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" |
9a8dd150 | 19 | |
e089f05c CM |
20 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
21 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
22 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
23 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
24 | int data_size, int extend); | |
5f39d397 | 25 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 26 | struct extent_buffer *dst, |
971a1f66 | 27 | struct extent_buffer *src, int empty); |
5f39d397 | 28 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
29 | struct extent_buffer *dst_buf, |
30 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
31 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
32 | int level, int slot); | |
d97e63b6 | 33 | |
af024ed2 JT |
34 | static const struct btrfs_csums { |
35 | u16 size; | |
59a0fcdb DS |
36 | const char name[10]; |
37 | const char driver[12]; | |
af024ed2 JT |
38 | } btrfs_csums[] = { |
39 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
3951e7f0 | 40 | [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, |
3831bf00 | 41 | [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, |
352ae07b DS |
42 | [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", |
43 | .driver = "blake2b-256" }, | |
af024ed2 JT |
44 | }; |
45 | ||
46 | int btrfs_super_csum_size(const struct btrfs_super_block *s) | |
47 | { | |
48 | u16 t = btrfs_super_csum_type(s); | |
49 | /* | |
50 | * csum type is validated at mount time | |
51 | */ | |
52 | return btrfs_csums[t].size; | |
53 | } | |
54 | ||
55 | const char *btrfs_super_csum_name(u16 csum_type) | |
56 | { | |
57 | /* csum type is validated at mount time */ | |
58 | return btrfs_csums[csum_type].name; | |
59 | } | |
60 | ||
b4e967be DS |
61 | /* |
62 | * Return driver name if defined, otherwise the name that's also a valid driver | |
63 | * name | |
64 | */ | |
65 | const char *btrfs_super_csum_driver(u16 csum_type) | |
66 | { | |
67 | /* csum type is validated at mount time */ | |
59a0fcdb DS |
68 | return btrfs_csums[csum_type].driver[0] ? |
69 | btrfs_csums[csum_type].driver : | |
b4e967be DS |
70 | btrfs_csums[csum_type].name; |
71 | } | |
72 | ||
604997b4 | 73 | size_t __attribute_const__ btrfs_get_num_csums(void) |
f7cea56c DS |
74 | { |
75 | return ARRAY_SIZE(btrfs_csums); | |
76 | } | |
77 | ||
df24a2b9 | 78 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 79 | { |
e2c89907 | 80 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
81 | } |
82 | ||
d352ac68 | 83 | /* this also releases the path */ |
df24a2b9 | 84 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 85 | { |
ff175d57 JJ |
86 | if (!p) |
87 | return; | |
b3b4aa74 | 88 | btrfs_release_path(p); |
df24a2b9 | 89 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
90 | } |
91 | ||
d352ac68 CM |
92 | /* |
93 | * path release drops references on the extent buffers in the path | |
94 | * and it drops any locks held by this path | |
95 | * | |
96 | * It is safe to call this on paths that no locks or extent buffers held. | |
97 | */ | |
b3b4aa74 | 98 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
99 | { |
100 | int i; | |
a2135011 | 101 | |
234b63a0 | 102 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 103 | p->slots[i] = 0; |
eb60ceac | 104 | if (!p->nodes[i]) |
925baedd CM |
105 | continue; |
106 | if (p->locks[i]) { | |
bd681513 | 107 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
108 | p->locks[i] = 0; |
109 | } | |
5f39d397 | 110 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 111 | p->nodes[i] = NULL; |
eb60ceac CM |
112 | } |
113 | } | |
114 | ||
d352ac68 CM |
115 | /* |
116 | * safely gets a reference on the root node of a tree. A lock | |
117 | * is not taken, so a concurrent writer may put a different node | |
118 | * at the root of the tree. See btrfs_lock_root_node for the | |
119 | * looping required. | |
120 | * | |
121 | * The extent buffer returned by this has a reference taken, so | |
122 | * it won't disappear. It may stop being the root of the tree | |
123 | * at any time because there are no locks held. | |
124 | */ | |
925baedd CM |
125 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
126 | { | |
127 | struct extent_buffer *eb; | |
240f62c8 | 128 | |
3083ee2e JB |
129 | while (1) { |
130 | rcu_read_lock(); | |
131 | eb = rcu_dereference(root->node); | |
132 | ||
133 | /* | |
134 | * RCU really hurts here, we could free up the root node because | |
01327610 | 135 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
136 | * the inc_not_zero dance and if it doesn't work then |
137 | * synchronize_rcu and try again. | |
138 | */ | |
139 | if (atomic_inc_not_zero(&eb->refs)) { | |
140 | rcu_read_unlock(); | |
141 | break; | |
142 | } | |
143 | rcu_read_unlock(); | |
144 | synchronize_rcu(); | |
145 | } | |
925baedd CM |
146 | return eb; |
147 | } | |
148 | ||
92a7cc42 QW |
149 | /* |
150 | * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), | |
151 | * just get put onto a simple dirty list. Transaction walks this list to make | |
152 | * sure they get properly updated on disk. | |
d352ac68 | 153 | */ |
0b86a832 CM |
154 | static void add_root_to_dirty_list(struct btrfs_root *root) |
155 | { | |
0b246afa JM |
156 | struct btrfs_fs_info *fs_info = root->fs_info; |
157 | ||
e7070be1 JB |
158 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
