]>
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> |
eb60ceac CM |
11 | #include "ctree.h" |
12 | #include "disk-io.h" | |
7f5c1516 | 13 | #include "transaction.h" |
5f39d397 | 14 | #include "print-tree.h" |
925baedd | 15 | #include "locking.h" |
de37aa51 | 16 | #include "volumes.h" |
f616f5cd | 17 | #include "qgroup.h" |
f3a84ccd | 18 | #include "tree-mod-log.h" |
88c602ab | 19 | #include "tree-checker.h" |
9a8dd150 | 20 | |
e089f05c CM |
21 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
22 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
23 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
24 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
25 | int data_size, int extend); | |
5f39d397 | 26 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 27 | struct extent_buffer *dst, |
971a1f66 | 28 | struct extent_buffer *src, int empty); |
5f39d397 | 29 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
30 | struct extent_buffer *dst_buf, |
31 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
32 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
33 | int level, int slot); | |
d97e63b6 | 34 | |
af024ed2 JT |
35 | static const struct btrfs_csums { |
36 | u16 size; | |
59a0fcdb DS |
37 | const char name[10]; |
38 | const char driver[12]; | |
af024ed2 JT |
39 | } btrfs_csums[] = { |
40 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
3951e7f0 | 41 | [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, |
3831bf00 | 42 | [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, |
352ae07b DS |
43 | [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", |
44 | .driver = "blake2b-256" }, | |
af024ed2 JT |
45 | }; |
46 | ||
47 | int btrfs_super_csum_size(const struct btrfs_super_block *s) | |
48 | { | |
49 | u16 t = btrfs_super_csum_type(s); | |
50 | /* | |
51 | * csum type is validated at mount time | |
52 | */ | |
53 | return btrfs_csums[t].size; | |
54 | } | |
55 | ||
56 | const char *btrfs_super_csum_name(u16 csum_type) | |
57 | { | |
58 | /* csum type is validated at mount time */ | |
59 | return btrfs_csums[csum_type].name; | |
60 | } | |
61 | ||
b4e967be DS |
62 | /* |
63 | * Return driver name if defined, otherwise the name that's also a valid driver | |
64 | * name | |
65 | */ | |
66 | const char *btrfs_super_csum_driver(u16 csum_type) | |
67 | { | |
68 | /* csum type is validated at mount time */ | |
59a0fcdb DS |
69 | return btrfs_csums[csum_type].driver[0] ? |
70 | btrfs_csums[csum_type].driver : | |
b4e967be DS |
71 | btrfs_csums[csum_type].name; |
72 | } | |
73 | ||
604997b4 | 74 | size_t __attribute_const__ btrfs_get_num_csums(void) |
f7cea56c DS |
75 | { |
76 | return ARRAY_SIZE(btrfs_csums); | |
77 | } | |
78 | ||
df24a2b9 | 79 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 80 | { |
e2c89907 | 81 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
82 | } |
83 | ||
d352ac68 | 84 | /* this also releases the path */ |
df24a2b9 | 85 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 86 | { |
ff175d57 JJ |
87 | if (!p) |
88 | return; | |
b3b4aa74 | 89 | btrfs_release_path(p); |
df24a2b9 | 90 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
91 | } |
92 | ||
d352ac68 CM |
93 | /* |
94 | * path release drops references on the extent buffers in the path | |
95 | * and it drops any locks held by this path | |
96 | * | |
97 | * It is safe to call this on paths that no locks or extent buffers held. | |
98 | */ | |
b3b4aa74 | 99 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
100 | { |
101 | int i; | |
a2135011 | 102 | |
234b63a0 | 103 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 104 | p->slots[i] = 0; |
eb60ceac | 105 | if (!p->nodes[i]) |
925baedd CM |
106 | continue; |
107 | if (p->locks[i]) { | |
bd681513 | 108 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
109 | p->locks[i] = 0; |
110 | } | |
5f39d397 | 111 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 112 | p->nodes[i] = NULL; |
eb60ceac CM |
113 | } |
114 | } | |
115 | ||
d352ac68 CM |
116 | /* |
117 | * safely gets a reference on the root node of a tree. A lock | |
118 | * is not taken, so a concurrent writer may put a different node | |
119 | * at the root of the tree. See btrfs_lock_root_node for the | |
120 | * looping required. | |
121 | * | |
122 | * The extent buffer returned by this has a reference taken, so | |
123 | * it won't disappear. It may stop being the root of the tree | |
124 | * at any time because there are no locks held. | |
125 | */ | |
925baedd CM |
126 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
127 | { | |
128 | struct extent_buffer *eb; | |
240f62c8 | 129 | |
3083ee2e JB |
130 | while (1) { |
131 | rcu_read_lock(); | |
132 | eb = rcu_dereference(root->node); | |
133 | ||
134 | /* | |
135 | * RCU really hurts here, we could free up the root node because | |
01327610 | 136 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
137 | * the inc_not_zero dance and if it doesn't work then |
138 | * synchronize_rcu and try again. | |
139 | */ | |
140 | if (atomic_inc_not_zero(&eb->refs)) { | |
141 | rcu_read_unlock(); | |
142 | break; | |
143 | } | |
144 | rcu_read_unlock(); | |
145 | synchronize_rcu(); | |
146 | } | |
925baedd CM |
147 | return eb; |
148 | } | |
149 | ||
92a7cc42 QW |
150 | /* |
151 | * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), | |
152 | * just get put onto a simple dirty list. Transaction walks this list to make | |
153 | * sure they get properly updated on disk. | |
d352ac68 | 154 | */ |
0b86a832 CM |
155 | static void add_root_to_dirty_list(struct btrfs_root *root) |
156 | { | |
0b246afa JM |
157 | struct btrfs_fs_info *fs_info = root->fs_info; |
158 | ||
e7070be1 JB |
159 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
160 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
161 | return; | |
162 | ||
0b246afa | 163 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
164 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
165 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 166 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 167 | list_move_tail(&root->dirty_list, |
0b246afa | 168 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
169 | else |
170 | list_move(&root->dirty_list, | |
0b246afa | 171 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 172 | } |
0b246afa | 173 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
174 | } |
175 | ||
d352ac68 CM |
176 | /* |
177 | * used by snapshot creation to make a copy of a root for a tree with | |
178 | * a given objectid. The buffer with the new root node is returned in | |
179 | * cow_ret, and this func returns zero on success or a negative error code. | |
180 | */ | |
be20aa9d CM |
181 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
182 | struct btrfs_root *root, | |
183 | struct extent_buffer *buf, | |
184 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
185 | { | |
0b246afa | 186 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 187 | struct extent_buffer *cow; |
be20aa9d CM |
188 | int ret = 0; |
189 | int level; | |
5d4f98a2 | 190 | struct btrfs_disk_key disk_key; |
be20aa9d | 191 | |
92a7cc42 | 192 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 193 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 194 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 195 | trans->transid != root->last_trans); |
be20aa9d CM |
196 | |
197 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
198 | if (level == 0) |
199 | btrfs_item_key(buf, &disk_key, 0); | |
200 | else | |
201 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 202 | |
4d75f8a9 | 203 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
cf6f34aa JB |
204 | &disk_key, level, buf->start, 0, |
205 | BTRFS_NESTING_NEW_ROOT); | |
5d4f98a2 | 206 | if (IS_ERR(cow)) |
be20aa9d CM |
207 | return PTR_ERR(cow); |
208 | ||
58e8012c | 209 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
210 | btrfs_set_header_bytenr(cow, cow->start); |
211 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
212 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
213 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
214 | BTRFS_HEADER_FLAG_RELOC); | |
215 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
216 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
217 | else | |
218 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 219 | |
de37aa51 | 220 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 221 | |
be20aa9d | 222 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 223 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 224 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 225 | else |
e339a6b0 | 226 | ret = btrfs_inc_ref(trans, root, cow, 0); |
867ed321 | 227 | if (ret) { |
72c9925f FM |
228 | btrfs_tree_unlock(cow); |
229 | free_extent_buffer(cow); | |
867ed321 | 230 | btrfs_abort_transaction(trans, ret); |
be20aa9d | 231 | return ret; |
867ed321 | 232 | } |
be20aa9d CM |
233 | |
234 | btrfs_mark_buffer_dirty(cow); | |
235 | *cow_ret = cow; | |
236 | return 0; | |
237 | } | |
238 | ||
5d4f98a2 YZ |
239 | /* |
240 | * check if the tree block can be shared by multiple trees | |
241 | */ | |
242 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
243 | struct extent_buffer *buf) | |
244 | { | |
245 | /* | |
92a7cc42 QW |
246 | * Tree blocks not in shareable trees and tree roots are never shared. |
247 | * If a block was allocated after the last snapshot and the block was | |
248 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 249 | */ |
92a7cc42 | 250 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
251 | buf != root->node && buf != root->commit_root && |
252 | (btrfs_header_generation(buf) <= | |
253 | btrfs_root_last_snapshot(&root->root_item) || | |
254 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
255 | return 1; | |
a79865c6 | 256 | |
5d4f98a2 YZ |
257 | return 0; |
258 | } | |
259 | ||
260 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
261 | struct btrfs_root *root, | |
262 | struct extent_buffer *buf, | |
f0486c68 YZ |
263 | struct extent_buffer *cow, |
264 | int *last_ref) | |
5d4f98a2 | 265 | { |
0b246afa | 266 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
267 | u64 refs; |
268 | u64 owner; | |
269 | u64 flags; | |
270 | u64 new_flags = 0; | |
271 | int ret; | |
272 | ||
273 | /* | |
274 | * Backrefs update rules: | |
275 | * | |
276 | * Always use full backrefs for extent pointers in tree block | |
277 | * allocated by tree relocation. | |
278 | * | |
279 | * If a shared tree block is no longer referenced by its owner | |
280 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
281 | * use full backrefs for extent pointers in tree block. | |
282 | * | |
283 | * If a tree block is been relocating | |
284 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
285 | * use full backrefs for extent pointers in tree block. | |
286 | * The reason for this is some operations (such as drop tree) | |
287 | * are only allowed for blocks use full backrefs. | |
288 | */ | |
289 | ||
290 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 291 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
292 | btrfs_header_level(buf), 1, |
293 | &refs, &flags); | |
be1a5564 MF |
294 | if (ret) |
295 | return ret; | |
e5df9573 MF |
296 | if (refs == 0) { |
297 | ret = -EROFS; | |
0b246afa | 298 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
299 | return ret; |
300 | } | |
5d4f98a2 YZ |
301 | } else { |
302 | refs = 1; | |
303 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
304 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
305 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
306 | else | |
307 | flags = 0; | |
308 | } | |
309 | ||
310 | owner = btrfs_header_owner(buf); | |
311 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
312 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
313 | ||
314 | if (refs > 1) { | |
315 | if ((owner == root->root_key.objectid || | |
316 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
317 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 318 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
319 | if (ret) |
320 | return ret; | |
5d4f98a2 YZ |
321 | |
322 | if (root->root_key.objectid == | |
323 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 324 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
325 | if (ret) |
326 | return ret; | |
e339a6b0 | 327 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
328 | if (ret) |
329 | return ret; | |
5d4f98a2 YZ |
330 | } |
331 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
332 | } else { | |
333 | ||
334 | if (root->root_key.objectid == | |
335 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 336 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 337 | else |
e339a6b0 | 338 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
339 | if (ret) |
340 | return ret; | |
5d4f98a2 YZ |
341 | } |
342 | if (new_flags != 0) { | |
b1c79e09 JB |
343 | int level = btrfs_header_level(buf); |
344 | ||
42c9d0b5 | 345 | ret = btrfs_set_disk_extent_flags(trans, buf, |
2fe6a5a1 | 346 | new_flags, level); |
be1a5564 MF |
347 | if (ret) |
348 | return ret; | |
5d4f98a2 YZ |
349 | } |
350 | } else { | |
351 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
352 | if (root->root_key.objectid == | |
353 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 354 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 355 | else |
e339a6b0 | 356 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
357 | if (ret) |
358 | return ret; | |
e339a6b0 | 359 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
360 | if (ret) |
361 | return ret; | |
5d4f98a2 | 362 | } |
6a884d7d | 363 | btrfs_clean_tree_block(buf); |
f0486c68 | 364 | *last_ref = 1; |
5d4f98a2 YZ |
365 | } |
366 | return 0; | |
367 | } | |
368 | ||
d352ac68 | 369 | /* |
d397712b CM |
370 | * does the dirty work in cow of a single block. The parent block (if |
371 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
372 | * dirty and returned locked. If you modify the block it needs to be marked | |
373 | * dirty again. | |
d352ac68 CM |
374 | * |
375 | * search_start -- an allocation hint for the new block | |
376 | * | |
d397712b CM |
377 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
378 | * bytes the allocator should try to find free next to the block it returns. | |
379 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 380 | */ |
d397712b | 381 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
382 | struct btrfs_root *root, |
383 | struct extent_buffer *buf, | |
384 | struct extent_buffer *parent, int parent_slot, | |
385 | struct extent_buffer **cow_ret, | |
9631e4cc JB |
386 | u64 search_start, u64 empty_size, |
387 | enum btrfs_lock_nesting nest) | |
02217ed2 | 388 | { |
0b246afa | 389 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 390 | struct btrfs_disk_key disk_key; |
5f39d397 | 391 | struct extent_buffer *cow; |
be1a5564 | 392 | int level, ret; |
f0486c68 | 393 | int last_ref = 0; |
925baedd | 394 | int unlock_orig = 0; |
0f5053eb | 395 | u64 parent_start = 0; |
7bb86316 | 396 | |
925baedd CM |
397 | if (*cow_ret == buf) |
398 | unlock_orig = 1; | |
399 | ||
49d0c642 | 400 | btrfs_assert_tree_write_locked(buf); |
925baedd | 401 | |
92a7cc42 | 402 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 403 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 404 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 405 | trans->transid != root->last_trans); |
5f39d397 | 406 | |
7bb86316 | 407 | level = btrfs_header_level(buf); |
31840ae1 | 408 | |
5d4f98a2 YZ |
409 | if (level == 0) |
410 | btrfs_item_key(buf, &disk_key, 0); | |
411 | else | |
412 | btrfs_node_key(buf, &disk_key, 0); | |
413 | ||
0f5053eb GR |
414 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
415 | parent_start = parent->start; | |
5d4f98a2 | 416 | |
79bd3712 FM |
417 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
418 | root->root_key.objectid, &disk_key, level, | |
419 | search_start, empty_size, nest); | |
54aa1f4d CM |
420 | if (IS_ERR(cow)) |
421 | return PTR_ERR(cow); | |
6702ed49 | 422 | |
b4ce94de CM |
423 | /* cow is set to blocking by btrfs_init_new_buffer */ |
424 | ||
58e8012c | 425 | copy_extent_buffer_full(cow, buf); |
db94535d | 426 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 427 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
428 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
429 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
430 | BTRFS_HEADER_FLAG_RELOC); | |
431 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
432 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
433 | else | |
434 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 435 | |
de37aa51 | 436 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 437 | |
be1a5564 | 438 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 439 | if (ret) { |
572c83ac JB |
440 | btrfs_tree_unlock(cow); |
441 | free_extent_buffer(cow); | |
66642832 | 442 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
443 | return ret; |
444 | } | |
1a40e23b | 445 | |
92a7cc42 | 446 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 447 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 448 | if (ret) { |
572c83ac JB |
449 | btrfs_tree_unlock(cow); |
450 | free_extent_buffer(cow); | |
66642832 | 451 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 452 | return ret; |
93314e3b | 453 | } |
83d4cfd4 | 454 | } |
3fd0a558 | 455 | |
02217ed2 | 456 | if (buf == root->node) { |
925baedd | 457 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
458 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
459 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
460 | parent_start = buf->start; | |
925baedd | 461 | |
67439dad | 462 | atomic_inc(&cow->refs); |
406808ab | 463 | ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); |
d9d19a01 | 464 | BUG_ON(ret < 0); |
240f62c8 | 465 | rcu_assign_pointer(root->node, cow); |
925baedd | 466 | |
7a163608 FM |
467 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
468 | parent_start, last_ref); | |
5f39d397 | 469 | free_extent_buffer(buf); |
0b86a832 | 470 | add_root_to_dirty_list(root); |
02217ed2 | 471 | } else { |
5d4f98a2 | 472 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
f3a84ccd FM |
473 | btrfs_tree_mod_log_insert_key(parent, parent_slot, |
474 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
5f39d397 | 475 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 476 | cow->start); |
74493f7a CM |
477 | btrfs_set_node_ptr_generation(parent, parent_slot, |
478 | trans->transid); | |
d6025579 | 479 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 480 | if (last_ref) { |
f3a84ccd | 481 | ret = btrfs_tree_mod_log_free_eb(buf); |
5de865ee | 482 | if (ret) { |
572c83ac JB |
483 | btrfs_tree_unlock(cow); |
484 | free_extent_buffer(cow); | |
66642832 | 485 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
486 | return ret; |
487 | } | |
488 | } | |
7a163608 FM |
489 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
490 | parent_start, last_ref); | |
02217ed2 | 491 | } |
925baedd CM |
492 | if (unlock_orig) |
493 | btrfs_tree_unlock(buf); | |
3083ee2e | 494 | free_extent_buffer_stale(buf); |
ccd467d6 | 495 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 496 | *cow_ret = cow; |
02217ed2 CM |
497 | return 0; |
498 | } | |
499 | ||
5d4f98a2 YZ |
500 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
501 | struct btrfs_root *root, | |
502 | struct extent_buffer *buf) | |
503 | { | |
f5ee5c9a | 504 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 505 | return 0; |
fccb84c9 | 506 | |
d1980131 DS |
507 | /* Ensure we can see the FORCE_COW bit */ |
508 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
509 | |
510 | /* | |
511 | * We do not need to cow a block if | |
512 | * 1) this block is not created or changed in this transaction; | |
513 | * 2) this block does not belong to TREE_RELOC tree; | |
514 | * 3) the root is not forced COW. | |
515 | * | |
516 | * What is forced COW: | |
01327610 | 517 | * when we create snapshot during committing the transaction, |
52042d8e | 518 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
519 | * block to ensure the metadata consistency. |
520 | */ | |
5d4f98a2 YZ |
521 | if (btrfs_header_generation(buf) == trans->transid && |
522 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
523 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 524 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 525 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
526 | return 0; |
527 | return 1; | |
528 | } | |
529 | ||
d352ac68 CM |
530 | /* |
531 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 532 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
533 | * once per transaction, as long as it hasn't been written yet |
534 | */ | |
d397712b | 535 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
536 | struct btrfs_root *root, struct extent_buffer *buf, |
537 | struct extent_buffer *parent, int parent_slot, | |
9631e4cc JB |
538 | struct extent_buffer **cow_ret, |
539 | enum btrfs_lock_nesting nest) | |
6702ed49 | 540 | { |
0b246afa | 541 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 542 | u64 search_start; |
f510cfec | 543 | int ret; |
dc17ff8f | 544 | |
83354f07 JB |
545 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
546 | btrfs_err(fs_info, | |
547 | "COW'ing blocks on a fs root that's being dropped"); | |
548 | ||
0b246afa | 549 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 550 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 551 | trans->transid, |
0b246afa | 552 | fs_info->running_transaction->transid); |
31b1a2bd | 553 | |
0b246afa | 554 | if (trans->transid != fs_info->generation) |
31b1a2bd | 555 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 556 | trans->transid, fs_info->generation); |
dc17ff8f | 557 | |
5d4f98a2 | 558 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
559 | *cow_ret = buf; |
560 | return 0; | |
561 | } | |
c487685d | 562 | |
ee22184b | 563 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de | 564 | |
f616f5cd QW |
565 | /* |
566 | * Before CoWing this block for later modification, check if it's | |
567 | * the subtree root and do the delayed subtree trace if needed. | |
568 | * | |
569 | * Also We don't care about the error, as it's handled internally. | |
570 | */ | |
571 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 572 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9631e4cc | 573 | parent_slot, cow_ret, search_start, 0, nest); |
1abe9b8a | 574 | |
575 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
576 | ||
f510cfec | 577 | return ret; |
6702ed49 | 578 | } |
f75e2b79 | 579 | ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); |
6702ed49 | 580 | |
d352ac68 CM |
581 | /* |
582 | * helper function for defrag to decide if two blocks pointed to by a | |
583 | * node are actually close by | |
584 | */ | |
6b80053d | 585 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 586 | { |
6b80053d | 587 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 588 | return 1; |
6b80053d | 589 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
590 | return 1; |
591 | return 0; | |
592 | } | |
593 | ||
ce6ef5ab DS |
594 | #ifdef __LITTLE_ENDIAN |
595 | ||
596 | /* | |
597 | * Compare two keys, on little-endian the disk order is same as CPU order and | |
