * the previous key we computed above. | |
2396 | */ | |
2397 | if (ret <= 0) | |
2398 | return 0; | |
2399 | return 1; | |
2400 | } | |
2401 | ||
2f38b3e1 AJ |
2402 | /* |
2403 | * helper to use instead of search slot if no exact match is needed but | |
2404 | * instead the next or previous item should be returned. | |
2405 | * When find_higher is true, the next higher item is returned, the next lower | |
2406 | * otherwise. | |
2407 | * When return_any and find_higher are both true, and no higher item is found, | |
2408 | * return the next lower instead. | |
2409 | * When return_any is true and find_higher is false, and no lower item is found, | |
2410 | * return the next higher instead. | |
2411 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2412 | * < 0 on error | |
2413 | */ | |
2414 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2415 | const struct btrfs_key *key, |
2416 | struct btrfs_path *p, int find_higher, | |
2417 | int return_any) | |
2f38b3e1 AJ |
2418 | { |
2419 | int ret; | |
2420 | struct extent_buffer *leaf; | |
2421 | ||
2422 | again: | |
2423 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2424 | if (ret <= 0) | |
2425 | return ret; | |
2426 | /* | |
2427 | * a return value of 1 means the path is at the position where the | |
2428 | * item should be inserted. Normally this is the next bigger item, | |
2429 | * but in case the previous item is the last in a leaf, path points | |
2430 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2431 | * item. | |
2432 | */ | |
2433 | leaf = p->nodes[0]; | |
2434 | ||
2435 | if (find_higher) { | |
2436 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2437 | ret = btrfs_next_leaf(root, p); | |
2438 | if (ret <= 0) | |
2439 | return ret; | |
2440 | if (!return_any) | |
2441 | return 1; | |
2442 | /* | |
2443 | * no higher item found, return the next | |
2444 | * lower instead | |
2445 | */ | |
2446 | return_any = 0; | |
2447 | find_higher = 0; | |
2448 | btrfs_release_path(p); | |
2449 | goto again; | |
2450 | } | |
2451 | } else { | |
e6793769 AJ |
2452 | if (p->slots[0] == 0) { |
2453 | ret = btrfs_prev_leaf(root, p); | |
2454 | if (ret < 0) | |
2455 | return ret; | |
2456 | if (!ret) { | |
23c6bf6a FDBM |
2457 | leaf = p->nodes[0]; |
2458 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
2459 | p->slots[0]--; | |
e6793769 | 2460 | return 0; |
2f38b3e1 | 2461 | } |
e6793769 AJ |
2462 | if (!return_any) |
2463 | return 1; | |
2464 | /* | |
2465 | * no lower item found, return the next | |
2466 | * higher instead | |
2467 | */ | |
2468 | return_any = 0; | |
2469 | find_higher = 1; | |
2470 | btrfs_release_path(p); | |
2471 | goto again; | |
2472 | } else { | |
2f38b3e1 AJ |
2473 | --p->slots[0]; |
2474 | } | |
2475 | } | |
2476 | return 0; | |
2477 | } | |
2478 | ||
0ff40a91 MPS |
2479 | /* |
2480 | * Execute search and call btrfs_previous_item to traverse backwards if the item | |
2481 | * was not found. | |
2482 | * | |
2483 | * Return 0 if found, 1 if not found and < 0 if error. | |
2484 | */ | |
2485 | int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, | |
2486 | struct btrfs_path *path) | |
2487 | { | |
2488 | int ret; | |
2489 | ||
2490 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
2491 | if (ret > 0) | |
2492 | ret = btrfs_previous_item(root, path, key->objectid, key->type); | |
2493 | ||
2494 | if (ret == 0) | |
2495 | btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); | |
2496 | ||
2497 | return ret; | |
2498 | } | |
2499 | ||
43dd529a | 2500 | /* |
62142be3 GN |
2501 | * Search for a valid slot for the given path. |
2502 | * | |
2503 | * @root: The root node of the tree. | |
2504 | * @key: Will contain a valid item if found. | |
2505 | * @path: The starting point to validate the slot. | |
2506 | * | |
2507 | * Return: 0 if the item is valid | |
2508 | * 1 if not found | |
2509 | * <0 if error. | |
2510 | */ | |
2511 | int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, | |
2512 | struct btrfs_path *path) | |
2513 | { | |
524f14bb | 2514 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
62142be3 | 2515 | int ret; |
62142be3 | 2516 | |
524f14bb FM |
2517 | ret = btrfs_next_leaf(root, path); |
2518 | if (ret) | |
2519 | return ret; | |
62142be3 | 2520 | } |
524f14bb FM |
2521 | |
2522 | btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); | |
62142be3 GN |
2523 | return 0; |
2524 | } | |
2525 | ||
74123bd7 CM |
2526 | /* |
2527 | * adjust the pointers going up the tree, starting at level | |
2528 | * making sure the right key of each node is points to 'key'. | |
2529 | * This is used after shifting pointers to the left, so it stops | |
2530 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2531 | * higher levels | |
aa5d6bed | 2532 | * |
74123bd7 | 2533 | */ |
50564b65 FM |
2534 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
2535 | struct btrfs_path *path, | |
143bede5 | 2536 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2537 | { |
2538 | int i; | |
5f39d397 | 2539 | struct extent_buffer *t; |
0e82bcfe | 2540 | int ret; |
5f39d397 | 2541 | |
234b63a0 | 2542 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2543 | int tslot = path->slots[i]; |
0e82bcfe | 2544 | |
eb60ceac | 2545 | if (!path->nodes[i]) |
be0e5c09 | 2546 | break; |
5f39d397 | 2547 | t = path->nodes[i]; |
f3a84ccd | 2548 | ret = btrfs_tree_mod_log_insert_key(t, tslot, |
33cff222 | 2549 | BTRFS_MOD_LOG_KEY_REPLACE); |
0e82bcfe | 2550 | BUG_ON(ret < 0); |
5f39d397 | 2551 | btrfs_set_node_key(t, key, tslot); |
50564b65 | 2552 | btrfs_mark_buffer_dirty(trans, path->nodes[i]); |
be0e5c09 CM |
2553 | if (tslot != 0) |
2554 | break; | |
2555 | } | |
2556 | } | |
2557 | ||
31840ae1 ZY |
2558 | /* |
2559 | * update item key. | |
2560 | * | |
2561 | * This function isn't completely safe. It's the caller's responsibility | |
2562 | * that the new key won't break the order | |
2563 | */ | |
50564b65 | 2564 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
b7a0365e | 2565 | struct btrfs_path *path, |
310712b2 | 2566 | const struct btrfs_key *new_key) |
31840ae1 | 2567 | { |
50564b65 | 2568 | struct btrfs_fs_info *fs_info = trans->fs_info; |
31840ae1 ZY |
2569 | struct btrfs_disk_key disk_key; |
2570 | struct extent_buffer *eb; | |
2571 | int slot; | |
2572 | ||
2573 | eb = path->nodes[0]; | |
2574 | slot = path->slots[0]; | |
2575 | if (slot > 0) { | |
2576 | btrfs_item_key(eb, &disk_key, slot - 1); | |
79d25df0 | 2577 | if (unlikely(btrfs_comp_keys(&disk_key, new_key) >= 0)) { |
eee3b811 | 2578 | btrfs_print_leaf(eb); |
7c15d410 QW |
2579 | btrfs_crit(fs_info, |
2580 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2581 | slot, btrfs_disk_key_objectid(&disk_key), | |
2582 | btrfs_disk_key_type(&disk_key), | |
2583 | btrfs_disk_key_offset(&disk_key), | |
2584 | new_key->objectid, new_key->type, | |
2585 | new_key->offset); | |
7c15d410 QW |
2586 | BUG(); |
2587 | } | |
31840ae1 ZY |
2588 | } |
2589 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2590 | btrfs_item_key(eb, &disk_key, slot + 1); | |
79d25df0 | 2591 | if (unlikely(btrfs_comp_keys(&disk_key, new_key) <= 0)) { |
eee3b811 | 2592 | btrfs_print_leaf(eb); |
7c15d410 QW |
2593 | btrfs_crit(fs_info, |
2594 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2595 | slot, btrfs_disk_key_objectid(&disk_key), | |
2596 | btrfs_disk_key_type(&disk_key), | |
2597 | btrfs_disk_key_offset(&disk_key), | |
2598 | new_key->objectid, new_key->type, | |
2599 | new_key->offset); | |
7c15d410 QW |
2600 | BUG(); |
2601 | } | |
31840ae1 ZY |
2602 | } |
2603 | ||
2604 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2605 | btrfs_set_item_key(eb, &disk_key, slot); | |
50564b65 | 2606 | btrfs_mark_buffer_dirty(trans, eb); |
31840ae1 | 2607 | if (slot == 0) |
50564b65 | 2608 | fixup_low_keys(trans, path, &disk_key, 1); |
31840ae1 ZY |
2609 | } |
2610 | ||
d16c702f QW |
2611 | /* |
2612 | * Check key order of two sibling extent buffers. | |
2613 | * | |
2614 | * Return true if something is wrong. | |
2615 | * Return false if everything is fine. | |
2616 | * | |
2617 | * Tree-checker only works inside one tree block, thus the following | |
2618 | * corruption can not be detected by tree-checker: | |
2619 | * | |
2620 | * Leaf @left | Leaf @right | |
2621 | * -------------------------------------------------------------- | |
2622 | * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | | |
2623 | * | |
2624 | * Key f6 in leaf @left itself is valid, but not valid when the next | |
2625 | * key in leaf @right is 7. | |
2626 | * This can only be checked at tree block merge time. | |
2627 | * And since tree checker has ensured all key order in each tree block | |
2628 | * is correct, we only need to bother the last key of @left and the first | |
2629 | * key of @right. | |
2630 | */ | |
2631 | static bool check_sibling_keys(struct extent_buffer *left, | |
2632 | struct extent_buffer *right) | |
2633 | { | |
2634 | struct btrfs_key left_last; | |
2635 | struct btrfs_key right_first; | |
2636 | int level = btrfs_header_level(left); | |
2637 | int nr_left = btrfs_header_nritems(left); | |
2638 | int nr_right = btrfs_header_nritems(right); | |
2639 | ||
2640 | /* No key to check in one of the tree blocks */ | |
2641 | if (!nr_left || !nr_right) | |
2642 | return false; | |
2643 | ||
2644 | if (level) { | |
2645 | btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); | |
2646 | btrfs_node_key_to_cpu(right, &right_first, 0); | |
2647 | } else { | |
2648 | btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); | |
2649 | btrfs_item_key_to_cpu(right, &right_first, 0); | |
2650 | } | |
2651 | ||
88ad95b0 | 2652 | if (unlikely(btrfs_comp_cpu_keys(&left_last, &right_first) >= 0)) { |
a2cea677 FM |
2653 | btrfs_crit(left->fs_info, "left extent buffer:"); |
2654 | btrfs_print_tree(left, false); | |
2655 | btrfs_crit(left->fs_info, "right extent buffer:"); | |
2656 | btrfs_print_tree(right, false); | |
d16c702f QW |
2657 | btrfs_crit(left->fs_info, |
2658 | "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", | |
2659 | left_last.objectid, left_last.type, | |
2660 | left_last.offset, right_first.objectid, | |
2661 | right_first.type, right_first.offset); | |
2662 | return true; | |
2663 | } | |
2664 | return false; | |
2665 | } | |
2666 | ||
74123bd7 CM |
2667 | /* |
2668 | * try to push data from one node into the next node left in the | |
79f95c82 | 2669 | * tree. |
aa5d6bed CM |
2670 | * |
2671 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2672 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2673 | */ |
98ed5174 | 2674 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 2675 | struct extent_buffer *dst, |
971a1f66 | 2676 | struct extent_buffer *src, int empty) |
be0e5c09 | 2677 | { |
d30a668f | 2678 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 2679 | int push_items = 0; |
bb803951 CM |
2680 | int src_nritems; |
2681 | int dst_nritems; | |
aa5d6bed | 2682 | int ret = 0; |
be0e5c09 | 2683 | |
5f39d397 CM |
2684 | src_nritems = btrfs_header_nritems(src); |
2685 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2686 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
2687 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2688 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2689 | |
bce4eae9 | 2690 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2691 | return 1; |
2692 | ||
d397712b | 2693 | if (push_items <= 0) |
be0e5c09 CM |
2694 | return 1; |
2695 | ||
bce4eae9 | 2696 | if (empty) { |
971a1f66 | 2697 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2698 | if (push_items < src_nritems) { |
2699 | /* leave at least 8 pointers in the node if | |
2700 | * we aren't going to empty it | |
2701 | */ | |
2702 | if (src_nritems - push_items < 8) { | |
2703 | if (push_items <= 8) | |
2704 | return 1; | |
2705 | push_items -= 8; | |
2706 | } | |
2707 | } | |
2708 | } else | |
2709 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2710 | |
d16c702f QW |
2711 | /* dst is the left eb, src is the middle eb */ |
2712 | if (check_sibling_keys(dst, src)) { | |
2713 | ret = -EUCLEAN; | |
2714 | btrfs_abort_transaction(trans, ret); | |
2715 | return ret; | |
2716 | } | |
f3a84ccd | 2717 | ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 2718 | if (ret) { |
66642832 | 2719 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2720 | return ret; |
2721 | } | |
5f39d397 | 2722 | copy_extent_buffer(dst, src, |
e23efd8e JB |
2723 | btrfs_node_key_ptr_offset(dst, dst_nritems), |
2724 | btrfs_node_key_ptr_offset(src, 0), | |
d397712b | 2725 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2726 | |
bb803951 | 2727 | if (push_items < src_nritems) { |
57911b8b | 2728 | /* |
5cead542 BB |
2729 | * btrfs_tree_mod_log_eb_copy handles logging the move, so we |
2730 | * don't need to do an explicit tree mod log operation for it. | |
57911b8b | 2731 | */ |
e23efd8e JB |
2732 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(src, 0), |
2733 | btrfs_node_key_ptr_offset(src, push_items), | |
5f39d397 CM |
2734 | (src_nritems - push_items) * |
2735 | sizeof(struct btrfs_key_ptr)); | |