159 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
160 | return; | |
161 | ||
0b246afa | 162 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
163 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
164 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 165 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 166 | list_move_tail(&root->dirty_list, |
0b246afa | 167 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
168 | else |
169 | list_move(&root->dirty_list, | |
0b246afa | 170 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 171 | } |
0b246afa | 172 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
173 | } |
174 | ||
d352ac68 CM |
175 | /* |
176 | * used by snapshot creation to make a copy of a root for a tree with | |
177 | * a given objectid. The buffer with the new root node is returned in | |
178 | * cow_ret, and this func returns zero on success or a negative error code. | |
179 | */ | |
be20aa9d CM |
180 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
181 | struct btrfs_root *root, | |
182 | struct extent_buffer *buf, | |
183 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
184 | { | |
0b246afa | 185 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 186 | struct extent_buffer *cow; |
be20aa9d CM |
187 | int ret = 0; |
188 | int level; | |
5d4f98a2 | 189 | struct btrfs_disk_key disk_key; |
be20aa9d | 190 | |
92a7cc42 | 191 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 192 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 193 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 194 | trans->transid != root->last_trans); |
be20aa9d CM |
195 | |
196 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
197 | if (level == 0) |
198 | btrfs_item_key(buf, &disk_key, 0); | |
199 | else | |
200 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 201 | |
4d75f8a9 | 202 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
cf6f34aa JB |
203 | &disk_key, level, buf->start, 0, |
204 | BTRFS_NESTING_NEW_ROOT); | |
5d4f98a2 | 205 | if (IS_ERR(cow)) |
be20aa9d CM |
206 | return PTR_ERR(cow); |
207 | ||
58e8012c | 208 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
209 | btrfs_set_header_bytenr(cow, cow->start); |
210 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
211 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
212 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
213 | BTRFS_HEADER_FLAG_RELOC); | |
214 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
215 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
216 | else | |
217 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 218 | |
de37aa51 | 219 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 220 | |
be20aa9d | 221 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 222 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 223 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 224 | else |
e339a6b0 | 225 | ret = btrfs_inc_ref(trans, root, cow, 0); |
867ed321 | 226 | if (ret) { |
72c9925f FM |
227 | btrfs_tree_unlock(cow); |
228 | free_extent_buffer(cow); | |
867ed321 | 229 | btrfs_abort_transaction(trans, ret); |
be20aa9d | 230 | return ret; |
867ed321 | 231 | } |
be20aa9d CM |
232 | |
233 | btrfs_mark_buffer_dirty(cow); | |
234 | *cow_ret = cow; | |
235 | return 0; | |
236 | } | |
237 | ||
5d4f98a2 YZ |
238 | /* |
239 | * check if the tree block can be shared by multiple trees | |
240 | */ | |
241 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
242 | struct extent_buffer *buf) | |
243 | { | |
244 | /* | |
92a7cc42 QW |
245 | * Tree blocks not in shareable trees and tree roots are never shared. |
246 | * If a block was allocated after the last snapshot and the block was | |
247 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 248 | */ |
92a7cc42 | 249 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
250 | buf != root->node && buf != root->commit_root && |
251 | (btrfs_header_generation(buf) <= | |
252 | btrfs_root_last_snapshot(&root->root_item) || | |
253 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
254 | return 1; | |
a79865c6 | 255 | |
5d4f98a2 YZ |
256 | return 0; |
257 | } | |
258 | ||
259 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
260 | struct btrfs_root *root, | |
261 | struct extent_buffer *buf, | |
f0486c68 YZ |
262 | struct extent_buffer *cow, |
263 | int *last_ref) | |
5d4f98a2 | 264 | { |
0b246afa | 265 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
266 | u64 refs; |
267 | u64 owner; | |
268 | u64 flags; | |
269 | u64 new_flags = 0; | |
270 | int ret; | |
271 | ||
272 | /* | |
273 | * Backrefs update rules: | |
274 | * | |
275 | * Always use full backrefs for extent pointers in tree block | |
276 | * allocated by tree relocation. | |
277 | * | |
278 | * If a shared tree block is no longer referenced by its owner | |
279 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
280 | * use full backrefs for extent pointers in tree block. | |
281 | * | |
282 | * If a tree block is been relocating | |
283 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
284 | * use full backrefs for extent pointers in tree block. | |
285 | * The reason for this is some operations (such as drop tree) | |
286 | * are only allowed for blocks use full backrefs. | |
287 | */ | |
288 | ||
289 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 290 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
291 | btrfs_header_level(buf), 1, |