598 | * we can avoid the conversion. | |
599 | */ | |
600 | static int comp_keys(const struct btrfs_disk_key *disk_key, | |
601 | const struct btrfs_key *k2) | |
602 | { | |
603 | const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; | |
604 | ||
605 | return btrfs_comp_cpu_keys(k1, k2); | |
606 | } | |
607 | ||
608 | #else | |
609 | ||
081e9573 CM |
610 | /* |
611 | * compare two keys in a memcmp fashion | |
612 | */ | |
310712b2 OS |
613 | static int comp_keys(const struct btrfs_disk_key *disk, |
614 | const struct btrfs_key *k2) | |
081e9573 CM |
615 | { |
616 | struct btrfs_key k1; | |
617 | ||
618 | btrfs_disk_key_to_cpu(&k1, disk); | |
619 | ||
20736aba | 620 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 | 621 | } |
ce6ef5ab | 622 | #endif |
081e9573 | 623 | |
f3465ca4 JB |
624 | /* |
625 | * same as comp_keys only with two btrfs_key's | |
626 | */ | |
e1f60a65 | 627 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
628 | { |
629 | if (k1->objectid > k2->objectid) | |
630 | return 1; | |
631 | if (k1->objectid < k2->objectid) | |
632 | return -1; | |
633 | if (k1->type > k2->type) | |
634 | return 1; | |
635 | if (k1->type < k2->type) | |
636 | return -1; | |
637 | if (k1->offset > k2->offset) | |
638 | return 1; | |
639 | if (k1->offset < k2->offset) | |
640 | return -1; | |
641 | return 0; | |
642 | } | |
081e9573 | 643 | |
d352ac68 CM |
644 | /* |
645 | * this is used by the defrag code to go through all the | |
646 | * leaves pointed to by a node and reallocate them so that | |
647 | * disk order is close to key order | |
648 | */ | |
6702ed49 | 649 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 650 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 651 | int start_slot, u64 *last_ret, |
a6b6e75e | 652 | struct btrfs_key *progress) |
6702ed49 | 653 | { |
0b246afa | 654 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 655 | struct extent_buffer *cur; |
6702ed49 | 656 | u64 blocknr; |
e9d0b13b CM |
657 | u64 search_start = *last_ret; |
658 | u64 last_block = 0; | |
6702ed49 CM |
659 | u64 other; |
660 | u32 parent_nritems; | |
6702ed49 CM |
661 | int end_slot; |
662 | int i; | |
663 | int err = 0; | |
6b80053d | 664 | u32 blocksize; |
081e9573 CM |
665 | int progress_passed = 0; |
666 | struct btrfs_disk_key disk_key; | |
6702ed49 | 667 | |
0b246afa JM |
668 | WARN_ON(trans->transaction != fs_info->running_transaction); |
669 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 670 | |
6b80053d | 671 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 672 | blocksize = fs_info->nodesize; |
5dfe2be7 | 673 | end_slot = parent_nritems - 1; |
6702ed49 | 674 | |
5dfe2be7 | 675 | if (parent_nritems <= 1) |
6702ed49 CM |
676 | return 0; |
677 | ||
5dfe2be7 | 678 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 679 | int close = 1; |
a6b6e75e | 680 | |
081e9573 CM |
681 | btrfs_node_key(parent, &disk_key, i); |
682 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
683 | continue; | |
684 | ||
685 | progress_passed = 1; | |
6b80053d | 686 | blocknr = btrfs_node_blockptr(parent, i); |
e9d0b13b CM |
687 | if (last_block == 0) |
688 | last_block = blocknr; | |
5708b959 | 689 | |
6702ed49 | 690 | if (i > 0) { |
6b80053d CM |
691 | other = btrfs_node_blockptr(parent, i - 1); |
692 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 693 | } |
5dfe2be7 | 694 | if (!close && i < end_slot) { |
6b80053d CM |
695 | other = btrfs_node_blockptr(parent, i + 1); |
696 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 697 | } |
e9d0b13b CM |
698 | if (close) { |
699 | last_block = blocknr; | |
6702ed49 | 700 | continue; |
e9d0b13b | 701 | } |
6702ed49 | 702 | |
206983b7 JB |
703 | cur = btrfs_read_node_slot(parent, i); |
704 | if (IS_ERR(cur)) | |
705 | return PTR_ERR(cur); | |
e9d0b13b | 706 | if (search_start == 0) |
6b80053d | 707 | search_start = last_block; |
e9d0b13b | 708 | |
e7a84565 | 709 | btrfs_tree_lock(cur); |
6b80053d | 710 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 711 | &cur, search_start, |
6b80053d | 712 | min(16 * blocksize, |
9631e4cc JB |
713 | (end_slot - i) * blocksize), |
714 | BTRFS_NESTING_COW); | |
252c38f0 | 715 | if (err) { |
e7a84565 | 716 | btrfs_tree_unlock(cur); |
6b80053d | 717 | free_extent_buffer(cur); |
6702ed49 | 718 | break; |
252c38f0 | 719 | } |
e7a84565 CM |
720 | search_start = cur->start; |
721 | last_block = cur->start; | |
f2183bde | 722 | *last_ret = search_start; |
e7a84565 CM |
723 | btrfs_tree_unlock(cur); |
724 | free_extent_buffer(cur); | |
6702ed49 CM |
725 | } |
726 | return err; | |
727 | } | |
728 | ||
74123bd7 | 729 | /* |
fb81212c | 730 | * Search for a key in the given extent_buffer. |
5f39d397 | 731 | * |
fb81212c FM |
732 | * The lower boundary for the search is specified by the slot number @low. Use a |
733 | * value of 0 to search over the whole extent buffer. | |
74123bd7 | 734 | * |
fb81212c FM |
735 | * The slot in the extent buffer is returned via @slot. If the key exists in the |
736 | * extent buffer, then @slot will point to the slot where the key is, otherwise | |
737 | * it points to the slot where you would insert the key. | |
738 | * | |
739 | * Slot may point to the total number of items (i.e. one position beyond the last | |
740 | * key) if the key is bigger than the last key in the extent buffer. | |
74123bd7 | 741 | */ |
fb81212c | 742 | static noinline int generic_bin_search(struct extent_buffer *eb, int low, |
67d5e289 | 743 | const struct btrfs_key *key, int *slot) |
be0e5c09 | 744 | { |
fb81212c FM |
745 | unsigned long p; |
746 | int item_size; | |
67d5e289 | 747 | int high = btrfs_header_nritems(eb); |
be0e5c09 | 748 | int ret; |
5cd17f34 | 749 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 750 | |
5e24e9af LB |
751 | if (low > high) { |
752 | btrfs_err(eb->fs_info, | |
753 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
754 | __func__, low, high, eb->start, | |
755 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
756 | return -EINVAL; | |
757 | } | |
758 | ||
fb81212c FM |
759 | if (btrfs_header_level(eb) == 0) { |
760 | p = offsetof(struct btrfs_leaf, items); | |
761 | item_size = sizeof(struct btrfs_item); | |
762 | } else { | |
763 | p = offsetof(struct btrfs_node, ptrs); | |
764 | item_size = sizeof(struct btrfs_key_ptr); | |
765 | } | |
766 | ||
d397712b | 767 | while (low < high) { |
5cd17f34 DS |
768 | unsigned long oip; |
769 | unsigned long offset; | |
770 | struct btrfs_disk_key *tmp; | |
771 | struct btrfs_disk_key unaligned; | |
772 | int mid; | |
773 | ||
be0e5c09 | 774 | mid = (low + high) / 2; |
5f39d397 | 775 | offset = p + mid * item_size; |
5cd17f34 | 776 | oip = offset_in_page(offset); |
5f39d397 | 777 | |
5cd17f34 | 778 | if (oip + key_size <= PAGE_SIZE) { |
884b07d0 | 779 | const unsigned long idx = get_eb_page_index(offset); |
5cd17f34 | 780 | char *kaddr = page_address(eb->pages[idx]); |
5f39d397 | 781 | |
884b07d0 | 782 | oip = get_eb_offset_in_page(eb, offset); |
5cd17f34 | 783 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 784 | } else { |
5cd17f34 DS |
785 | read_extent_buffer(eb, &unaligned, offset, key_size); |
786 | tmp = &unaligned; | |
5f39d397 | 787 | } |
5cd17f34 | 788 | |
be0e5c09 CM |
789 | ret = comp_keys(tmp, key); |
790 | ||
791 | if (ret < 0) | |
792 | low = mid + 1; | |
793 | else if (ret > 0) | |
794 | high = mid; | |
795 | else { | |
796 | *slot = mid; | |
797 | return 0; | |
798 | } | |
799 | } | |
800 | *slot = low; | |
801 | return 1; | |
802 | } | |
803 | ||
97571fd0 | 804 | /* |
fb81212c FM |
805 | * Simple binary search on an extent buffer. Works for both leaves and nodes, and |
806 | * always searches over the whole range of keys (slot 0 to slot 'nritems - 1'). | |
97571fd0 | 807 | */ |
a74b35ec | 808 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
e3b83361 | 809 | int *slot) |
be0e5c09 | 810 | { |
fb81212c | 811 | return generic_bin_search(eb, 0, key, slot); |
be0e5c09 CM |
812 | } |
813 | ||
f0486c68 YZ |
814 | static void root_add_used(struct btrfs_root *root, u32 size) |
815 | { | |
816 | spin_lock(&root->accounting_lock); | |
817 | btrfs_set_root_used(&root->root_item, | |
818 | btrfs_root_used(&root->root_item) + size); | |
819 | spin_unlock(&root->accounting_lock); | |
820 | } | |
821 | ||
822 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
823 | { | |
824 | spin_lock(&root->accounting_lock); | |
825 | btrfs_set_root_used(&root->root_item, | |
826 | btrfs_root_used(&root->root_item) - size); | |
827 | spin_unlock(&root->accounting_lock); | |
828 | } | |
829 | ||
d352ac68 CM |
830 | /* given a node and slot number, this reads the blocks it points to. The |
831 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 832 | */ |
4b231ae4 DS |
833 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
834 | int slot) | |
bb803951 | 835 | { |
ca7a79ad | 836 | int level = btrfs_header_level(parent); |
416bc658 | 837 | struct extent_buffer *eb; |
581c1760 | 838 | struct btrfs_key first_key; |
416bc658 | 839 | |
fb770ae4 LB |
840 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
841 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
842 | |
843 | BUG_ON(level == 0); | |
844 | ||
581c1760 | 845 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 846 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
1b7ec85e | 847 | btrfs_header_owner(parent), |
581c1760 QW |
848 | btrfs_node_ptr_generation(parent, slot), |
849 | level - 1, &first_key); | |
4eb150d6 QW |
850 | if (IS_ERR(eb)) |
851 | return eb; | |
852 | if (!extent_buffer_uptodate(eb)) { | |
fb770ae4 | 853 | free_extent_buffer(eb); |
4eb150d6 | 854 | return ERR_PTR(-EIO); |
416bc658 JB |
855 | } |
856 | ||
857 | return eb; | |
bb803951 CM |
858 | } |
859 | ||
d352ac68 CM |
860 | /* |
861 | * node level balancing, used to make sure nodes are in proper order for | |
862 | * item deletion. We balance from the top down, so we have to make sure | |
863 | * that a deletion won't leave an node completely empty later on. | |
864 | */ | |
e02119d5 | 865 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
866 | struct btrfs_root *root, |
867 | struct btrfs_path *path, int level) | |
bb803951 | 868 | { |
0b246afa | 869 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
870 | struct extent_buffer *right = NULL; |
871 | struct extent_buffer *mid; | |
872 | struct extent_buffer *left = NULL; | |
873 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
874 | int ret = 0; |
875 | int wret; | |
876 | int pslot; | |
bb803951 | 877 | int orig_slot = path->slots[level]; |
79f95c82 | 878 | u64 orig_ptr; |
bb803951 | 879 | |
98e6b1eb | 880 | ASSERT(level > 0); |
bb803951 | 881 | |
5f39d397 | 882 | mid = path->nodes[level]; |
b4ce94de | 883 | |
ac5887c8 | 884 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); |
7bb86316 CM |
885 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
886 | ||
1d4f8a0c | 887 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 888 | |
a05a9bb1 | 889 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 890 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
891 | pslot = path->slots[level + 1]; |
892 | } | |
bb803951 | 893 | |
40689478 CM |
894 | /* |
895 | * deal with the case where there is only one pointer in the root | |
896 | * by promoting the node below to a root | |
897 | */ | |
5f39d397 CM |
898 | if (!parent) { |
899 | struct extent_buffer *child; | |
bb803951 | 900 | |
5f39d397 | 901 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
902 | return 0; |
903 | ||
904 | /* promote the child to a root */ | |
4b231ae4 | 905 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
906 | if (IS_ERR(child)) { |
907 | ret = PTR_ERR(child); | |
0b246afa | 908 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
909 | goto enospc; |
910 | } | |
911 | ||
925baedd | 912 | btrfs_tree_lock(child); |
9631e4cc JB |
913 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, |
914 | BTRFS_NESTING_COW); | |
f0486c68 YZ |
915 | if (ret) { |
916 | btrfs_tree_unlock(child); | |
917 | free_extent_buffer(child); | |
918 | goto enospc; | |
919 | } | |
2f375ab9 | 920 | |
406808ab | 921 | ret = btrfs_tree_mod_log_insert_root(root->node, child, true); |
d9d19a01 | 922 | BUG_ON(ret < 0); |
240f62c8 | 923 | rcu_assign_pointer(root->node, child); |
925baedd | 924 | |
0b86a832 | 925 | add_root_to_dirty_list(root); |
925baedd | 926 | btrfs_tree_unlock(child); |
b4ce94de | 927 | |
925baedd | 928 | path->locks[level] = 0; |
bb803951 | 929 | path->nodes[level] = NULL; |
6a884d7d | 930 | btrfs_clean_tree_block(mid); |
925baedd | 931 | btrfs_tree_unlock(mid); |
bb803951 | 932 | /* once for the path */ |
5f39d397 | 933 | free_extent_buffer(mid); |
f0486c68 YZ |
934 | |
935 | root_sub_used(root, mid->len); | |
7a163608 | 936 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
bb803951 | 937 | /* once for the root ptr */ |
3083ee2e | 938 | free_extent_buffer_stale(mid); |
f0486c68 | 939 | return 0; |
bb803951 | 940 | } |
5f39d397 | 941 | if (btrfs_header_nritems(mid) > |
0b246afa | 942 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
943 | return 0; |
944 | ||
4b231ae4 | 945 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
946 | if (IS_ERR(left)) |
947 | left = NULL; | |
948 | ||
5f39d397 | 949 | if (left) { |
bf77467a | 950 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
5f39d397 | 951 | wret = btrfs_cow_block(trans, root, left, |
9631e4cc | 952 | parent, pslot - 1, &left, |
bf59a5a2 | 953 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
954 | if (wret) { |
955 | ret = wret; | |
956 | goto enospc; | |
957 | } | |
2cc58cf2 | 958 | } |
fb770ae4 | 959 | |
4b231ae4 | 960 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
961 | if (IS_ERR(right)) |
962 | right = NULL; | |
963 | ||
5f39d397 | 964 | if (right) { |
bf77467a | 965 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
5f39d397 | 966 | wret = btrfs_cow_block(trans, root, right, |
9631e4cc | 967 | parent, pslot + 1, &right, |
bf59a5a2 | 968 | BTRFS_NESTING_RIGHT_COW); |
2cc58cf2 CM |
969 | if (wret) { |
970 | ret = wret; | |
971 | goto enospc; | |
972 | } | |
973 | } | |
974 | ||
975 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
976 | if (left) { |
977 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 978 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
979 | if (wret < 0) |
980 | ret = wret; | |
bb803951 | 981 | } |
79f95c82 CM |
982 | |
983 | /* | |
984 | * then try to empty the right most buffer into the middle | |
985 | */ | |
5f39d397 | 986 | if (right) { |
d30a668f | 987 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 988 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 989 | ret = wret; |
5f39d397 | 990 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 991 | btrfs_clean_tree_block(right); |
925baedd | 992 | btrfs_tree_unlock(right); |
afe5fea7 | 993 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 994 | root_sub_used(root, right->len); |
7a163608 FM |
995 | btrfs_free_tree_block(trans, btrfs_root_id(root), right, |
996 | 0, 1); | |
3083ee2e | 997 | free_extent_buffer_stale(right); |
f0486c68 | 998 | right = NULL; |
bb803951 | 999 | } else { |
5f39d397 CM |
1000 | struct btrfs_disk_key right_key; |
1001 | btrfs_node_key(right, &right_key, 0); | |
f3a84ccd FM |
1002 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
1003 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1004 | BUG_ON(ret < 0); |
5f39d397 CM |
1005 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1006 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1007 | } |
1008 | } | |
5f39d397 | 1009 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1010 | /* |
1011 | * we're not allowed to leave a node with one item in the | |
1012 | * tree during a delete. A deletion from lower in the tree | |
1013 | * could try to delete the only pointer in this node. | |
1014 | * So, pull some keys from the left. | |
1015 | * There has to be a left pointer at this point because | |
1016 | * otherwise we would have pulled some pointers from the | |
1017 | * right | |
1018 | */ | |
305a26af MF |
1019 | if (!left) { |
1020 | ret = -EROFS; | |
0b246afa | 1021 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1022 | goto enospc; |
1023 | } | |
55d32ed8 | 1024 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1025 | if (wret < 0) { |
79f95c82 | 1026 | ret = wret; |
54aa1f4d CM |
1027 | goto enospc; |
1028 | } | |
bce4eae9 | 1029 | if (wret == 1) { |
d30a668f | 1030 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1031 | if (wret < 0) |
1032 | ret = wret; | |
1033 | } | |
79f95c82 CM |
1034 | BUG_ON(wret == 1); |
1035 | } | |
5f39d397 | 1036 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1037 | btrfs_clean_tree_block(mid); |
925baedd | 1038 | btrfs_tree_unlock(mid); |
afe5fea7 | 1039 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1040 | root_sub_used(root, mid->len); |
7a163608 | 1041 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
3083ee2e | 1042 | free_extent_buffer_stale(mid); |
f0486c68 | 1043 | mid = NULL; |
79f95c82 CM |
1044 | } else { |
1045 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1046 | struct btrfs_disk_key mid_key; |
1047 | btrfs_node_key(mid, &mid_key, 0); | |
f3a84ccd FM |
1048 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1049 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1050 | BUG_ON(ret < 0); |
5f39d397 CM |
1051 | btrfs_set_node_key(parent, &mid_key, pslot); |
1052 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1053 | } |
bb803951 | 1054 | |
79f95c82 | 1055 | /* update the path */ |
5f39d397 CM |
1056 | if (left) { |
1057 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1058 | atomic_inc(&left->refs); |
925baedd | 1059 | /* left was locked after cow */ |
5f39d397 | 1060 | path->nodes[level] = left; |
bb803951 CM |
1061 | path->slots[level + 1] -= 1; |
1062 | path->slots[level] = orig_slot; | |
925baedd CM |
1063 | if (mid) { |
1064 | btrfs_tree_unlock(mid); | |
5f39d397 | 1065 | free_extent_buffer(mid); |
925baedd | 1066 | } |
bb803951 | 1067 | } else { |
5f39d397 | 1068 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1069 | path->slots[level] = orig_slot; |
1070 | } | |
1071 | } | |
79f95c82 | 1072 | /* double check we haven't messed things up */ |
e20d96d6 | 1073 | if (orig_ptr != |
5f39d397 | 1074 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1075 | BUG(); |
54aa1f4d | 1076 | enospc: |
925baedd CM |
1077 | if (right) { |
1078 | btrfs_tree_unlock(right); | |
5f39d397 | 1079 | free_extent_buffer(right); |
925baedd CM |
1080 | } |
1081 | if (left) { | |
1082 | if (path->nodes[level] != left) | |
1083 | btrfs_tree_unlock(left); | |
5f39d397 | 1084 | free_extent_buffer(left); |
925baedd | 1085 | } |
bb803951 CM |
1086 | return ret; |
1087 | } | |
1088 | ||
d352ac68 CM |
1089 | /* Node balancing for insertion. Here we only split or push nodes around |
1090 | * when they are completely full. This is also done top down, so we | |
1091 | * have to be pessimistic. | |
1092 | */ | |
d397712b | 1093 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1094 | struct btrfs_root *root, |
1095 | struct btrfs_path *path, int level) | |
e66f709b | 1096 | { |
0b246afa | 1097 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1098 | struct extent_buffer *right = NULL; |
1099 | struct extent_buffer *mid; | |
1100 | struct extent_buffer *left = NULL; | |
1101 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1102 | int ret = 0; |
1103 | int wret; | |
1104 | int pslot; | |
1105 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1106 | |
1107 | if (level == 0) | |
1108 | return 1; | |
1109 | ||
5f39d397 | 1110 | mid = path->nodes[level]; |
7bb86316 | 1111 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1112 | |
a05a9bb1 | 1113 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1114 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1115 | pslot = path->slots[level + 1]; |
1116 | } | |
e66f709b | 1117 | |
5f39d397 | 1118 | if (!parent) |
e66f709b | 1119 | return 1; |
e66f709b | 1120 | |
4b231ae4 | 1121 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1122 | if (IS_ERR(left)) |
1123 | left = NULL; | |
e66f709b CM |
1124 | |
1125 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1126 | if (left) { |
e66f709b | 1127 | u32 left_nr; |
925baedd | 1128 | |
bf77467a | 1129 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
b4ce94de | 1130 | |
5f39d397 | 1131 | left_nr = btrfs_header_nritems(left); |
0b246afa | 1132 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1133 | wret = 1; |
1134 | } else { | |
5f39d397 | 1135 | ret = btrfs_cow_block(trans, root, left, parent, |
9631e4cc | 1136 | pslot - 1, &left, |
bf59a5a2 | 1137 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
1138 | if (ret) |
1139 | wret = 1; | |
1140 | else { | |
d30a668f | 1141 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 1142 | } |
33ade1f8 | 1143 | } |
e66f709b CM |
1144 | if (wret < 0) |
1145 | ret = wret; | |
1146 | if (wret == 0) { | |
5f39d397 | 1147 | struct btrfs_disk_key disk_key; |
e66f709b | 1148 | orig_slot += left_nr; |
5f39d397 | 1149 | btrfs_node_key(mid, &disk_key, 0); |
f3a84ccd FM |
1150 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1151 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1152 | BUG_ON(ret < 0); |
5f39d397 CM |
1153 | btrfs_set_node_key(parent, &disk_key, pslot); |
1154 | btrfs_mark_buffer_dirty(parent); | |
1155 | if (btrfs_header_nritems(left) > orig_slot) { | |
1156 | path->nodes[level] = left; | |
e66f709b CM |
1157 | path->slots[level + 1] -= 1; |
1158 | path->slots[level] = orig_slot; | |
925baedd | 1159 | btrfs_tree_unlock(mid); |
5f39d397 | 1160 | free_extent_buffer(mid); |
e66f709b CM |
1161 | } else { |
1162 | orig_slot -= | |
5f39d397 | 1163 | btrfs_header_nritems(left); |
e66f709b | 1164 | path->slots[level] = orig_slot; |
925baedd | 1165 | btrfs_tree_unlock(left); |
5f39d397 | 1166 | free_extent_buffer(left); |
e66f709b | 1167 | } |
e66f709b CM |
1168 | return 0; |
1169 | } | |
925baedd | 1170 | btrfs_tree_unlock(left); |
5f39d397 | 1171 | free_extent_buffer(left); |
e66f709b | 1172 | } |
4b231ae4 | 1173 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1174 | if (IS_ERR(right)) |
1175 | right = NULL; | |
e66f709b CM |
1176 | |
1177 | /* | |
1178 | * then try to empty the right most buffer into the middle | |
1179 | */ | |
5f39d397 | 1180 | if (right) { |
33ade1f8 | 1181 | u32 right_nr; |
b4ce94de | 1182 | |
bf77467a | 1183 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
b4ce94de | 1184 | |
5f39d397 | 1185 | right_nr = btrfs_header_nritems(right); |
0b246afa | 1186 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1187 | wret = 1; |
1188 | } else { | |
5f39d397 CM |
1189 | ret = btrfs_cow_block(trans, root, right, |
1190 | parent, pslot + 1, | |
bf59a5a2 | 1191 | &right, BTRFS_NESTING_RIGHT_COW); |
54aa1f4d CM |
1192 | if (ret) |
1193 | wret = 1; | |
1194 | else { | |
55d32ed8 | 1195 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 1196 | } |
33ade1f8 | 1197 | } |
e66f709b CM |
1198 | if (wret < 0) |
1199 | ret = wret; | |
1200 | if (wret == 0) { | |
5f39d397 CM |
1201 | struct btrfs_disk_key disk_key; |
1202 | ||
1203 | btrfs_node_key(right, &disk_key, 0); | |
f3a84ccd FM |
1204 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
1205 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1206 | BUG_ON(ret < 0); |