2736 | } | |
2737 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2738 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
50564b65 FM |
2739 | btrfs_mark_buffer_dirty(trans, src); |
2740 | btrfs_mark_buffer_dirty(trans, dst); | |
31840ae1 | 2741 | |
79f95c82 CM |
2742 | return ret; |
2743 | } | |
2744 | ||
2745 | /* | |
2746 | * try to push data from one node into the next node right in the | |
2747 | * tree. | |
2748 | * | |
2749 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2750 | * error, and > 0 if there was no room in the right hand block. | |
2751 | * | |
2752 | * this will only push up to 1/2 the contents of the left node over | |
2753 | */ | |
5f39d397 | 2754 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2755 | struct extent_buffer *dst, |
2756 | struct extent_buffer *src) | |
79f95c82 | 2757 | { |
55d32ed8 | 2758 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
2759 | int push_items = 0; |
2760 | int max_push; | |
2761 | int src_nritems; | |
2762 | int dst_nritems; | |
2763 | int ret = 0; | |
79f95c82 | 2764 | |
7bb86316 CM |
2765 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2766 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2767 | ||
5f39d397 CM |
2768 | src_nritems = btrfs_header_nritems(src); |
2769 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2770 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 2771 | if (push_items <= 0) |
79f95c82 | 2772 | return 1; |
bce4eae9 | 2773 | |
d397712b | 2774 | if (src_nritems < 4) |
bce4eae9 | 2775 | return 1; |
79f95c82 CM |
2776 | |
2777 | max_push = src_nritems / 2 + 1; | |
2778 | /* don't try to empty the node */ | |
d397712b | 2779 | if (max_push >= src_nritems) |
79f95c82 | 2780 | return 1; |
252c38f0 | 2781 | |
79f95c82 CM |
2782 | if (max_push < push_items) |
2783 | push_items = max_push; | |
2784 | ||
d16c702f QW |
2785 | /* dst is the right eb, src is the middle eb */ |
2786 | if (check_sibling_keys(src, dst)) { | |
2787 | ret = -EUCLEAN; | |
2788 | btrfs_abort_transaction(trans, ret); | |
2789 | return ret; | |
2790 | } | |
5cead542 BB |
2791 | |
2792 | /* | |
2793 | * btrfs_tree_mod_log_eb_copy handles logging the move, so we don't | |
2794 | * need to do an explicit tree mod log operation for it. | |
2795 | */ | |
e23efd8e JB |
2796 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(dst, push_items), |
2797 | btrfs_node_key_ptr_offset(dst, 0), | |
5f39d397 CM |
2798 | (dst_nritems) * |
2799 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2800 | |
f3a84ccd FM |
2801 | ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
2802 | push_items); | |
5de865ee | 2803 | if (ret) { |
66642832 | 2804 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2805 | return ret; |
2806 | } | |
5f39d397 | 2807 | copy_extent_buffer(dst, src, |
e23efd8e JB |
2808 | btrfs_node_key_ptr_offset(dst, 0), |
2809 | btrfs_node_key_ptr_offset(src, src_nritems - push_items), | |
d397712b | 2810 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2811 | |
5f39d397 CM |
2812 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2813 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2814 | |
50564b65 FM |
2815 | btrfs_mark_buffer_dirty(trans, src); |
2816 | btrfs_mark_buffer_dirty(trans, dst); | |
31840ae1 | 2817 | |
aa5d6bed | 2818 | return ret; |
be0e5c09 CM |
2819 | } |
2820 | ||
97571fd0 CM |
2821 | /* |
2822 | * helper function to insert a new root level in the tree. | |
2823 | * A new node is allocated, and a single item is inserted to | |
2824 | * point to the existing root | |
aa5d6bed CM |
2825 | * |
2826 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2827 | */ |
d397712b | 2828 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 2829 | struct btrfs_root *root, |
fdd99c72 | 2830 | struct btrfs_path *path, int level) |
5c680ed6 | 2831 | { |
7bb86316 | 2832 | u64 lower_gen; |
5f39d397 CM |
2833 | struct extent_buffer *lower; |
2834 | struct extent_buffer *c; | |
925baedd | 2835 | struct extent_buffer *old; |
5f39d397 | 2836 | struct btrfs_disk_key lower_key; |
d9d19a01 | 2837 | int ret; |
5c680ed6 CM |
2838 | |
2839 | BUG_ON(path->nodes[level]); | |
2840 | BUG_ON(path->nodes[level-1] != root->node); | |
2841 | ||
7bb86316 CM |
2842 | lower = path->nodes[level-1]; |
2843 | if (level == 1) | |
2844 | btrfs_item_key(lower, &lower_key, 0); | |
2845 | else | |
2846 | btrfs_node_key(lower, &lower_key, 0); | |
2847 | ||
79bd3712 FM |
2848 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2849 | &lower_key, level, root->node->start, 0, | |
60ea105a | 2850 | 0, BTRFS_NESTING_NEW_ROOT); |
5f39d397 CM |
2851 | if (IS_ERR(c)) |
2852 | return PTR_ERR(c); | |
925baedd | 2853 | |
02cd00fa | 2854 | root_add_used_bytes(root); |
f0486c68 | 2855 | |
5f39d397 | 2856 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 2857 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2858 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2859 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2860 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2861 | |
2862 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2863 | |
50564b65 | 2864 | btrfs_mark_buffer_dirty(trans, c); |
d5719762 | 2865 | |
925baedd | 2866 | old = root->node; |
406808ab | 2867 | ret = btrfs_tree_mod_log_insert_root(root->node, c, false); |
f61aa7ba FM |
2868 | if (ret < 0) { |
2869 | btrfs_free_tree_block(trans, btrfs_root_id(root), c, 0, 1); | |
2870 | btrfs_tree_unlock(c); | |
2871 | free_extent_buffer(c); | |
2872 | return ret; | |
2873 | } | |
240f62c8 | 2874 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2875 | |
2876 | /* the super has an extra ref to root->node */ | |
2877 | free_extent_buffer(old); | |
2878 | ||
0b86a832 | 2879 | add_root_to_dirty_list(root); |
67439dad | 2880 | atomic_inc(&c->refs); |
5f39d397 | 2881 | path->nodes[level] = c; |
ac5887c8 | 2882 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2883 | path->slots[level] = 0; |
2884 | return 0; | |
2885 | } | |
2886 | ||
74123bd7 CM |
2887 | /* |
2888 | * worker function to insert a single pointer in a node. | |
2889 | * the node should have enough room for the pointer already | |
97571fd0 | 2890 | * |
74123bd7 CM |
2891 | * slot and level indicate where you want the key to go, and |
2892 | * blocknr is the block the key points to. | |
2893 | */ | |
50b5d1fc FM |
2894 | static int insert_ptr(struct btrfs_trans_handle *trans, |
2895 | struct btrfs_path *path, | |
2896 | struct btrfs_disk_key *key, u64 bytenr, | |
2897 | int slot, int level) | |
74123bd7 | 2898 | { |
5f39d397 | 2899 | struct extent_buffer *lower; |
74123bd7 | 2900 | int nritems; |
f3ea38da | 2901 | int ret; |
5c680ed6 CM |
2902 | |
2903 | BUG_ON(!path->nodes[level]); | |
49d0c642 | 2904 | btrfs_assert_tree_write_locked(path->nodes[level]); |
5f39d397 CM |
2905 | lower = path->nodes[level]; |
2906 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2907 | BUG_ON(slot > nritems); |
6ad3cf6d | 2908 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 2909 | if (slot != nritems) { |
bf1d3425 | 2910 | if (level) { |
f3a84ccd FM |
2911 | ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, |
2912 | slot, nritems - slot); | |
50b5d1fc FM |
2913 | if (ret < 0) { |
2914 | btrfs_abort_transaction(trans, ret); | |
2915 | return ret; | |
2916 | } | |
bf1d3425 | 2917 | } |
5f39d397 | 2918 | memmove_extent_buffer(lower, |
e23efd8e JB |
2919 | btrfs_node_key_ptr_offset(lower, slot + 1), |
2920 | btrfs_node_key_ptr_offset(lower, slot), | |
d6025579 | 2921 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2922 | } |
c3e06965 | 2923 | if (level) { |
f3a84ccd | 2924 | ret = btrfs_tree_mod_log_insert_key(lower, slot, |
33cff222 | 2925 | BTRFS_MOD_LOG_KEY_ADD); |
50b5d1fc FM |
2926 | if (ret < 0) { |
2927 | btrfs_abort_transaction(trans, ret); | |
2928 | return ret; | |
2929 | } | |
f3ea38da | 2930 | } |
5f39d397 | 2931 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2932 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2933 | WARN_ON(trans->transid == 0); |
2934 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 | 2935 | btrfs_set_header_nritems(lower, nritems + 1); |
50564b65 | 2936 | btrfs_mark_buffer_dirty(trans, lower); |
50b5d1fc FM |
2937 | |
2938 | return 0; | |
74123bd7 CM |
2939 | } |
2940 | ||
97571fd0 CM |
2941 | /* |
2942 | * split the node at the specified level in path in two. | |
2943 | * The path is corrected to point to the appropriate node after the split | |
2944 | * | |
2945 | * Before splitting this tries to make some room in the node by pushing | |
2946 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2947 | * |
2948 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2949 | */ |
e02119d5 CM |
2950 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2951 | struct btrfs_root *root, | |
2952 | struct btrfs_path *path, int level) | |
be0e5c09 | 2953 | { |
0b246afa | 2954 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2955 | struct extent_buffer *c; |
2956 | struct extent_buffer *split; | |
2957 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2958 | int mid; |
5c680ed6 | 2959 | int ret; |
7518a238 | 2960 | u32 c_nritems; |
eb60ceac | 2961 | |
5f39d397 | 2962 | c = path->nodes[level]; |
7bb86316 | 2963 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2964 | if (c == root->node) { |
d9abbf1c | 2965 | /* |
90f8d62e JS |
2966 | * trying to split the root, lets make a new one |
2967 | * | |
fdd99c72 | 2968 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
2969 | * insert_new_root, because that root buffer will be kept as a |
2970 | * normal node. We are going to log removal of half of the | |
f3a84ccd FM |
2971 | * elements below with btrfs_tree_mod_log_eb_copy(). We're |
2972 | * holding a tree lock on the buffer, which is why we cannot | |
2973 | * race with other tree_mod_log users. | |
d9abbf1c | 2974 | */ |
fdd99c72 | 2975 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2976 | if (ret) |
2977 | return ret; | |
b3612421 | 2978 | } else { |
e66f709b | 2979 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2980 | c = path->nodes[level]; |
2981 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 2982 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 2983 | return 0; |
54aa1f4d CM |
2984 | if (ret < 0) |
2985 | return ret; | |
be0e5c09 | 2986 | } |
e66f709b | 2987 | |
5f39d397 | 2988 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2989 | mid = (c_nritems + 1) / 2; |
2990 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2991 | |
79bd3712 FM |
2992 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2993 | &disk_key, level, c->start, 0, | |
60ea105a | 2994 | 0, BTRFS_NESTING_SPLIT); |
5f39d397 CM |
2995 | if (IS_ERR(split)) |
2996 | return PTR_ERR(split); | |
2997 | ||
02cd00fa | 2998 | root_add_used_bytes(root); |
bc877d28 | 2999 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 3000 | |
f3a84ccd | 3001 | ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 3002 | if (ret) { |
ede600e4 FM |
3003 | btrfs_tree_unlock(split); |
3004 | free_extent_buffer(split); | |
66642832 | 3005 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3006 | return ret; |
3007 | } | |
5f39d397 | 3008 | copy_extent_buffer(split, c, |
e23efd8e JB |
3009 | btrfs_node_key_ptr_offset(split, 0), |
3010 | btrfs_node_key_ptr_offset(c, mid), | |
5f39d397 CM |
3011 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); |
3012 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3013 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed | 3014 | |
50564b65 FM |
3015 | btrfs_mark_buffer_dirty(trans, c); |
3016 | btrfs_mark_buffer_dirty(trans, split); | |
5f39d397 | 3017 | |
50b5d1fc FM |
3018 | ret = insert_ptr(trans, path, &disk_key, split->start, |
3019 | path->slots[level + 1] + 1, level + 1); | |
3020 | if (ret < 0) { | |
3021 | btrfs_tree_unlock(split); | |
3022 | free_extent_buffer(split); | |
3023 | return ret; | |
3024 | } | |
aa5d6bed | 3025 | |
5de08d7d | 3026 | if (path->slots[level] >= mid) { |
5c680ed6 | 3027 | path->slots[level] -= mid; |
925baedd | 3028 | btrfs_tree_unlock(c); |
5f39d397 CM |
3029 | free_extent_buffer(c); |
3030 | path->nodes[level] = split; | |
5c680ed6 CM |
3031 | path->slots[level + 1] += 1; |
3032 | } else { | |
925baedd | 3033 | btrfs_tree_unlock(split); |
5f39d397 | 3034 | free_extent_buffer(split); |
be0e5c09 | 3035 | } |
d5286a92 | 3036 | return 0; |
be0e5c09 CM |
3037 | } |
3038 | ||
74123bd7 CM |
3039 | /* |
3040 | * how many bytes are required to store the items in a leaf. start | |
3041 | * and nr indicate which items in the leaf to check. This totals up the | |
3042 | * space used both by the item structs and the item data | |
3043 | */ | |
6c75a589 | 3044 | static int leaf_space_used(const struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
3045 | { |
3046 | int data_len; | |
5f39d397 | 3047 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3048 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3049 | |
3050 | if (!nr) | |
3051 | return 0; | |
3212fa14 JB |
3052 | data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); |