292 | &refs, &flags); | |
be1a5564 MF |
293 | if (ret) |
294 | return ret; | |
e5df9573 MF |
295 | if (refs == 0) { |
296 | ret = -EROFS; | |
0b246afa | 297 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
298 | return ret; |
299 | } | |
5d4f98a2 YZ |
300 | } else { |
301 | refs = 1; | |
302 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
303 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
304 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
305 | else | |
306 | flags = 0; | |
307 | } | |
308 | ||
309 | owner = btrfs_header_owner(buf); | |
310 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
311 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
312 | ||
313 | if (refs > 1) { | |
314 | if ((owner == root->root_key.objectid || | |
315 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
316 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 317 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
318 | if (ret) |
319 | return ret; | |
5d4f98a2 YZ |
320 | |
321 | if (root->root_key.objectid == | |
322 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 323 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
324 | if (ret) |
325 | return ret; | |
e339a6b0 | 326 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
327 | if (ret) |
328 | return ret; | |
5d4f98a2 YZ |
329 | } |
330 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
331 | } else { | |
332 | ||
333 | if (root->root_key.objectid == | |
334 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 335 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 336 | else |
e339a6b0 | 337 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
338 | if (ret) |
339 | return ret; | |
5d4f98a2 YZ |
340 | } |
341 | if (new_flags != 0) { | |
b1c79e09 JB |
342 | int level = btrfs_header_level(buf); |
343 | ||
42c9d0b5 | 344 | ret = btrfs_set_disk_extent_flags(trans, buf, |
b1c79e09 | 345 | new_flags, level, 0); |
be1a5564 MF |
346 | if (ret) |
347 | return ret; | |
5d4f98a2 YZ |
348 | } |
349 | } else { | |
350 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
351 | if (root->root_key.objectid == | |
352 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 353 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 354 | else |
e339a6b0 | 355 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
356 | if (ret) |
357 | return ret; | |
e339a6b0 | 358 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
359 | if (ret) |
360 | return ret; | |
5d4f98a2 | 361 | } |
6a884d7d | 362 | btrfs_clean_tree_block(buf); |
f0486c68 | 363 | *last_ref = 1; |
5d4f98a2 YZ |
364 | } |
365 | return 0; | |
366 | } | |
367 | ||
d352ac68 | 368 | /* |
d397712b CM |
369 | * does the dirty work in cow of a single block. The parent block (if |
370 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
371 | * dirty and returned locked. If you modify the block it needs to be marked | |
372 | * dirty again. | |
d352ac68 CM |
373 | * |
374 | * search_start -- an allocation hint for the new block | |
375 | * | |
d397712b CM |
376 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
377 | * bytes the allocator should try to find free next to the block it returns. | |
378 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 379 | */ |
d397712b | 380 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
381 | struct btrfs_root *root, |
382 | struct extent_buffer *buf, | |
383 | struct extent_buffer *parent, int parent_slot, | |
384 | struct extent_buffer **cow_ret, | |
9631e4cc JB |
385 | u64 search_start, u64 empty_size, |
386 | enum btrfs_lock_nesting nest) | |
02217ed2 | 387 | { |
0b246afa | 388 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 389 | struct btrfs_disk_key disk_key; |
5f39d397 | 390 | struct extent_buffer *cow; |
be1a5564 | 391 | int level, ret; |
f0486c68 | 392 | int last_ref = 0; |
925baedd | 393 | int unlock_orig = 0; |
0f5053eb | 394 | u64 parent_start = 0; |
7bb86316 | 395 | |
925baedd CM |
396 | if (*cow_ret == buf) |
397 | unlock_orig = 1; | |
398 | ||
49d0c642 | 399 | btrfs_assert_tree_write_locked(buf); |
925baedd | 400 | |
92a7cc42 | 401 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 402 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 403 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 404 | trans->transid != root->last_trans); |
5f39d397 | 405 | |
7bb86316 | 406 | level = btrfs_header_level(buf); |
31840ae1 | 407 | |
5d4f98a2 YZ |
408 | if (level == 0) |
409 | btrfs_item_key(buf, &disk_key, 0); | |
410 | else | |
411 | btrfs_node_key(buf, &disk_key, 0); | |
412 | ||
0f5053eb GR |
413 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
414 | parent_start = parent->start; | |
5d4f98a2 | 415 | |
79bd3712 FM |
416 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
417 | root->root_key.objectid, &disk_key, level, | |
418 | search_start, empty_size, nest); | |
54aa1f4d CM |
419 | if (IS_ERR(cow)) |
420 | return PTR_ERR(cow); | |
6702ed49 | 421 | |
b4ce94de CM |
422 | /* cow is set to blocking by btrfs_init_new_buffer */ |
423 | ||
58e8012c | 424 | copy_extent_buffer_full(cow, buf); |
db94535d | 425 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 426 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
427 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
428 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
429 | BTRFS_HEADER_FLAG_RELOC); | |
430 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