5f39d397 CM |
1207 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
1208 | btrfs_mark_buffer_dirty(parent); | |
1209 | ||
1210 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1211 | path->nodes[level] = right; | |
e66f709b CM |
1212 | path->slots[level + 1] += 1; |
1213 | path->slots[level] = orig_slot - | |
5f39d397 | 1214 | btrfs_header_nritems(mid); |
925baedd | 1215 | btrfs_tree_unlock(mid); |
5f39d397 | 1216 | free_extent_buffer(mid); |
e66f709b | 1217 | } else { |
925baedd | 1218 | btrfs_tree_unlock(right); |
5f39d397 | 1219 | free_extent_buffer(right); |
e66f709b | 1220 | } |
e66f709b CM |
1221 | return 0; |
1222 | } | |
925baedd | 1223 | btrfs_tree_unlock(right); |
5f39d397 | 1224 | free_extent_buffer(right); |
e66f709b | 1225 | } |
e66f709b CM |
1226 | return 1; |
1227 | } | |
1228 | ||
3c69faec | 1229 | /* |
d352ac68 CM |
1230 | * readahead one full node of leaves, finding things that are close |
1231 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1232 | */ |
2ff7e61e | 1233 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
1234 | struct btrfs_path *path, |
1235 | int level, int slot, u64 objectid) | |
3c69faec | 1236 | { |
5f39d397 | 1237 | struct extent_buffer *node; |
01f46658 | 1238 | struct btrfs_disk_key disk_key; |
3c69faec | 1239 | u32 nritems; |
3c69faec | 1240 | u64 search; |
a7175319 | 1241 | u64 target; |
6b80053d | 1242 | u64 nread = 0; |
ace75066 | 1243 | u64 nread_max; |
6b80053d CM |
1244 | u32 nr; |
1245 | u32 blocksize; | |
1246 | u32 nscan = 0; | |
db94535d | 1247 | |
ace75066 | 1248 | if (level != 1 && path->reada != READA_FORWARD_ALWAYS) |
6702ed49 CM |
1249 | return; |
1250 | ||
1251 | if (!path->nodes[level]) | |
3c69faec CM |
1252 | return; |
1253 | ||
5f39d397 | 1254 | node = path->nodes[level]; |
925baedd | 1255 | |
ace75066 FM |
1256 | /* |
1257 | * Since the time between visiting leaves is much shorter than the time | |
1258 | * between visiting nodes, limit read ahead of nodes to 1, to avoid too | |
1259 | * much IO at once (possibly random). | |
1260 | */ | |
1261 | if (path->reada == READA_FORWARD_ALWAYS) { | |
1262 | if (level > 1) | |
1263 | nread_max = node->fs_info->nodesize; | |
1264 | else | |
1265 | nread_max = SZ_128K; | |
1266 | } else { | |
1267 | nread_max = SZ_64K; | |
1268 | } | |
1269 | ||
3c69faec | 1270 | search = btrfs_node_blockptr(node, slot); |
0b246afa | 1271 | blocksize = fs_info->nodesize; |
069a2e37 FM |
1272 | if (path->reada != READA_FORWARD_ALWAYS) { |
1273 | struct extent_buffer *eb; | |
1274 | ||
1275 | eb = find_extent_buffer(fs_info, search); | |
1276 | if (eb) { | |
1277 | free_extent_buffer(eb); | |
1278 | return; | |
1279 | } | |
3c69faec CM |
1280 | } |
1281 | ||
a7175319 | 1282 | target = search; |
6b80053d | 1283 | |
5f39d397 | 1284 | nritems = btrfs_header_nritems(node); |
6b80053d | 1285 | nr = slot; |
25b8b936 | 1286 | |
d397712b | 1287 | while (1) { |
e4058b54 | 1288 | if (path->reada == READA_BACK) { |
6b80053d CM |
1289 | if (nr == 0) |
1290 | break; | |
1291 | nr--; | |
ace75066 FM |
1292 | } else if (path->reada == READA_FORWARD || |
1293 | path->reada == READA_FORWARD_ALWAYS) { | |
6b80053d CM |
1294 | nr++; |
1295 | if (nr >= nritems) | |
1296 | break; | |
3c69faec | 1297 | } |
e4058b54 | 1298 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
1299 | btrfs_node_key(node, &disk_key, nr); |
1300 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1301 | break; | |
1302 | } | |
6b80053d | 1303 | search = btrfs_node_blockptr(node, nr); |
ace75066 FM |
1304 | if (path->reada == READA_FORWARD_ALWAYS || |
1305 | (search <= target && target - search <= 65536) || | |
a7175319 | 1306 | (search > target && search - target <= 65536)) { |
bfb484d9 | 1307 | btrfs_readahead_node_child(node, nr); |
6b80053d CM |
1308 | nread += blocksize; |
1309 | } | |
1310 | nscan++; | |
ace75066 | 1311 | if (nread > nread_max || nscan > 32) |
6b80053d | 1312 | break; |
3c69faec CM |
1313 | } |
1314 | } | |
925baedd | 1315 | |
bfb484d9 | 1316 | static noinline void reada_for_balance(struct btrfs_path *path, int level) |
b4ce94de | 1317 | { |
bfb484d9 | 1318 | struct extent_buffer *parent; |
b4ce94de CM |
1319 | int slot; |
1320 | int nritems; | |
b4ce94de | 1321 | |
8c594ea8 | 1322 | parent = path->nodes[level + 1]; |
b4ce94de | 1323 | if (!parent) |
0b08851f | 1324 | return; |
b4ce94de CM |
1325 | |
1326 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1327 | slot = path->slots[level + 1]; |
b4ce94de | 1328 | |
bfb484d9 JB |
1329 | if (slot > 0) |
1330 | btrfs_readahead_node_child(parent, slot - 1); | |
1331 | if (slot + 1 < nritems) | |
1332 | btrfs_readahead_node_child(parent, slot + 1); | |
b4ce94de CM |
1333 | } |
1334 | ||
1335 | ||
d352ac68 | 1336 | /* |
d397712b CM |
1337 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1338 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1339 | * operations on the tree might require changing key pointers higher up in the | |
1340 | * tree. | |
d352ac68 | 1341 | * |
d397712b CM |
1342 | * callers might also have set path->keep_locks, which tells this code to keep |
1343 | * the lock if the path points to the last slot in the block. This is part of | |
1344 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1345 | * |
d397712b CM |
1346 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1347 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1348 | */ |
e02119d5 | 1349 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1350 | int lowest_unlock, int min_write_lock_level, |
1351 | int *write_lock_level) | |
925baedd CM |
1352 | { |
1353 | int i; | |
1354 | int skip_level = level; | |
c1227996 | 1355 | bool check_skip = true; |
925baedd CM |
1356 | |
1357 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1358 | if (!path->nodes[i]) | |
1359 | break; | |
1360 | if (!path->locks[i]) | |
1361 | break; | |
c1227996 NB |
1362 | |
1363 | if (check_skip) { | |
1364 | if (path->slots[i] == 0) { | |
925baedd CM |
1365 | skip_level = i + 1; |
1366 | continue; | |
1367 | } | |
c1227996 NB |
1368 | |
1369 | if (path->keep_locks) { | |
1370 | u32 nritems; | |
1371 | ||
1372 | nritems = btrfs_header_nritems(path->nodes[i]); | |
1373 | if (nritems < 1 || path->slots[i] >= nritems - 1) { | |
1374 | skip_level = i + 1; | |
1375 | continue; | |
1376 | } | |
1377 | } | |
925baedd | 1378 | } |
051e1b9f | 1379 | |
d80bb3f9 | 1380 | if (i >= lowest_unlock && i > skip_level) { |
c1227996 NB |
1381 | check_skip = false; |
1382 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); | |
925baedd | 1383 | path->locks[i] = 0; |
f7c79f30 CM |
1384 | if (write_lock_level && |
1385 | i > min_write_lock_level && | |
1386 | i <= *write_lock_level) { | |
1387 | *write_lock_level = i - 1; | |
1388 | } | |
925baedd CM |
1389 | } |
1390 | } | |
1391 | } | |
1392 | ||
c8c42864 | 1393 | /* |
376a21d7 FM |
1394 | * Helper function for btrfs_search_slot() and other functions that do a search |
1395 | * on a btree. The goal is to find a tree block in the cache (the radix tree at | |
1396 | * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read | |
1397 | * its pages from disk. | |
c8c42864 | 1398 | * |
376a21d7 FM |
1399 | * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the |
1400 | * whole btree search, starting again from the current root node. | |
c8c42864 CM |
1401 | */ |
1402 | static int | |
d07b8528 LB |
1403 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
1404 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 1405 | const struct btrfs_key *key) |
c8c42864 | 1406 | { |
0b246afa | 1407 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
1408 | u64 blocknr; |
1409 | u64 gen; | |
c8c42864 | 1410 | struct extent_buffer *tmp; |
581c1760 | 1411 | struct btrfs_key first_key; |
76a05b35 | 1412 | int ret; |
581c1760 | 1413 | int parent_level; |
b246666e | 1414 | bool unlock_up; |
c8c42864 | 1415 | |
b246666e | 1416 | unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); |
213ff4b7 NB |
1417 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
1418 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
1419 | parent_level = btrfs_header_level(*eb_ret); | |
1420 | btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); | |
c8c42864 | 1421 | |
b246666e FM |
1422 | /* |
1423 | * If we need to read an extent buffer from disk and we are holding locks | |
1424 | * on upper level nodes, we unlock all the upper nodes before reading the | |
1425 | * extent buffer, and then return -EAGAIN to the caller as it needs to | |
1426 | * restart the search. We don't release the lock on the current level | |
1427 | * because we need to walk this node to figure out which blocks to read. | |
1428 | */ | |
0b246afa | 1429 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 1430 | if (tmp) { |
ace75066 FM |
1431 | if (p->reada == READA_FORWARD_ALWAYS) |
1432 | reada_for_search(fs_info, p, level, slot, key->objectid); | |
1433 | ||
b9fab919 | 1434 | /* first we do an atomic uptodate check */ |
bdf7c00e | 1435 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
1436 | /* |
1437 | * Do extra check for first_key, eb can be stale due to | |
1438 | * being cached, read from scrub, or have multiple | |
1439 | * parents (shared tree blocks). | |
1440 | */ | |
e064d5e9 | 1441 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
1442 | parent_level - 1, &first_key, gen)) { |
1443 | free_extent_buffer(tmp); | |
1444 | return -EUCLEAN; | |
1445 | } | |
bdf7c00e JB |
1446 | *eb_ret = tmp; |
1447 | return 0; | |
1448 | } | |
1449 | ||
b246666e FM |
1450 | if (unlock_up) |
1451 | btrfs_unlock_up_safe(p, level + 1); | |
1452 | ||
bdf7c00e | 1453 | /* now we're allowed to do a blocking uptodate check */ |
6a2e9dc4 | 1454 | ret = btrfs_read_extent_buffer(tmp, gen, parent_level - 1, &first_key); |
9a4ffa1b QW |
1455 | if (ret) { |
1456 | free_extent_buffer(tmp); | |
1457 | btrfs_release_path(p); | |
1458 | return -EIO; | |
cb44921a | 1459 | } |
88c602ab QW |
1460 | if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { |
1461 | free_extent_buffer(tmp); | |
1462 | btrfs_release_path(p); | |
1463 | return -EUCLEAN; | |
1464 | } | |
b246666e FM |
1465 | |
1466 | if (unlock_up) | |
1467 | ret = -EAGAIN; | |
1468 | ||
1469 | goto out; | |
c8c42864 CM |
1470 | } |
1471 | ||
b246666e | 1472 | if (unlock_up) { |
4bb59055 FM |
1473 | btrfs_unlock_up_safe(p, level + 1); |
1474 | ret = -EAGAIN; | |
1475 | } else { | |
1476 | ret = 0; | |
1477 | } | |
8c594ea8 | 1478 | |
e4058b54 | 1479 | if (p->reada != READA_NONE) |
2ff7e61e | 1480 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 1481 | |
1b7ec85e JB |
1482 | tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, |
1483 | gen, parent_level - 1, &first_key); | |
4eb150d6 QW |
1484 | if (IS_ERR(tmp)) { |
1485 | btrfs_release_path(p); | |
1486 | return PTR_ERR(tmp); | |
76a05b35 | 1487 | } |
4eb150d6 QW |
1488 | /* |
1489 | * If the read above didn't mark this buffer up to date, | |
1490 | * it will never end up being up to date. Set ret to EIO now | |
1491 | * and give up so that our caller doesn't loop forever | |
1492 | * on our EAGAINs. | |
1493 | */ | |
1494 | if (!extent_buffer_uptodate(tmp)) | |
1495 | ret = -EIO; | |
02a3307a | 1496 | |
b246666e | 1497 | out: |
4bb59055 FM |
1498 | if (ret == 0) { |
1499 | *eb_ret = tmp; | |
1500 | } else { | |
1501 | free_extent_buffer(tmp); | |
1502 | btrfs_release_path(p); | |
1503 | } | |
1504 | ||
76a05b35 | 1505 | return ret; |
c8c42864 CM |
1506 | } |
1507 | ||
1508 | /* | |
1509 | * helper function for btrfs_search_slot. This does all of the checks | |
1510 | * for node-level blocks and does any balancing required based on | |
1511 | * the ins_len. | |
1512 | * | |
1513 | * If no extra work was required, zero is returned. If we had to | |
1514 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1515 | * start over | |
1516 | */ | |
1517 | static int | |
1518 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1519 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1520 | struct extent_buffer *b, int level, int ins_len, |
1521 | int *write_lock_level) | |
c8c42864 | 1522 | { |
0b246afa | 1523 | struct btrfs_fs_info *fs_info = root->fs_info; |
95b982de | 1524 | int ret = 0; |
0b246afa | 1525 | |
c8c42864 | 1526 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 1527 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 | 1528 | |
bd681513 CM |
1529 | if (*write_lock_level < level + 1) { |
1530 | *write_lock_level = level + 1; | |
1531 | btrfs_release_path(p); | |
95b982de | 1532 | return -EAGAIN; |
bd681513 CM |
1533 | } |
1534 | ||
bfb484d9 | 1535 | reada_for_balance(p, level); |
95b982de | 1536 | ret = split_node(trans, root, p, level); |
c8c42864 | 1537 | |
c8c42864 CM |
1538 | b = p->nodes[level]; |
1539 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 1540 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 | 1541 | |
bd681513 CM |
1542 | if (*write_lock_level < level + 1) { |
1543 | *write_lock_level = level + 1; | |
1544 | btrfs_release_path(p); | |
95b982de | 1545 | return -EAGAIN; |
bd681513 CM |
1546 | } |
1547 | ||
bfb484d9 | 1548 | reada_for_balance(p, level); |
95b982de NB |
1549 | ret = balance_level(trans, root, p, level); |
1550 | if (ret) | |
1551 | return ret; | |
c8c42864 | 1552 | |
c8c42864 CM |
1553 | b = p->nodes[level]; |
1554 | if (!b) { | |
b3b4aa74 | 1555 | btrfs_release_path(p); |
95b982de | 1556 | return -EAGAIN; |
c8c42864 CM |
1557 | } |
1558 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1559 | } | |
c8c42864 CM |
1560 | return ret; |
1561 | } | |
1562 | ||
381cf658 | 1563 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
1564 | u64 iobjectid, u64 ioff, u8 key_type, |
1565 | struct btrfs_key *found_key) | |
1566 | { | |
1567 | int ret; | |
1568 | struct btrfs_key key; | |
1569 | struct extent_buffer *eb; | |
381cf658 DS |
1570 | |
1571 | ASSERT(path); | |
1d4c08e0 | 1572 | ASSERT(found_key); |
e33d5c3d KN |
1573 | |
1574 | key.type = key_type; | |
1575 | key.objectid = iobjectid; | |
1576 | key.offset = ioff; | |
1577 | ||
1578 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 1579 | if (ret < 0) |
e33d5c3d KN |
1580 | return ret; |
1581 | ||
1582 | eb = path->nodes[0]; | |
1583 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
1584 | ret = btrfs_next_leaf(fs_root, path); | |
1585 | if (ret) | |
1586 | return ret; | |
1587 | eb = path->nodes[0]; | |
1588 | } | |
1589 | ||
1590 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
1591 | if (found_key->type != key.type || | |
1592 | found_key->objectid != key.objectid) | |
1593 | return 1; | |
1594 | ||
1595 | return 0; | |
1596 | } | |
1597 | ||
1fc28d8e LB |
1598 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
1599 | struct btrfs_path *p, | |
1600 | int write_lock_level) | |
1601 | { | |
1fc28d8e | 1602 | struct extent_buffer *b; |
120de408 | 1603 | int root_lock = 0; |
1fc28d8e LB |
1604 | int level = 0; |
1605 | ||
1fc28d8e | 1606 | if (p->search_commit_root) { |
d96b3424 FM |
1607 | b = root->commit_root; |
1608 | atomic_inc(&b->refs); | |
be6821f8 | 1609 | level = btrfs_header_level(b); |
f9ddfd05 LB |
1610 | /* |
1611 | * Ensure that all callers have set skip_locking when | |
1612 | * p->search_commit_root = 1. | |
1613 | */ | |
1614 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
1615 | |
1616 | goto out; | |
1617 | } | |
1618 | ||
1619 | if (p->skip_locking) { | |
1620 | b = btrfs_root_node(root); | |
1621 | level = btrfs_header_level(b); | |
1622 | goto out; | |
1623 | } | |
1624 | ||
120de408 JB |
1625 | /* We try very hard to do read locks on the root */ |
1626 | root_lock = BTRFS_READ_LOCK; | |
1627 | ||
1fc28d8e | 1628 | /* |
662c653b LB |
1629 | * If the level is set to maximum, we can skip trying to get the read |
1630 | * lock. | |
1fc28d8e | 1631 | */ |
662c653b LB |
1632 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
1633 | /* | |
1634 | * We don't know the level of the root node until we actually | |
1635 | * have it read locked | |
1636 | */ | |
1bb96598 | 1637 | b = btrfs_read_lock_root_node(root); |
662c653b LB |
1638 | level = btrfs_header_level(b); |
1639 | if (level > write_lock_level) | |
1640 | goto out; | |
1641 | ||
1642 | /* Whoops, must trade for write lock */ | |
1643 | btrfs_tree_read_unlock(b); | |
1644 | free_extent_buffer(b); | |
1645 | } | |
1fc28d8e | 1646 | |
1fc28d8e LB |
1647 | b = btrfs_lock_root_node(root); |
1648 | root_lock = BTRFS_WRITE_LOCK; | |
1649 | ||
1650 | /* The level might have changed, check again */ | |
1651 | level = btrfs_header_level(b); | |
1652 | ||
1653 | out: | |
120de408 JB |
1654 | /* |
1655 | * The root may have failed to write out at some point, and thus is no | |
1656 | * longer valid, return an error in this case. | |
1657 | */ | |
1658 | if (!extent_buffer_uptodate(b)) { | |
1659 | if (root_lock) | |
1660 | btrfs_tree_unlock_rw(b, root_lock); | |
1661 | free_extent_buffer(b); | |
1662 | return ERR_PTR(-EIO); | |
1663 | } | |
1664 | ||
1fc28d8e LB |
1665 | p->nodes[level] = b; |
1666 | if (!p->skip_locking) | |
1667 | p->locks[level] = root_lock; | |
1668 | /* | |
1669 | * Callers are responsible for dropping b's references. | |
1670 | */ | |
1671 | return b; | |
1672 | } | |
1673 | ||
d96b3424 FM |
1674 | /* |
1675 | * Replace the extent buffer at the lowest level of the path with a cloned | |
1676 | * version. The purpose is to be able to use it safely, after releasing the | |
1677 | * commit root semaphore, even if relocation is happening in parallel, the | |
1678 | * transaction used for relocation is committed and the extent buffer is | |
1679 | * reallocated in the next transaction. | |
1680 | * | |
1681 | * This is used in a context where the caller does not prevent transaction | |
1682 | * commits from happening, either by holding a transaction handle or holding | |
1683 | * some lock, while it's doing searches through a commit root. | |
1684 | * At the moment it's only used for send operations. | |
1685 | */ | |
1686 | static int finish_need_commit_sem_search(struct btrfs_path *path) | |
1687 | { | |
1688 | const int i = path->lowest_level; | |
1689 | const int slot = path->slots[i]; | |
1690 | struct extent_buffer *lowest = path->nodes[i]; | |
1691 | struct extent_buffer *clone; | |
1692 | ||
1693 | ASSERT(path->need_commit_sem); | |
1694 | ||
1695 | if (!lowest) | |
1696 | return 0; | |
1697 | ||
1698 | lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); | |
1699 | ||
1700 | clone = btrfs_clone_extent_buffer(lowest); | |
1701 | if (!clone) | |
1702 | return -ENOMEM; | |
1703 | ||
1704 | btrfs_release_path(path); | |
1705 | path->nodes[i] = clone; | |
1706 | path->slots[i] = slot; | |
1707 | ||
1708 | return 0; | |
1709 | } | |
1fc28d8e | 1710 | |
e2e58d0f FM |
1711 | static inline int search_for_key_slot(struct extent_buffer *eb, |
1712 | int search_low_slot, | |
1713 | const struct btrfs_key *key, | |
1714 | int prev_cmp, | |
1715 | int *slot) | |
1716 | { | |
1717 | /* | |
1718 | * If a previous call to btrfs_bin_search() on a parent node returned an | |
1719 | * exact match (prev_cmp == 0), we can safely assume the target key will | |
1720 | * always be at slot 0 on lower levels, since each key pointer | |
1721 | * (struct btrfs_key_ptr) refers to the lowest key accessible from the | |
1722 | * subtree it points to. Thus we can skip searching lower levels. | |
1723 | */ | |
1724 | if (prev_cmp == 0) { | |
1725 | *slot = 0; | |
1726 | return 0; | |
1727 | } | |
1728 | ||
1729 | return generic_bin_search(eb, search_low_slot, key, slot); | |
1730 | } | |
1731 | ||
109324cf FM |
1732 | static int search_leaf(struct btrfs_trans_handle *trans, |
1733 | struct btrfs_root *root, | |
1734 | const struct btrfs_key *key, | |
1735 | struct btrfs_path *path, | |
1736 | int ins_len, | |
1737 | int prev_cmp) | |
1738 | { | |
1739 | struct extent_buffer *leaf = path->nodes[0]; | |
1740 | int leaf_free_space = -1; | |
1741 | int search_low_slot = 0; | |
1742 | int ret; | |
1743 | bool do_bin_search = true; | |
1744 | ||
1745 | /* | |
1746 | * If we are doing an insertion, the leaf has enough free space and the | |
1747 | * destination slot for the key is not slot 0, then we can unlock our | |
1748 | * write lock on the parent, and any other upper nodes, before doing the | |
1749 | * binary search on the leaf (with search_for_key_slot()), allowing other | |
1750 | * tasks to lock the parent and any other upper nodes. | |
1751 | */ | |
1752 | if (ins_len > 0) { | |
1753 | /* | |
1754 | * Cache the leaf free space, since we will need it later and it | |
1755 | * will not change until then. | |
1756 | */ | |
1757 | leaf_free_space = btrfs_leaf_free_space(leaf); | |
1758 | ||
1759 | /* | |
1760 | * !path->locks[1] means we have a single node tree, the leaf is | |
1761 | * the root of the tree. | |
1762 | */ | |
1763 | if (path->locks[1] && leaf_free_space >= ins_len) { | |
1764 | struct btrfs_disk_key first_key; | |
1765 | ||
1766 | ASSERT(btrfs_header_nritems(leaf) > 0); | |
1767 | btrfs_item_key(leaf, &first_key, 0); | |
1768 | ||
1769 | /* | |
1770 | * Doing the extra comparison with the first key is cheap, | |
1771 | * taking into account that the first key is very likely | |
1772 | * already in a cache line because it immediately follows | |
1773 | * the extent buffer's header and we have recently accessed | |
1774 | * the header's level field. | |
1775 | */ | |
1776 | ret = comp_keys(&first_key, key); | |
1777 | if (ret < 0) { | |
1778 | /* | |
1779 | * The first key is smaller than the key we want | |
1780 | * to insert, so we are safe to unlock all upper | |
1781 | * nodes and we have to do the binary search. | |
1782 | * | |
1783 | * We do use btrfs_unlock_up_safe() and not | |
1784 | * unlock_up() because the later does not unlock | |
1785 | * nodes with a slot of 0 - we can safely unlock | |
1786 | * any node even if its slot is 0 since in this | |
1787 | * case the key does not end up at slot 0 of the | |
1788 | * leaf and there's no need to split the leaf. | |
1789 | */ | |
1790 | btrfs_unlock_up_safe(path, 1); | |
1791 | search_low_slot = 1; | |
1792 | } else { | |
1793 | /* | |
1794 | * The first key is >= then the key we want to | |
1795 | * insert, so we can skip the binary search as | |
1796 | * the target key will be at slot 0. | |
1797 | * | |
1798 | * We can not unlock upper nodes when the key is | |
1799 | * less than the first key, because we will need | |
1800 | * to update the key at slot 0 of the parent node | |
1801 | * and possibly of other upper nodes too. | |
1802 | * If the key matches the first key, then we can | |
1803 | * unlock all the upper nodes, using | |
1804 | * btrfs_unlock_up_safe() instead of unlock_up() | |
1805 | * as stated above. | |
1806 | */ | |
1807 | if (ret == 0) | |
1808 | btrfs_unlock_up_safe(path, 1); | |
1809 | /* | |
1810 | * ret is already 0 or 1, matching the result of | |
1811 | * a btrfs_bin_search() call, so there is no need | |
1812 | * to adjust it. | |
1813 | */ | |
1814 | do_bin_search = false; | |
1815 | path->slots[0] = 0; | |
1816 | } | |
1817 | } | |
1818 | } | |
1819 | ||
1820 | if (do_bin_search) { | |
1821 | ret = search_for_key_slot(leaf, search_low_slot, key, | |
1822 | prev_cmp, &path->slots[0]); | |
1823 | if (ret < 0) | |
1824 | return ret; | |
1825 | } | |
1826 | ||
1827 | if (ins_len > 0) { | |