3053 | data_len = data_len - btrfs_item_offset(l, end); | |
0783fcfc | 3054 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3055 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3056 | return data_len; |
3057 | } | |
3058 | ||
d4dbff95 CM |
3059 | /* |
3060 | * The space between the end of the leaf items and | |
3061 | * the start of the leaf data. IOW, how much room | |
3062 | * the leaf has left for both items and data | |
3063 | */ | |
6c75a589 | 3064 | int btrfs_leaf_free_space(const struct extent_buffer *leaf) |
d4dbff95 | 3065 | { |
e902baac | 3066 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
3067 | int nritems = btrfs_header_nritems(leaf); |
3068 | int ret; | |
0b246afa JM |
3069 | |
3070 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3071 | if (ret < 0) { |
0b246afa JM |
3072 | btrfs_crit(fs_info, |
3073 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3074 | ret, | |
3075 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3076 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3077 | } |
3078 | return ret; | |
d4dbff95 CM |
3079 | } |
3080 | ||
99d8f83c CM |
3081 | /* |
3082 | * min slot controls the lowest index we're willing to push to the | |
3083 | * right. We'll push up to and including min_slot, but no lower | |
3084 | */ | |
ed25dab3 JB |
3085 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3086 | struct btrfs_path *path, | |
44871b1b CM |
3087 | int data_size, int empty, |
3088 | struct extent_buffer *right, | |
99d8f83c CM |
3089 | int free_space, u32 left_nritems, |
3090 | u32 min_slot) | |
00ec4c51 | 3091 | { |
f72f0010 | 3092 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 3093 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3094 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3095 | struct btrfs_map_token token; |
5f39d397 | 3096 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3097 | int slot; |
34a38218 | 3098 | u32 i; |
00ec4c51 CM |
3099 | int push_space = 0; |
3100 | int push_items = 0; | |
34a38218 | 3101 | u32 nr; |
7518a238 | 3102 | u32 right_nritems; |
5f39d397 | 3103 | u32 data_end; |
db94535d | 3104 | u32 this_item_size; |
00ec4c51 | 3105 | |
34a38218 CM |
3106 | if (empty) |
3107 | nr = 0; | |
3108 | else | |
99d8f83c | 3109 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3110 | |
31840ae1 | 3111 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3112 | push_space += data_size; |
31840ae1 | 3113 | |
44871b1b | 3114 | slot = path->slots[1]; |
34a38218 CM |
3115 | i = left_nritems - 1; |
3116 | while (i >= nr) { | |
31840ae1 ZY |
3117 | if (!empty && push_items > 0) { |
3118 | if (path->slots[0] > i) | |
3119 | break; | |
3120 | if (path->slots[0] == i) { | |
e902baac DS |
3121 | int space = btrfs_leaf_free_space(left); |
3122 | ||
31840ae1 ZY |
3123 | if (space + push_space * 2 > free_space) |
3124 | break; | |
3125 | } | |
3126 | } | |
3127 | ||
00ec4c51 | 3128 | if (path->slots[0] == i) |
87b29b20 | 3129 | push_space += data_size; |
db94535d | 3130 | |
3212fa14 | 3131 | this_item_size = btrfs_item_size(left, i); |
74794207 JB |
3132 | if (this_item_size + sizeof(struct btrfs_item) + |
3133 | push_space > free_space) | |
00ec4c51 | 3134 | break; |
31840ae1 | 3135 | |
00ec4c51 | 3136 | push_items++; |
74794207 | 3137 | push_space += this_item_size + sizeof(struct btrfs_item); |
34a38218 CM |
3138 | if (i == 0) |
3139 | break; | |
3140 | i--; | |
db94535d | 3141 | } |
5f39d397 | 3142 | |
925baedd CM |
3143 | if (push_items == 0) |
3144 | goto out_unlock; | |
5f39d397 | 3145 | |
6c1500f2 | 3146 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3147 | |
00ec4c51 | 3148 | /* push left to right */ |
5f39d397 | 3149 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3150 | |
dc2e724e | 3151 | push_space = btrfs_item_data_end(left, left_nritems - push_items); |
8f881e8c | 3152 | push_space -= leaf_data_end(left); |
5f39d397 | 3153 | |
00ec4c51 | 3154 | /* make room in the right data area */ |
8f881e8c | 3155 | data_end = leaf_data_end(right); |
637e3b48 JB |
3156 | memmove_leaf_data(right, data_end - push_space, data_end, |
3157 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); | |
5f39d397 | 3158 | |
00ec4c51 | 3159 | /* copy from the left data area */ |
637e3b48 JB |
3160 | copy_leaf_data(right, left, BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3161 | leaf_data_end(left), push_space); | |
5f39d397 | 3162 | |
637e3b48 | 3163 | memmove_leaf_items(right, push_items, 0, right_nritems); |
5f39d397 | 3164 | |
00ec4c51 | 3165 | /* copy the items from left to right */ |
637e3b48 | 3166 | copy_leaf_items(right, left, 0, left_nritems - push_items, push_items); |
00ec4c51 CM |
3167 | |
3168 | /* update the item pointers */ | |
c82f823c | 3169 | btrfs_init_map_token(&token, right); |
7518a238 | 3170 | right_nritems += push_items; |
5f39d397 | 3171 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3172 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3173 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
3174 | push_space -= btrfs_token_item_size(&token, i); |
3175 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d CM |
3176 | } |
3177 | ||
7518a238 | 3178 | left_nritems -= push_items; |
5f39d397 | 3179 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3180 | |
34a38218 | 3181 | if (left_nritems) |
50564b65 | 3182 | btrfs_mark_buffer_dirty(trans, left); |
f0486c68 | 3183 | else |
190a8339 | 3184 | btrfs_clear_buffer_dirty(trans, left); |
f0486c68 | 3185 | |
50564b65 | 3186 | btrfs_mark_buffer_dirty(trans, right); |
a429e513 | 3187 | |
5f39d397 CM |
3188 | btrfs_item_key(right, &disk_key, 0); |
3189 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
50564b65 | 3190 | btrfs_mark_buffer_dirty(trans, upper); |
02217ed2 | 3191 | |
00ec4c51 | 3192 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3193 | if (path->slots[0] >= left_nritems) { |
3194 | path->slots[0] -= left_nritems; | |
925baedd | 3195 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
190a8339 | 3196 | btrfs_clear_buffer_dirty(trans, path->nodes[0]); |
925baedd | 3197 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3198 | free_extent_buffer(path->nodes[0]); |
3199 | path->nodes[0] = right; | |
00ec4c51 CM |
3200 | path->slots[1] += 1; |
3201 | } else { | |
925baedd | 3202 | btrfs_tree_unlock(right); |
5f39d397 | 3203 | free_extent_buffer(right); |
00ec4c51 CM |
3204 | } |
3205 | return 0; | |
925baedd CM |
3206 | |
3207 | out_unlock: | |
3208 | btrfs_tree_unlock(right); | |
3209 | free_extent_buffer(right); | |
3210 | return 1; | |
00ec4c51 | 3211 | } |
925baedd | 3212 | |
44871b1b CM |
3213 | /* |
3214 | * push some data in the path leaf to the right, trying to free up at | |
3215 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3216 | * | |
3217 | * returns 1 if the push failed because the other node didn't have enough | |
3218 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3219 | * |
3220 | * this will push starting from min_slot to the end of the leaf. It won't | |
3221 | * push any slot lower than min_slot | |
44871b1b CM |
3222 | */ |
3223 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3224 | *root, struct btrfs_path *path, |
3225 | int min_data_size, int data_size, | |
3226 | int empty, u32 min_slot) | |
44871b1b CM |
3227 | { |
3228 | struct extent_buffer *left = path->nodes[0]; | |
3229 | struct extent_buffer *right; | |
3230 | struct extent_buffer *upper; | |
3231 | int slot; | |
3232 | int free_space; | |
3233 | u32 left_nritems; | |
3234 | int ret; | |
3235 | ||
3236 | if (!path->nodes[1]) | |
3237 | return 1; | |
3238 | ||
3239 | slot = path->slots[1]; | |
3240 | upper = path->nodes[1]; | |
3241 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3242 | return 1; | |
3243 | ||
49d0c642 | 3244 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3245 | |
4b231ae4 | 3246 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 | 3247 | if (IS_ERR(right)) |
9cf14029 | 3248 | return PTR_ERR(right); |
91ca338d | 3249 | |
bf77467a | 3250 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
44871b1b | 3251 | |
e902baac | 3252 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3253 | if (free_space < data_size) |
3254 | goto out_unlock; | |
3255 | ||
44871b1b | 3256 | ret = btrfs_cow_block(trans, root, right, upper, |
bf59a5a2 | 3257 | slot + 1, &right, BTRFS_NESTING_RIGHT_COW); |
44871b1b CM |
3258 | if (ret) |
3259 | goto out_unlock; | |
3260 | ||
44871b1b CM |
3261 | left_nritems = btrfs_header_nritems(left); |
3262 | if (left_nritems == 0) | |
3263 | goto out_unlock; | |
3264 | ||
d16c702f QW |
3265 | if (check_sibling_keys(left, right)) { |
3266 | ret = -EUCLEAN; | |
9ae5afd0 | 3267 | btrfs_abort_transaction(trans, ret); |
d16c702f QW |
3268 | btrfs_tree_unlock(right); |
3269 | free_extent_buffer(right); | |
3270 | return ret; | |
3271 | } | |
2ef1fed2 FDBM |
3272 | if (path->slots[0] == left_nritems && !empty) { |
3273 | /* Key greater than all keys in the leaf, right neighbor has | |
3274 | * enough room for it and we're not emptying our leaf to delete | |
3275 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3276 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3277 | btrfs_tree_unlock(left); |
3278 | free_extent_buffer(left); | |
3279 | path->nodes[0] = right; | |
3280 | path->slots[0] = 0; | |
3281 | path->slots[1]++; | |
3282 | return 0; | |
3283 | } | |
3284 | ||
ed25dab3 JB |
3285 | return __push_leaf_right(trans, path, min_data_size, empty, right, |
3286 | free_space, left_nritems, min_slot); | |
44871b1b CM |
3287 | out_unlock: |
3288 | btrfs_tree_unlock(right); | |
3289 | free_extent_buffer(right); | |
3290 | return 1; | |
3291 | } | |
3292 | ||
74123bd7 CM |
3293 | /* |
3294 | * push some data in the path leaf to the left, trying to free up at | |
3295 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3296 | * |
3297 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3298 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3299 | * items | |
74123bd7 | 3300 | */ |
ed25dab3 JB |
3301 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3302 | struct btrfs_path *path, int data_size, | |
44871b1b | 3303 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3304 | int free_space, u32 right_nritems, |
3305 | u32 max_slot) | |
be0e5c09 | 3306 | { |
8087c193 | 3307 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3308 | struct btrfs_disk_key disk_key; |
3309 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3310 | int i; |
be0e5c09 CM |
3311 | int push_space = 0; |
3312 | int push_items = 0; | |
7518a238 | 3313 | u32 old_left_nritems; |
34a38218 | 3314 | u32 nr; |
aa5d6bed | 3315 | int ret = 0; |
db94535d CM |
3316 | u32 this_item_size; |
3317 | u32 old_left_item_size; | |
cfed81a0 CM |
3318 | struct btrfs_map_token token; |
3319 | ||
34a38218 | 3320 | if (empty) |
99d8f83c | 3321 | nr = min(right_nritems, max_slot); |
34a38218 | 3322 | else |
99d8f83c | 3323 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3324 | |
3325 | for (i = 0; i < nr; i++) { | |
31840ae1 ZY |
3326 | if (!empty && push_items > 0) { |
3327 | if (path->slots[0] < i) | |
3328 | break; | |
3329 | if (path->slots[0] == i) { | |
e902baac DS |
3330 | int space = btrfs_leaf_free_space(right); |
3331 | ||
31840ae1 ZY |
3332 | if (space + push_space * 2 > free_space) |
3333 | break; | |
3334 | } | |
3335 | } | |
3336 | ||
be0e5c09 | 3337 | if (path->slots[0] == i) |
87b29b20 | 3338 | push_space += data_size; |
db94535d | 3339 | |
3212fa14 | 3340 | this_item_size = btrfs_item_size(right, i); |
74794207 JB |
3341 | if (this_item_size + sizeof(struct btrfs_item) + push_space > |
3342 | free_space) | |
be0e5c09 | 3343 | break; |
db94535d | 3344 | |
be0e5c09 | 3345 | push_items++; |
74794207 | 3346 | push_space += this_item_size + sizeof(struct btrfs_item); |
db94535d CM |
3347 | } |
3348 | ||
be0e5c09 | 3349 | if (push_items == 0) { |
925baedd CM |
3350 | ret = 1; |
3351 | goto out; | |
be0e5c09 | 3352 | } |
fae7f21c | 3353 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3354 | |
be0e5c09 | 3355 | /* push data from right to left */ |
637e3b48 | 3356 | copy_leaf_items(left, right, btrfs_header_nritems(left), 0, push_items); |
5f39d397 | 3357 | |
0b246afa | 3358 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
3212fa14 | 3359 | btrfs_item_offset(right, push_items - 1); |
5f39d397 | 3360 | |
637e3b48 JB |
3361 | copy_leaf_data(left, right, leaf_data_end(left) - push_space, |
3362 | btrfs_item_offset(right, push_items - 1), push_space); | |
5f39d397 | 3363 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3364 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3365 | |
c82f823c | 3366 | btrfs_init_map_token(&token, left); |
3212fa14 | 3367 | old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); |
0783fcfc | 3368 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3369 | u32 ioff; |
db94535d | 3370 | |
3212fa14 JB |
3371 | ioff = btrfs_token_item_offset(&token, i); |
3372 | btrfs_set_token_item_offset(&token, i, | |
cc4c13d5 | 3373 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); |
be0e5c09 | 3374 | } |