431 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
432 | else | |
433 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 434 | |
de37aa51 | 435 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 436 | |
be1a5564 | 437 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 438 | if (ret) { |
572c83ac JB |
439 | btrfs_tree_unlock(cow); |
440 | free_extent_buffer(cow); | |
66642832 | 441 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
442 | return ret; |
443 | } | |
1a40e23b | 444 | |
92a7cc42 | 445 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 446 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 447 | if (ret) { |
572c83ac JB |
448 | btrfs_tree_unlock(cow); |
449 | free_extent_buffer(cow); | |
66642832 | 450 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 451 | return ret; |
93314e3b | 452 | } |
83d4cfd4 | 453 | } |
3fd0a558 | 454 | |
02217ed2 | 455 | if (buf == root->node) { |
925baedd | 456 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
457 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
458 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
459 | parent_start = buf->start; | |
925baedd | 460 | |
67439dad | 461 | atomic_inc(&cow->refs); |
406808ab | 462 | ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); |
d9d19a01 | 463 | BUG_ON(ret < 0); |
240f62c8 | 464 | rcu_assign_pointer(root->node, cow); |
925baedd | 465 | |
7a163608 FM |
466 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
467 | parent_start, last_ref); | |
5f39d397 | 468 | free_extent_buffer(buf); |
0b86a832 | 469 | add_root_to_dirty_list(root); |
02217ed2 | 470 | } else { |
5d4f98a2 | 471 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
f3a84ccd FM |
472 | btrfs_tree_mod_log_insert_key(parent, parent_slot, |
473 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
5f39d397 | 474 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 475 | cow->start); |
74493f7a CM |
476 | btrfs_set_node_ptr_generation(parent, parent_slot, |
477 | trans->transid); | |
d6025579 | 478 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 479 | if (last_ref) { |
f3a84ccd | 480 | ret = btrfs_tree_mod_log_free_eb(buf); |
5de865ee | 481 | if (ret) { |
572c83ac JB |
482 | btrfs_tree_unlock(cow); |
483 | free_extent_buffer(cow); | |
66642832 | 484 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
485 | return ret; |
486 | } | |
487 | } | |
7a163608 FM |
488 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
489 | parent_start, last_ref); | |
02217ed2 | 490 | } |
925baedd CM |
491 | if (unlock_orig) |
492 | btrfs_tree_unlock(buf); | |
3083ee2e | 493 | free_extent_buffer_stale(buf); |
ccd467d6 | 494 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 495 | *cow_ret = cow; |
02217ed2 CM |
496 | return 0; |
497 | } | |
498 | ||
5d4f98a2 YZ |
499 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
500 | struct btrfs_root *root, | |
501 | struct extent_buffer *buf) | |
502 | { | |
f5ee5c9a | 503 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 504 | return 0; |
fccb84c9 | 505 | |
d1980131 DS |
506 | /* Ensure we can see the FORCE_COW bit */ |
507 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
508 | |
509 | /* | |
510 | * We do not need to cow a block if | |
511 | * 1) this block is not created or changed in this transaction; | |
512 | * 2) this block does not belong to TREE_RELOC tree; | |
513 | * 3) the root is not forced COW. | |
514 | * | |
515 | * What is forced COW: | |
01327610 | 516 | * when we create snapshot during committing the transaction, |
52042d8e | 517 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
518 | * block to ensure the metadata consistency. |
519 | */ | |
5d4f98a2 YZ |
520 | if (btrfs_header_generation(buf) == trans->transid && |
521 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
522 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 523 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 524 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
525 | return 0; |
526 | return 1; | |
527 | } | |
528 | ||
d352ac68 CM |
529 | /* |
530 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 531 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
532 | * once per transaction, as long as it hasn't been written yet |
533 | */ | |
d397712b | 534 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
535 | struct btrfs_root *root, struct extent_buffer *buf, |
536 | struct extent_buffer *parent, int parent_slot, | |
9631e4cc JB |
537 | struct extent_buffer **cow_ret, |
538 | enum btrfs_lock_nesting nest) | |
6702ed49 | 539 | { |
0b246afa | 540 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 541 | u64 search_start; |
f510cfec | 542 | int ret; |
dc17ff8f | 543 | |
83354f07 JB |
544 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
545 | btrfs_err(fs_info, | |
546 | "COW'ing blocks on a fs root that's being dropped"); | |
547 | ||
0b246afa | 548 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 549 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 550 | trans->transid, |
0b246afa | 551 | fs_info->running_transaction->transid); |
31b1a2bd | 552 | |
0b246afa | 553 | if (trans->transid != fs_info->generation) |
31b1a2bd | 554 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 555 | trans->transid, fs_info->generation); |
dc17ff8f | 556 | |
5d4f98a2 | 557 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