1828 | /* | |
1829 | * Item key already exists. In this case, if we are allowed to | |
1830 | * insert the item (for example, in dir_item case, item key | |
1831 | * collision is allowed), it will be merged with the original | |
1832 | * item. Only the item size grows, no new btrfs item will be | |
1833 | * added. If search_for_extension is not set, ins_len already | |
1834 | * accounts the size btrfs_item, deduct it here so leaf space | |
1835 | * check will be correct. | |
1836 | */ | |
1837 | if (ret == 0 && !path->search_for_extension) { | |
1838 | ASSERT(ins_len >= sizeof(struct btrfs_item)); | |
1839 | ins_len -= sizeof(struct btrfs_item); | |
1840 | } | |
1841 | ||
1842 | ASSERT(leaf_free_space >= 0); | |
1843 | ||
1844 | if (leaf_free_space < ins_len) { | |
1845 | int err; | |
1846 | ||
1847 | err = split_leaf(trans, root, key, path, ins_len, | |
1848 | (ret == 0)); | |
bb8e9a60 FM |
1849 | ASSERT(err <= 0); |
1850 | if (WARN_ON(err > 0)) | |
1851 | err = -EUCLEAN; | |
109324cf FM |
1852 | if (err) |
1853 | ret = err; | |
1854 | } | |
1855 | } | |
1856 | ||
1857 | return ret; | |
1858 | } | |
1859 | ||
74123bd7 | 1860 | /* |
4271ecea NB |
1861 | * btrfs_search_slot - look for a key in a tree and perform necessary |
1862 | * modifications to preserve tree invariants. | |
74123bd7 | 1863 | * |
4271ecea NB |
1864 | * @trans: Handle of transaction, used when modifying the tree |
1865 | * @p: Holds all btree nodes along the search path | |
1866 | * @root: The root node of the tree | |
1867 | * @key: The key we are looking for | |
9a664971 | 1868 | * @ins_len: Indicates purpose of search: |
1869 | * >0 for inserts it's size of item inserted (*) | |
1870 | * <0 for deletions | |
1871 | * 0 for plain searches, not modifying the tree | |
1872 | * | |
1873 | * (*) If size of item inserted doesn't include | |
1874 | * sizeof(struct btrfs_item), then p->search_for_extension must | |
1875 | * be set. | |
4271ecea NB |
1876 | * @cow: boolean should CoW operations be performed. Must always be 1 |
1877 | * when modifying the tree. | |
97571fd0 | 1878 | * |
4271ecea NB |
1879 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
1880 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
1881 | * | |
1882 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
1883 | * of the path (level 0) | |
1884 | * | |
1885 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
1886 | * points to the slot where it should be inserted | |
1887 | * | |
1888 | * If an error is encountered while searching the tree a negative error number | |
1889 | * is returned | |
74123bd7 | 1890 | */ |
310712b2 OS |
1891 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1892 | const struct btrfs_key *key, struct btrfs_path *p, | |
1893 | int ins_len, int cow) | |
be0e5c09 | 1894 | { |
d96b3424 | 1895 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 1896 | struct extent_buffer *b; |
be0e5c09 CM |
1897 | int slot; |
1898 | int ret; | |
33c66f43 | 1899 | int err; |
be0e5c09 | 1900 | int level; |
925baedd | 1901 | int lowest_unlock = 1; |
bd681513 CM |
1902 | /* everything at write_lock_level or lower must be write locked */ |
1903 | int write_lock_level = 0; | |
9f3a7427 | 1904 | u8 lowest_level = 0; |
f7c79f30 | 1905 | int min_write_lock_level; |
d7396f07 | 1906 | int prev_cmp; |
9f3a7427 | 1907 | |
6702ed49 | 1908 | lowest_level = p->lowest_level; |
323ac95b | 1909 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1910 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 1911 | BUG_ON(!cow && ins_len); |
25179201 | 1912 | |
bd681513 | 1913 | if (ins_len < 0) { |
925baedd | 1914 | lowest_unlock = 2; |
65b51a00 | 1915 | |
bd681513 CM |
1916 | /* when we are removing items, we might have to go up to level |
1917 | * two as we update tree pointers Make sure we keep write | |
1918 | * for those levels as well | |
1919 | */ | |
1920 | write_lock_level = 2; | |
1921 | } else if (ins_len > 0) { | |
1922 | /* | |
1923 | * for inserting items, make sure we have a write lock on | |
1924 | * level 1 so we can update keys | |
1925 | */ | |
1926 | write_lock_level = 1; | |
1927 | } | |
1928 | ||
1929 | if (!cow) | |
1930 | write_lock_level = -1; | |
1931 | ||
09a2a8f9 | 1932 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
1933 | write_lock_level = BTRFS_MAX_LEVEL; |
1934 | ||
f7c79f30 CM |
1935 | min_write_lock_level = write_lock_level; |
1936 | ||
d96b3424 FM |
1937 | if (p->need_commit_sem) { |
1938 | ASSERT(p->search_commit_root); | |
1939 | down_read(&fs_info->commit_root_sem); | |
1940 | } | |
1941 | ||
bb803951 | 1942 | again: |
d7396f07 | 1943 | prev_cmp = -1; |
1fc28d8e | 1944 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
1945 | if (IS_ERR(b)) { |
1946 | ret = PTR_ERR(b); | |
1947 | goto done; | |
1948 | } | |
925baedd | 1949 | |
eb60ceac | 1950 | while (b) { |
f624d976 QW |
1951 | int dec = 0; |
1952 | ||
5f39d397 | 1953 | level = btrfs_header_level(b); |
65b51a00 | 1954 | |
02217ed2 | 1955 | if (cow) { |
9ea2c7c9 NB |
1956 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
1957 | ||
c8c42864 CM |
1958 | /* |
1959 | * if we don't really need to cow this block | |
1960 | * then we don't want to set the path blocking, | |
1961 | * so we test it here | |
1962 | */ | |
5963ffca | 1963 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 1964 | goto cow_done; |
5d4f98a2 | 1965 | |
bd681513 CM |
1966 | /* |
1967 | * must have write locks on this node and the | |
1968 | * parent | |
1969 | */ | |
5124e00e JB |
1970 | if (level > write_lock_level || |
1971 | (level + 1 > write_lock_level && | |
1972 | level + 1 < BTRFS_MAX_LEVEL && | |
1973 | p->nodes[level + 1])) { | |
bd681513 CM |
1974 | write_lock_level = level + 1; |
1975 | btrfs_release_path(p); | |
1976 | goto again; | |
1977 | } | |
1978 | ||
9ea2c7c9 NB |
1979 | if (last_level) |
1980 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
9631e4cc JB |
1981 | &b, |
1982 | BTRFS_NESTING_COW); | |
9ea2c7c9 NB |
1983 | else |
1984 | err = btrfs_cow_block(trans, root, b, | |
1985 | p->nodes[level + 1], | |
9631e4cc JB |
1986 | p->slots[level + 1], &b, |
1987 | BTRFS_NESTING_COW); | |
33c66f43 | 1988 | if (err) { |
33c66f43 | 1989 | ret = err; |
65b51a00 | 1990 | goto done; |
54aa1f4d | 1991 | } |
02217ed2 | 1992 | } |
65b51a00 | 1993 | cow_done: |
eb60ceac | 1994 | p->nodes[level] = b; |
b4ce94de CM |
1995 | |
1996 | /* | |
1997 | * we have a lock on b and as long as we aren't changing | |
1998 | * the tree, there is no way to for the items in b to change. | |
1999 | * It is safe to drop the lock on our parent before we | |
2000 | * go through the expensive btree search on b. | |
2001 | * | |
eb653de1 FDBM |
2002 | * If we're inserting or deleting (ins_len != 0), then we might |
2003 | * be changing slot zero, which may require changing the parent. | |
2004 | * So, we can't drop the lock until after we know which slot | |
2005 | * we're operating on. | |
b4ce94de | 2006 | */ |
eb653de1 FDBM |
2007 | if (!ins_len && !p->keep_locks) { |
2008 | int u = level + 1; | |
2009 | ||
2010 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2011 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2012 | p->locks[u] = 0; | |
2013 | } | |
2014 | } | |
b4ce94de | 2015 | |
e2e58d0f | 2016 | if (level == 0) { |
109324cf | 2017 | if (ins_len > 0) |
e5e1c174 | 2018 | ASSERT(write_lock_level >= 1); |
bd681513 | 2019 | |
109324cf | 2020 | ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); |
459931ec | 2021 | if (!p->search_for_split) |
f7c79f30 | 2022 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2023 | min_write_lock_level, NULL); |
65b51a00 | 2024 | goto done; |
be0e5c09 | 2025 | } |
e2e58d0f FM |
2026 | |
2027 | ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); | |
2028 | if (ret < 0) | |
2029 | goto done; | |
2030 | prev_cmp = ret; | |
2031 | ||
f624d976 QW |
2032 | if (ret && slot > 0) { |
2033 | dec = 1; | |
2034 | slot--; | |
2035 | } | |
2036 | p->slots[level] = slot; | |
2037 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2038 | &write_lock_level); | |
2039 | if (err == -EAGAIN) | |
2040 | goto again; | |
2041 | if (err) { | |
2042 | ret = err; | |
2043 | goto done; | |
2044 | } | |
2045 | b = p->nodes[level]; | |
2046 | slot = p->slots[level]; | |
2047 | ||
2048 | /* | |
2049 | * Slot 0 is special, if we change the key we have to update | |
2050 | * the parent pointer which means we must have a write lock on | |
2051 | * the parent | |
2052 | */ | |
2053 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2054 | write_lock_level = level + 1; | |
2055 | btrfs_release_path(p); | |
2056 | goto again; | |
2057 | } | |
2058 | ||
2059 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2060 | &write_lock_level); | |
2061 | ||
2062 | if (level == lowest_level) { | |
2063 | if (dec) | |
2064 | p->slots[level]++; | |
2065 | goto done; | |
2066 | } | |
2067 | ||
2068 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2069 | if (err == -EAGAIN) | |
2070 | goto again; | |
2071 | if (err) { | |
2072 | ret = err; | |
2073 | goto done; | |
2074 | } | |
2075 | ||
2076 | if (!p->skip_locking) { | |
2077 | level = btrfs_header_level(b); | |
b40130b2 JB |
2078 | |
2079 | btrfs_maybe_reset_lockdep_class(root, b); | |
2080 | ||
f624d976 | 2081 | if (level <= write_lock_level) { |
ac5887c8 | 2082 | btrfs_tree_lock(b); |
f624d976 QW |
2083 | p->locks[level] = BTRFS_WRITE_LOCK; |
2084 | } else { | |
fe596ca3 | 2085 | btrfs_tree_read_lock(b); |
f624d976 QW |
2086 | p->locks[level] = BTRFS_READ_LOCK; |
2087 | } | |
2088 | p->nodes[level] = b; | |
2089 | } | |
be0e5c09 | 2090 | } |
65b51a00 CM |
2091 | ret = 1; |
2092 | done: | |
5f5bc6b1 | 2093 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2094 | btrfs_release_path(p); |
d96b3424 FM |
2095 | |
2096 | if (p->need_commit_sem) { | |
2097 | int ret2; | |
2098 | ||
2099 | ret2 = finish_need_commit_sem_search(p); | |
2100 | up_read(&fs_info->commit_root_sem); | |
2101 | if (ret2) | |
2102 | ret = ret2; | |
2103 | } | |
2104 | ||
65b51a00 | 2105 | return ret; |
be0e5c09 | 2106 | } |
f75e2b79 | 2107 | ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); |
be0e5c09 | 2108 | |
5d9e75c4 JS |
2109 | /* |
2110 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2111 | * current state of the tree together with the operations recorded in the tree | |
2112 | * modification log to search for the key in a previous version of this tree, as | |
2113 | * denoted by the time_seq parameter. | |
2114 | * | |
2115 | * Naturally, there is no support for insert, delete or cow operations. | |
2116 | * | |
2117 | * The resulting path and return value will be set up as if we called | |
2118 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2119 | */ | |
310712b2 | 2120 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2121 | struct btrfs_path *p, u64 time_seq) |
2122 | { | |
0b246afa | 2123 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2124 | struct extent_buffer *b; |
2125 | int slot; | |
2126 | int ret; | |
2127 | int err; | |
2128 | int level; | |
2129 | int lowest_unlock = 1; | |
2130 | u8 lowest_level = 0; | |
2131 | ||
2132 | lowest_level = p->lowest_level; | |
2133 | WARN_ON(p->nodes[0] != NULL); | |
2134 | ||
2135 | if (p->search_commit_root) { | |
2136 | BUG_ON(time_seq); | |
2137 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2138 | } | |
2139 | ||
2140 | again: | |
f3a84ccd | 2141 | b = btrfs_get_old_root(root, time_seq); |
315bed43 NB |
2142 | if (!b) { |
2143 | ret = -EIO; | |
2144 | goto done; | |
2145 | } | |
5d9e75c4 | 2146 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2147 | p->locks[level] = BTRFS_READ_LOCK; |
2148 | ||
2149 | while (b) { | |
abe9339d QW |
2150 | int dec = 0; |
2151 | ||
5d9e75c4 JS |
2152 | level = btrfs_header_level(b); |
2153 | p->nodes[level] = b; | |
5d9e75c4 JS |
2154 | |
2155 | /* | |
2156 | * we have a lock on b and as long as we aren't changing | |
2157 | * the tree, there is no way to for the items in b to change. | |
2158 | * It is safe to drop the lock on our parent before we | |
2159 | * go through the expensive btree search on b. | |
2160 | */ | |
2161 | btrfs_unlock_up_safe(p, level + 1); | |
2162 | ||
995e9a16 | 2163 | ret = btrfs_bin_search(b, key, &slot); |
cbca7d59 FM |
2164 | if (ret < 0) |
2165 | goto done; | |
5d9e75c4 | 2166 | |
abe9339d | 2167 | if (level == 0) { |
5d9e75c4 JS |
2168 | p->slots[level] = slot; |
2169 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2170 | goto done; |
2171 | } | |
5d9e75c4 | 2172 | |
abe9339d QW |
2173 | if (ret && slot > 0) { |
2174 | dec = 1; | |
2175 | slot--; | |
2176 | } | |
2177 | p->slots[level] = slot; | |
2178 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2179 | |
abe9339d QW |
2180 | if (level == lowest_level) { |
2181 | if (dec) | |
2182 | p->slots[level]++; | |
2183 | goto done; | |
2184 | } | |
5d9e75c4 | 2185 | |
abe9339d QW |
2186 | err = read_block_for_search(root, p, &b, level, slot, key); |
2187 | if (err == -EAGAIN) | |
2188 | goto again; | |
2189 | if (err) { | |
2190 | ret = err; | |
5d9e75c4 JS |
2191 | goto done; |
2192 | } | |
abe9339d QW |
2193 | |
2194 | level = btrfs_header_level(b); | |
ac5887c8 | 2195 | btrfs_tree_read_lock(b); |
f3a84ccd | 2196 | b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); |
abe9339d QW |
2197 | if (!b) { |
2198 | ret = -ENOMEM; | |
2199 | goto done; | |
2200 | } | |
2201 | p->locks[level] = BTRFS_READ_LOCK; | |
2202 | p->nodes[level] = b; | |
5d9e75c4 JS |
2203 | } |
2204 | ret = 1; | |
2205 | done: | |
5d9e75c4 JS |
2206 | if (ret < 0) |
2207 | btrfs_release_path(p); | |
2208 | ||
2209 | return ret; | |
2210 | } | |
2211 | ||
2f38b3e1 AJ |
2212 | /* |
2213 | * helper to use instead of search slot if no exact match is needed but | |
2214 | * instead the next or previous item should be returned. | |
2215 | * When find_higher is true, the next higher item is returned, the next lower | |
2216 | * otherwise. | |
2217 | * When return_any and find_higher are both true, and no higher item is found, | |
2218 | * return the next lower instead. | |
2219 | * When return_any is true and find_higher is false, and no lower item is found, | |
2220 | * return the next higher instead. | |
2221 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2222 | * < 0 on error | |
2223 | */ | |
2224 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2225 | const struct btrfs_key *key, |
2226 | struct btrfs_path *p, int find_higher, | |
2227 | int return_any) | |
2f38b3e1 AJ |
2228 | { |
2229 | int ret; | |
2230 | struct extent_buffer *leaf; | |
2231 | ||
2232 | again: | |
2233 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2234 | if (ret <= 0) | |
2235 | return ret; | |
2236 | /* | |
2237 | * a return value of 1 means the path is at the position where the | |
2238 | * item should be inserted. Normally this is the next bigger item, | |
2239 | * but in case the previous item is the last in a leaf, path points | |
2240 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2241 | * item. | |
2242 | */ | |
2243 | leaf = p->nodes[0]; | |
2244 | ||
2245 | if (find_higher) { | |
2246 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2247 | ret = btrfs_next_leaf(root, p); | |
2248 | if (ret <= 0) | |
2249 | return ret; | |
2250 | if (!return_any) | |
2251 | return 1; | |
2252 | /* | |
2253 | * no higher item found, return the next | |
2254 | * lower instead | |
2255 | */ | |
2256 | return_any = 0; | |
2257 | find_higher = 0; | |
2258 | btrfs_release_path(p); | |
2259 | goto again; | |
2260 | } | |
2261 | } else { | |
e6793769 AJ |
2262 | if (p->slots[0] == 0) { |
2263 | ret = btrfs_prev_leaf(root, p); | |
2264 | if (ret < 0) | |
2265 | return ret; | |
2266 | if (!ret) { | |
23c6bf6a FDBM |
2267 | leaf = p->nodes[0]; |
2268 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
2269 | p->slots[0]--; | |
e6793769 | 2270 | return 0; |
2f38b3e1 | 2271 | } |
e6793769 AJ |
2272 | if (!return_any) |
2273 | return 1; | |
2274 | /* | |
2275 | * no lower item found, return the next | |
2276 | * higher instead | |
2277 | */ | |
2278 | return_any = 0; | |
2279 | find_higher = 1; | |
2280 | btrfs_release_path(p); | |
2281 | goto again; | |
2282 | } else { | |
2f38b3e1 AJ |
2283 | --p->slots[0]; |
2284 | } | |
2285 | } | |
2286 | return 0; | |
2287 | } | |
2288 | ||
0ff40a91 MPS |
2289 | /* |
2290 | * Execute search and call btrfs_previous_item to traverse backwards if the item | |
2291 | * was not found. | |
2292 | * | |
2293 | * Return 0 if found, 1 if not found and < 0 if error. | |
2294 | */ | |
2295 | int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, | |
2296 | struct btrfs_path *path) | |
2297 | { | |
2298 | int ret; | |
2299 | ||
2300 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
2301 | if (ret > 0) | |
2302 | ret = btrfs_previous_item(root, path, key->objectid, key->type); | |
2303 | ||
2304 | if (ret == 0) | |
2305 | btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); | |
2306 | ||
2307 | return ret; | |
2308 | } | |
2309 | ||
62142be3 GN |
2310 | /** |
2311 | * Search for a valid slot for the given path. | |
2312 | * | |
2313 | * @root: The root node of the tree. | |
2314 | * @key: Will contain a valid item if found. | |
2315 | * @path: The starting point to validate the slot. | |
2316 | * | |
2317 | * Return: 0 if the item is valid | |
2318 | * 1 if not found | |
2319 | * <0 if error. | |
2320 | */ | |
2321 | int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, | |
2322 | struct btrfs_path *path) | |
2323 | { | |
2324 | while (1) { | |
2325 | int ret; | |
2326 | const int slot = path->slots[0]; | |
2327 | const struct extent_buffer *leaf = path->nodes[0]; | |
2328 | ||
2329 | /* This is where we start walking the path. */ | |
2330 | if (slot >= btrfs_header_nritems(leaf)) { | |
2331 | /* | |
2332 | * If we've reached the last slot in this leaf we need | |
2333 | * to go to the next leaf and reset the path. | |
2334 | */ | |
2335 | ret = btrfs_next_leaf(root, path); | |
2336 | if (ret) | |
2337 | return ret; | |
2338 | continue; | |
2339 | } | |
2340 | /* Store the found, valid item in @key. */ | |
2341 | btrfs_item_key_to_cpu(leaf, key, slot); | |
2342 | break; | |
2343 | } | |
2344 | return 0; | |
2345 | } | |
2346 | ||
74123bd7 CM |
2347 | /* |
2348 | * adjust the pointers going up the tree, starting at level | |
2349 | * making sure the right key of each node is points to 'key'. | |
2350 | * This is used after shifting pointers to the left, so it stops | |
2351 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2352 | * higher levels | |
aa5d6bed | 2353 | * |
74123bd7 | 2354 | */ |
b167fa91 | 2355 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 2356 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2357 | { |
2358 | int i; | |
5f39d397 | 2359 | struct extent_buffer *t; |
0e82bcfe | 2360 | int ret; |
5f39d397 | 2361 | |
234b63a0 | 2362 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2363 | int tslot = path->slots[i]; |
0e82bcfe | 2364 | |
eb60ceac | 2365 | if (!path->nodes[i]) |
be0e5c09 | 2366 | break; |
5f39d397 | 2367 | t = path->nodes[i]; |
f3a84ccd FM |
2368 | ret = btrfs_tree_mod_log_insert_key(t, tslot, |
2369 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_ATOMIC); | |
0e82bcfe | 2370 | BUG_ON(ret < 0); |
5f39d397 | 2371 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2372 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2373 | if (tslot != 0) |
2374 | break; | |
2375 | } | |
2376 | } | |
2377 | ||
31840ae1 ZY |
2378 | /* |
2379 | * update item key. | |
2380 | * | |
2381 | * This function isn't completely safe. It's the caller's responsibility | |
2382 | * that the new key won't break the order | |
2383 | */ | |
b7a0365e DD |
2384 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
2385 | struct btrfs_path *path, | |
310712b2 | 2386 | const struct btrfs_key *new_key) |
31840ae1 ZY |
2387 | { |
2388 | struct btrfs_disk_key disk_key; | |
2389 | struct extent_buffer *eb; | |
2390 | int slot; | |
2391 | ||
2392 | eb = path->nodes[0]; | |
2393 | slot = path->slots[0]; | |
2394 | if (slot > 0) { | |
2395 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
2396 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
2397 | btrfs_crit(fs_info, | |
2398 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2399 | slot, btrfs_disk_key_objectid(&disk_key), | |
2400 | btrfs_disk_key_type(&disk_key), | |
2401 | btrfs_disk_key_offset(&disk_key), | |
2402 | new_key->objectid, new_key->type, | |
2403 | new_key->offset); | |
2404 | btrfs_print_leaf(eb); | |
2405 | BUG(); | |
2406 | } | |
31840ae1 ZY |
2407 | } |
2408 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2409 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
2410 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
2411 | btrfs_crit(fs_info, | |
2412 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2413 | slot, btrfs_disk_key_objectid(&disk_key), | |
2414 | btrfs_disk_key_type(&disk_key), | |
2415 | btrfs_disk_key_offset(&disk_key), | |
2416 | new_key->objectid, new_key->type, | |
2417 | new_key->offset); | |
2418 | btrfs_print_leaf(eb); | |
2419 | BUG(); | |
2420 | } | |
31840ae1 ZY |
2421 | } |
2422 | ||
2423 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2424 | btrfs_set_item_key(eb, &disk_key, slot); | |
2425 | btrfs_mark_buffer_dirty(eb); | |
2426 | if (slot == 0) | |
b167fa91 | 2427 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
2428 | } |
2429 | ||
d16c702f QW |
2430 | /* |
2431 | * Check key order of two sibling extent buffers. | |
2432 | * | |
2433 | * Return true if something is wrong. | |
2434 | * Return false if everything is fine. | |
2435 | * | |
2436 | * Tree-checker only works inside one tree block, thus the following | |
2437 | * corruption can not be detected by tree-checker: | |
2438 | * | |
2439 | * Leaf @left | Leaf @right | |
2440 | * -------------------------------------------------------------- | |
2441 | * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | | |
2442 | * | |
2443 | * Key f6 in leaf @left itself is valid, but not valid when the next | |
2444 | * key in leaf @right is 7. | |
2445 | * This can only be checked at tree block merge time. | |
2446 | * And since tree checker has ensured all key order in each tree block | |
2447 | * is correct, we only need to bother the last key of @left and the first | |
2448 | * key of @right. | |
2449 | */ | |
2450 | static bool check_sibling_keys(struct extent_buffer *left, | |
2451 | struct extent_buffer *right) | |
2452 | { | |
2453 | struct btrfs_key left_last; | |
2454 | struct btrfs_key right_first; | |
2455 | int level = btrfs_header_level(left); | |
2456 | int nr_left = btrfs_header_nritems(left); | |
2457 | int nr_right = btrfs_header_nritems(right); | |
2458 | ||
2459 | /* No key to check in one of the tree blocks */ | |
2460 | if (!nr_left || !nr_right) | |
2461 | return false; | |
2462 | ||
2463 | if (level) { | |
2464 | btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); | |
2465 | btrfs_node_key_to_cpu(right, &right_first, 0); | |
2466 | } else { | |
2467 | btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); | |
2468 | btrfs_item_key_to_cpu(right, &right_first, 0); | |
2469 | } | |
2470 | ||
2471 | if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { | |
2472 | btrfs_crit(left->fs_info, | |
2473 | "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", | |