5f39d397 | 3375 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3376 | |
3377 | /* fixup right node */ | |
31b1a2bd JL |
3378 | if (push_items > right_nritems) |
3379 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3380 | right_nritems); |
34a38218 CM |
3381 | |
3382 | if (push_items < right_nritems) { | |
3212fa14 | 3383 | push_space = btrfs_item_offset(right, push_items - 1) - |
8f881e8c | 3384 | leaf_data_end(right); |
637e3b48 JB |
3385 | memmove_leaf_data(right, |
3386 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, | |
3387 | leaf_data_end(right), push_space); | |
3388 | ||
3389 | memmove_leaf_items(right, 0, push_items, | |
3390 | btrfs_header_nritems(right) - push_items); | |
34a38218 | 3391 | } |
c82f823c DS |
3392 | |
3393 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3394 | right_nritems -= push_items; |
3395 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3396 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3397 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
3398 | push_space = push_space - btrfs_token_item_size(&token, i); |
3399 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d | 3400 | } |
eb60ceac | 3401 | |
50564b65 | 3402 | btrfs_mark_buffer_dirty(trans, left); |
34a38218 | 3403 | if (right_nritems) |
50564b65 | 3404 | btrfs_mark_buffer_dirty(trans, right); |
f0486c68 | 3405 | else |
190a8339 | 3406 | btrfs_clear_buffer_dirty(trans, right); |
098f59c2 | 3407 | |
5f39d397 | 3408 | btrfs_item_key(right, &disk_key, 0); |
50564b65 | 3409 | fixup_low_keys(trans, path, &disk_key, 1); |
be0e5c09 CM |
3410 | |
3411 | /* then fixup the leaf pointer in the path */ | |
3412 | if (path->slots[0] < push_items) { | |
3413 | path->slots[0] += old_left_nritems; | |
925baedd | 3414 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3415 | free_extent_buffer(path->nodes[0]); |
3416 | path->nodes[0] = left; | |
be0e5c09 CM |
3417 | path->slots[1] -= 1; |
3418 | } else { | |
925baedd | 3419 | btrfs_tree_unlock(left); |
5f39d397 | 3420 | free_extent_buffer(left); |
be0e5c09 CM |
3421 | path->slots[0] -= push_items; |
3422 | } | |
eb60ceac | 3423 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3424 | return ret; |
925baedd CM |
3425 | out: |
3426 | btrfs_tree_unlock(left); | |
3427 | free_extent_buffer(left); | |
3428 | return ret; | |
be0e5c09 CM |
3429 | } |
3430 | ||
44871b1b CM |
3431 | /* |
3432 | * push some data in the path leaf to the left, trying to free up at | |
3433 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3434 | * |
3435 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3436 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3437 | * items | |
44871b1b CM |
3438 | */ |
3439 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3440 | *root, struct btrfs_path *path, int min_data_size, |
3441 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3442 | { |
3443 | struct extent_buffer *right = path->nodes[0]; | |
3444 | struct extent_buffer *left; | |
3445 | int slot; | |
3446 | int free_space; | |
3447 | u32 right_nritems; | |
3448 | int ret = 0; | |
3449 | ||
3450 | slot = path->slots[1]; | |
3451 | if (slot == 0) | |
3452 | return 1; | |
3453 | if (!path->nodes[1]) | |
3454 | return 1; | |
3455 | ||
3456 | right_nritems = btrfs_header_nritems(right); | |
3457 | if (right_nritems == 0) | |
3458 | return 1; | |
3459 | ||
49d0c642 | 3460 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3461 | |
4b231ae4 | 3462 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 | 3463 | if (IS_ERR(left)) |
9cf14029 | 3464 | return PTR_ERR(left); |
91ca338d | 3465 | |
bf77467a | 3466 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
44871b1b | 3467 | |
e902baac | 3468 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3469 | if (free_space < data_size) { |
3470 | ret = 1; | |
3471 | goto out; | |
3472 | } | |
3473 | ||
44871b1b | 3474 | ret = btrfs_cow_block(trans, root, left, |
9631e4cc | 3475 | path->nodes[1], slot - 1, &left, |
bf59a5a2 | 3476 | BTRFS_NESTING_LEFT_COW); |
44871b1b CM |
3477 | if (ret) { |
3478 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3479 | if (ret == -ENOSPC) |
3480 | ret = 1; | |
44871b1b CM |
3481 | goto out; |
3482 | } | |
3483 | ||
d16c702f QW |
3484 | if (check_sibling_keys(left, right)) { |
3485 | ret = -EUCLEAN; | |
9ae5afd0 | 3486 | btrfs_abort_transaction(trans, ret); |
d16c702f QW |
3487 | goto out; |
3488 | } | |
ed25dab3 JB |
3489 | return __push_leaf_left(trans, path, min_data_size, empty, left, |
3490 | free_space, right_nritems, max_slot); | |
44871b1b CM |
3491 | out: |
3492 | btrfs_tree_unlock(left); | |
3493 | free_extent_buffer(left); | |
3494 | return ret; | |
3495 | } | |
3496 | ||
3497 | /* | |
3498 | * split the path's leaf in two, making sure there is at least data_size | |
3499 | * available for the resulting leaf level of the path. | |
44871b1b | 3500 | */ |
50b5d1fc FM |
3501 | static noinline int copy_for_split(struct btrfs_trans_handle *trans, |
3502 | struct btrfs_path *path, | |
3503 | struct extent_buffer *l, | |
3504 | struct extent_buffer *right, | |
3505 | int slot, int mid, int nritems) | |
44871b1b | 3506 | { |
94f94ad9 | 3507 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
3508 | int data_copy_size; |
3509 | int rt_data_off; | |
3510 | int i; | |
50b5d1fc | 3511 | int ret; |
44871b1b | 3512 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3513 | struct btrfs_map_token token; |
3514 | ||
44871b1b CM |
3515 | nritems = nritems - mid; |
3516 | btrfs_set_header_nritems(right, nritems); | |
dc2e724e | 3517 | data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); |
44871b1b | 3518 | |
637e3b48 | 3519 | copy_leaf_items(right, l, 0, mid, nritems); |
44871b1b | 3520 | |
637e3b48 JB |
3521 | copy_leaf_data(right, l, BTRFS_LEAF_DATA_SIZE(fs_info) - data_copy_size, |
3522 | leaf_data_end(l), data_copy_size); | |
44871b1b | 3523 | |
dc2e724e | 3524 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); |
44871b1b | 3525 | |
c82f823c | 3526 | btrfs_init_map_token(&token, right); |
44871b1b | 3527 | for (i = 0; i < nritems; i++) { |
44871b1b CM |
3528 | u32 ioff; |
3529 | ||
3212fa14 JB |
3530 | ioff = btrfs_token_item_offset(&token, i); |
3531 | btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); | |
44871b1b CM |
3532 | } |
3533 | ||
44871b1b | 3534 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3535 | btrfs_item_key(right, &disk_key, 0); |
50b5d1fc FM |
3536 | ret = insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
3537 | if (ret < 0) | |
3538 | return ret; | |
44871b1b | 3539 | |
50564b65 FM |
3540 | btrfs_mark_buffer_dirty(trans, right); |
3541 | btrfs_mark_buffer_dirty(trans, l); | |
44871b1b CM |
3542 | BUG_ON(path->slots[0] != slot); |
3543 | ||
44871b1b CM |
3544 | if (mid <= slot) { |
3545 | btrfs_tree_unlock(path->nodes[0]); | |
3546 | free_extent_buffer(path->nodes[0]); | |
3547 | path->nodes[0] = right; | |
3548 | path->slots[0] -= mid; | |
3549 | path->slots[1] += 1; | |
3550 | } else { | |
3551 | btrfs_tree_unlock(right); | |
3552 | free_extent_buffer(right); | |
3553 | } | |
3554 | ||
3555 | BUG_ON(path->slots[0] < 0); | |
50b5d1fc FM |
3556 | |
3557 | return 0; | |
44871b1b CM |
3558 | } |
3559 | ||
99d8f83c CM |
3560 | /* |
3561 | * double splits happen when we need to insert a big item in the middle | |
3562 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3563 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3564 | * A B C | |
3565 | * | |
3566 | * We avoid this by trying to push the items on either side of our target | |
3567 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3568 | * completely. | |
3569 | */ | |
3570 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3571 | struct btrfs_root *root, | |
3572 | struct btrfs_path *path, | |
3573 | int data_size) | |
3574 | { | |
3575 | int ret; | |
3576 | int progress = 0; | |
3577 | int slot; | |
3578 | u32 nritems; | |
5a4267ca | 3579 | int space_needed = data_size; |
99d8f83c CM |
3580 | |
3581 | slot = path->slots[0]; | |
5a4267ca | 3582 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 3583 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
3584 | |
3585 | /* | |
3586 | * try to push all the items after our slot into the | |
3587 | * right leaf | |
3588 | */ | |
5a4267ca | 3589 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3590 | if (ret < 0) |
3591 | return ret; | |
3592 | ||
3593 | if (ret == 0) | |
3594 | progress++; | |
3595 | ||
3596 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3597 | /* | |
3598 | * our goal is to get our slot at the start or end of a leaf. If | |
3599 | * we've done so we're done | |
3600 | */ | |
3601 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3602 | return 0; | |
3603 | ||
e902baac | 3604 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3605 | return 0; |
3606 | ||
3607 | /* try to push all the items before our slot into the next leaf */ | |
3608 | slot = path->slots[0]; | |
263d3995 FM |
3609 | space_needed = data_size; |
3610 | if (slot > 0) | |
e902baac | 3611 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 3612 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3613 | if (ret < 0) |
3614 | return ret; | |
3615 | ||
3616 | if (ret == 0) | |
3617 | progress++; | |
3618 | ||
3619 | if (progress) | |
3620 | return 0; | |
3621 | return 1; | |
3622 | } | |
3623 | ||
74123bd7 CM |
3624 | /* |
3625 | * split the path's leaf in two, making sure there is at least data_size | |
3626 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3627 | * |
3628 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3629 | */ |
e02119d5 CM |
3630 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3631 | struct btrfs_root *root, | |
310712b2 | 3632 | const struct btrfs_key *ins_key, |
e02119d5 CM |
3633 | struct btrfs_path *path, int data_size, |
3634 | int extend) | |
be0e5c09 | 3635 | { |
5d4f98a2 | 3636 | struct btrfs_disk_key disk_key; |
5f39d397 | 3637 | struct extent_buffer *l; |
7518a238 | 3638 | u32 nritems; |
eb60ceac CM |
3639 | int mid; |
3640 | int slot; | |
5f39d397 | 3641 | struct extent_buffer *right; |
b7a0365e | 3642 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 3643 | int ret = 0; |
aa5d6bed | 3644 | int wret; |
5d4f98a2 | 3645 | int split; |
cc0c5538 | 3646 | int num_doubles = 0; |
99d8f83c | 3647 | int tried_avoid_double = 0; |
aa5d6bed | 3648 | |
a5719521 YZ |
3649 | l = path->nodes[0]; |
3650 | slot = path->slots[0]; | |
3212fa14 | 3651 | if (extend && data_size + btrfs_item_size(l, slot) + |
0b246afa | 3652 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
3653 | return -EOVERFLOW; |
3654 | ||
40689478 | 3655 | /* first try to make some room by pushing left and right */ |
33157e05 | 3656 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
3657 | int space_needed = data_size; |
3658 | ||
3659 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 3660 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3661 | |
3662 | wret = push_leaf_right(trans, root, path, space_needed, | |
3663 | space_needed, 0, 0); | |
d397712b | 3664 | if (wret < 0) |
eaee50e8 | 3665 | return wret; |
3685f791 | 3666 | if (wret) { |
263d3995 FM |
3667 | space_needed = data_size; |
3668 | if (slot > 0) | |
e902baac | 3669 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3670 | wret = push_leaf_left(trans, root, path, space_needed, |
3671 | space_needed, 0, (u32)-1); | |
3685f791 CM |
3672 | if (wret < 0) |
3673 | return wret; | |
3674 | } | |
3675 | l = path->nodes[0]; | |
aa5d6bed | 3676 | |
3685f791 | 3677 | /* did the pushes work? */ |
e902baac | 3678 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 3679 | return 0; |
3326d1b0 | 3680 | } |
aa5d6bed | 3681 | |
5c680ed6 | 3682 | if (!path->nodes[1]) { |
fdd99c72 | 3683 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3684 | if (ret) |
3685 | return ret; | |
3686 | } | |
cc0c5538 | 3687 | again: |
5d4f98a2 | 3688 | split = 1; |
cc0c5538 | 3689 | l = path->nodes[0]; |
eb60ceac | 3690 | slot = path->slots[0]; |
5f39d397 | 3691 | nritems = btrfs_header_nritems(l); |
d397712b | 3692 | mid = (nritems + 1) / 2; |
54aa1f4d | 3693 | |
5d4f98a2 YZ |
3694 | if (mid <= slot) { |
3695 | if (nritems == 1 || | |
3696 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 3697 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3698 | if (slot >= nritems) { |
3699 | split = 0; | |
3700 | } else { | |
3701 | mid = slot; | |
3702 | if (mid != nritems && | |
3703 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3704 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3705 | if (data_size && !tried_avoid_double) |
3706 | goto push_for_double; | |
5d4f98a2 YZ |
3707 | split = 2; |
3708 | } | |
3709 | } | |
3710 | } | |
3711 | } else { | |
3712 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 3713 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3714 | if (!extend && data_size && slot == 0) { |
3715 | split = 0; | |
3716 | } else if ((extend || !data_size) && slot == 0) { | |
3717 | mid = 1; | |
3718 | } else { | |
3719 | mid = slot; | |
3720 | if (mid != nritems && | |