558 | *cow_ret = buf; |
559 | return 0; | |
560 | } | |
c487685d | 561 | |
ee22184b | 562 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de | 563 | |
f616f5cd QW |
564 | /* |
565 | * Before CoWing this block for later modification, check if it's | |
566 | * the subtree root and do the delayed subtree trace if needed. | |
567 | * | |
568 | * Also We don't care about the error, as it's handled internally. | |
569 | */ | |
570 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 571 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9631e4cc | 572 | parent_slot, cow_ret, search_start, 0, nest); |
1abe9b8a | 573 | |
574 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
575 | ||
f510cfec | 576 | return ret; |
6702ed49 | 577 | } |
f75e2b79 | 578 | ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); |
6702ed49 | 579 | |
d352ac68 CM |
580 | /* |
581 | * helper function for defrag to decide if two blocks pointed to by a | |
582 | * node are actually close by | |
583 | */ | |
6b80053d | 584 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 585 | { |
6b80053d | 586 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 587 | return 1; |
6b80053d | 588 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
589 | return 1; |
590 | return 0; | |
591 | } | |
592 | ||
ce6ef5ab DS |
593 | #ifdef __LITTLE_ENDIAN |
594 | ||
595 | /* | |
596 | * Compare two keys, on little-endian the disk order is same as CPU order and | |
597 | * we can avoid the conversion. | |
598 | */ | |
599 | static int comp_keys(const struct btrfs_disk_key *disk_key, | |
600 | const struct btrfs_key *k2) | |
601 | { | |
602 | const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; | |
603 | ||
604 | return btrfs_comp_cpu_keys(k1, k2); | |
605 | } | |
606 | ||
607 | #else | |
608 | ||
081e9573 CM |
609 | /* |
610 | * compare two keys in a memcmp fashion | |
611 | */ | |
310712b2 OS |
612 | static int comp_keys(const struct btrfs_disk_key *disk, |
613 | const struct btrfs_key *k2) | |
081e9573 CM |
614 | { |
615 | struct btrfs_key k1; | |
616 | ||
617 | btrfs_disk_key_to_cpu(&k1, disk); | |
618 | ||
20736aba | 619 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 | 620 | } |
ce6ef5ab | 621 | #endif |
081e9573 | 622 | |
f3465ca4 JB |
623 | /* |
624 | * same as comp_keys only with two btrfs_key's | |
625 | */ | |
e1f60a65 | 626 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
627 | { |
628 | if (k1->objectid > k2->objectid) | |
629 | return 1; | |
630 | if (k1->objectid < k2->objectid) | |
631 | return -1; | |
632 | if (k1->type > k2->type) | |
633 | return 1; | |
634 | if (k1->type < k2->type) | |
635 | return -1; | |
636 | if (k1->offset > k2->offset) | |
637 | return 1; | |
638 | if (k1->offset < k2->offset) | |
639 | return -1; | |
640 | return 0; | |
641 | } | |
081e9573 | 642 | |
d352ac68 CM |
643 | /* |
644 | * this is used by the defrag code to go through all the | |
645 | * leaves pointed to by a node and reallocate them so that | |
646 | * disk order is close to key order | |
647 | */ | |
6702ed49 | 648 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 649 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 650 | int start_slot, u64 *last_ret, |
a6b6e75e | 651 | struct btrfs_key *progress) |
6702ed49 | 652 | { |
0b246afa | 653 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 654 | struct extent_buffer *cur; |
6702ed49 | 655 | u64 blocknr; |
e9d0b13b CM |
656 | u64 search_start = *last_ret; |
657 | u64 last_block = 0; | |
6702ed49 CM |
658 | u64 other; |
659 | u32 parent_nritems; | |
6702ed49 CM |
660 | int end_slot; |
661 | int i; | |
662 | int err = 0; | |
6b80053d | 663 | u32 blocksize; |
081e9573 CM |
664 | int progress_passed = 0; |
665 | struct btrfs_disk_key disk_key; | |
6702ed49 | 666 | |
0b246afa JM |
667 | WARN_ON(trans->transaction != fs_info->running_transaction); |
668 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 669 | |
6b80053d | 670 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 671 | blocksize = fs_info->nodesize; |
5dfe2be7 | 672 | end_slot = parent_nritems - 1; |
6702ed49 | 673 | |
5dfe2be7 | 674 | if (parent_nritems <= 1) |
6702ed49 CM |
675 | return 0; |
676 | ||
5dfe2be7 | 677 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 678 | int close = 1; |
a6b6e75e | 679 | |
081e9573 CM |
680 | btrfs_node_key(parent, &disk_key, i); |
681 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
682 | continue; | |
683 | ||
684 | progress_passed = 1; | |
6b80053d | 685 | blocknr = btrfs_node_blockptr(parent, i); |
e9d0b13b CM |
686 | if (last_block == 0) |
687 | last_block = blocknr; | |
5708b959 | 688 | |
6702ed49 | 689 | if (i > 0) { |
6b80053d CM |
690 | other = btrfs_node_blockptr(parent, i - 1); |
691 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 692 | } |
5dfe2be7 | 693 | if (!close && i < end_slot) { |
6b80053d CM |
694 | other = btrfs_node_blockptr(parent, i + 1); |
695 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 696 | } |
e9d0b13b CM |
697 | if (close) { |
698 | last_block = blocknr; | |
6702ed49 | 699 | continue; |
e9d0b13b | 700 | } |
6702ed49 | 701 | |
206983b7 JB |
702 | cur = btrfs_read_node_slot(parent, i); |
703 | if (IS_ERR(cur)) | |
704 | return PTR_ERR(cur); | |
e9d0b13b | 705 | if (search_start == 0) |
6b80053d | 706 | search_start = last_block; |
e9d0b13b | 707 | |
e7a84565 | 708 | btrfs_tree_lock(cur); |