2474 | left_last.objectid, left_last.type, | |
2475 | left_last.offset, right_first.objectid, | |
2476 | right_first.type, right_first.offset); | |
2477 | return true; | |
2478 | } | |
2479 | return false; | |
2480 | } | |
2481 | ||
74123bd7 CM |
2482 | /* |
2483 | * try to push data from one node into the next node left in the | |
79f95c82 | 2484 | * tree. |
aa5d6bed CM |
2485 | * |
2486 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2487 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2488 | */ |
98ed5174 | 2489 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 2490 | struct extent_buffer *dst, |
971a1f66 | 2491 | struct extent_buffer *src, int empty) |
be0e5c09 | 2492 | { |
d30a668f | 2493 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 2494 | int push_items = 0; |
bb803951 CM |
2495 | int src_nritems; |
2496 | int dst_nritems; | |
aa5d6bed | 2497 | int ret = 0; |
be0e5c09 | 2498 | |
5f39d397 CM |
2499 | src_nritems = btrfs_header_nritems(src); |
2500 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2501 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
2502 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2503 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2504 | |
bce4eae9 | 2505 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2506 | return 1; |
2507 | ||
d397712b | 2508 | if (push_items <= 0) |
be0e5c09 CM |
2509 | return 1; |
2510 | ||
bce4eae9 | 2511 | if (empty) { |
971a1f66 | 2512 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2513 | if (push_items < src_nritems) { |
2514 | /* leave at least 8 pointers in the node if | |
2515 | * we aren't going to empty it | |
2516 | */ | |
2517 | if (src_nritems - push_items < 8) { | |
2518 | if (push_items <= 8) | |
2519 | return 1; | |
2520 | push_items -= 8; | |
2521 | } | |
2522 | } | |
2523 | } else | |
2524 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2525 | |
d16c702f QW |
2526 | /* dst is the left eb, src is the middle eb */ |
2527 | if (check_sibling_keys(dst, src)) { | |
2528 | ret = -EUCLEAN; | |
2529 | btrfs_abort_transaction(trans, ret); | |
2530 | return ret; | |
2531 | } | |
f3a84ccd | 2532 | ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 2533 | if (ret) { |
66642832 | 2534 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2535 | return ret; |
2536 | } | |
5f39d397 CM |
2537 | copy_extent_buffer(dst, src, |
2538 | btrfs_node_key_ptr_offset(dst_nritems), | |
2539 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2540 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2541 | |
bb803951 | 2542 | if (push_items < src_nritems) { |
57911b8b | 2543 | /* |
f3a84ccd FM |
2544 | * Don't call btrfs_tree_mod_log_insert_move() here, key removal |
2545 | * was already fully logged by btrfs_tree_mod_log_eb_copy() above. | |
57911b8b | 2546 | */ |
5f39d397 CM |
2547 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2548 | btrfs_node_key_ptr_offset(push_items), | |
2549 | (src_nritems - push_items) * | |
2550 | sizeof(struct btrfs_key_ptr)); | |
2551 | } | |
2552 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2553 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2554 | btrfs_mark_buffer_dirty(src); | |
2555 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2556 | |
79f95c82 CM |
2557 | return ret; |
2558 | } | |
2559 | ||
2560 | /* | |
2561 | * try to push data from one node into the next node right in the | |
2562 | * tree. | |
2563 | * | |
2564 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2565 | * error, and > 0 if there was no room in the right hand block. | |
2566 | * | |
2567 | * this will only push up to 1/2 the contents of the left node over | |
2568 | */ | |
5f39d397 | 2569 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2570 | struct extent_buffer *dst, |
2571 | struct extent_buffer *src) | |
79f95c82 | 2572 | { |
55d32ed8 | 2573 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
2574 | int push_items = 0; |
2575 | int max_push; | |
2576 | int src_nritems; | |
2577 | int dst_nritems; | |
2578 | int ret = 0; | |
79f95c82 | 2579 | |
7bb86316 CM |
2580 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2581 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2582 | ||
5f39d397 CM |
2583 | src_nritems = btrfs_header_nritems(src); |
2584 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2585 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 2586 | if (push_items <= 0) |
79f95c82 | 2587 | return 1; |
bce4eae9 | 2588 | |
d397712b | 2589 | if (src_nritems < 4) |
bce4eae9 | 2590 | return 1; |
79f95c82 CM |
2591 | |
2592 | max_push = src_nritems / 2 + 1; | |
2593 | /* don't try to empty the node */ | |
d397712b | 2594 | if (max_push >= src_nritems) |
79f95c82 | 2595 | return 1; |
252c38f0 | 2596 | |
79f95c82 CM |
2597 | if (max_push < push_items) |
2598 | push_items = max_push; | |
2599 | ||
d16c702f QW |
2600 | /* dst is the right eb, src is the middle eb */ |
2601 | if (check_sibling_keys(src, dst)) { | |
2602 | ret = -EUCLEAN; | |
2603 | btrfs_abort_transaction(trans, ret); | |
2604 | return ret; | |
2605 | } | |
f3a84ccd | 2606 | ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
bf1d3425 | 2607 | BUG_ON(ret < 0); |
5f39d397 CM |
2608 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2609 | btrfs_node_key_ptr_offset(0), | |
2610 | (dst_nritems) * | |
2611 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2612 | |
f3a84ccd FM |
2613 | ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
2614 | push_items); | |
5de865ee | 2615 | if (ret) { |
66642832 | 2616 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2617 | return ret; |
2618 | } | |
5f39d397 CM |
2619 | copy_extent_buffer(dst, src, |
2620 | btrfs_node_key_ptr_offset(0), | |
2621 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2622 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2623 | |
5f39d397 CM |
2624 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2625 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2626 | |
5f39d397 CM |
2627 | btrfs_mark_buffer_dirty(src); |
2628 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2629 | |
aa5d6bed | 2630 | return ret; |
be0e5c09 CM |
2631 | } |
2632 | ||
97571fd0 CM |
2633 | /* |
2634 | * helper function to insert a new root level in the tree. | |
2635 | * A new node is allocated, and a single item is inserted to | |
2636 | * point to the existing root | |
aa5d6bed CM |
2637 | * |
2638 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2639 | */ |
d397712b | 2640 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 2641 | struct btrfs_root *root, |
fdd99c72 | 2642 | struct btrfs_path *path, int level) |
5c680ed6 | 2643 | { |
0b246afa | 2644 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 2645 | u64 lower_gen; |
5f39d397 CM |
2646 | struct extent_buffer *lower; |
2647 | struct extent_buffer *c; | |
925baedd | 2648 | struct extent_buffer *old; |
5f39d397 | 2649 | struct btrfs_disk_key lower_key; |
d9d19a01 | 2650 | int ret; |
5c680ed6 CM |
2651 | |
2652 | BUG_ON(path->nodes[level]); | |
2653 | BUG_ON(path->nodes[level-1] != root->node); | |
2654 | ||
7bb86316 CM |
2655 | lower = path->nodes[level-1]; |
2656 | if (level == 1) | |
2657 | btrfs_item_key(lower, &lower_key, 0); | |
2658 | else | |
2659 | btrfs_node_key(lower, &lower_key, 0); | |
2660 | ||
79bd3712 FM |
2661 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2662 | &lower_key, level, root->node->start, 0, | |
2663 | BTRFS_NESTING_NEW_ROOT); | |
5f39d397 CM |
2664 | if (IS_ERR(c)) |
2665 | return PTR_ERR(c); | |
925baedd | 2666 | |
0b246afa | 2667 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 2668 | |
5f39d397 | 2669 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 2670 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2671 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2672 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2673 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2674 | |
2675 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2676 | |
5f39d397 | 2677 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2678 | |
925baedd | 2679 | old = root->node; |
406808ab | 2680 | ret = btrfs_tree_mod_log_insert_root(root->node, c, false); |
d9d19a01 | 2681 | BUG_ON(ret < 0); |
240f62c8 | 2682 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2683 | |
2684 | /* the super has an extra ref to root->node */ | |
2685 | free_extent_buffer(old); | |
2686 | ||
0b86a832 | 2687 | add_root_to_dirty_list(root); |
67439dad | 2688 | atomic_inc(&c->refs); |
5f39d397 | 2689 | path->nodes[level] = c; |
ac5887c8 | 2690 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2691 | path->slots[level] = 0; |
2692 | return 0; | |
2693 | } | |
2694 | ||
74123bd7 CM |
2695 | /* |
2696 | * worker function to insert a single pointer in a node. | |
2697 | * the node should have enough room for the pointer already | |
97571fd0 | 2698 | * |
74123bd7 CM |
2699 | * slot and level indicate where you want the key to go, and |
2700 | * blocknr is the block the key points to. | |
2701 | */ | |
143bede5 | 2702 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 2703 | struct btrfs_path *path, |
143bede5 | 2704 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 2705 | int slot, int level) |
74123bd7 | 2706 | { |
5f39d397 | 2707 | struct extent_buffer *lower; |
74123bd7 | 2708 | int nritems; |
f3ea38da | 2709 | int ret; |
5c680ed6 CM |
2710 | |
2711 | BUG_ON(!path->nodes[level]); | |
49d0c642 | 2712 | btrfs_assert_tree_write_locked(path->nodes[level]); |
5f39d397 CM |
2713 | lower = path->nodes[level]; |
2714 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2715 | BUG_ON(slot > nritems); |
6ad3cf6d | 2716 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 2717 | if (slot != nritems) { |
bf1d3425 | 2718 | if (level) { |
f3a84ccd FM |
2719 | ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, |
2720 | slot, nritems - slot); | |
bf1d3425 DS |
2721 | BUG_ON(ret < 0); |
2722 | } | |
5f39d397 CM |
2723 | memmove_extent_buffer(lower, |
2724 | btrfs_node_key_ptr_offset(slot + 1), | |
2725 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2726 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2727 | } |
c3e06965 | 2728 | if (level) { |
f3a84ccd FM |
2729 | ret = btrfs_tree_mod_log_insert_key(lower, slot, |
2730 | BTRFS_MOD_LOG_KEY_ADD, GFP_NOFS); | |
f3ea38da JS |
2731 | BUG_ON(ret < 0); |
2732 | } | |
5f39d397 | 2733 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2734 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2735 | WARN_ON(trans->transid == 0); |
2736 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2737 | btrfs_set_header_nritems(lower, nritems + 1); |
2738 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2739 | } |
2740 | ||
97571fd0 CM |
2741 | /* |
2742 | * split the node at the specified level in path in two. | |
2743 | * The path is corrected to point to the appropriate node after the split | |
2744 | * | |
2745 | * Before splitting this tries to make some room in the node by pushing | |
2746 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2747 | * |
2748 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2749 | */ |
e02119d5 CM |
2750 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2751 | struct btrfs_root *root, | |
2752 | struct btrfs_path *path, int level) | |
be0e5c09 | 2753 | { |
0b246afa | 2754 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2755 | struct extent_buffer *c; |
2756 | struct extent_buffer *split; | |
2757 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2758 | int mid; |
5c680ed6 | 2759 | int ret; |
7518a238 | 2760 | u32 c_nritems; |
eb60ceac | 2761 | |
5f39d397 | 2762 | c = path->nodes[level]; |
7bb86316 | 2763 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2764 | if (c == root->node) { |
d9abbf1c | 2765 | /* |
90f8d62e JS |
2766 | * trying to split the root, lets make a new one |
2767 | * | |
fdd99c72 | 2768 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
2769 | * insert_new_root, because that root buffer will be kept as a |
2770 | * normal node. We are going to log removal of half of the | |
f3a84ccd FM |
2771 | * elements below with btrfs_tree_mod_log_eb_copy(). We're |
2772 | * holding a tree lock on the buffer, which is why we cannot | |
2773 | * race with other tree_mod_log users. | |
d9abbf1c | 2774 | */ |
fdd99c72 | 2775 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2776 | if (ret) |
2777 | return ret; | |
b3612421 | 2778 | } else { |
e66f709b | 2779 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2780 | c = path->nodes[level]; |
2781 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 2782 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 2783 | return 0; |
54aa1f4d CM |
2784 | if (ret < 0) |
2785 | return ret; | |
be0e5c09 | 2786 | } |
e66f709b | 2787 | |
5f39d397 | 2788 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2789 | mid = (c_nritems + 1) / 2; |
2790 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2791 | |
79bd3712 FM |
2792 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2793 | &disk_key, level, c->start, 0, | |
2794 | BTRFS_NESTING_SPLIT); | |
5f39d397 CM |
2795 | if (IS_ERR(split)) |
2796 | return PTR_ERR(split); | |
2797 | ||
0b246afa | 2798 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 2799 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 2800 | |
f3a84ccd | 2801 | ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 2802 | if (ret) { |
66642832 | 2803 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2804 | return ret; |
2805 | } | |
5f39d397 CM |
2806 | copy_extent_buffer(split, c, |
2807 | btrfs_node_key_ptr_offset(0), | |
2808 | btrfs_node_key_ptr_offset(mid), | |
2809 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2810 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2811 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed | 2812 | |
5f39d397 CM |
2813 | btrfs_mark_buffer_dirty(c); |
2814 | btrfs_mark_buffer_dirty(split); | |
2815 | ||
6ad3cf6d | 2816 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 2817 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 2818 | |
5de08d7d | 2819 | if (path->slots[level] >= mid) { |
5c680ed6 | 2820 | path->slots[level] -= mid; |
925baedd | 2821 | btrfs_tree_unlock(c); |
5f39d397 CM |
2822 | free_extent_buffer(c); |
2823 | path->nodes[level] = split; | |
5c680ed6 CM |
2824 | path->slots[level + 1] += 1; |
2825 | } else { | |
925baedd | 2826 | btrfs_tree_unlock(split); |
5f39d397 | 2827 | free_extent_buffer(split); |
be0e5c09 | 2828 | } |
d5286a92 | 2829 | return 0; |
be0e5c09 CM |
2830 | } |
2831 | ||
74123bd7 CM |
2832 | /* |
2833 | * how many bytes are required to store the items in a leaf. start | |
2834 | * and nr indicate which items in the leaf to check. This totals up the | |
2835 | * space used both by the item structs and the item data | |
2836 | */ | |
5f39d397 | 2837 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2838 | { |
2839 | int data_len; | |
5f39d397 | 2840 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2841 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2842 | |
2843 | if (!nr) | |
2844 | return 0; | |
3212fa14 JB |
2845 | data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); |
2846 | data_len = data_len - btrfs_item_offset(l, end); | |
0783fcfc | 2847 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2848 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2849 | return data_len; |
2850 | } | |
2851 | ||
d4dbff95 CM |
2852 | /* |
2853 | * The space between the end of the leaf items and | |
2854 | * the start of the leaf data. IOW, how much room | |
2855 | * the leaf has left for both items and data | |
2856 | */ | |
e902baac | 2857 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 2858 | { |
e902baac | 2859 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
2860 | int nritems = btrfs_header_nritems(leaf); |
2861 | int ret; | |
0b246afa JM |
2862 | |
2863 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 2864 | if (ret < 0) { |
0b246afa JM |
2865 | btrfs_crit(fs_info, |
2866 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
2867 | ret, | |
2868 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
2869 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
2870 | } |
2871 | return ret; | |
d4dbff95 CM |
2872 | } |
2873 | ||
99d8f83c CM |
2874 | /* |
2875 | * min slot controls the lowest index we're willing to push to the | |
2876 | * right. We'll push up to and including min_slot, but no lower | |
2877 | */ | |
f72f0010 | 2878 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
2879 | int data_size, int empty, |
2880 | struct extent_buffer *right, | |
99d8f83c CM |
2881 | int free_space, u32 left_nritems, |
2882 | u32 min_slot) | |
00ec4c51 | 2883 | { |
f72f0010 | 2884 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 2885 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2886 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2887 | struct btrfs_map_token token; |
5f39d397 | 2888 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2889 | int slot; |
34a38218 | 2890 | u32 i; |
00ec4c51 CM |
2891 | int push_space = 0; |
2892 | int push_items = 0; | |
34a38218 | 2893 | u32 nr; |
7518a238 | 2894 | u32 right_nritems; |
5f39d397 | 2895 | u32 data_end; |
db94535d | 2896 | u32 this_item_size; |
00ec4c51 | 2897 | |
34a38218 CM |
2898 | if (empty) |
2899 | nr = 0; | |
2900 | else | |
99d8f83c | 2901 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2902 | |
31840ae1 | 2903 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2904 | push_space += data_size; |
31840ae1 | 2905 | |
44871b1b | 2906 | slot = path->slots[1]; |
34a38218 CM |
2907 | i = left_nritems - 1; |
2908 | while (i >= nr) { | |
31840ae1 ZY |
2909 | if (!empty && push_items > 0) { |
2910 | if (path->slots[0] > i) | |
2911 | break; | |
2912 | if (path->slots[0] == i) { | |
e902baac DS |
2913 | int space = btrfs_leaf_free_space(left); |
2914 | ||
31840ae1 ZY |
2915 | if (space + push_space * 2 > free_space) |
2916 | break; | |
2917 | } | |
2918 | } | |
2919 | ||
00ec4c51 | 2920 | if (path->slots[0] == i) |
87b29b20 | 2921 | push_space += data_size; |
db94535d | 2922 | |
3212fa14 | 2923 | this_item_size = btrfs_item_size(left, i); |
74794207 JB |
2924 | if (this_item_size + sizeof(struct btrfs_item) + |
2925 | push_space > free_space) | |
00ec4c51 | 2926 | break; |
31840ae1 | 2927 | |
00ec4c51 | 2928 | push_items++; |
74794207 | 2929 | push_space += this_item_size + sizeof(struct btrfs_item); |
34a38218 CM |
2930 | if (i == 0) |
2931 | break; | |
2932 | i--; | |
db94535d | 2933 | } |
5f39d397 | 2934 | |
925baedd CM |
2935 | if (push_items == 0) |
2936 | goto out_unlock; | |
5f39d397 | 2937 | |
6c1500f2 | 2938 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 2939 | |
00ec4c51 | 2940 | /* push left to right */ |
5f39d397 | 2941 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2942 | |
dc2e724e | 2943 | push_space = btrfs_item_data_end(left, left_nritems - push_items); |
8f881e8c | 2944 | push_space -= leaf_data_end(left); |
5f39d397 | 2945 | |
00ec4c51 | 2946 | /* make room in the right data area */ |
8f881e8c | 2947 | data_end = leaf_data_end(right); |
5f39d397 | 2948 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
2949 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
2950 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 2951 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 2952 | |
00ec4c51 | 2953 | /* copy from the left data area */ |
3d9ec8c4 | 2954 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 2955 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 2956 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 2957 | push_space); |
5f39d397 CM |
2958 | |
2959 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2960 | btrfs_item_nr_offset(0), | |
2961 | right_nritems * sizeof(struct btrfs_item)); | |
2962 | ||
00ec4c51 | 2963 | /* copy the items from left to right */ |
5f39d397 CM |
2964 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2965 | btrfs_item_nr_offset(left_nritems - push_items), | |
2966 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2967 | |
2968 | /* update the item pointers */ | |
c82f823c | 2969 | btrfs_init_map_token(&token, right); |
7518a238 | 2970 | right_nritems += push_items; |
5f39d397 | 2971 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 2972 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 2973 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
2974 | push_space -= btrfs_token_item_size(&token, i); |
2975 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d CM |
2976 | } |
2977 | ||
7518a238 | 2978 | left_nritems -= push_items; |
5f39d397 | 2979 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2980 | |
34a38218 CM |
2981 | if (left_nritems) |
2982 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 2983 | else |
6a884d7d | 2984 | btrfs_clean_tree_block(left); |
f0486c68 | 2985 | |
5f39d397 | 2986 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2987 | |
5f39d397 CM |
2988 | btrfs_item_key(right, &disk_key, 0); |
2989 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2990 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2991 | |
00ec4c51 | 2992 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2993 | if (path->slots[0] >= left_nritems) { |
2994 | path->slots[0] -= left_nritems; | |
925baedd | 2995 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 2996 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 2997 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2998 | free_extent_buffer(path->nodes[0]); |
2999 | path->nodes[0] = right; | |
00ec4c51 CM |
3000 | path->slots[1] += 1; |
3001 | } else { | |
925baedd | 3002 | btrfs_tree_unlock(right); |
5f39d397 | 3003 | free_extent_buffer(right); |
00ec4c51 CM |
3004 | } |
3005 | return 0; | |
925baedd CM |
3006 | |
3007 | out_unlock: | |
3008 | btrfs_tree_unlock(right); | |
3009 | free_extent_buffer(right); | |
3010 | return 1; | |
00ec4c51 | 3011 | } |
925baedd | 3012 | |
44871b1b CM |
3013 | /* |
3014 | * push some data in the path leaf to the right, trying to free up at | |
3015 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3016 | * | |
3017 | * returns 1 if the push failed because the other node didn't have enough | |
3018 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3019 | * |
3020 | * this will push starting from min_slot to the end of the leaf. It won't | |
3021 | * push any slot lower than min_slot | |
44871b1b CM |
3022 | */ |
3023 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3024 | *root, struct btrfs_path *path, |
3025 | int min_data_size, int data_size, | |
3026 | int empty, u32 min_slot) | |
44871b1b CM |
3027 | { |
3028 | struct extent_buffer *left = path->nodes[0]; | |
3029 | struct extent_buffer *right; | |
3030 | struct extent_buffer *upper; | |
3031 | int slot; | |
3032 | int free_space; | |
3033 | u32 left_nritems; | |
3034 | int ret; | |
3035 | ||
3036 | if (!path->nodes[1]) | |
3037 | return 1; | |
3038 | ||
3039 | slot = path->slots[1]; | |
3040 | upper = path->nodes[1]; | |
3041 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3042 | return 1; | |
3043 | ||
49d0c642 | 3044 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3045 | |
4b231ae4 | 3046 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3047 | /* |
3048 | * slot + 1 is not valid or we fail to read the right node, | |