3721 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3722 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3723 | if (data_size && !tried_avoid_double) |
3724 | goto push_for_double; | |
67871254 | 3725 | split = 2; |
5d4f98a2 YZ |
3726 | } |
3727 | } | |
3728 | } | |
3729 | } | |
3730 | ||
3731 | if (split == 0) | |
3732 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3733 | else | |
3734 | btrfs_item_key(l, &disk_key, mid); | |
3735 | ||
ca9d473a JB |
3736 | /* |
3737 | * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double | |
3738 | * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES | |
3739 | * subclasses, which is 8 at the time of this patch, and we've maxed it | |
3740 | * out. In the future we could add a | |
3741 | * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just | |
3742 | * use BTRFS_NESTING_NEW_ROOT. | |
3743 | */ | |
79bd3712 | 3744 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
60ea105a | 3745 | &disk_key, 0, l->start, 0, 0, |
79bd3712 FM |
3746 | num_doubles ? BTRFS_NESTING_NEW_ROOT : |
3747 | BTRFS_NESTING_SPLIT); | |
f0486c68 | 3748 | if (IS_ERR(right)) |
5f39d397 | 3749 | return PTR_ERR(right); |
f0486c68 | 3750 | |
02cd00fa | 3751 | root_add_used_bytes(root); |
5f39d397 | 3752 | |
5d4f98a2 YZ |
3753 | if (split == 0) { |
3754 | if (mid <= slot) { | |
3755 | btrfs_set_header_nritems(right, 0); | |
50b5d1fc FM |
3756 | ret = insert_ptr(trans, path, &disk_key, |
3757 | right->start, path->slots[1] + 1, 1); | |
3758 | if (ret < 0) { | |
3759 | btrfs_tree_unlock(right); | |
3760 | free_extent_buffer(right); | |
3761 | return ret; | |
3762 | } | |
5d4f98a2 YZ |
3763 | btrfs_tree_unlock(path->nodes[0]); |
3764 | free_extent_buffer(path->nodes[0]); | |
3765 | path->nodes[0] = right; | |
3766 | path->slots[0] = 0; | |
3767 | path->slots[1] += 1; | |
3768 | } else { | |
3769 | btrfs_set_header_nritems(right, 0); | |
50b5d1fc FM |
3770 | ret = insert_ptr(trans, path, &disk_key, |
3771 | right->start, path->slots[1], 1); | |
3772 | if (ret < 0) { | |
3773 | btrfs_tree_unlock(right); | |
3774 | free_extent_buffer(right); | |
3775 | return ret; | |
3776 | } | |
5d4f98a2 YZ |
3777 | btrfs_tree_unlock(path->nodes[0]); |
3778 | free_extent_buffer(path->nodes[0]); | |
3779 | path->nodes[0] = right; | |
3780 | path->slots[0] = 0; | |
143bede5 | 3781 | if (path->slots[1] == 0) |
50564b65 | 3782 | fixup_low_keys(trans, path, &disk_key, 1); |
d4dbff95 | 3783 | } |
196e0249 LB |
3784 | /* |
3785 | * We create a new leaf 'right' for the required ins_len and | |
3786 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
3787 | * the content of ins_len to 'right'. | |
3788 | */ | |
5d4f98a2 | 3789 | return ret; |
d4dbff95 | 3790 | } |
74123bd7 | 3791 | |
50b5d1fc FM |
3792 | ret = copy_for_split(trans, path, l, right, slot, mid, nritems); |
3793 | if (ret < 0) { | |
3794 | btrfs_tree_unlock(right); | |
3795 | free_extent_buffer(right); | |
3796 | return ret; | |
3797 | } | |
31840ae1 | 3798 | |
5d4f98a2 | 3799 | if (split == 2) { |
cc0c5538 CM |
3800 | BUG_ON(num_doubles != 0); |
3801 | num_doubles++; | |
3802 | goto again; | |
a429e513 | 3803 | } |
44871b1b | 3804 | |
143bede5 | 3805 | return 0; |
99d8f83c CM |
3806 | |
3807 | push_for_double: | |
3808 | push_for_double_split(trans, root, path, data_size); | |
3809 | tried_avoid_double = 1; | |
e902baac | 3810 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3811 | return 0; |
3812 | goto again; | |
be0e5c09 CM |
3813 | } |
3814 | ||
ad48fd75 YZ |
3815 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3816 | struct btrfs_root *root, | |
3817 | struct btrfs_path *path, int ins_len) | |
459931ec | 3818 | { |
ad48fd75 | 3819 | struct btrfs_key key; |
459931ec | 3820 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3821 | struct btrfs_file_extent_item *fi; |
3822 | u64 extent_len = 0; | |
3823 | u32 item_size; | |
3824 | int ret; | |
459931ec CM |
3825 | |
3826 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3827 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3828 | ||
3829 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3830 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3831 | ||
e902baac | 3832 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 3833 | return 0; |
459931ec | 3834 | |
3212fa14 | 3835 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ad48fd75 YZ |
3836 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3837 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3838 | struct btrfs_file_extent_item); | |
3839 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3840 | } | |
b3b4aa74 | 3841 | btrfs_release_path(path); |
459931ec | 3842 | |
459931ec | 3843 | path->keep_locks = 1; |
ad48fd75 YZ |
3844 | path->search_for_split = 1; |
3845 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3846 | path->search_for_split = 0; |
a8df6fe6 FM |
3847 | if (ret > 0) |
3848 | ret = -EAGAIN; | |
ad48fd75 YZ |
3849 | if (ret < 0) |
3850 | goto err; | |
459931ec | 3851 | |
ad48fd75 YZ |
3852 | ret = -EAGAIN; |
3853 | leaf = path->nodes[0]; | |
a8df6fe6 | 3854 | /* if our item isn't there, return now */ |
3212fa14 | 3855 | if (item_size != btrfs_item_size(leaf, path->slots[0])) |
ad48fd75 YZ |
3856 | goto err; |
3857 | ||
109f6aef | 3858 | /* the leaf has changed, it now has room. return now */ |
e902baac | 3859 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
3860 | goto err; |
3861 | ||
ad48fd75 YZ |
3862 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3863 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3864 | struct btrfs_file_extent_item); | |
3865 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3866 | goto err; | |
459931ec CM |
3867 | } |
3868 | ||
ad48fd75 | 3869 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3870 | if (ret) |
3871 | goto err; | |
459931ec | 3872 | |
ad48fd75 | 3873 | path->keep_locks = 0; |
b9473439 | 3874 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3875 | return 0; |
3876 | err: | |
3877 | path->keep_locks = 0; | |
3878 | return ret; | |
3879 | } | |
3880 | ||
50564b65 FM |
3881 | static noinline int split_item(struct btrfs_trans_handle *trans, |
3882 | struct btrfs_path *path, | |
310712b2 | 3883 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3884 | unsigned long split_offset) |
3885 | { | |
3886 | struct extent_buffer *leaf; | |
c91666b1 | 3887 | int orig_slot, slot; |
ad48fd75 YZ |
3888 | char *buf; |
3889 | u32 nritems; | |
3890 | u32 item_size; | |
3891 | u32 orig_offset; | |
3892 | struct btrfs_disk_key disk_key; | |
3893 | ||
b9473439 | 3894 | leaf = path->nodes[0]; |
7569141e FM |
3895 | /* |
3896 | * Shouldn't happen because the caller must have previously called | |
3897 | * setup_leaf_for_split() to make room for the new item in the leaf. | |
3898 | */ | |
3899 | if (WARN_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item))) | |
3900 | return -ENOSPC; | |
b9473439 | 3901 | |
c91666b1 | 3902 | orig_slot = path->slots[0]; |
3212fa14 JB |
3903 | orig_offset = btrfs_item_offset(leaf, path->slots[0]); |
3904 | item_size = btrfs_item_size(leaf, path->slots[0]); | |
459931ec | 3905 | |
459931ec | 3906 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3907 | if (!buf) |
3908 | return -ENOMEM; | |
3909 | ||
459931ec CM |
3910 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3911 | path->slots[0]), item_size); | |
459931ec | 3912 | |
ad48fd75 | 3913 | slot = path->slots[0] + 1; |
459931ec | 3914 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3915 | if (slot != nritems) { |
3916 | /* shift the items */ | |
637e3b48 | 3917 | memmove_leaf_items(leaf, slot + 1, slot, nritems - slot); |
459931ec CM |
3918 | } |
3919 | ||
3920 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3921 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3922 | ||
3212fa14 JB |
3923 | btrfs_set_item_offset(leaf, slot, orig_offset); |
3924 | btrfs_set_item_size(leaf, slot, item_size - split_offset); | |
459931ec | 3925 | |
3212fa14 | 3926 | btrfs_set_item_offset(leaf, orig_slot, |
c91666b1 | 3927 | orig_offset + item_size - split_offset); |
3212fa14 | 3928 | btrfs_set_item_size(leaf, orig_slot, split_offset); |
459931ec CM |
3929 | |
3930 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3931 | ||
3932 | /* write the data for the start of the original item */ | |
3933 | write_extent_buffer(leaf, buf, | |
3934 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3935 | split_offset); | |
3936 | ||
3937 | /* write the data for the new item */ | |
3938 | write_extent_buffer(leaf, buf + split_offset, | |
3939 | btrfs_item_ptr_offset(leaf, slot), | |
3940 | item_size - split_offset); | |
50564b65 | 3941 | btrfs_mark_buffer_dirty(trans, leaf); |
459931ec | 3942 | |
e902baac | 3943 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 3944 | kfree(buf); |
ad48fd75 YZ |
3945 | return 0; |
3946 | } | |
3947 | ||
3948 | /* | |
3949 | * This function splits a single item into two items, | |
3950 | * giving 'new_key' to the new item and splitting the | |
3951 | * old one at split_offset (from the start of the item). | |
3952 | * | |
3953 | * The path may be released by this operation. After | |
3954 | * the split, the path is pointing to the old item. The | |
3955 | * new item is going to be in the same node as the old one. | |
3956 | * | |
3957 | * Note, the item being split must be smaller enough to live alone on | |
3958 | * a tree block with room for one extra struct btrfs_item | |
3959 | * | |
3960 | * This allows us to split the item in place, keeping a lock on the | |
3961 | * leaf the entire time. | |
3962 | */ | |
3963 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3964 | struct btrfs_root *root, | |
3965 | struct btrfs_path *path, | |
310712b2 | 3966 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3967 | unsigned long split_offset) |
3968 | { | |
3969 | int ret; | |
3970 | ret = setup_leaf_for_split(trans, root, path, | |
3971 | sizeof(struct btrfs_item)); | |
3972 | if (ret) | |
3973 | return ret; | |
3974 | ||
50564b65 | 3975 | ret = split_item(trans, path, new_key, split_offset); |
459931ec CM |
3976 | return ret; |
3977 | } | |
3978 | ||
d352ac68 CM |
3979 | /* |
3980 | * make the item pointed to by the path smaller. new_size indicates | |
3981 | * how small to make it, and from_end tells us if we just chop bytes | |
3982 | * off the end of the item or if we shift the item to chop bytes off | |
3983 | * the front. | |
3984 | */ | |
50564b65 FM |
3985 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
3986 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 3987 | { |
b18c6685 | 3988 | int slot; |
5f39d397 | 3989 | struct extent_buffer *leaf; |
b18c6685 CM |
3990 | u32 nritems; |
3991 | unsigned int data_end; | |
3992 | unsigned int old_data_start; | |
3993 | unsigned int old_size; | |
3994 | unsigned int size_diff; | |
3995 | int i; | |
cfed81a0 CM |
3996 | struct btrfs_map_token token; |
3997 | ||
5f39d397 | 3998 | leaf = path->nodes[0]; |
179e29e4 CM |
3999 | slot = path->slots[0]; |
4000 | ||
3212fa14 | 4001 | old_size = btrfs_item_size(leaf, slot); |
179e29e4 | 4002 | if (old_size == new_size) |
143bede5 | 4003 | return; |
b18c6685 | 4004 | |
5f39d397 | 4005 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4006 | data_end = leaf_data_end(leaf); |
b18c6685 | 4007 | |
3212fa14 | 4008 | old_data_start = btrfs_item_offset(leaf, slot); |
179e29e4 | 4009 | |
b18c6685 CM |
4010 | size_diff = old_size - new_size; |
4011 | ||
4012 | BUG_ON(slot < 0); | |
4013 | BUG_ON(slot >= nritems); | |
4014 | ||
4015 | /* | |
4016 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4017 | */ | |
4018 | /* first correct the data pointers */ | |
c82f823c | 4019 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 4020 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4021 | u32 ioff; |
db94535d | 4022 | |
3212fa14 JB |
4023 | ioff = btrfs_token_item_offset(&token, i); |
4024 | btrfs_set_token_item_offset(&token, i, ioff + size_diff); | |
b18c6685 | 4025 | } |
db94535d | 4026 | |
b18c6685 | 4027 | /* shift the data */ |
179e29e4 | 4028 | if (from_end) { |
637e3b48 JB |
4029 | memmove_leaf_data(leaf, data_end + size_diff, data_end, |
4030 | old_data_start + new_size - data_end); | |
179e29e4 CM |
4031 | } else { |
4032 | struct btrfs_disk_key disk_key; | |
4033 | u64 offset; | |
4034 | ||
4035 | btrfs_item_key(leaf, &disk_key, slot); | |
4036 | ||
4037 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4038 | unsigned long ptr; | |
4039 | struct btrfs_file_extent_item *fi; | |
4040 | ||
4041 | fi = btrfs_item_ptr(leaf, slot, | |
4042 | struct btrfs_file_extent_item); | |
4043 | fi = (struct btrfs_file_extent_item *)( | |
4044 | (unsigned long)fi - size_diff); | |
4045 | ||
4046 | if (btrfs_file_extent_type(leaf, fi) == | |
4047 | BTRFS_FILE_EXTENT_INLINE) { | |
4048 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4049 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4050 | (unsigned long)fi, |
7ec20afb | 4051 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4052 | } |
4053 | } | |
4054 | ||
637e3b48 JB |
4055 | memmove_leaf_data(leaf, data_end + size_diff, data_end, |