6b80053d | 709 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 710 | &cur, search_start, |
6b80053d | 711 | min(16 * blocksize, |
9631e4cc JB |
712 | (end_slot - i) * blocksize), |
713 | BTRFS_NESTING_COW); | |
252c38f0 | 714 | if (err) { |
e7a84565 | 715 | btrfs_tree_unlock(cur); |
6b80053d | 716 | free_extent_buffer(cur); |
6702ed49 | 717 | break; |
252c38f0 | 718 | } |
e7a84565 CM |
719 | search_start = cur->start; |
720 | last_block = cur->start; | |
f2183bde | 721 | *last_ret = search_start; |
e7a84565 CM |
722 | btrfs_tree_unlock(cur); |
723 | free_extent_buffer(cur); | |
6702ed49 CM |
724 | } |
725 | return err; | |
726 | } | |
727 | ||
74123bd7 | 728 | /* |
5f39d397 | 729 | * search for key in the extent_buffer. The items start at offset p, |
67d5e289 | 730 | * and they are item_size apart. |
5f39d397 | 731 | * |
74123bd7 CM |
732 | * the slot in the array is returned via slot, and it points to |
733 | * the place where you would insert key if it is not found in | |
734 | * the array. | |
735 | * | |
67d5e289 MPS |
736 | * Slot may point to total number of items if the key is bigger than |
737 | * all of the keys | |
74123bd7 | 738 | */ |
e02119d5 | 739 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 | 740 | unsigned long p, int item_size, |
67d5e289 | 741 | const struct btrfs_key *key, int *slot) |
be0e5c09 CM |
742 | { |
743 | int low = 0; | |
67d5e289 | 744 | int high = btrfs_header_nritems(eb); |
be0e5c09 | 745 | int ret; |
5cd17f34 | 746 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 747 | |
5e24e9af LB |
748 | if (low > high) { |
749 | btrfs_err(eb->fs_info, | |
750 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
751 | __func__, low, high, eb->start, | |
752 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
753 | return -EINVAL; | |
754 | } | |
755 | ||
d397712b | 756 | while (low < high) { |
5cd17f34 DS |
757 | unsigned long oip; |
758 | unsigned long offset; | |
759 | struct btrfs_disk_key *tmp; | |
760 | struct btrfs_disk_key unaligned; | |
761 | int mid; | |
762 | ||
be0e5c09 | 763 | mid = (low + high) / 2; |
5f39d397 | 764 | offset = p + mid * item_size; |
5cd17f34 | 765 | oip = offset_in_page(offset); |
5f39d397 | 766 | |
5cd17f34 | 767 | if (oip + key_size <= PAGE_SIZE) { |
884b07d0 | 768 | const unsigned long idx = get_eb_page_index(offset); |
5cd17f34 | 769 | char *kaddr = page_address(eb->pages[idx]); |
5f39d397 | 770 | |
884b07d0 | 771 | oip = get_eb_offset_in_page(eb, offset); |
5cd17f34 | 772 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 773 | } else { |
5cd17f34 DS |
774 | read_extent_buffer(eb, &unaligned, offset, key_size); |
775 | tmp = &unaligned; | |
5f39d397 | 776 | } |
5cd17f34 | 777 | |
be0e5c09 CM |
778 | ret = comp_keys(tmp, key); |
779 | ||
780 | if (ret < 0) | |
781 | low = mid + 1; | |
782 | else if (ret > 0) | |
783 | high = mid; | |
784 | else { | |
785 | *slot = mid; | |
786 | return 0; | |
787 | } | |
788 | } | |
789 | *slot = low; | |
790 | return 1; | |
791 | } | |
792 | ||
97571fd0 CM |
793 | /* |
794 | * simple bin_search frontend that does the right thing for | |
795 | * leaves vs nodes | |
796 | */ | |
a74b35ec | 797 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
e3b83361 | 798 | int *slot) |
be0e5c09 | 799 | { |
e3b83361 | 800 | if (btrfs_header_level(eb) == 0) |
5f39d397 CM |
801 | return generic_bin_search(eb, |
802 | offsetof(struct btrfs_leaf, items), | |
67d5e289 | 803 | sizeof(struct btrfs_item), key, slot); |
f775738f | 804 | else |
5f39d397 CM |
805 | return generic_bin_search(eb, |
806 | offsetof(struct btrfs_node, ptrs), | |
67d5e289 | 807 | sizeof(struct btrfs_key_ptr), key, slot); |
be0e5c09 CM |
808 | } |
809 | ||
f0486c68 YZ |
810 | static void root_add_used(struct btrfs_root *root, u32 size) |
811 | { | |
812 | spin_lock(&root->accounting_lock); | |
813 | btrfs_set_root_used(&root->root_item, | |
814 | btrfs_root_used(&root->root_item) + size); | |
815 | spin_unlock(&root->accounting_lock); | |
816 | } | |
817 | ||
818 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
819 | { | |
820 | spin_lock(&root->accounting_lock); | |
821 | btrfs_set_root_used(&root->root_item, | |
822 | btrfs_root_used(&root->root_item) - size); | |
823 | spin_unlock(&root->accounting_lock); | |
824 | } | |
825 | ||
d352ac68 CM |
826 | /* given a node and slot number, this reads the blocks it points to. The |
827 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 828 | */ |
4b231ae4 DS |
829 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
830 | int slot) | |
bb803951 | 831 | { |
ca7a79ad | 832 | int level = btrfs_header_level(parent); |
416bc658 | 833 | struct extent_buffer *eb; |
581c1760 | 834 | struct btrfs_key first_key; |
416bc658 | 835 | |
fb770ae4 LB |
836 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
837 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
838 | |
839 | BUG_ON(level == 0); | |
840 | ||
581c1760 | 841 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 842 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
1b7ec85e | 843 | btrfs_header_owner(parent), |
581c1760 QW |
844 | btrfs_node_ptr_generation(parent, slot), |
845 | level - 1, &first_key); | |
fb770ae4 LB |
846 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
847 | free_extent_buffer(eb); | |
848 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
849 | } |
850 | ||
851 | return eb; | |
bb803951 CM |
852 | } |
853 | ||
d352ac68 CM |