3049 | * no big deal, just return. | |
3050 | */ | |
3051 | if (IS_ERR(right)) | |
91ca338d TI |
3052 | return 1; |
3053 | ||
bf77467a | 3054 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
44871b1b | 3055 | |
e902baac | 3056 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3057 | if (free_space < data_size) |
3058 | goto out_unlock; | |
3059 | ||
44871b1b | 3060 | ret = btrfs_cow_block(trans, root, right, upper, |
bf59a5a2 | 3061 | slot + 1, &right, BTRFS_NESTING_RIGHT_COW); |
44871b1b CM |
3062 | if (ret) |
3063 | goto out_unlock; | |
3064 | ||
44871b1b CM |
3065 | left_nritems = btrfs_header_nritems(left); |
3066 | if (left_nritems == 0) | |
3067 | goto out_unlock; | |
3068 | ||
d16c702f QW |
3069 | if (check_sibling_keys(left, right)) { |
3070 | ret = -EUCLEAN; | |
3071 | btrfs_tree_unlock(right); | |
3072 | free_extent_buffer(right); | |
3073 | return ret; | |
3074 | } | |
2ef1fed2 FDBM |
3075 | if (path->slots[0] == left_nritems && !empty) { |
3076 | /* Key greater than all keys in the leaf, right neighbor has | |
3077 | * enough room for it and we're not emptying our leaf to delete | |
3078 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3079 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3080 | btrfs_tree_unlock(left); |
3081 | free_extent_buffer(left); | |
3082 | path->nodes[0] = right; | |
3083 | path->slots[0] = 0; | |
3084 | path->slots[1]++; | |
3085 | return 0; | |
3086 | } | |
3087 | ||
f72f0010 | 3088 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3089 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3090 | out_unlock: |
3091 | btrfs_tree_unlock(right); | |
3092 | free_extent_buffer(right); | |
3093 | return 1; | |
3094 | } | |
3095 | ||
74123bd7 CM |
3096 | /* |
3097 | * push some data in the path leaf to the left, trying to free up at | |
3098 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3099 | * |
3100 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3101 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3102 | * items | |
74123bd7 | 3103 | */ |
8087c193 | 3104 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3105 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3106 | int free_space, u32 right_nritems, |
3107 | u32 max_slot) | |
be0e5c09 | 3108 | { |
8087c193 | 3109 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3110 | struct btrfs_disk_key disk_key; |
3111 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3112 | int i; |
be0e5c09 CM |
3113 | int push_space = 0; |
3114 | int push_items = 0; | |
7518a238 | 3115 | u32 old_left_nritems; |
34a38218 | 3116 | u32 nr; |
aa5d6bed | 3117 | int ret = 0; |
db94535d CM |
3118 | u32 this_item_size; |
3119 | u32 old_left_item_size; | |
cfed81a0 CM |
3120 | struct btrfs_map_token token; |
3121 | ||
34a38218 | 3122 | if (empty) |
99d8f83c | 3123 | nr = min(right_nritems, max_slot); |
34a38218 | 3124 | else |
99d8f83c | 3125 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3126 | |
3127 | for (i = 0; i < nr; i++) { | |
31840ae1 ZY |
3128 | if (!empty && push_items > 0) { |
3129 | if (path->slots[0] < i) | |
3130 | break; | |
3131 | if (path->slots[0] == i) { | |
e902baac DS |
3132 | int space = btrfs_leaf_free_space(right); |
3133 | ||
31840ae1 ZY |
3134 | if (space + push_space * 2 > free_space) |
3135 | break; | |
3136 | } | |
3137 | } | |
3138 | ||
be0e5c09 | 3139 | if (path->slots[0] == i) |
87b29b20 | 3140 | push_space += data_size; |
db94535d | 3141 | |
3212fa14 | 3142 | this_item_size = btrfs_item_size(right, i); |
74794207 JB |
3143 | if (this_item_size + sizeof(struct btrfs_item) + push_space > |
3144 | free_space) | |
be0e5c09 | 3145 | break; |
db94535d | 3146 | |
be0e5c09 | 3147 | push_items++; |
74794207 | 3148 | push_space += this_item_size + sizeof(struct btrfs_item); |
db94535d CM |
3149 | } |
3150 | ||
be0e5c09 | 3151 | if (push_items == 0) { |
925baedd CM |
3152 | ret = 1; |
3153 | goto out; | |
be0e5c09 | 3154 | } |
fae7f21c | 3155 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3156 | |
be0e5c09 | 3157 | /* push data from right to left */ |
5f39d397 CM |
3158 | copy_extent_buffer(left, right, |
3159 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3160 | btrfs_item_nr_offset(0), | |
3161 | push_items * sizeof(struct btrfs_item)); | |
3162 | ||
0b246afa | 3163 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
3212fa14 | 3164 | btrfs_item_offset(right, push_items - 1); |
5f39d397 | 3165 | |
3d9ec8c4 | 3166 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3167 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3168 | BTRFS_LEAF_DATA_OFFSET + |
3212fa14 | 3169 | btrfs_item_offset(right, push_items - 1), |
d6025579 | 3170 | push_space); |
5f39d397 | 3171 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3172 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3173 | |
c82f823c | 3174 | btrfs_init_map_token(&token, left); |
3212fa14 | 3175 | old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); |
0783fcfc | 3176 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3177 | u32 ioff; |
db94535d | 3178 | |
3212fa14 JB |
3179 | ioff = btrfs_token_item_offset(&token, i); |
3180 | btrfs_set_token_item_offset(&token, i, | |
cc4c13d5 | 3181 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); |
be0e5c09 | 3182 | } |
5f39d397 | 3183 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3184 | |
3185 | /* fixup right node */ | |
31b1a2bd JL |
3186 | if (push_items > right_nritems) |
3187 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3188 | right_nritems); |
34a38218 CM |
3189 | |
3190 | if (push_items < right_nritems) { | |
3212fa14 | 3191 | push_space = btrfs_item_offset(right, push_items - 1) - |
8f881e8c | 3192 | leaf_data_end(right); |
3d9ec8c4 | 3193 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3194 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3195 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3196 | leaf_data_end(right), push_space); |
34a38218 CM |
3197 | |
3198 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3199 | btrfs_item_nr_offset(push_items), |
3200 | (btrfs_header_nritems(right) - push_items) * | |
3201 | sizeof(struct btrfs_item)); | |
34a38218 | 3202 | } |
c82f823c DS |
3203 | |
3204 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3205 | right_nritems -= push_items; |
3206 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3207 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3208 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
3209 | push_space = push_space - btrfs_token_item_size(&token, i); |
3210 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d | 3211 | } |
eb60ceac | 3212 | |
5f39d397 | 3213 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3214 | if (right_nritems) |
3215 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3216 | else |
6a884d7d | 3217 | btrfs_clean_tree_block(right); |
098f59c2 | 3218 | |
5f39d397 | 3219 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3220 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3221 | |
3222 | /* then fixup the leaf pointer in the path */ | |
3223 | if (path->slots[0] < push_items) { | |
3224 | path->slots[0] += old_left_nritems; | |
925baedd | 3225 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3226 | free_extent_buffer(path->nodes[0]); |
3227 | path->nodes[0] = left; | |
be0e5c09 CM |
3228 | path->slots[1] -= 1; |
3229 | } else { | |
925baedd | 3230 | btrfs_tree_unlock(left); |
5f39d397 | 3231 | free_extent_buffer(left); |
be0e5c09 CM |
3232 | path->slots[0] -= push_items; |
3233 | } | |
eb60ceac | 3234 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3235 | return ret; |
925baedd CM |
3236 | out: |
3237 | btrfs_tree_unlock(left); | |
3238 | free_extent_buffer(left); | |
3239 | return ret; | |
be0e5c09 CM |
3240 | } |
3241 | ||
44871b1b CM |
3242 | /* |
3243 | * push some data in the path leaf to the left, trying to free up at | |
3244 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3245 | * |
3246 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3247 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3248 | * items | |
44871b1b CM |
3249 | */ |
3250 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3251 | *root, struct btrfs_path *path, int min_data_size, |
3252 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3253 | { |
3254 | struct extent_buffer *right = path->nodes[0]; | |
3255 | struct extent_buffer *left; | |
3256 | int slot; | |
3257 | int free_space; | |
3258 | u32 right_nritems; | |
3259 | int ret = 0; | |
3260 | ||
3261 | slot = path->slots[1]; | |
3262 | if (slot == 0) | |
3263 | return 1; | |
3264 | if (!path->nodes[1]) | |
3265 | return 1; | |
3266 | ||
3267 | right_nritems = btrfs_header_nritems(right); | |
3268 | if (right_nritems == 0) | |
3269 | return 1; | |
3270 | ||
49d0c642 | 3271 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3272 | |
4b231ae4 | 3273 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3274 | /* |
3275 | * slot - 1 is not valid or we fail to read the left node, | |
3276 | * no big deal, just return. | |
3277 | */ | |
3278 | if (IS_ERR(left)) | |
91ca338d TI |
3279 | return 1; |
3280 | ||
bf77467a | 3281 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
44871b1b | 3282 | |
e902baac | 3283 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3284 | if (free_space < data_size) { |
3285 | ret = 1; | |
3286 | goto out; | |
3287 | } | |
3288 | ||
44871b1b | 3289 | ret = btrfs_cow_block(trans, root, left, |
9631e4cc | 3290 | path->nodes[1], slot - 1, &left, |
bf59a5a2 | 3291 | BTRFS_NESTING_LEFT_COW); |
44871b1b CM |
3292 | if (ret) { |
3293 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3294 | if (ret == -ENOSPC) |
3295 | ret = 1; | |
44871b1b CM |
3296 | goto out; |
3297 | } | |
3298 | ||
d16c702f QW |
3299 | if (check_sibling_keys(left, right)) { |
3300 | ret = -EUCLEAN; | |
3301 | goto out; | |
3302 | } | |
8087c193 | 3303 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
3304 | empty, left, free_space, right_nritems, |
3305 | max_slot); | |
44871b1b CM |
3306 | out: |
3307 | btrfs_tree_unlock(left); | |
3308 | free_extent_buffer(left); | |
3309 | return ret; | |
3310 | } | |
3311 | ||
3312 | /* | |
3313 | * split the path's leaf in two, making sure there is at least data_size | |
3314 | * available for the resulting leaf level of the path. | |
44871b1b | 3315 | */ |
143bede5 | 3316 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
3317 | struct btrfs_path *path, |
3318 | struct extent_buffer *l, | |
3319 | struct extent_buffer *right, | |
3320 | int slot, int mid, int nritems) | |
44871b1b | 3321 | { |
94f94ad9 | 3322 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
3323 | int data_copy_size; |
3324 | int rt_data_off; | |
3325 | int i; | |
44871b1b | 3326 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3327 | struct btrfs_map_token token; |
3328 | ||
44871b1b CM |
3329 | nritems = nritems - mid; |
3330 | btrfs_set_header_nritems(right, nritems); | |
dc2e724e | 3331 | data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); |
44871b1b CM |
3332 | |
3333 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3334 | btrfs_item_nr_offset(mid), | |
3335 | nritems * sizeof(struct btrfs_item)); | |
3336 | ||
3337 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
3338 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
3339 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 3340 | leaf_data_end(l), data_copy_size); |
44871b1b | 3341 | |
dc2e724e | 3342 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); |
44871b1b | 3343 | |
c82f823c | 3344 | btrfs_init_map_token(&token, right); |
44871b1b | 3345 | for (i = 0; i < nritems; i++) { |
44871b1b CM |
3346 | u32 ioff; |
3347 | ||
3212fa14 JB |
3348 | ioff = btrfs_token_item_offset(&token, i); |
3349 | btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); | |
44871b1b CM |
3350 | } |
3351 | ||
44871b1b | 3352 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3353 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 3354 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
3355 | |
3356 | btrfs_mark_buffer_dirty(right); | |
3357 | btrfs_mark_buffer_dirty(l); | |
3358 | BUG_ON(path->slots[0] != slot); | |
3359 | ||
44871b1b CM |
3360 | if (mid <= slot) { |
3361 | btrfs_tree_unlock(path->nodes[0]); | |
3362 | free_extent_buffer(path->nodes[0]); | |
3363 | path->nodes[0] = right; | |
3364 | path->slots[0] -= mid; | |
3365 | path->slots[1] += 1; | |
3366 | } else { | |
3367 | btrfs_tree_unlock(right); | |
3368 | free_extent_buffer(right); | |
3369 | } | |
3370 | ||
3371 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3372 | } |
3373 | ||
99d8f83c CM |
3374 | /* |
3375 | * double splits happen when we need to insert a big item in the middle | |
3376 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3377 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3378 | * A B C | |
3379 | * | |
3380 | * We avoid this by trying to push the items on either side of our target | |
3381 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3382 | * completely. | |
3383 | */ | |
3384 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3385 | struct btrfs_root *root, | |
3386 | struct btrfs_path *path, | |
3387 | int data_size) | |
3388 | { | |
3389 | int ret; | |
3390 | int progress = 0; | |
3391 | int slot; | |
3392 | u32 nritems; | |
5a4267ca | 3393 | int space_needed = data_size; |
99d8f83c CM |
3394 | |
3395 | slot = path->slots[0]; | |
5a4267ca | 3396 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 3397 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
3398 | |
3399 | /* | |
3400 | * try to push all the items after our slot into the | |
3401 | * right leaf | |
3402 | */ | |
5a4267ca | 3403 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3404 | if (ret < 0) |
3405 | return ret; | |
3406 | ||
3407 | if (ret == 0) | |
3408 | progress++; | |
3409 | ||
3410 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3411 | /* | |
3412 | * our goal is to get our slot at the start or end of a leaf. If | |
3413 | * we've done so we're done | |
3414 | */ | |
3415 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3416 | return 0; | |
3417 | ||
e902baac | 3418 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3419 | return 0; |
3420 | ||
3421 | /* try to push all the items before our slot into the next leaf */ | |
3422 | slot = path->slots[0]; | |
263d3995 FM |
3423 | space_needed = data_size; |
3424 | if (slot > 0) | |
e902baac | 3425 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 3426 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3427 | if (ret < 0) |
3428 | return ret; | |
3429 | ||
3430 | if (ret == 0) | |
3431 | progress++; | |
3432 | ||
3433 | if (progress) | |
3434 | return 0; | |
3435 | return 1; | |
3436 | } | |
3437 | ||
74123bd7 CM |
3438 | /* |
3439 | * split the path's leaf in two, making sure there is at least data_size | |
3440 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3441 | * |
3442 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3443 | */ |
e02119d5 CM |
3444 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3445 | struct btrfs_root *root, | |
310712b2 | 3446 | const struct btrfs_key *ins_key, |
e02119d5 CM |
3447 | struct btrfs_path *path, int data_size, |
3448 | int extend) | |
be0e5c09 | 3449 | { |
5d4f98a2 | 3450 | struct btrfs_disk_key disk_key; |
5f39d397 | 3451 | struct extent_buffer *l; |
7518a238 | 3452 | u32 nritems; |
eb60ceac CM |
3453 | int mid; |
3454 | int slot; | |
5f39d397 | 3455 | struct extent_buffer *right; |
b7a0365e | 3456 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 3457 | int ret = 0; |
aa5d6bed | 3458 | int wret; |
5d4f98a2 | 3459 | int split; |
cc0c5538 | 3460 | int num_doubles = 0; |
99d8f83c | 3461 | int tried_avoid_double = 0; |
aa5d6bed | 3462 | |
a5719521 YZ |
3463 | l = path->nodes[0]; |
3464 | slot = path->slots[0]; | |
3212fa14 | 3465 | if (extend && data_size + btrfs_item_size(l, slot) + |
0b246afa | 3466 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
3467 | return -EOVERFLOW; |
3468 | ||
40689478 | 3469 | /* first try to make some room by pushing left and right */ |
33157e05 | 3470 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
3471 | int space_needed = data_size; |
3472 | ||
3473 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 3474 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3475 | |
3476 | wret = push_leaf_right(trans, root, path, space_needed, | |
3477 | space_needed, 0, 0); | |
d397712b | 3478 | if (wret < 0) |
eaee50e8 | 3479 | return wret; |
3685f791 | 3480 | if (wret) { |
263d3995 FM |
3481 | space_needed = data_size; |
3482 | if (slot > 0) | |
e902baac | 3483 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3484 | wret = push_leaf_left(trans, root, path, space_needed, |
3485 | space_needed, 0, (u32)-1); | |
3685f791 CM |
3486 | if (wret < 0) |
3487 | return wret; | |
3488 | } | |
3489 | l = path->nodes[0]; | |
aa5d6bed | 3490 | |
3685f791 | 3491 | /* did the pushes work? */ |
e902baac | 3492 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 3493 | return 0; |
3326d1b0 | 3494 | } |
aa5d6bed | 3495 | |
5c680ed6 | 3496 | if (!path->nodes[1]) { |
fdd99c72 | 3497 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3498 | if (ret) |
3499 | return ret; | |
3500 | } | |
cc0c5538 | 3501 | again: |
5d4f98a2 | 3502 | split = 1; |
cc0c5538 | 3503 | l = path->nodes[0]; |
eb60ceac | 3504 | slot = path->slots[0]; |
5f39d397 | 3505 | nritems = btrfs_header_nritems(l); |
d397712b | 3506 | mid = (nritems + 1) / 2; |
54aa1f4d | 3507 | |
5d4f98a2 YZ |
3508 | if (mid <= slot) { |
3509 | if (nritems == 1 || | |
3510 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 3511 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3512 | if (slot >= nritems) { |
3513 | split = 0; | |
3514 | } else { | |
3515 | mid = slot; | |
3516 | if (mid != nritems && | |
3517 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3518 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3519 | if (data_size && !tried_avoid_double) |
3520 | goto push_for_double; | |
5d4f98a2 YZ |
3521 | split = 2; |
3522 | } | |
3523 | } | |
3524 | } | |
3525 | } else { | |
3526 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 3527 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3528 | if (!extend && data_size && slot == 0) { |
3529 | split = 0; | |
3530 | } else if ((extend || !data_size) && slot == 0) { | |
3531 | mid = 1; | |
3532 | } else { | |
3533 | mid = slot; | |
3534 | if (mid != nritems && | |
3535 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3536 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3537 | if (data_size && !tried_avoid_double) |
3538 | goto push_for_double; | |
67871254 | 3539 | split = 2; |
5d4f98a2 YZ |
3540 | } |
3541 | } | |
3542 | } | |
3543 | } | |
3544 | ||
3545 | if (split == 0) | |
3546 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3547 | else | |
3548 | btrfs_item_key(l, &disk_key, mid); | |
3549 | ||
ca9d473a JB |
3550 | /* |
3551 | * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double | |
3552 | * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES | |
3553 | * subclasses, which is 8 at the time of this patch, and we've maxed it | |
3554 | * out. In the future we could add a | |
3555 | * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just | |
3556 | * use BTRFS_NESTING_NEW_ROOT. | |
3557 | */ | |
79bd3712 FM |
3558 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3559 | &disk_key, 0, l->start, 0, | |
3560 | num_doubles ? BTRFS_NESTING_NEW_ROOT : | |
3561 | BTRFS_NESTING_SPLIT); | |
f0486c68 | 3562 | if (IS_ERR(right)) |
5f39d397 | 3563 | return PTR_ERR(right); |
f0486c68 | 3564 | |
0b246afa | 3565 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 3566 | |
5d4f98a2 YZ |
3567 | if (split == 0) { |
3568 | if (mid <= slot) { | |
3569 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3570 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3571 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
3572 | btrfs_tree_unlock(path->nodes[0]); |
3573 | free_extent_buffer(path->nodes[0]); | |
3574 | path->nodes[0] = right; | |
3575 | path->slots[0] = 0; | |
3576 | path->slots[1] += 1; | |
3577 | } else { | |
3578 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3579 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3580 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
3581 | btrfs_tree_unlock(path->nodes[0]); |
3582 | free_extent_buffer(path->nodes[0]); | |
3583 | path->nodes[0] = right; | |
3584 | path->slots[0] = 0; | |
143bede5 | 3585 | if (path->slots[1] == 0) |
b167fa91 | 3586 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 3587 | } |
196e0249 LB |
3588 | /* |
3589 | * We create a new leaf 'right' for the required ins_len and | |
3590 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
3591 | * the content of ins_len to 'right'. | |
3592 | */ | |
5d4f98a2 | 3593 | return ret; |
d4dbff95 | 3594 | } |
74123bd7 | 3595 | |
94f94ad9 | 3596 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 3597 | |
5d4f98a2 | 3598 | if (split == 2) { |
cc0c5538 CM |
3599 | BUG_ON(num_doubles != 0); |
3600 | num_doubles++; | |
3601 | goto again; | |
a429e513 | 3602 | } |
44871b1b | 3603 | |
143bede5 | 3604 | return 0; |
99d8f83c CM |
3605 | |
3606 | push_for_double: | |
3607 | push_for_double_split(trans, root, path, data_size); | |
3608 | tried_avoid_double = 1; | |
e902baac | 3609 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3610 | return 0; |
3611 | goto again; | |
be0e5c09 CM |
3612 | } |
3613 | ||
ad48fd75 YZ |
3614 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3615 | struct btrfs_root *root, | |
3616 | struct btrfs_path *path, int ins_len) | |
459931ec | 3617 | { |
ad48fd75 | 3618 | struct btrfs_key key; |
459931ec | 3619 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3620 | struct btrfs_file_extent_item *fi; |
3621 | u64 extent_len = 0; | |
3622 | u32 item_size; | |
3623 | int ret; | |
459931ec CM |
3624 | |
3625 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3626 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3627 | ||
3628 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3629 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3630 | ||
e902baac | 3631 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 3632 | return 0; |
459931ec | 3633 | |
3212fa14 | 3634 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ad48fd75 YZ |
3635 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3636 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3637 | struct btrfs_file_extent_item); | |
3638 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3639 | } | |
b3b4aa74 | 3640 | btrfs_release_path(path); |
459931ec | 3641 | |
459931ec | 3642 | path->keep_locks = 1; |
ad48fd75 YZ |
3643 | path->search_for_split = 1; |
3644 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3645 | path->search_for_split = 0; |
a8df6fe6 FM |
3646 | if (ret > 0) |
3647 | ret = -EAGAIN; | |
ad48fd75 YZ |
3648 | if (ret < 0) |
3649 | goto err; | |
459931ec | 3650 | |
ad48fd75 YZ |
3651 | ret = -EAGAIN; |
3652 | leaf = path->nodes[0]; | |
a8df6fe6 | 3653 | /* if our item isn't there, return now */ |
3212fa14 | 3654 | if (item_size != btrfs_item_size(leaf, path->slots[0])) |
ad48fd75 YZ |
3655 | goto err; |
3656 | ||