4056 | old_data_start - data_end); | |
179e29e4 CM |
4057 | |
4058 | offset = btrfs_disk_key_offset(&disk_key); | |
4059 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4060 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4061 | if (slot == 0) | |
50564b65 | 4062 | fixup_low_keys(trans, path, &disk_key, 1); |
179e29e4 | 4063 | } |
5f39d397 | 4064 | |
3212fa14 | 4065 | btrfs_set_item_size(leaf, slot, new_size); |
50564b65 | 4066 | btrfs_mark_buffer_dirty(trans, leaf); |
b18c6685 | 4067 | |
e902baac | 4068 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4069 | btrfs_print_leaf(leaf); |
b18c6685 | 4070 | BUG(); |
5f39d397 | 4071 | } |
b18c6685 CM |
4072 | } |
4073 | ||
d352ac68 | 4074 | /* |
8f69dbd2 | 4075 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4076 | */ |
50564b65 FM |
4077 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
4078 | struct btrfs_path *path, u32 data_size) | |
6567e837 | 4079 | { |
6567e837 | 4080 | int slot; |
5f39d397 | 4081 | struct extent_buffer *leaf; |
6567e837 CM |
4082 | u32 nritems; |
4083 | unsigned int data_end; | |
4084 | unsigned int old_data; | |
4085 | unsigned int old_size; | |
4086 | int i; | |
cfed81a0 CM |
4087 | struct btrfs_map_token token; |
4088 | ||
5f39d397 | 4089 | leaf = path->nodes[0]; |
6567e837 | 4090 | |
5f39d397 | 4091 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4092 | data_end = leaf_data_end(leaf); |
6567e837 | 4093 | |
e902baac | 4094 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 4095 | btrfs_print_leaf(leaf); |
6567e837 | 4096 | BUG(); |
5f39d397 | 4097 | } |
6567e837 | 4098 | slot = path->slots[0]; |
dc2e724e | 4099 | old_data = btrfs_item_data_end(leaf, slot); |
6567e837 CM |
4100 | |
4101 | BUG_ON(slot < 0); | |
3326d1b0 | 4102 | if (slot >= nritems) { |
a4f78750 | 4103 | btrfs_print_leaf(leaf); |
c71dd880 | 4104 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 4105 | slot, nritems); |
290342f6 | 4106 | BUG(); |
3326d1b0 | 4107 | } |
6567e837 CM |
4108 | |
4109 | /* | |
4110 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4111 | */ | |
4112 | /* first correct the data pointers */ | |
c82f823c | 4113 | btrfs_init_map_token(&token, leaf); |
6567e837 | 4114 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4115 | u32 ioff; |
db94535d | 4116 | |
3212fa14 JB |
4117 | ioff = btrfs_token_item_offset(&token, i); |
4118 | btrfs_set_token_item_offset(&token, i, ioff - data_size); | |
6567e837 | 4119 | } |
5f39d397 | 4120 | |
6567e837 | 4121 | /* shift the data */ |
637e3b48 JB |
4122 | memmove_leaf_data(leaf, data_end - data_size, data_end, |
4123 | old_data - data_end); | |
5f39d397 | 4124 | |
6567e837 | 4125 | data_end = old_data; |
3212fa14 JB |
4126 | old_size = btrfs_item_size(leaf, slot); |
4127 | btrfs_set_item_size(leaf, slot, old_size + data_size); | |
50564b65 | 4128 | btrfs_mark_buffer_dirty(trans, leaf); |
6567e837 | 4129 | |
e902baac | 4130 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4131 | btrfs_print_leaf(leaf); |
6567e837 | 4132 | BUG(); |
5f39d397 | 4133 | } |
6567e837 CM |
4134 | } |
4135 | ||
43dd529a DS |
4136 | /* |
4137 | * Make space in the node before inserting one or more items. | |
da9ffb24 | 4138 | * |
50564b65 | 4139 | * @trans: transaction handle |
da9ffb24 NB |
4140 | * @root: root we are inserting items to |
4141 | * @path: points to the leaf/slot where we are going to insert new items | |
b7ef5f3a | 4142 | * @batch: information about the batch of items to insert |
43dd529a DS |
4143 | * |
4144 | * Main purpose is to save stack depth by doing the bulk of the work in a | |
4145 | * function that doesn't call btrfs_search_slot | |
74123bd7 | 4146 | */ |
50564b65 FM |
4147 | static void setup_items_for_insert(struct btrfs_trans_handle *trans, |
4148 | struct btrfs_root *root, struct btrfs_path *path, | |
f0641656 | 4149 | const struct btrfs_item_batch *batch) |
be0e5c09 | 4150 | { |
0b246afa | 4151 | struct btrfs_fs_info *fs_info = root->fs_info; |
9c58309d | 4152 | int i; |
7518a238 | 4153 | u32 nritems; |
be0e5c09 | 4154 | unsigned int data_end; |
e2fa7227 | 4155 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4156 | struct extent_buffer *leaf; |
4157 | int slot; | |
cfed81a0 | 4158 | struct btrfs_map_token token; |
fc0d82e1 | 4159 | u32 total_size; |
cfed81a0 | 4160 | |
b7ef5f3a FM |
4161 | /* |
4162 | * Before anything else, update keys in the parent and other ancestors | |
4163 | * if needed, then release the write locks on them, so that other tasks | |
4164 | * can use them while we modify the leaf. | |
4165 | */ | |
24cdc847 | 4166 | if (path->slots[0] == 0) { |
b7ef5f3a | 4167 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); |
50564b65 | 4168 | fixup_low_keys(trans, path, &disk_key, 1); |
24cdc847 FM |
4169 | } |
4170 | btrfs_unlock_up_safe(path, 1); | |
4171 | ||
5f39d397 | 4172 | leaf = path->nodes[0]; |
44871b1b | 4173 | slot = path->slots[0]; |
74123bd7 | 4174 | |
5f39d397 | 4175 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4176 | data_end = leaf_data_end(leaf); |
b7ef5f3a | 4177 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
eb60ceac | 4178 | |
e902baac | 4179 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4180 | btrfs_print_leaf(leaf); |
0b246afa | 4181 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4182 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4183 | BUG(); |
d4dbff95 | 4184 | } |
5f39d397 | 4185 | |
c82f823c | 4186 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4187 | if (slot != nritems) { |
dc2e724e | 4188 | unsigned int old_data = btrfs_item_data_end(leaf, slot); |
be0e5c09 | 4189 | |
5f39d397 | 4190 | if (old_data < data_end) { |
a4f78750 | 4191 | btrfs_print_leaf(leaf); |
7269ddd2 NB |
4192 | btrfs_crit(fs_info, |
4193 | "item at slot %d with data offset %u beyond data end of leaf %u", | |
5d163e0e | 4194 | slot, old_data, data_end); |
290342f6 | 4195 | BUG(); |
5f39d397 | 4196 | } |
be0e5c09 CM |
4197 | /* |
4198 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4199 | */ | |
4200 | /* first correct the data pointers */ | |
0783fcfc | 4201 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4202 | u32 ioff; |
db94535d | 4203 | |
3212fa14 JB |
4204 | ioff = btrfs_token_item_offset(&token, i); |
4205 | btrfs_set_token_item_offset(&token, i, | |
74794207 | 4206 | ioff - batch->total_data_size); |
0783fcfc | 4207 | } |
be0e5c09 | 4208 | /* shift the items */ |
637e3b48 | 4209 | memmove_leaf_items(leaf, slot + batch->nr, slot, nritems - slot); |
be0e5c09 CM |
4210 | |
4211 | /* shift the data */ | |
637e3b48 JB |
4212 | memmove_leaf_data(leaf, data_end - batch->total_data_size, |
4213 | data_end, old_data - data_end); | |
be0e5c09 CM |
4214 | data_end = old_data; |
4215 | } | |
5f39d397 | 4216 | |
62e2749e | 4217 | /* setup the item for the new data */ |
b7ef5f3a FM |
4218 | for (i = 0; i < batch->nr; i++) { |
4219 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); | |
9c58309d | 4220 | btrfs_set_item_key(leaf, &disk_key, slot + i); |
b7ef5f3a | 4221 | data_end -= batch->data_sizes[i]; |
3212fa14 JB |
4222 | btrfs_set_token_item_offset(&token, slot + i, data_end); |
4223 | btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); | |
9c58309d | 4224 | } |
44871b1b | 4225 | |
b7ef5f3a | 4226 | btrfs_set_header_nritems(leaf, nritems + batch->nr); |
50564b65 | 4227 | btrfs_mark_buffer_dirty(trans, leaf); |
aa5d6bed | 4228 | |
e902baac | 4229 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4230 | btrfs_print_leaf(leaf); |
be0e5c09 | 4231 | BUG(); |
5f39d397 | 4232 | } |
44871b1b CM |
4233 | } |
4234 | ||
f0641656 FM |
4235 | /* |
4236 | * Insert a new item into a leaf. | |
4237 | * | |
50564b65 | 4238 | * @trans: Transaction handle. |
f0641656 FM |
4239 | * @root: The root of the btree. |
4240 | * @path: A path pointing to the target leaf and slot. | |
4241 | * @key: The key of the new item. | |
4242 | * @data_size: The size of the data associated with the new key. | |
4243 | */ | |
50564b65 FM |
4244 | void btrfs_setup_item_for_insert(struct btrfs_trans_handle *trans, |
4245 | struct btrfs_root *root, | |
f0641656 FM |
4246 | struct btrfs_path *path, |
4247 | const struct btrfs_key *key, | |
4248 | u32 data_size) | |
4249 | { | |
4250 | struct btrfs_item_batch batch; | |
4251 | ||
4252 | batch.keys = key; | |
4253 | batch.data_sizes = &data_size; | |
4254 | batch.total_data_size = data_size; | |
4255 | batch.nr = 1; | |
4256 | ||
50564b65 | 4257 | setup_items_for_insert(trans, root, path, &batch); |
f0641656 FM |
4258 | } |
4259 | ||
44871b1b CM |
4260 | /* |
4261 | * Given a key and some data, insert items into the tree. | |
4262 | * This does all the path init required, making room in the tree if needed. | |
4263 | */ | |
4264 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4265 | struct btrfs_root *root, | |
4266 | struct btrfs_path *path, | |
b7ef5f3a | 4267 | const struct btrfs_item_batch *batch) |
44871b1b | 4268 | { |
44871b1b CM |
4269 | int ret = 0; |
4270 | int slot; | |
b7ef5f3a | 4271 | u32 total_size; |
44871b1b | 4272 | |
b7ef5f3a FM |
4273 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
4274 | ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); | |
44871b1b CM |
4275 | if (ret == 0) |
4276 | return -EEXIST; | |
4277 | if (ret < 0) | |
143bede5 | 4278 | return ret; |
44871b1b | 4279 | |
44871b1b CM |
4280 | slot = path->slots[0]; |
4281 | BUG_ON(slot < 0); | |
4282 | ||
50564b65 | 4283 | setup_items_for_insert(trans, root, path, batch); |
143bede5 | 4284 | return 0; |
62e2749e CM |
4285 | } |
4286 | ||
4287 | /* | |
4288 | * Given a key and some data, insert an item into the tree. | |
4289 | * This does all the path init required, making room in the tree if needed. | |
4290 | */ | |
310712b2 OS |
4291 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4292 | const struct btrfs_key *cpu_key, void *data, | |
4293 | u32 data_size) | |
62e2749e CM |
4294 | { |
4295 | int ret = 0; | |
2c90e5d6 | 4296 | struct btrfs_path *path; |
5f39d397 CM |
4297 | struct extent_buffer *leaf; |
4298 | unsigned long ptr; | |
62e2749e | 4299 | |
2c90e5d6 | 4300 | path = btrfs_alloc_path(); |
db5b493a TI |
4301 | if (!path) |
4302 | return -ENOMEM; | |
2c90e5d6 | 4303 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4304 | if (!ret) { |
5f39d397 CM |
4305 | leaf = path->nodes[0]; |
4306 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4307 | write_extent_buffer(leaf, data, ptr, data_size); | |
50564b65 | 4308 | btrfs_mark_buffer_dirty(trans, leaf); |
62e2749e | 4309 | } |
2c90e5d6 | 4310 | btrfs_free_path(path); |
aa5d6bed | 4311 | return ret; |
be0e5c09 CM |
4312 | } |
4313 | ||
f0641656 FM |
4314 | /* |
4315 | * This function duplicates an item, giving 'new_key' to the new item. | |
4316 | * It guarantees both items live in the same tree leaf and the new item is | |
4317 | * contiguous with the original item. | |
4318 | * | |
4319 | * This allows us to split a file extent in place, keeping a lock on the leaf | |
4320 | * the entire time. | |
4321 | */ | |
4322 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4323 | struct btrfs_root *root, | |
4324 | struct btrfs_path *path, | |
4325 | const struct btrfs_key *new_key) | |
4326 | { | |
4327 | struct extent_buffer *leaf; | |
4328 | int ret; | |
4329 | u32 item_size; | |
4330 | ||
4331 | leaf = path->nodes[0]; | |
3212fa14 | 4332 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f0641656 FM |
4333 | ret = setup_leaf_for_split(trans, root, path, |
4334 | item_size + sizeof(struct btrfs_item)); | |
4335 | if (ret) | |
4336 | return ret; | |
4337 | ||
4338 | path->slots[0]++; | |
50564b65 | 4339 | btrfs_setup_item_for_insert(trans, root, path, new_key, item_size); |
f0641656 FM |
4340 | leaf = path->nodes[0]; |
4341 | memcpy_extent_buffer(leaf, | |
4342 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4343 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4344 | item_size); | |
4345 | return 0; | |
4346 | } | |
4347 | ||
74123bd7 | 4348 | /* |
5de08d7d | 4349 | * delete the pointer from a given node. |
74123bd7 | 4350 | * |
d352ac68 CM |
4351 | * the tree should have been previously balanced so the deletion does not |
4352 | * empty a node. | |
016f9d0b JB |
4353 | * |
4354 | * This is exported for use inside btrfs-progs, don't un-export it. | |
74123bd7 | 4355 | */ |
751a2761 FM |
4356 | int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4357 | struct btrfs_path *path, int level, int slot) | |
be0e5c09 | 4358 | { |
5f39d397 | 4359 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4360 | u32 nritems; |
f3ea38da | 4361 | int ret; |
be0e5c09 | 4362 | |
5f39d397 | 4363 | nritems = btrfs_header_nritems(parent); |
d397712b | 4364 | if (slot != nritems - 1) { |
bf1d3425 | 4365 | if (level) { |
f3a84ccd FM |
4366 | ret = btrfs_tree_mod_log_insert_move(parent, slot, |
4367 | slot + 1, nritems - slot - 1); | |
751a2761 FM |
4368 | if (ret < 0) { |
4369 | btrfs_abort_transaction(trans, ret); | |
4370 | return ret; | |
4371 | } | |
bf1d3425 | 4372 | } |
5f39d397 | 4373 | memmove_extent_buffer(parent, |
e23efd8e JB |
4374 | btrfs_node_key_ptr_offset(parent, slot), |