854 | /* |
855 | * node level balancing, used to make sure nodes are in proper order for | |
856 | * item deletion. We balance from the top down, so we have to make sure | |
857 | * that a deletion won't leave an node completely empty later on. | |
858 | */ | |
e02119d5 | 859 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
860 | struct btrfs_root *root, |
861 | struct btrfs_path *path, int level) | |
bb803951 | 862 | { |
0b246afa | 863 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
864 | struct extent_buffer *right = NULL; |
865 | struct extent_buffer *mid; | |
866 | struct extent_buffer *left = NULL; | |
867 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
868 | int ret = 0; |
869 | int wret; | |
870 | int pslot; | |
bb803951 | 871 | int orig_slot = path->slots[level]; |
79f95c82 | 872 | u64 orig_ptr; |
bb803951 | 873 | |
98e6b1eb | 874 | ASSERT(level > 0); |
bb803951 | 875 | |
5f39d397 | 876 | mid = path->nodes[level]; |
b4ce94de | 877 | |
ac5887c8 | 878 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); |
7bb86316 CM |
879 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
880 | ||
1d4f8a0c | 881 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 882 | |
a05a9bb1 | 883 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 884 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
885 | pslot = path->slots[level + 1]; |
886 | } | |
bb803951 | 887 | |
40689478 CM |
888 | /* |
889 | * deal with the case where there is only one pointer in the root | |
890 | * by promoting the node below to a root | |
891 | */ | |
5f39d397 CM |
892 | if (!parent) { |
893 | struct extent_buffer *child; | |
bb803951 | 894 | |
5f39d397 | 895 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
896 | return 0; |
897 | ||
898 | /* promote the child to a root */ | |
4b231ae4 | 899 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
900 | if (IS_ERR(child)) { |
901 | ret = PTR_ERR(child); | |
0b246afa | 902 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
903 | goto enospc; |
904 | } | |
905 | ||
925baedd | 906 | btrfs_tree_lock(child); |
9631e4cc JB |
907 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, |
908 | BTRFS_NESTING_COW); | |
f0486c68 YZ |
909 | if (ret) { |
910 | btrfs_tree_unlock(child); | |
911 | free_extent_buffer(child); | |
912 | goto enospc; | |
913 | } | |
2f375ab9 | 914 | |
406808ab | 915 | ret = btrfs_tree_mod_log_insert_root(root->node, child, true); |
d9d19a01 | 916 | BUG_ON(ret < 0); |
240f62c8 | 917 | rcu_assign_pointer(root->node, child); |
925baedd | 918 | |
0b86a832 | 919 | add_root_to_dirty_list(root); |
925baedd | 920 | btrfs_tree_unlock(child); |
b4ce94de | 921 | |
925baedd | 922 | path->locks[level] = 0; |
bb803951 | 923 | path->nodes[level] = NULL; |
6a884d7d | 924 | btrfs_clean_tree_block(mid); |
925baedd | 925 | btrfs_tree_unlock(mid); |
bb803951 | 926 | /* once for the path */ |
5f39d397 | 927 | free_extent_buffer(mid); |
f0486c68 YZ |
928 | |
929 | root_sub_used(root, mid->len); | |
7a163608 | 930 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
bb803951 | 931 | /* once for the root ptr */ |
3083ee2e | 932 | free_extent_buffer_stale(mid); |
f0486c68 | 933 | return 0; |
bb803951 | 934 | } |
5f39d397 | 935 | if (btrfs_header_nritems(mid) > |
0b246afa | 936 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
937 | return 0; |
938 | ||
4b231ae4 | 939 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
940 | if (IS_ERR(left)) |
941 | left = NULL; | |
942 | ||
5f39d397 | 943 | if (left) { |
bf77467a | 944 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
5f39d397 | 945 | wret = btrfs_cow_block(trans, root, left, |
9631e4cc | 946 | parent, pslot - 1, &left, |
bf59a5a2 | 947 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
948 | if (wret) { |
949 | ret = wret; | |
950 | goto enospc; | |
951 | } | |
2cc58cf2 | 952 | } |
fb770ae4 | 953 | |
4b231ae4 | 954 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
955 | if (IS_ERR(right)) |
956 | right = NULL; | |
957 | ||
5f39d397 | 958 | if (right) { |
bf77467a | 959 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
5f39d397 | 960 | wret = btrfs_cow_block(trans, root, right, |
9631e4cc | 961 | parent, pslot + 1, &right, |
bf59a5a2 | 962 | BTRFS_NESTING_RIGHT_COW); |
2cc58cf2 CM |
963 | if (wret) { |
964 | ret = wret; | |
965 | goto enospc; | |
966 | } | |
967 | } | |
968 | ||
969 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
970 | if (left) { |
971 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 972 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
973 | if (wret < 0) |
974 | ret = wret; | |
bb803951 | 975 | } |
79f95c82 CM |
976 | |
977 | /* | |
978 | * then try to empty the right most buffer into the middle | |
979 | */ | |
5f39d397 | 980 | if (right) { |
d30a668f | 981 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 982 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 983 | ret = wret; |
5f39d397 | 984 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 985 | btrfs_clean_tree_block(right); |
925baedd | 986 | btrfs_tree_unlock(right); |
afe5fea7 | 987 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 988 | root_sub_used(root, right->len); |
7a163608 FM |
989 | btrfs_free_tree_block(trans, btrfs_root_id(root), right, |
990 | 0, 1); | |
3083ee2e | 991 | free_extent_buffer_stale(right); |
f0486c68 | 992 | right = NULL; |
bb803951 | 993 | } else { |