109f6aef | 3657 | /* the leaf has changed, it now has room. return now */ |
e902baac | 3658 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
3659 | goto err; |
3660 | ||
ad48fd75 YZ |
3661 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3662 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3663 | struct btrfs_file_extent_item); | |
3664 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3665 | goto err; | |
459931ec CM |
3666 | } |
3667 | ||
ad48fd75 | 3668 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3669 | if (ret) |
3670 | goto err; | |
459931ec | 3671 | |
ad48fd75 | 3672 | path->keep_locks = 0; |
b9473439 | 3673 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3674 | return 0; |
3675 | err: | |
3676 | path->keep_locks = 0; | |
3677 | return ret; | |
3678 | } | |
3679 | ||
25263cd7 | 3680 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 3681 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3682 | unsigned long split_offset) |
3683 | { | |
3684 | struct extent_buffer *leaf; | |
c91666b1 | 3685 | int orig_slot, slot; |
ad48fd75 YZ |
3686 | char *buf; |
3687 | u32 nritems; | |
3688 | u32 item_size; | |
3689 | u32 orig_offset; | |
3690 | struct btrfs_disk_key disk_key; | |
3691 | ||
b9473439 | 3692 | leaf = path->nodes[0]; |
e902baac | 3693 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 3694 | |
c91666b1 | 3695 | orig_slot = path->slots[0]; |
3212fa14 JB |
3696 | orig_offset = btrfs_item_offset(leaf, path->slots[0]); |
3697 | item_size = btrfs_item_size(leaf, path->slots[0]); | |
459931ec | 3698 | |
459931ec | 3699 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3700 | if (!buf) |
3701 | return -ENOMEM; | |
3702 | ||
459931ec CM |
3703 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3704 | path->slots[0]), item_size); | |
459931ec | 3705 | |
ad48fd75 | 3706 | slot = path->slots[0] + 1; |
459931ec | 3707 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3708 | if (slot != nritems) { |
3709 | /* shift the items */ | |
3710 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3711 | btrfs_item_nr_offset(slot), |
3712 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3713 | } |
3714 | ||
3715 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3716 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3717 | ||
3212fa14 JB |
3718 | btrfs_set_item_offset(leaf, slot, orig_offset); |
3719 | btrfs_set_item_size(leaf, slot, item_size - split_offset); | |
459931ec | 3720 | |
3212fa14 | 3721 | btrfs_set_item_offset(leaf, orig_slot, |
c91666b1 | 3722 | orig_offset + item_size - split_offset); |
3212fa14 | 3723 | btrfs_set_item_size(leaf, orig_slot, split_offset); |
459931ec CM |
3724 | |
3725 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3726 | ||
3727 | /* write the data for the start of the original item */ | |
3728 | write_extent_buffer(leaf, buf, | |
3729 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3730 | split_offset); | |
3731 | ||
3732 | /* write the data for the new item */ | |
3733 | write_extent_buffer(leaf, buf + split_offset, | |
3734 | btrfs_item_ptr_offset(leaf, slot), | |
3735 | item_size - split_offset); | |
3736 | btrfs_mark_buffer_dirty(leaf); | |
3737 | ||
e902baac | 3738 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 3739 | kfree(buf); |
ad48fd75 YZ |
3740 | return 0; |
3741 | } | |
3742 | ||
3743 | /* | |
3744 | * This function splits a single item into two items, | |
3745 | * giving 'new_key' to the new item and splitting the | |
3746 | * old one at split_offset (from the start of the item). | |
3747 | * | |
3748 | * The path may be released by this operation. After | |
3749 | * the split, the path is pointing to the old item. The | |
3750 | * new item is going to be in the same node as the old one. | |
3751 | * | |
3752 | * Note, the item being split must be smaller enough to live alone on | |
3753 | * a tree block with room for one extra struct btrfs_item | |
3754 | * | |
3755 | * This allows us to split the item in place, keeping a lock on the | |
3756 | * leaf the entire time. | |
3757 | */ | |
3758 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3759 | struct btrfs_root *root, | |
3760 | struct btrfs_path *path, | |
310712b2 | 3761 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3762 | unsigned long split_offset) |
3763 | { | |
3764 | int ret; | |
3765 | ret = setup_leaf_for_split(trans, root, path, | |
3766 | sizeof(struct btrfs_item)); | |
3767 | if (ret) | |
3768 | return ret; | |
3769 | ||
25263cd7 | 3770 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
3771 | return ret; |
3772 | } | |
3773 | ||
d352ac68 CM |
3774 | /* |
3775 | * make the item pointed to by the path smaller. new_size indicates | |
3776 | * how small to make it, and from_end tells us if we just chop bytes | |
3777 | * off the end of the item or if we shift the item to chop bytes off | |
3778 | * the front. | |
3779 | */ | |
78ac4f9e | 3780 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 3781 | { |
b18c6685 | 3782 | int slot; |
5f39d397 | 3783 | struct extent_buffer *leaf; |
b18c6685 CM |
3784 | u32 nritems; |
3785 | unsigned int data_end; | |
3786 | unsigned int old_data_start; | |
3787 | unsigned int old_size; | |
3788 | unsigned int size_diff; | |
3789 | int i; | |
cfed81a0 CM |
3790 | struct btrfs_map_token token; |
3791 | ||
5f39d397 | 3792 | leaf = path->nodes[0]; |
179e29e4 CM |
3793 | slot = path->slots[0]; |
3794 | ||
3212fa14 | 3795 | old_size = btrfs_item_size(leaf, slot); |
179e29e4 | 3796 | if (old_size == new_size) |
143bede5 | 3797 | return; |
b18c6685 | 3798 | |
5f39d397 | 3799 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3800 | data_end = leaf_data_end(leaf); |
b18c6685 | 3801 | |
3212fa14 | 3802 | old_data_start = btrfs_item_offset(leaf, slot); |
179e29e4 | 3803 | |
b18c6685 CM |
3804 | size_diff = old_size - new_size; |
3805 | ||
3806 | BUG_ON(slot < 0); | |
3807 | BUG_ON(slot >= nritems); | |
3808 | ||
3809 | /* | |
3810 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3811 | */ | |
3812 | /* first correct the data pointers */ | |
c82f823c | 3813 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 3814 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3815 | u32 ioff; |
db94535d | 3816 | |
3212fa14 JB |
3817 | ioff = btrfs_token_item_offset(&token, i); |
3818 | btrfs_set_token_item_offset(&token, i, ioff + size_diff); | |
b18c6685 | 3819 | } |
db94535d | 3820 | |
b18c6685 | 3821 | /* shift the data */ |
179e29e4 | 3822 | if (from_end) { |
3d9ec8c4 NB |
3823 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3824 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3825 | data_end, old_data_start + new_size - data_end); |
3826 | } else { | |
3827 | struct btrfs_disk_key disk_key; | |
3828 | u64 offset; | |
3829 | ||
3830 | btrfs_item_key(leaf, &disk_key, slot); | |
3831 | ||
3832 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3833 | unsigned long ptr; | |
3834 | struct btrfs_file_extent_item *fi; | |
3835 | ||
3836 | fi = btrfs_item_ptr(leaf, slot, | |
3837 | struct btrfs_file_extent_item); | |
3838 | fi = (struct btrfs_file_extent_item *)( | |
3839 | (unsigned long)fi - size_diff); | |
3840 | ||
3841 | if (btrfs_file_extent_type(leaf, fi) == | |
3842 | BTRFS_FILE_EXTENT_INLINE) { | |
3843 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3844 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 3845 | (unsigned long)fi, |
7ec20afb | 3846 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
3847 | } |
3848 | } | |
3849 | ||
3d9ec8c4 NB |
3850 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3851 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3852 | data_end, old_data_start - data_end); |
3853 | ||
3854 | offset = btrfs_disk_key_offset(&disk_key); | |
3855 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3856 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3857 | if (slot == 0) | |
b167fa91 | 3858 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 3859 | } |
5f39d397 | 3860 | |
3212fa14 | 3861 | btrfs_set_item_size(leaf, slot, new_size); |
5f39d397 | 3862 | btrfs_mark_buffer_dirty(leaf); |
b18c6685 | 3863 | |
e902baac | 3864 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3865 | btrfs_print_leaf(leaf); |
b18c6685 | 3866 | BUG(); |
5f39d397 | 3867 | } |
b18c6685 CM |
3868 | } |
3869 | ||
d352ac68 | 3870 | /* |
8f69dbd2 | 3871 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 3872 | */ |
c71dd880 | 3873 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 3874 | { |
6567e837 | 3875 | int slot; |
5f39d397 | 3876 | struct extent_buffer *leaf; |
6567e837 CM |
3877 | u32 nritems; |
3878 | unsigned int data_end; | |
3879 | unsigned int old_data; | |
3880 | unsigned int old_size; | |
3881 | int i; | |
cfed81a0 CM |
3882 | struct btrfs_map_token token; |
3883 | ||
5f39d397 | 3884 | leaf = path->nodes[0]; |
6567e837 | 3885 | |
5f39d397 | 3886 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3887 | data_end = leaf_data_end(leaf); |
6567e837 | 3888 | |
e902baac | 3889 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 3890 | btrfs_print_leaf(leaf); |
6567e837 | 3891 | BUG(); |
5f39d397 | 3892 | } |
6567e837 | 3893 | slot = path->slots[0]; |
dc2e724e | 3894 | old_data = btrfs_item_data_end(leaf, slot); |
6567e837 CM |
3895 | |
3896 | BUG_ON(slot < 0); | |
3326d1b0 | 3897 | if (slot >= nritems) { |
a4f78750 | 3898 | btrfs_print_leaf(leaf); |
c71dd880 | 3899 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 3900 | slot, nritems); |
290342f6 | 3901 | BUG(); |
3326d1b0 | 3902 | } |
6567e837 CM |
3903 | |
3904 | /* | |
3905 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3906 | */ | |
3907 | /* first correct the data pointers */ | |
c82f823c | 3908 | btrfs_init_map_token(&token, leaf); |
6567e837 | 3909 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3910 | u32 ioff; |
db94535d | 3911 | |
3212fa14 JB |
3912 | ioff = btrfs_token_item_offset(&token, i); |
3913 | btrfs_set_token_item_offset(&token, i, ioff - data_size); | |
6567e837 | 3914 | } |
5f39d397 | 3915 | |
6567e837 | 3916 | /* shift the data */ |
3d9ec8c4 NB |
3917 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3918 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 3919 | data_end, old_data - data_end); |
5f39d397 | 3920 | |
6567e837 | 3921 | data_end = old_data; |
3212fa14 JB |
3922 | old_size = btrfs_item_size(leaf, slot); |
3923 | btrfs_set_item_size(leaf, slot, old_size + data_size); | |
5f39d397 | 3924 | btrfs_mark_buffer_dirty(leaf); |
6567e837 | 3925 | |
e902baac | 3926 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3927 | btrfs_print_leaf(leaf); |
6567e837 | 3928 | BUG(); |
5f39d397 | 3929 | } |
6567e837 CM |
3930 | } |
3931 | ||
da9ffb24 NB |
3932 | /** |
3933 | * setup_items_for_insert - Helper called before inserting one or more items | |
3934 | * to a leaf. Main purpose is to save stack depth by doing the bulk of the work | |
3935 | * in a function that doesn't call btrfs_search_slot | |
3936 | * | |
3937 | * @root: root we are inserting items to | |
3938 | * @path: points to the leaf/slot where we are going to insert new items | |
b7ef5f3a | 3939 | * @batch: information about the batch of items to insert |
74123bd7 | 3940 | */ |
f0641656 FM |
3941 | static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
3942 | const struct btrfs_item_batch *batch) | |
be0e5c09 | 3943 | { |
0b246afa | 3944 | struct btrfs_fs_info *fs_info = root->fs_info; |
9c58309d | 3945 | int i; |
7518a238 | 3946 | u32 nritems; |
be0e5c09 | 3947 | unsigned int data_end; |
e2fa7227 | 3948 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3949 | struct extent_buffer *leaf; |
3950 | int slot; | |
cfed81a0 | 3951 | struct btrfs_map_token token; |
fc0d82e1 | 3952 | u32 total_size; |
cfed81a0 | 3953 | |
b7ef5f3a FM |
3954 | /* |
3955 | * Before anything else, update keys in the parent and other ancestors | |
3956 | * if needed, then release the write locks on them, so that other tasks | |
3957 | * can use them while we modify the leaf. | |
3958 | */ | |
24cdc847 | 3959 | if (path->slots[0] == 0) { |
b7ef5f3a | 3960 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); |
b167fa91 | 3961 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
3962 | } |
3963 | btrfs_unlock_up_safe(path, 1); | |
3964 | ||
5f39d397 | 3965 | leaf = path->nodes[0]; |
44871b1b | 3966 | slot = path->slots[0]; |
74123bd7 | 3967 | |
5f39d397 | 3968 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3969 | data_end = leaf_data_end(leaf); |
b7ef5f3a | 3970 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
eb60ceac | 3971 | |
e902baac | 3972 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 3973 | btrfs_print_leaf(leaf); |
0b246afa | 3974 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 3975 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 3976 | BUG(); |
d4dbff95 | 3977 | } |
5f39d397 | 3978 | |
c82f823c | 3979 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 3980 | if (slot != nritems) { |
dc2e724e | 3981 | unsigned int old_data = btrfs_item_data_end(leaf, slot); |
be0e5c09 | 3982 | |
5f39d397 | 3983 | if (old_data < data_end) { |
a4f78750 | 3984 | btrfs_print_leaf(leaf); |
7269ddd2 NB |
3985 | btrfs_crit(fs_info, |
3986 | "item at slot %d with data offset %u beyond data end of leaf %u", | |
5d163e0e | 3987 | slot, old_data, data_end); |
290342f6 | 3988 | BUG(); |
5f39d397 | 3989 | } |
be0e5c09 CM |
3990 | /* |
3991 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3992 | */ | |
3993 | /* first correct the data pointers */ | |
0783fcfc | 3994 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3995 | u32 ioff; |
db94535d | 3996 | |
3212fa14 JB |
3997 | ioff = btrfs_token_item_offset(&token, i); |
3998 | btrfs_set_token_item_offset(&token, i, | |
74794207 | 3999 | ioff - batch->total_data_size); |
0783fcfc | 4000 | } |
be0e5c09 | 4001 | /* shift the items */ |
b7ef5f3a | 4002 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + batch->nr), |
5f39d397 | 4003 | btrfs_item_nr_offset(slot), |
d6025579 | 4004 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4005 | |
4006 | /* shift the data */ | |
3d9ec8c4 | 4007 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
b7ef5f3a FM |
4008 | data_end - batch->total_data_size, |
4009 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
4010 | old_data - data_end); | |
be0e5c09 CM |
4011 | data_end = old_data; |
4012 | } | |
5f39d397 | 4013 | |
62e2749e | 4014 | /* setup the item for the new data */ |
b7ef5f3a FM |
4015 | for (i = 0; i < batch->nr; i++) { |
4016 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); | |
9c58309d | 4017 | btrfs_set_item_key(leaf, &disk_key, slot + i); |
b7ef5f3a | 4018 | data_end -= batch->data_sizes[i]; |
3212fa14 JB |
4019 | btrfs_set_token_item_offset(&token, slot + i, data_end); |
4020 | btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); | |
9c58309d | 4021 | } |
44871b1b | 4022 | |
b7ef5f3a | 4023 | btrfs_set_header_nritems(leaf, nritems + batch->nr); |
b9473439 | 4024 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4025 | |
e902baac | 4026 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4027 | btrfs_print_leaf(leaf); |
be0e5c09 | 4028 | BUG(); |
5f39d397 | 4029 | } |
44871b1b CM |
4030 | } |
4031 | ||
f0641656 FM |
4032 | /* |
4033 | * Insert a new item into a leaf. | |
4034 | * | |
4035 | * @root: The root of the btree. | |
4036 | * @path: A path pointing to the target leaf and slot. | |
4037 | * @key: The key of the new item. | |
4038 | * @data_size: The size of the data associated with the new key. | |
4039 | */ | |
4040 | void btrfs_setup_item_for_insert(struct btrfs_root *root, | |
4041 | struct btrfs_path *path, | |
4042 | const struct btrfs_key *key, | |
4043 | u32 data_size) | |
4044 | { | |
4045 | struct btrfs_item_batch batch; | |
4046 | ||
4047 | batch.keys = key; | |
4048 | batch.data_sizes = &data_size; | |
4049 | batch.total_data_size = data_size; | |
4050 | batch.nr = 1; | |
4051 | ||
4052 | setup_items_for_insert(root, path, &batch); | |
4053 | } | |
4054 | ||
44871b1b CM |
4055 | /* |
4056 | * Given a key and some data, insert items into the tree. | |
4057 | * This does all the path init required, making room in the tree if needed. | |
4058 | */ | |
4059 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4060 | struct btrfs_root *root, | |
4061 | struct btrfs_path *path, | |
b7ef5f3a | 4062 | const struct btrfs_item_batch *batch) |
44871b1b | 4063 | { |
44871b1b CM |
4064 | int ret = 0; |
4065 | int slot; | |
b7ef5f3a | 4066 | u32 total_size; |
44871b1b | 4067 | |
b7ef5f3a FM |
4068 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
4069 | ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); | |
44871b1b CM |
4070 | if (ret == 0) |
4071 | return -EEXIST; | |
4072 | if (ret < 0) | |
143bede5 | 4073 | return ret; |
44871b1b | 4074 | |
44871b1b CM |
4075 | slot = path->slots[0]; |
4076 | BUG_ON(slot < 0); | |
4077 | ||
b7ef5f3a | 4078 | setup_items_for_insert(root, path, batch); |
143bede5 | 4079 | return 0; |
62e2749e CM |
4080 | } |
4081 | ||
4082 | /* | |
4083 | * Given a key and some data, insert an item into the tree. | |
4084 | * This does all the path init required, making room in the tree if needed. | |
4085 | */ | |
310712b2 OS |
4086 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4087 | const struct btrfs_key *cpu_key, void *data, | |
4088 | u32 data_size) | |
62e2749e CM |
4089 | { |
4090 | int ret = 0; | |
2c90e5d6 | 4091 | struct btrfs_path *path; |
5f39d397 CM |
4092 | struct extent_buffer *leaf; |
4093 | unsigned long ptr; | |
62e2749e | 4094 | |
2c90e5d6 | 4095 | path = btrfs_alloc_path(); |
db5b493a TI |
4096 | if (!path) |
4097 | return -ENOMEM; | |
2c90e5d6 | 4098 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4099 | if (!ret) { |
5f39d397 CM |
4100 | leaf = path->nodes[0]; |
4101 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4102 | write_extent_buffer(leaf, data, ptr, data_size); | |
4103 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4104 | } |
2c90e5d6 | 4105 | btrfs_free_path(path); |
aa5d6bed | 4106 | return ret; |
be0e5c09 CM |
4107 | } |
4108 | ||
f0641656 FM |
4109 | /* |
4110 | * This function duplicates an item, giving 'new_key' to the new item. | |
4111 | * It guarantees both items live in the same tree leaf and the new item is | |
4112 | * contiguous with the original item. | |
4113 | * | |
4114 | * This allows us to split a file extent in place, keeping a lock on the leaf | |
4115 | * the entire time. | |
4116 | */ | |
4117 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4118 | struct btrfs_root *root, | |
4119 | struct btrfs_path *path, | |
4120 | const struct btrfs_key *new_key) | |
4121 | { | |
4122 | struct extent_buffer *leaf; | |
4123 | int ret; | |
4124 | u32 item_size; | |
4125 | ||
4126 | leaf = path->nodes[0]; | |
3212fa14 | 4127 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f0641656 FM |
4128 | ret = setup_leaf_for_split(trans, root, path, |
4129 | item_size + sizeof(struct btrfs_item)); | |
4130 | if (ret) | |
4131 | return ret; | |
4132 | ||
4133 | path->slots[0]++; | |
4134 | btrfs_setup_item_for_insert(root, path, new_key, item_size); | |
4135 | leaf = path->nodes[0]; | |
4136 | memcpy_extent_buffer(leaf, | |
4137 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4138 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4139 | item_size); | |
4140 | return 0; | |
4141 | } | |
4142 | ||
74123bd7 | 4143 | /* |
5de08d7d | 4144 | * delete the pointer from a given node. |
74123bd7 | 4145 | * |
d352ac68 CM |
4146 | * the tree should have been previously balanced so the deletion does not |
4147 | * empty a node. | |
74123bd7 | 4148 | */ |
afe5fea7 TI |
4149 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4150 | int level, int slot) | |
be0e5c09 | 4151 | { |
5f39d397 | 4152 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4153 | u32 nritems; |
f3ea38da | 4154 | int ret; |
be0e5c09 | 4155 | |
5f39d397 | 4156 | nritems = btrfs_header_nritems(parent); |
d397712b | 4157 | if (slot != nritems - 1) { |
bf1d3425 | 4158 | if (level) { |
f3a84ccd FM |
4159 | ret = btrfs_tree_mod_log_insert_move(parent, slot, |
4160 | slot + 1, nritems - slot - 1); | |
bf1d3425 DS |
4161 | BUG_ON(ret < 0); |
4162 | } | |
5f39d397 CM |
4163 | memmove_extent_buffer(parent, |
4164 | btrfs_node_key_ptr_offset(slot), | |
4165 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4166 | sizeof(struct btrfs_key_ptr) * |
4167 | (nritems - slot - 1)); | |
57ba86c0 | 4168 | } else if (level) { |
f3a84ccd FM |
4169 | ret = btrfs_tree_mod_log_insert_key(parent, slot, |
4170 | BTRFS_MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
57ba86c0 | 4171 | BUG_ON(ret < 0); |
bb803951 | 4172 | } |
f3ea38da | 4173 | |
7518a238 | 4174 | nritems--; |
5f39d397 | 4175 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4176 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4177 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4178 | /* just turn the root into a leaf and break */ |
5f39d397 | 4179 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4180 | } else if (slot == 0) { |
5f39d397 CM |
4181 | struct btrfs_disk_key disk_key; |
4182 | ||
4183 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4184 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4185 | } |
d6025579 | 4186 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4187 | } |
4188 | ||
323ac95b CM |
4189 | /* |
4190 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4191 | * path->nodes[1]. |
323ac95b CM |
4192 | * |
4193 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4194 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4195 | * | |
4196 | * The path must have already been setup for deleting the leaf, including | |
4197 | * all the proper balancing. path->nodes[1] must be locked. | |
4198 | */ | |
143bede5 JM |
4199 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4200 | struct btrfs_root *root, | |
4201 | struct btrfs_path *path, | |
4202 | struct extent_buffer *leaf) | |
323ac95b | 4203 | { |
5d4f98a2 | 4204 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4205 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4206 | |
4d081c41 CM |
4207 | /* |
4208 | * btrfs_free_extent is expensive, we want to make sure we | |
4209 | * aren't holding any locks when we call it | |
4210 | */ | |
4211 | btrfs_unlock_up_safe(path, 0); | |
4212 | ||
f0486c68 YZ |
4213 | root_sub_used(root, leaf->len); |
4214 | ||
67439dad | 4215 | atomic_inc(&leaf->refs); |
7a163608 | 4216 | btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); |
3083ee2e | 4217 | free_extent_buffer_stale(leaf); |
323ac95b | 4218 | } |
74123bd7 CM |
4219 | /* |
4220 | * delete the item at the leaf level in path. If that empties | |
4221 | * the leaf, remove it from the tree | |
4222 | */ | |
85e21bac CM |
4223 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4224 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4225 | { |
0b246afa | 4226 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4227 | struct extent_buffer *leaf; |
aa5d6bed CM |
4228 | int ret = 0; |
4229 | int wret; | |
7518a238 | 4230 | u32 nritems; |
be0e5c09 | 4231 | |
5f39d397 | 4232 | leaf = path->nodes[0]; |
5f39d397 | 4233 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4234 | |
85e21bac | 4235 | if (slot + nr != nritems) { |
0cae23b6 FM |
4236 | const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); |