4375 | btrfs_node_key_ptr_offset(parent, slot + 1), | |
d6025579 CM |
4376 | sizeof(struct btrfs_key_ptr) * |
4377 | (nritems - slot - 1)); | |
57ba86c0 | 4378 | } else if (level) { |
f3a84ccd | 4379 | ret = btrfs_tree_mod_log_insert_key(parent, slot, |
33cff222 | 4380 | BTRFS_MOD_LOG_KEY_REMOVE); |
751a2761 FM |
4381 | if (ret < 0) { |
4382 | btrfs_abort_transaction(trans, ret); | |
4383 | return ret; | |
4384 | } | |
bb803951 | 4385 | } |
f3ea38da | 4386 | |
7518a238 | 4387 | nritems--; |
5f39d397 | 4388 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4389 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4390 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4391 | /* just turn the root into a leaf and break */ |
5f39d397 | 4392 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4393 | } else if (slot == 0) { |
5f39d397 CM |
4394 | struct btrfs_disk_key disk_key; |
4395 | ||
4396 | btrfs_node_key(parent, &disk_key, 0); | |
50564b65 | 4397 | fixup_low_keys(trans, path, &disk_key, level + 1); |
be0e5c09 | 4398 | } |
50564b65 | 4399 | btrfs_mark_buffer_dirty(trans, parent); |
751a2761 | 4400 | return 0; |
be0e5c09 CM |
4401 | } |
4402 | ||
323ac95b CM |
4403 | /* |
4404 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4405 | * path->nodes[1]. |
323ac95b CM |
4406 | * |
4407 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4408 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4409 | * | |
4410 | * The path must have already been setup for deleting the leaf, including | |
4411 | * all the proper balancing. path->nodes[1] must be locked. | |
4412 | */ | |
751a2761 FM |
4413 | static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4414 | struct btrfs_root *root, | |
4415 | struct btrfs_path *path, | |
4416 | struct extent_buffer *leaf) | |
323ac95b | 4417 | { |
751a2761 FM |
4418 | int ret; |
4419 | ||
5d4f98a2 | 4420 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
751a2761 FM |
4421 | ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]); |
4422 | if (ret < 0) | |
4423 | return ret; | |
323ac95b | 4424 | |
4d081c41 CM |
4425 | /* |
4426 | * btrfs_free_extent is expensive, we want to make sure we | |
4427 | * aren't holding any locks when we call it | |
4428 | */ | |
4429 | btrfs_unlock_up_safe(path, 0); | |
4430 | ||
02cd00fa | 4431 | root_sub_used_bytes(root); |
f0486c68 | 4432 | |
67439dad | 4433 | atomic_inc(&leaf->refs); |
7a163608 | 4434 | btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); |
3083ee2e | 4435 | free_extent_buffer_stale(leaf); |
751a2761 | 4436 | return 0; |
323ac95b | 4437 | } |
74123bd7 CM |
4438 | /* |
4439 | * delete the item at the leaf level in path. If that empties | |
4440 | * the leaf, remove it from the tree | |
4441 | */ | |
85e21bac CM |
4442 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4443 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4444 | { |
0b246afa | 4445 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4446 | struct extent_buffer *leaf; |
aa5d6bed CM |
4447 | int ret = 0; |
4448 | int wret; | |
7518a238 | 4449 | u32 nritems; |
be0e5c09 | 4450 | |
5f39d397 | 4451 | leaf = path->nodes[0]; |
5f39d397 | 4452 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4453 | |
85e21bac | 4454 | if (slot + nr != nritems) { |
0cae23b6 FM |
4455 | const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); |
4456 | const int data_end = leaf_data_end(leaf); | |
c82f823c | 4457 | struct btrfs_map_token token; |
0cae23b6 FM |
4458 | u32 dsize = 0; |
4459 | int i; | |
4460 | ||
4461 | for (i = 0; i < nr; i++) | |
4462 | dsize += btrfs_item_size(leaf, slot + i); | |
5f39d397 | 4463 | |
637e3b48 JB |
4464 | memmove_leaf_data(leaf, data_end + dsize, data_end, |
4465 | last_off - data_end); | |
5f39d397 | 4466 | |
c82f823c | 4467 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4468 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4469 | u32 ioff; |
db94535d | 4470 | |
3212fa14 JB |
4471 | ioff = btrfs_token_item_offset(&token, i); |
4472 | btrfs_set_token_item_offset(&token, i, ioff + dsize); | |
0783fcfc | 4473 | } |
db94535d | 4474 | |
637e3b48 | 4475 | memmove_leaf_items(leaf, slot, slot + nr, nritems - slot - nr); |
be0e5c09 | 4476 | } |
85e21bac CM |
4477 | btrfs_set_header_nritems(leaf, nritems - nr); |
4478 | nritems -= nr; | |
5f39d397 | 4479 | |
74123bd7 | 4480 | /* delete the leaf if we've emptied it */ |
7518a238 | 4481 | if (nritems == 0) { |
5f39d397 CM |
4482 | if (leaf == root->node) { |
4483 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4484 | } else { |
190a8339 | 4485 | btrfs_clear_buffer_dirty(trans, leaf); |
751a2761 FM |
4486 | ret = btrfs_del_leaf(trans, root, path, leaf); |
4487 | if (ret < 0) | |
4488 | return ret; | |
9a8dd150 | 4489 | } |
be0e5c09 | 4490 | } else { |
7518a238 | 4491 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4492 | if (slot == 0) { |
5f39d397 CM |
4493 | struct btrfs_disk_key disk_key; |
4494 | ||
4495 | btrfs_item_key(leaf, &disk_key, 0); | |
50564b65 | 4496 | fixup_low_keys(trans, path, &disk_key, 1); |
aa5d6bed | 4497 | } |
aa5d6bed | 4498 | |
7c4063d1 FM |
4499 | /* |
4500 | * Try to delete the leaf if it is mostly empty. We do this by | |
4501 | * trying to move all its items into its left and right neighbours. | |
4502 | * If we can't move all the items, then we don't delete it - it's | |
4503 | * not ideal, but future insertions might fill the leaf with more | |
4504 | * items, or items from other leaves might be moved later into our | |
4505 | * leaf due to deletions on those leaves. | |
4506 | */ | |
0b246afa | 4507 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
7c4063d1 FM |
4508 | u32 min_push_space; |
4509 | ||
be0e5c09 CM |
4510 | /* push_leaf_left fixes the path. |
4511 | * make sure the path still points to our leaf | |
016f9d0b | 4512 | * for possible call to btrfs_del_ptr below |
be0e5c09 | 4513 | */ |
4920c9ac | 4514 | slot = path->slots[1]; |
67439dad | 4515 | atomic_inc(&leaf->refs); |
7c4063d1 FM |
4516 | /* |
4517 | * We want to be able to at least push one item to the | |
4518 | * left neighbour leaf, and that's the first item. | |
4519 | */ | |
4520 | min_push_space = sizeof(struct btrfs_item) + | |
4521 | btrfs_item_size(leaf, 0); | |
4522 | wret = push_leaf_left(trans, root, path, 0, | |
4523 | min_push_space, 1, (u32)-1); | |
54aa1f4d | 4524 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4525 | ret = wret; |
5f39d397 CM |
4526 | |
4527 | if (path->nodes[0] == leaf && | |
4528 | btrfs_header_nritems(leaf)) { | |
7c4063d1 FM |
4529 | /* |
4530 | * If we were not able to push all items from our | |
4531 | * leaf to its left neighbour, then attempt to | |
4532 | * either push all the remaining items to the | |
4533 | * right neighbour or none. There's no advantage | |
4534 | * in pushing only some items, instead of all, as | |
4535 | * it's pointless to end up with a leaf having | |
4536 | * too few items while the neighbours can be full | |
4537 | * or nearly full. | |
4538 | */ | |
4539 | nritems = btrfs_header_nritems(leaf); | |
4540 | min_push_space = leaf_space_used(leaf, 0, nritems); | |
4541 | wret = push_leaf_right(trans, root, path, 0, | |
4542 | min_push_space, 1, 0); | |
54aa1f4d | 4543 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4544 | ret = wret; |
4545 | } | |
5f39d397 CM |
4546 | |
4547 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4548 | path->slots[1] = slot; |
751a2761 FM |
4549 | ret = btrfs_del_leaf(trans, root, path, leaf); |
4550 | if (ret < 0) | |
4551 | return ret; | |
5f39d397 | 4552 | free_extent_buffer(leaf); |
143bede5 | 4553 | ret = 0; |
5de08d7d | 4554 | } else { |
925baedd CM |
4555 | /* if we're still in the path, make sure |
4556 | * we're dirty. Otherwise, one of the | |
4557 | * push_leaf functions must have already | |
4558 | * dirtied this buffer | |
4559 | */ | |
4560 | if (path->nodes[0] == leaf) | |
50564b65 | 4561 | btrfs_mark_buffer_dirty(trans, leaf); |
5f39d397 | 4562 | free_extent_buffer(leaf); |
be0e5c09 | 4563 | } |
d5719762 | 4564 | } else { |
50564b65 | 4565 | btrfs_mark_buffer_dirty(trans, leaf); |
be0e5c09 CM |
4566 | } |
4567 | } | |
aa5d6bed | 4568 | return ret; |
be0e5c09 CM |
4569 | } |
4570 | ||
3f157a2f CM |
4571 | /* |
4572 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4573 | * for nodes or leaves that are have a minimum transaction id. |
4574 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4575 | * |
4576 | * This does not cow, but it does stuff the starting key it finds back | |
4577 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4578 | * key and get a writable path. | |
4579 | * | |
3f157a2f CM |
4580 | * This honors path->lowest_level to prevent descent past a given level |
4581 | * of the tree. | |
4582 | * | |
d352ac68 CM |
4583 | * min_trans indicates the oldest transaction that you are interested |
4584 | * in walking through. Any nodes or leaves older than min_trans are | |
4585 | * skipped over (without reading them). | |
4586 | * | |
3f157a2f CM |
4587 | * returns zero if something useful was found, < 0 on error and 1 if there |
4588 | * was nothing in the tree that matched the search criteria. | |
4589 | */ | |
4590 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 4591 | struct btrfs_path *path, |
3f157a2f CM |
4592 | u64 min_trans) |
4593 | { | |
4594 | struct extent_buffer *cur; | |
4595 | struct btrfs_key found_key; | |
4596 | int slot; | |
9652480b | 4597 | int sret; |
3f157a2f CM |
4598 | u32 nritems; |
4599 | int level; | |
4600 | int ret = 1; | |
f98de9b9 | 4601 | int keep_locks = path->keep_locks; |
3f157a2f | 4602 | |
c922b016 | 4603 | ASSERT(!path->nowait); |
f98de9b9 | 4604 | path->keep_locks = 1; |
3f157a2f | 4605 | again: |
bd681513 | 4606 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4607 | level = btrfs_header_level(cur); |
e02119d5 | 4608 | WARN_ON(path->nodes[level]); |
3f157a2f | 4609 | path->nodes[level] = cur; |
bd681513 | 4610 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4611 | |
4612 | if (btrfs_header_generation(cur) < min_trans) { | |
4613 | ret = 1; | |
4614 | goto out; | |
4615 | } | |
d397712b | 4616 | while (1) { |
3f157a2f CM |
4617 | nritems = btrfs_header_nritems(cur); |
4618 | level = btrfs_header_level(cur); | |
fdf8d595 | 4619 | sret = btrfs_bin_search(cur, 0, min_key, &slot); |
cbca7d59 FM |
4620 | if (sret < 0) { |
4621 | ret = sret; | |
4622 | goto out; | |
4623 | } | |
3f157a2f | 4624 | |
323ac95b CM |
4625 | /* at the lowest level, we're done, setup the path and exit */ |
4626 | if (level == path->lowest_level) { | |
e02119d5 CM |
4627 | if (slot >= nritems) |
4628 | goto find_next_key; | |
3f157a2f CM |
4629 | ret = 0; |
4630 | path->slots[level] = slot; | |
4631 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4632 | goto out; | |
4633 | } | |
9652480b Y |
4634 | if (sret && slot > 0) |
4635 | slot--; | |
3f157a2f | 4636 | /* |
de78b51a | 4637 | * check this node pointer against the min_trans parameters. |
260db43c | 4638 | * If it is too old, skip to the next one. |
3f157a2f | 4639 | */ |
d397712b | 4640 | while (slot < nritems) { |
3f157a2f | 4641 | u64 gen; |
e02119d5 | 4642 | |
3f157a2f CM |
4643 | gen = btrfs_node_ptr_generation(cur, slot); |
4644 | if (gen < min_trans) { | |
4645 | slot++; | |
4646 | continue; | |
4647 | } | |
de78b51a | 4648 | break; |
3f157a2f | 4649 | } |
e02119d5 | 4650 | find_next_key: |
3f157a2f CM |
4651 | /* |
4652 | * we didn't find a candidate key in this node, walk forward | |
4653 | * and find another one | |
4654 | */ | |
4655 | if (slot >= nritems) { | |
e02119d5 CM |
4656 | path->slots[level] = slot; |
4657 | sret = btrfs_find_next_key(root, path, min_key, level, | |
de78b51a | 4658 | min_trans); |
e02119d5 | 4659 | if (sret == 0) { |
b3b4aa74 | 4660 | btrfs_release_path(path); |
3f157a2f CM |
4661 | goto again; |
4662 | } else { | |
4663 | goto out; | |
4664 | } | |
4665 | } | |
4666 | /* save our key for returning back */ | |
4667 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4668 | path->slots[level] = slot; | |
4669 | if (level == path->lowest_level) { | |
4670 | ret = 0; | |
3f157a2f CM |
4671 | goto out; |
4672 | } | |
4b231ae4 | 4673 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
4674 | if (IS_ERR(cur)) { |
4675 | ret = PTR_ERR(cur); | |
4676 | goto out; | |
4677 | } | |
3f157a2f | 4678 | |
bd681513 | 4679 | btrfs_tree_read_lock(cur); |
b4ce94de | 4680 | |
bd681513 | 4681 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4682 | path->nodes[level - 1] = cur; |
f7c79f30 | 4683 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4684 | } |
4685 | out: | |
f98de9b9 FM |
4686 | path->keep_locks = keep_locks; |
4687 | if (ret == 0) { | |
4688 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
3f157a2f | 4689 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 4690 | } |
3f157a2f CM |
4691 | return ret; |
4692 | } | |
4693 | ||
4694 | /* | |
4695 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4696 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 4697 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
4698 | * |