4237 | const int data_end = leaf_data_end(leaf); | |
c82f823c | 4238 | struct btrfs_map_token token; |
0cae23b6 FM |
4239 | u32 dsize = 0; |
4240 | int i; | |
4241 | ||
4242 | for (i = 0; i < nr; i++) | |
4243 | dsize += btrfs_item_size(leaf, slot + i); | |
5f39d397 | 4244 | |
3d9ec8c4 | 4245 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4246 | data_end + dsize, |
3d9ec8c4 | 4247 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4248 | last_off - data_end); |
5f39d397 | 4249 | |
c82f823c | 4250 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4251 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4252 | u32 ioff; |
db94535d | 4253 | |
3212fa14 JB |
4254 | ioff = btrfs_token_item_offset(&token, i); |
4255 | btrfs_set_token_item_offset(&token, i, ioff + dsize); | |
0783fcfc | 4256 | } |
db94535d | 4257 | |
5f39d397 | 4258 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4259 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4260 | sizeof(struct btrfs_item) * |
85e21bac | 4261 | (nritems - slot - nr)); |
be0e5c09 | 4262 | } |
85e21bac CM |
4263 | btrfs_set_header_nritems(leaf, nritems - nr); |
4264 | nritems -= nr; | |
5f39d397 | 4265 | |
74123bd7 | 4266 | /* delete the leaf if we've emptied it */ |
7518a238 | 4267 | if (nritems == 0) { |
5f39d397 CM |
4268 | if (leaf == root->node) { |
4269 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4270 | } else { |
6a884d7d | 4271 | btrfs_clean_tree_block(leaf); |
143bede5 | 4272 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4273 | } |
be0e5c09 | 4274 | } else { |
7518a238 | 4275 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4276 | if (slot == 0) { |
5f39d397 CM |
4277 | struct btrfs_disk_key disk_key; |
4278 | ||
4279 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4280 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4281 | } |
aa5d6bed | 4282 | |
7c4063d1 FM |
4283 | /* |
4284 | * Try to delete the leaf if it is mostly empty. We do this by | |
4285 | * trying to move all its items into its left and right neighbours. | |
4286 | * If we can't move all the items, then we don't delete it - it's | |
4287 | * not ideal, but future insertions might fill the leaf with more | |
4288 | * items, or items from other leaves might be moved later into our | |
4289 | * leaf due to deletions on those leaves. | |
4290 | */ | |
0b246afa | 4291 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
7c4063d1 FM |
4292 | u32 min_push_space; |
4293 | ||
be0e5c09 CM |
4294 | /* push_leaf_left fixes the path. |
4295 | * make sure the path still points to our leaf | |
4296 | * for possible call to del_ptr below | |
4297 | */ | |
4920c9ac | 4298 | slot = path->slots[1]; |
67439dad | 4299 | atomic_inc(&leaf->refs); |
7c4063d1 FM |
4300 | /* |
4301 | * We want to be able to at least push one item to the | |
4302 | * left neighbour leaf, and that's the first item. | |
4303 | */ | |
4304 | min_push_space = sizeof(struct btrfs_item) + | |
4305 | btrfs_item_size(leaf, 0); | |
4306 | wret = push_leaf_left(trans, root, path, 0, | |
4307 | min_push_space, 1, (u32)-1); | |
54aa1f4d | 4308 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4309 | ret = wret; |
5f39d397 CM |
4310 | |
4311 | if (path->nodes[0] == leaf && | |
4312 | btrfs_header_nritems(leaf)) { | |
7c4063d1 FM |
4313 | /* |
4314 | * If we were not able to push all items from our | |
4315 | * leaf to its left neighbour, then attempt to | |
4316 | * either push all the remaining items to the | |
4317 | * right neighbour or none. There's no advantage | |
4318 | * in pushing only some items, instead of all, as | |
4319 | * it's pointless to end up with a leaf having | |
4320 | * too few items while the neighbours can be full | |
4321 | * or nearly full. | |
4322 | */ | |
4323 | nritems = btrfs_header_nritems(leaf); | |
4324 | min_push_space = leaf_space_used(leaf, 0, nritems); | |
4325 | wret = push_leaf_right(trans, root, path, 0, | |
4326 | min_push_space, 1, 0); | |
54aa1f4d | 4327 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4328 | ret = wret; |
4329 | } | |
5f39d397 CM |
4330 | |
4331 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4332 | path->slots[1] = slot; |
143bede5 | 4333 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4334 | free_extent_buffer(leaf); |
143bede5 | 4335 | ret = 0; |
5de08d7d | 4336 | } else { |
925baedd CM |
4337 | /* if we're still in the path, make sure |
4338 | * we're dirty. Otherwise, one of the | |
4339 | * push_leaf functions must have already | |
4340 | * dirtied this buffer | |
4341 | */ | |
4342 | if (path->nodes[0] == leaf) | |
4343 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4344 | free_extent_buffer(leaf); |
be0e5c09 | 4345 | } |
d5719762 | 4346 | } else { |
5f39d397 | 4347 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4348 | } |
4349 | } | |
aa5d6bed | 4350 | return ret; |
be0e5c09 CM |
4351 | } |
4352 | ||
7bb86316 | 4353 | /* |
925baedd | 4354 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4355 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4356 | * returns < 0 on io errors. | |
d352ac68 CM |
4357 | * |
4358 | * This may release the path, and so you may lose any locks held at the | |
4359 | * time you call it. | |
7bb86316 | 4360 | */ |
16e7549f | 4361 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4362 | { |
925baedd CM |
4363 | struct btrfs_key key; |
4364 | struct btrfs_disk_key found_key; | |
4365 | int ret; | |
7bb86316 | 4366 | |
925baedd | 4367 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4368 | |
e8b0d724 | 4369 | if (key.offset > 0) { |
925baedd | 4370 | key.offset--; |
e8b0d724 | 4371 | } else if (key.type > 0) { |
925baedd | 4372 | key.type--; |
e8b0d724 FDBM |
4373 | key.offset = (u64)-1; |
4374 | } else if (key.objectid > 0) { | |
925baedd | 4375 | key.objectid--; |
e8b0d724 FDBM |
4376 | key.type = (u8)-1; |
4377 | key.offset = (u64)-1; | |
4378 | } else { | |
925baedd | 4379 | return 1; |
e8b0d724 | 4380 | } |
7bb86316 | 4381 | |
b3b4aa74 | 4382 | btrfs_release_path(path); |
925baedd CM |
4383 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4384 | if (ret < 0) | |
4385 | return ret; | |
4386 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4387 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
4388 | /* |
4389 | * We might have had an item with the previous key in the tree right | |
4390 | * before we released our path. And after we released our path, that | |
4391 | * item might have been pushed to the first slot (0) of the leaf we | |
4392 | * were holding due to a tree balance. Alternatively, an item with the | |
4393 | * previous key can exist as the only element of a leaf (big fat item). | |
4394 | * Therefore account for these 2 cases, so that our callers (like | |
4395 | * btrfs_previous_item) don't miss an existing item with a key matching | |
4396 | * the previous key we computed above. | |
4397 | */ | |
4398 | if (ret <= 0) | |
925baedd CM |
4399 | return 0; |
4400 | return 1; | |
7bb86316 CM |
4401 | } |
4402 | ||
3f157a2f CM |
4403 | /* |
4404 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4405 | * for nodes or leaves that are have a minimum transaction id. |
4406 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4407 | * |
4408 | * This does not cow, but it does stuff the starting key it finds back | |
4409 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4410 | * key and get a writable path. | |
4411 | * | |
3f157a2f CM |
4412 | * This honors path->lowest_level to prevent descent past a given level |
4413 | * of the tree. | |
4414 | * | |
d352ac68 CM |
4415 | * min_trans indicates the oldest transaction that you are interested |
4416 | * in walking through. Any nodes or leaves older than min_trans are | |
4417 | * skipped over (without reading them). | |
4418 | * | |
3f157a2f CM |
4419 | * returns zero if something useful was found, < 0 on error and 1 if there |
4420 | * was nothing in the tree that matched the search criteria. | |
4421 | */ | |
4422 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 4423 | struct btrfs_path *path, |
3f157a2f CM |
4424 | u64 min_trans) |
4425 | { | |
4426 | struct extent_buffer *cur; | |
4427 | struct btrfs_key found_key; | |
4428 | int slot; | |
9652480b | 4429 | int sret; |
3f157a2f CM |
4430 | u32 nritems; |
4431 | int level; | |
4432 | int ret = 1; | |
f98de9b9 | 4433 | int keep_locks = path->keep_locks; |
3f157a2f | 4434 | |
f98de9b9 | 4435 | path->keep_locks = 1; |
3f157a2f | 4436 | again: |
bd681513 | 4437 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4438 | level = btrfs_header_level(cur); |
e02119d5 | 4439 | WARN_ON(path->nodes[level]); |
3f157a2f | 4440 | path->nodes[level] = cur; |
bd681513 | 4441 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4442 | |
4443 | if (btrfs_header_generation(cur) < min_trans) { | |
4444 | ret = 1; | |
4445 | goto out; | |
4446 | } | |
d397712b | 4447 | while (1) { |
3f157a2f CM |
4448 | nritems = btrfs_header_nritems(cur); |
4449 | level = btrfs_header_level(cur); | |
e3b83361 | 4450 | sret = btrfs_bin_search(cur, min_key, &slot); |
cbca7d59 FM |
4451 | if (sret < 0) { |
4452 | ret = sret; | |
4453 | goto out; | |
4454 | } | |
3f157a2f | 4455 | |
323ac95b CM |
4456 | /* at the lowest level, we're done, setup the path and exit */ |
4457 | if (level == path->lowest_level) { | |
e02119d5 CM |
4458 | if (slot >= nritems) |
4459 | goto find_next_key; | |
3f157a2f CM |
4460 | ret = 0; |
4461 | path->slots[level] = slot; | |
4462 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4463 | goto out; | |
4464 | } | |
9652480b Y |
4465 | if (sret && slot > 0) |
4466 | slot--; | |
3f157a2f | 4467 | /* |
de78b51a | 4468 | * check this node pointer against the min_trans parameters. |
260db43c | 4469 | * If it is too old, skip to the next one. |
3f157a2f | 4470 | */ |
d397712b | 4471 | while (slot < nritems) { |
3f157a2f | 4472 | u64 gen; |
e02119d5 | 4473 | |
3f157a2f CM |
4474 | gen = btrfs_node_ptr_generation(cur, slot); |
4475 | if (gen < min_trans) { | |
4476 | slot++; | |
4477 | continue; | |
4478 | } | |
de78b51a | 4479 | break; |
3f157a2f | 4480 | } |
e02119d5 | 4481 | find_next_key: |
3f157a2f CM |
4482 | /* |
4483 | * we didn't find a candidate key in this node, walk forward | |
4484 | * and find another one | |
4485 | */ | |
4486 | if (slot >= nritems) { | |
e02119d5 CM |
4487 | path->slots[level] = slot; |
4488 | sret = btrfs_find_next_key(root, path, min_key, level, | |
de78b51a | 4489 | min_trans); |
e02119d5 | 4490 | if (sret == 0) { |
b3b4aa74 | 4491 | btrfs_release_path(path); |
3f157a2f CM |
4492 | goto again; |
4493 | } else { | |
4494 | goto out; | |
4495 | } | |
4496 | } | |
4497 | /* save our key for returning back */ | |
4498 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4499 | path->slots[level] = slot; | |
4500 | if (level == path->lowest_level) { | |
4501 | ret = 0; | |
3f157a2f CM |
4502 | goto out; |
4503 | } | |
4b231ae4 | 4504 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
4505 | if (IS_ERR(cur)) { |
4506 | ret = PTR_ERR(cur); | |
4507 | goto out; | |
4508 | } | |
3f157a2f | 4509 | |
bd681513 | 4510 | btrfs_tree_read_lock(cur); |
b4ce94de | 4511 | |
bd681513 | 4512 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4513 | path->nodes[level - 1] = cur; |
f7c79f30 | 4514 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4515 | } |
4516 | out: | |
f98de9b9 FM |
4517 | path->keep_locks = keep_locks; |
4518 | if (ret == 0) { | |
4519 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
3f157a2f | 4520 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 4521 | } |
3f157a2f CM |
4522 | return ret; |
4523 | } | |
4524 | ||
4525 | /* | |
4526 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4527 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 4528 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
4529 | * |
4530 | * 0 is returned if another key is found, < 0 if there are any errors | |
4531 | * and 1 is returned if there are no higher keys in the tree | |
4532 | * | |
4533 | * path->keep_locks should be set to 1 on the search made before | |
4534 | * calling this function. | |
4535 | */ | |
e7a84565 | 4536 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 4537 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 4538 | { |
e7a84565 CM |
4539 | int slot; |
4540 | struct extent_buffer *c; | |
4541 | ||
6a9fb468 | 4542 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 4543 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4544 | if (!path->nodes[level]) |
4545 | return 1; | |
4546 | ||
4547 | slot = path->slots[level] + 1; | |
4548 | c = path->nodes[level]; | |
3f157a2f | 4549 | next: |
e7a84565 | 4550 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4551 | int ret; |
4552 | int orig_lowest; | |
4553 | struct btrfs_key cur_key; | |
4554 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4555 | !path->nodes[level + 1]) | |
e7a84565 | 4556 | return 1; |
33c66f43 | 4557 | |
6a9fb468 | 4558 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
4559 | level++; |
4560 | continue; | |
4561 | } | |
4562 | ||
4563 | slot = btrfs_header_nritems(c) - 1; | |
4564 | if (level == 0) | |
4565 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4566 | else | |
4567 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4568 | ||
4569 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4570 | btrfs_release_path(path); |
33c66f43 YZ |
4571 | path->lowest_level = level; |
4572 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4573 | 0, 0); | |
4574 | path->lowest_level = orig_lowest; | |
4575 | if (ret < 0) | |
4576 | return ret; | |
4577 | ||
4578 | c = path->nodes[level]; | |
4579 | slot = path->slots[level]; | |
4580 | if (ret == 0) | |
4581 | slot++; | |
4582 | goto next; | |
e7a84565 | 4583 | } |
33c66f43 | 4584 | |
e7a84565 CM |
4585 | if (level == 0) |
4586 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 4587 | else { |
3f157a2f CM |
4588 | u64 gen = btrfs_node_ptr_generation(c, slot); |
4589 | ||
3f157a2f CM |
4590 | if (gen < min_trans) { |
4591 | slot++; | |
4592 | goto next; | |
4593 | } | |
e7a84565 | 4594 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4595 | } |
e7a84565 CM |
4596 | return 0; |
4597 | } | |
4598 | return 1; | |
4599 | } | |
4600 | ||
3d7806ec JS |
4601 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, |
4602 | u64 time_seq) | |
d97e63b6 CM |
4603 | { |
4604 | int slot; | |
8e73f275 | 4605 | int level; |
5f39d397 | 4606 | struct extent_buffer *c; |
8e73f275 | 4607 | struct extent_buffer *next; |
d96b3424 | 4608 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 4609 | struct btrfs_key key; |
d96b3424 | 4610 | bool need_commit_sem = false; |
925baedd CM |
4611 | u32 nritems; |
4612 | int ret; | |
0e46318d | 4613 | int i; |
925baedd CM |
4614 | |
4615 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4616 | if (nritems == 0) |
925baedd | 4617 | return 1; |
925baedd | 4618 | |
8e73f275 CM |
4619 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4620 | again: | |
4621 | level = 1; | |
4622 | next = NULL; | |
b3b4aa74 | 4623 | btrfs_release_path(path); |
8e73f275 | 4624 | |
a2135011 | 4625 | path->keep_locks = 1; |
8e73f275 | 4626 | |
d96b3424 | 4627 | if (time_seq) { |
3d7806ec | 4628 | ret = btrfs_search_old_slot(root, &key, path, time_seq); |
d96b3424 FM |
4629 | } else { |
4630 | if (path->need_commit_sem) { | |
4631 | path->need_commit_sem = 0; | |
4632 | need_commit_sem = true; | |
4633 | down_read(&fs_info->commit_root_sem); | |
4634 | } | |
3d7806ec | 4635 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
d96b3424 | 4636 | } |
925baedd CM |
4637 | path->keep_locks = 0; |
4638 | ||
4639 | if (ret < 0) | |
d96b3424 | 4640 | goto done; |
925baedd | 4641 | |
a2135011 | 4642 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4643 | /* |
4644 | * by releasing the path above we dropped all our locks. A balance | |
4645 | * could have added more items next to the key that used to be | |
4646 | * at the very end of the block. So, check again here and | |
4647 | * advance the path if there are now more items available. | |
4648 | */ | |
a2135011 | 4649 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4650 | if (ret == 0) |
4651 | path->slots[0]++; | |
8e73f275 | 4652 | ret = 0; |
925baedd CM |
4653 | goto done; |
4654 | } | |
0b43e04f LB |
4655 | /* |
4656 | * So the above check misses one case: | |
4657 | * - after releasing the path above, someone has removed the item that | |
4658 | * used to be at the very end of the block, and balance between leafs | |
4659 | * gets another one with bigger key.offset to replace it. | |
4660 | * | |
4661 | * This one should be returned as well, or we can get leaf corruption | |
4662 | * later(esp. in __btrfs_drop_extents()). | |
4663 | * | |
4664 | * And a bit more explanation about this check, | |
4665 | * with ret > 0, the key isn't found, the path points to the slot | |
4666 | * where it should be inserted, so the path->slots[0] item must be the | |
4667 | * bigger one. | |
4668 | */ | |
4669 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
4670 | ret = 0; | |
4671 | goto done; | |
4672 | } | |
d97e63b6 | 4673 | |
d397712b | 4674 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4675 | if (!path->nodes[level]) { |
4676 | ret = 1; | |
4677 | goto done; | |
4678 | } | |
5f39d397 | 4679 | |
d97e63b6 CM |
4680 | slot = path->slots[level] + 1; |
4681 | c = path->nodes[level]; | |
5f39d397 | 4682 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4683 | level++; |
8e73f275 CM |
4684 | if (level == BTRFS_MAX_LEVEL) { |
4685 | ret = 1; | |
4686 | goto done; | |
4687 | } | |
d97e63b6 CM |
4688 | continue; |
4689 | } | |
5f39d397 | 4690 | |
0e46318d JB |
4691 | |
4692 | /* | |
4693 | * Our current level is where we're going to start from, and to | |
4694 | * make sure lockdep doesn't complain we need to drop our locks | |
4695 | * and nodes from 0 to our current level. | |
4696 | */ | |
4697 | for (i = 0; i < level; i++) { | |
4698 | if (path->locks[level]) { | |
4699 | btrfs_tree_read_unlock(path->nodes[i]); | |
4700 | path->locks[i] = 0; | |
4701 | } | |
4702 | free_extent_buffer(path->nodes[i]); | |
4703 | path->nodes[i] = NULL; | |
925baedd | 4704 | } |
5f39d397 | 4705 | |
8e73f275 | 4706 | next = c; |
d07b8528 | 4707 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4708 | slot, &key); |
8e73f275 CM |
4709 | if (ret == -EAGAIN) |
4710 | goto again; | |
5f39d397 | 4711 | |
76a05b35 | 4712 | if (ret < 0) { |
b3b4aa74 | 4713 | btrfs_release_path(path); |
76a05b35 CM |
4714 | goto done; |
4715 | } | |
4716 | ||
5cd57b2c | 4717 | if (!path->skip_locking) { |
bd681513 | 4718 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
4719 | if (!ret && time_seq) { |
4720 | /* | |
4721 | * If we don't get the lock, we may be racing | |
4722 | * with push_leaf_left, holding that lock while | |
4723 | * itself waiting for the leaf we've currently | |
4724 | * locked. To solve this situation, we give up | |
4725 | * on our lock and cycle. | |
4726 | */ | |
cf538830 | 4727 | free_extent_buffer(next); |
d42244a0 JS |
4728 | btrfs_release_path(path); |
4729 | cond_resched(); | |
4730 | goto again; | |
4731 | } | |
0e46318d JB |
4732 | if (!ret) |
4733 | btrfs_tree_read_lock(next); | |
5cd57b2c | 4734 | } |
d97e63b6 CM |
4735 | break; |
4736 | } | |
4737 | path->slots[level] = slot; | |
d397712b | 4738 | while (1) { |
d97e63b6 | 4739 | level--; |
d97e63b6 CM |
4740 | path->nodes[level] = next; |
4741 | path->slots[level] = 0; | |
a74a4b97 | 4742 | if (!path->skip_locking) |
ffeb03cf | 4743 | path->locks[level] = BTRFS_READ_LOCK; |
d97e63b6 CM |
4744 | if (!level) |
4745 | break; | |
b4ce94de | 4746 | |
d07b8528 | 4747 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4748 | 0, &key); |
8e73f275 CM |
4749 | if (ret == -EAGAIN) |
4750 | goto again; | |
4751 | ||
76a05b35 | 4752 | if (ret < 0) { |
b3b4aa74 | 4753 | btrfs_release_path(path); |
76a05b35 CM |
4754 | goto done; |
4755 | } | |
4756 | ||
ffeb03cf | 4757 | if (!path->skip_locking) |
0e46318d | 4758 | btrfs_tree_read_lock(next); |
d97e63b6 | 4759 | } |
8e73f275 | 4760 | ret = 0; |
925baedd | 4761 | done: |
f7c79f30 | 4762 | unlock_up(path, 0, 1, 0, NULL); |
d96b3424 FM |
4763 | if (need_commit_sem) { |
4764 | int ret2; | |
4765 | ||
4766 | path->need_commit_sem = 1; | |
4767 | ret2 = finish_need_commit_sem_search(path); | |
4768 | up_read(&fs_info->commit_root_sem); | |
4769 | if (ret2) | |
4770 | ret = ret2; | |
4771 | } | |
8e73f275 CM |
4772 | |
4773 | return ret; | |
d97e63b6 | 4774 | } |
0b86a832 | 4775 | |
3f157a2f CM |
4776 | /* |
4777 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4778 | * searching until it gets past min_objectid or finds an item of 'type' | |
4779 | * | |
4780 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4781 | */ | |
0b86a832 CM |
4782 | int btrfs_previous_item(struct btrfs_root *root, |
4783 | struct btrfs_path *path, u64 min_objectid, | |
4784 | int type) | |
4785 | { | |
4786 | struct btrfs_key found_key; | |
4787 | struct extent_buffer *leaf; | |
e02119d5 | 4788 | u32 nritems; |
0b86a832 CM |
4789 | int ret; |
4790 | ||
d397712b | 4791 | while (1) { |
0b86a832 CM |
4792 | if (path->slots[0] == 0) { |
4793 | ret = btrfs_prev_leaf(root, path); | |
4794 | if (ret != 0) | |
4795 | return ret; | |
4796 | } else { | |
4797 | path->slots[0]--; | |
4798 | } | |
4799 | leaf = path->nodes[0]; | |
e02119d5 CM |
4800 | nritems = btrfs_header_nritems(leaf); |
4801 | if (nritems == 0) | |
4802 | return 1; | |
4803 | if (path->slots[0] == nritems) | |
4804 | path->slots[0]--; | |
4805 | ||
0b86a832 | 4806 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4807 | if (found_key.objectid < min_objectid) |
4808 | break; | |
0a4eefbb YZ |
4809 | if (found_key.type == type) |
4810 | return 0; | |
e02119d5 CM |
4811 | if (found_key.objectid == min_objectid && |
4812 | found_key.type < type) | |
4813 | break; | |
0b86a832 CM |
4814 | } |
4815 | return 1; | |
4816 | } | |
ade2e0b3 WS |
4817 | |
4818 | /* | |
4819 | * search in extent tree to find a previous Metadata/Data extent item with | |
4820 | * min objecitd. | |
4821 | * | |
4822 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4823 | */ | |
4824 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
4825 | struct btrfs_path *path, u64 min_objectid) | |
4826 | { | |
4827 | struct btrfs_key found_key; | |
4828 | struct extent_buffer *leaf; | |
4829 | u32 nritems; | |
4830 | int ret; | |
4831 | ||
4832 | while (1) { | |
4833 | if (path->slots[0] == 0) { | |
ade2e0b3 WS |
4834 | ret = btrfs_prev_leaf(root, path); |
4835 | if (ret != 0) | |
4836 | return ret; | |
4837 | } else { | |
4838 | path->slots[0]--; | |
4839 | } | |
4840 | leaf = path->nodes[0]; | |
4841 | nritems = btrfs_header_nritems(leaf); | |
4842 | if (nritems == 0) | |
4843 | return 1; | |
4844 | if (path->slots[0] == nritems) | |
4845 | path->slots[0]--; | |
4846 | ||
4847 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4848 | if (found_key.objectid < min_objectid) | |
4849 | break; | |
4850 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
4851 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
4852 | return 0; | |
4853 | if (found_key.objectid == min_objectid && | |
4854 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
4855 | break; | |
4856 | } | |
4857 | return 1; | |
4858 | } |