4699 | * 0 is returned if another key is found, < 0 if there are any errors | |
4700 | * and 1 is returned if there are no higher keys in the tree | |
4701 | * | |
4702 | * path->keep_locks should be set to 1 on the search made before | |
4703 | * calling this function. | |
4704 | */ | |
e7a84565 | 4705 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 4706 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 4707 | { |
e7a84565 CM |
4708 | int slot; |
4709 | struct extent_buffer *c; | |
4710 | ||
6a9fb468 | 4711 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 4712 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4713 | if (!path->nodes[level]) |
4714 | return 1; | |
4715 | ||
4716 | slot = path->slots[level] + 1; | |
4717 | c = path->nodes[level]; | |
3f157a2f | 4718 | next: |
e7a84565 | 4719 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4720 | int ret; |
4721 | int orig_lowest; | |
4722 | struct btrfs_key cur_key; | |
4723 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4724 | !path->nodes[level + 1]) | |
e7a84565 | 4725 | return 1; |
33c66f43 | 4726 | |
6a9fb468 | 4727 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
4728 | level++; |
4729 | continue; | |
4730 | } | |
4731 | ||
4732 | slot = btrfs_header_nritems(c) - 1; | |
4733 | if (level == 0) | |
4734 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4735 | else | |
4736 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4737 | ||
4738 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4739 | btrfs_release_path(path); |
33c66f43 YZ |
4740 | path->lowest_level = level; |
4741 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4742 | 0, 0); | |
4743 | path->lowest_level = orig_lowest; | |
4744 | if (ret < 0) | |
4745 | return ret; | |
4746 | ||
4747 | c = path->nodes[level]; | |
4748 | slot = path->slots[level]; | |
4749 | if (ret == 0) | |
4750 | slot++; | |
4751 | goto next; | |
e7a84565 | 4752 | } |
33c66f43 | 4753 | |
e7a84565 CM |
4754 | if (level == 0) |
4755 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 4756 | else { |
3f157a2f CM |
4757 | u64 gen = btrfs_node_ptr_generation(c, slot); |
4758 | ||
3f157a2f CM |
4759 | if (gen < min_trans) { |
4760 | slot++; | |
4761 | goto next; | |
4762 | } | |
e7a84565 | 4763 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4764 | } |
e7a84565 CM |
4765 | return 0; |
4766 | } | |
4767 | return 1; | |
4768 | } | |
4769 | ||
3d7806ec JS |
4770 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, |
4771 | u64 time_seq) | |
d97e63b6 CM |
4772 | { |
4773 | int slot; | |
8e73f275 | 4774 | int level; |
5f39d397 | 4775 | struct extent_buffer *c; |
8e73f275 | 4776 | struct extent_buffer *next; |
d96b3424 | 4777 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 4778 | struct btrfs_key key; |
d96b3424 | 4779 | bool need_commit_sem = false; |
925baedd CM |
4780 | u32 nritems; |
4781 | int ret; | |
0e46318d | 4782 | int i; |
925baedd | 4783 | |
bdcdd86c FM |
4784 | /* |
4785 | * The nowait semantics are used only for write paths, where we don't | |
4786 | * use the tree mod log and sequence numbers. | |
4787 | */ | |
4788 | if (time_seq) | |
4789 | ASSERT(!path->nowait); | |
c922b016 | 4790 | |
925baedd | 4791 | nritems = btrfs_header_nritems(path->nodes[0]); |
d397712b | 4792 | if (nritems == 0) |
925baedd | 4793 | return 1; |
925baedd | 4794 | |
8e73f275 CM |
4795 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4796 | again: | |
4797 | level = 1; | |
4798 | next = NULL; | |
b3b4aa74 | 4799 | btrfs_release_path(path); |
8e73f275 | 4800 | |
a2135011 | 4801 | path->keep_locks = 1; |
8e73f275 | 4802 | |
d96b3424 | 4803 | if (time_seq) { |
3d7806ec | 4804 | ret = btrfs_search_old_slot(root, &key, path, time_seq); |
d96b3424 FM |
4805 | } else { |
4806 | if (path->need_commit_sem) { | |
4807 | path->need_commit_sem = 0; | |
4808 | need_commit_sem = true; | |
bdcdd86c FM |
4809 | if (path->nowait) { |
4810 | if (!down_read_trylock(&fs_info->commit_root_sem)) { | |
4811 | ret = -EAGAIN; | |
4812 | goto done; | |
4813 | } | |
4814 | } else { | |
4815 | down_read(&fs_info->commit_root_sem); | |
4816 | } | |
d96b3424 | 4817 | } |
3d7806ec | 4818 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
d96b3424 | 4819 | } |
925baedd CM |
4820 | path->keep_locks = 0; |
4821 | ||
4822 | if (ret < 0) | |
d96b3424 | 4823 | goto done; |
925baedd | 4824 | |
a2135011 | 4825 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4826 | /* |
4827 | * by releasing the path above we dropped all our locks. A balance | |
4828 | * could have added more items next to the key that used to be | |
4829 | * at the very end of the block. So, check again here and | |
4830 | * advance the path if there are now more items available. | |
4831 | */ | |
a2135011 | 4832 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4833 | if (ret == 0) |
4834 | path->slots[0]++; | |
8e73f275 | 4835 | ret = 0; |
925baedd CM |
4836 | goto done; |
4837 | } | |
0b43e04f LB |
4838 | /* |
4839 | * So the above check misses one case: | |
4840 | * - after releasing the path above, someone has removed the item that | |
4841 | * used to be at the very end of the block, and balance between leafs | |
4842 | * gets another one with bigger key.offset to replace it. | |
4843 | * | |
4844 | * This one should be returned as well, or we can get leaf corruption | |
4845 | * later(esp. in __btrfs_drop_extents()). | |
4846 | * | |
4847 | * And a bit more explanation about this check, | |
4848 | * with ret > 0, the key isn't found, the path points to the slot | |
4849 | * where it should be inserted, so the path->slots[0] item must be the | |
4850 | * bigger one. | |
4851 | */ | |
4852 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
4853 | ret = 0; | |
4854 | goto done; | |
4855 | } | |
d97e63b6 | 4856 | |
d397712b | 4857 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4858 | if (!path->nodes[level]) { |
4859 | ret = 1; | |
4860 | goto done; | |
4861 | } | |
5f39d397 | 4862 | |
d97e63b6 CM |
4863 | slot = path->slots[level] + 1; |
4864 | c = path->nodes[level]; | |
5f39d397 | 4865 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4866 | level++; |
8e73f275 CM |
4867 | if (level == BTRFS_MAX_LEVEL) { |
4868 | ret = 1; | |
4869 | goto done; | |
4870 | } | |
d97e63b6 CM |
4871 | continue; |
4872 | } | |
5f39d397 | 4873 | |
0e46318d JB |
4874 | |
4875 | /* | |
4876 | * Our current level is where we're going to start from, and to | |
4877 | * make sure lockdep doesn't complain we need to drop our locks | |
4878 | * and nodes from 0 to our current level. | |
4879 | */ | |
4880 | for (i = 0; i < level; i++) { | |
4881 | if (path->locks[level]) { | |
4882 | btrfs_tree_read_unlock(path->nodes[i]); | |
4883 | path->locks[i] = 0; | |
4884 | } | |
4885 | free_extent_buffer(path->nodes[i]); | |
4886 | path->nodes[i] = NULL; | |
925baedd | 4887 | } |
5f39d397 | 4888 | |
8e73f275 | 4889 | next = c; |
d07b8528 | 4890 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4891 | slot, &key); |
bdcdd86c | 4892 | if (ret == -EAGAIN && !path->nowait) |
8e73f275 | 4893 | goto again; |
5f39d397 | 4894 | |
76a05b35 | 4895 | if (ret < 0) { |
b3b4aa74 | 4896 | btrfs_release_path(path); |
76a05b35 CM |
4897 | goto done; |
4898 | } | |
4899 | ||
5cd57b2c | 4900 | if (!path->skip_locking) { |
bd681513 | 4901 | ret = btrfs_try_tree_read_lock(next); |
bdcdd86c FM |
4902 | if (!ret && path->nowait) { |
4903 | ret = -EAGAIN; | |
4904 | goto done; | |
4905 | } | |
d42244a0 JS |
4906 | if (!ret && time_seq) { |
4907 | /* | |
4908 | * If we don't get the lock, we may be racing | |
4909 | * with push_leaf_left, holding that lock while | |
4910 | * itself waiting for the leaf we've currently | |
4911 | * locked. To solve this situation, we give up | |
4912 | * on our lock and cycle. | |
4913 | */ | |
cf538830 | 4914 | free_extent_buffer(next); |
d42244a0 JS |
4915 | btrfs_release_path(path); |
4916 | cond_resched(); | |
4917 | goto again; | |
4918 | } | |
0e46318d JB |
4919 | if (!ret) |
4920 | btrfs_tree_read_lock(next); | |
5cd57b2c | 4921 | } |
d97e63b6 CM |
4922 | break; |
4923 | } | |
4924 | path->slots[level] = slot; | |
d397712b | 4925 | while (1) { |
d97e63b6 | 4926 | level--; |
d97e63b6 CM |
4927 | path->nodes[level] = next; |
4928 | path->slots[level] = 0; | |
a74a4b97 | 4929 | if (!path->skip_locking) |
ffeb03cf | 4930 | path->locks[level] = BTRFS_READ_LOCK; |
d97e63b6 CM |
4931 | if (!level) |
4932 | break; | |
b4ce94de | 4933 | |
d07b8528 | 4934 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4935 | 0, &key); |
bdcdd86c | 4936 | if (ret == -EAGAIN && !path->nowait) |
8e73f275 CM |
4937 | goto again; |
4938 | ||
76a05b35 | 4939 | if (ret < 0) { |
b3b4aa74 | 4940 | btrfs_release_path(path); |
76a05b35 CM |
4941 | goto done; |
4942 | } | |
4943 | ||
bdcdd86c FM |
4944 | if (!path->skip_locking) { |
4945 | if (path->nowait) { | |
4946 | if (!btrfs_try_tree_read_lock(next)) { | |
4947 | ret = -EAGAIN; | |
4948 | goto done; | |
4949 | } | |
4950 | } else { | |
4951 | btrfs_tree_read_lock(next); | |
4952 | } | |
4953 | } | |
d97e63b6 | 4954 | } |
8e73f275 | 4955 | ret = 0; |
925baedd | 4956 | done: |
f7c79f30 | 4957 | unlock_up(path, 0, 1, 0, NULL); |
d96b3424 FM |
4958 | if (need_commit_sem) { |
4959 | int ret2; | |
4960 | ||
4961 | path->need_commit_sem = 1; | |
4962 | ret2 = finish_need_commit_sem_search(path); | |
4963 | up_read(&fs_info->commit_root_sem); | |
4964 | if (ret2) | |
4965 | ret = ret2; | |
4966 | } | |
8e73f275 CM |
4967 | |
4968 | return ret; | |
d97e63b6 | 4969 | } |
0b86a832 | 4970 | |
890d2b1a JB |
4971 | int btrfs_next_old_item(struct btrfs_root *root, struct btrfs_path *path, u64 time_seq) |
4972 | { | |
4973 | path->slots[0]++; | |
4974 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) | |
4975 | return btrfs_next_old_leaf(root, path, time_seq); | |
4976 | return 0; | |
4977 | } | |
4978 | ||
3f157a2f CM |
4979 | /* |
4980 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4981 | * searching until it gets past min_objectid or finds an item of 'type' | |
4982 | * | |
4983 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4984 | */ | |
0b86a832 CM |
4985 | int btrfs_previous_item(struct btrfs_root *root, |
4986 | struct btrfs_path *path, u64 min_objectid, | |
4987 | int type) | |
4988 | { | |
4989 | struct btrfs_key found_key; | |
4990 | struct extent_buffer *leaf; | |
e02119d5 | 4991 | u32 nritems; |
0b86a832 CM |
4992 | int ret; |
4993 | ||
d397712b | 4994 | while (1) { |
0b86a832 CM |
4995 | if (path->slots[0] == 0) { |
4996 | ret = btrfs_prev_leaf(root, path); | |
4997 | if (ret != 0) | |
4998 | return ret; | |
4999 | } else { | |
5000 | path->slots[0]--; | |
5001 | } | |
5002 | leaf = path->nodes[0]; | |
e02119d5 CM |
5003 | nritems = btrfs_header_nritems(leaf); |
5004 | if (nritems == 0) | |
5005 | return 1; | |
5006 | if (path->slots[0] == nritems) | |
5007 | path->slots[0]--; | |
5008 | ||
0b86a832 | 5009 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5010 | if (found_key.objectid < min_objectid) |
5011 | break; | |
0a4eefbb YZ |
5012 | if (found_key.type == type) |
5013 | return 0; | |
e02119d5 CM |
5014 | if (found_key.objectid == min_objectid && |
5015 | found_key.type < type) | |
5016 | break; | |
0b86a832 CM |
5017 | } |
5018 | return 1; | |
5019 | } | |
ade2e0b3 WS |
5020 | |
5021 | /* | |
5022 | * search in extent tree to find a previous Metadata/Data extent item with | |
5023 | * min objecitd. | |
5024 | * | |
5025 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5026 | */ | |
5027 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5028 | struct btrfs_path *path, u64 min_objectid) | |
5029 | { | |
5030 | struct btrfs_key found_key; | |
5031 | struct extent_buffer *leaf; | |
5032 | u32 nritems; | |
5033 | int ret; | |
5034 | ||
5035 | while (1) { | |
5036 | if (path->slots[0] == 0) { | |
ade2e0b3 WS |
5037 | ret = btrfs_prev_leaf(root, path); |
5038 | if (ret != 0) | |
5039 | return ret; | |
5040 | } else { | |
5041 | path->slots[0]--; | |
5042 | } | |
5043 | leaf = path->nodes[0]; | |
5044 | nritems = btrfs_header_nritems(leaf); | |
5045 | if (nritems == 0) | |
5046 | return 1; | |
5047 | if (path->slots[0] == nritems) | |
5048 | path->slots[0]--; | |
5049 | ||
5050 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5051 | if (found_key.objectid < min_objectid) | |
5052 | break; | |
5053 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5054 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5055 | return 0; | |
5056 | if (found_key.objectid == min_objectid && | |
5057 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5058 | break; | |
5059 | } | |
5060 | return 1; | |
5061 | } | |
226463d7 JB |
5062 | |
5063 | int __init btrfs_ctree_init(void) | |
5064 | { | |
5065 | btrfs_path_cachep = kmem_cache_create("btrfs_path", | |
5066 | sizeof(struct btrfs_path), 0, | |
5067 | SLAB_MEM_SPREAD, NULL); | |
5068 | if (!btrfs_path_cachep) | |
5069 | return -ENOMEM; | |
5070 | return 0; | |
5071 | } | |
5072 | ||
5073 | void __cold btrfs_ctree_exit(void) | |
5074 | { | |
5075 | kmem_cache_destroy(btrfs_path_cachep); | |
5076 | } |