5f39d397 CM |
994 | struct btrfs_disk_key right_key; |
995 | btrfs_node_key(right, &right_key, 0); | |
f3a84ccd FM |
996 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
997 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 998 | BUG_ON(ret < 0); |
5f39d397 CM |
999 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1000 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1001 | } |
1002 | } | |
5f39d397 | 1003 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1004 | /* |
1005 | * we're not allowed to leave a node with one item in the | |
1006 | * tree during a delete. A deletion from lower in the tree | |
1007 | * could try to delete the only pointer in this node. | |
1008 | * So, pull some keys from the left. | |
1009 | * There has to be a left pointer at this point because | |
1010 | * otherwise we would have pulled some pointers from the | |
1011 | * right | |
1012 | */ | |
305a26af MF |
1013 | if (!left) { |
1014 | ret = -EROFS; | |
0b246afa | 1015 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1016 | goto enospc; |
1017 | } | |
55d32ed8 | 1018 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1019 | if (wret < 0) { |
79f95c82 | 1020 | ret = wret; |
54aa1f4d CM |
1021 | goto enospc; |
1022 | } | |
bce4eae9 | 1023 | if (wret == 1) { |
d30a668f | 1024 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1025 | if (wret < 0) |
1026 | ret = wret; | |
1027 | } | |
79f95c82 CM |
1028 | BUG_ON(wret == 1); |
1029 | } | |
5f39d397 | 1030 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1031 | btrfs_clean_tree_block(mid); |
925baedd | 1032 | btrfs_tree_unlock(mid); |
afe5fea7 | 1033 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1034 | root_sub_used(root, mid->len); |
7a163608 | 1035 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
3083ee2e | 1036 | free_extent_buffer_stale(mid); |
f0486c68 | 1037 | mid = NULL; |
79f95c82 CM |
1038 | } else { |
1039 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1040 | struct btrfs_disk_key mid_key; |
1041 | btrfs_node_key(mid, &mid_key, 0); | |
f3a84ccd FM |
1042 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1043 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1044 | BUG_ON(ret < 0); |
5f39d397 CM |
1045 | btrfs_set_node_key(parent, &mid_key, pslot); |
1046 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1047 | } |
bb803951 | 1048 | |
79f95c82 | 1049 | /* update the path */ |
5f39d397 CM |
1050 | if (left) { |
1051 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1052 | atomic_inc(&left->refs); |
925baedd | 1053 | /* left was locked after cow */ |
5f39d397 | 1054 | path->nodes[level] = left; |
bb803951 CM |
1055 | path->slots[level + 1] -= 1; |
1056 | path->slots[level] = orig_slot; | |
925baedd CM |
1057 | if (mid) { |
1058 | btrfs_tree_unlock(mid); | |
5f39d397 | 1059 | free_extent_buffer(mid); |
925baedd | 1060 | } |
bb803951 | 1061 | } else { |
5f39d397 | 1062 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1063 | path->slots[level] = orig_slot; |
1064 | } | |
1065 | } | |
79f95c82 | 1066 | /* double check we haven't messed things up */ |
e20d96d6 | 1067 | if (orig_ptr != |
5f39d397 | 1068 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1069 | BUG(); |
54aa1f4d | 1070 | enospc: |
925baedd CM |
1071 | if (right) { |
1072 | btrfs_tree_unlock(right); | |
5f39d397 | 1073 | free_extent_buffer(right); |
925baedd CM |
1074 | } |
1075 | if (left) { | |
1076 | if (path->nodes[level] != left) | |
1077 | btrfs_tree_unlock(left); | |
5f39d397 | 1078 | free_extent_buffer(left); |
925baedd | 1079 | } |
bb803951 CM |
1080 | return ret; |
1081 | } | |
1082 | ||
d352ac68 CM |
1083 | /* Node balancing for insertion. Here we only split or push nodes around |
1084 | * when they are completely full. This is also done top down, so we | |
1085 | * have to be pessimistic. | |
1086 | */ | |
d397712b | 1087 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1088 | struct btrfs_root *root, |
1089 | struct btrfs_path *path, int level) | |
e66f709b | 1090 | { |
0b246afa | 1091 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1092 | struct extent_buffer *right = NULL; |
1093 | struct extent_buffer *mid; | |
1094 | struct extent_buffer *left = NULL; | |
1095 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1096 | int ret = 0; |
1097 | int wret; | |
1098 | int pslot; | |
1099 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1100 | |
1101 | if (level == 0) | |
1102 | return 1; | |
1103 | ||
5f39d397 | 1104 | mid = path->nodes[level]; |
7bb86316 | 1105 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1106 | |
a05a9bb1 | 1107 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1108 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1109 | pslot = path->slots[level + 1]; |
1110 | } | |
e66f709b | 1111 | |
5f39d397 | 1112 | if (!parent) |
e66f709b | 1113 | return 1; |
e66f709b | 1114 | |
4b231ae4 | 1115 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1116 | if (IS_ERR(left)) |
1117 | left = NULL; | |
e66f709b CM |
1118 | |
1119 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1120 | if (left) { |
e66f709b | 1121 | u32 left_nr; |
925baedd | 1122 | |
bf77467a | 1123 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
b4ce94de | 1124 | |
5f39d397 | 1125 | left_nr = btrfs_header_nritems(left); |
0b246afa | 1126 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1127 | wret = 1; |
1128 | } else { | |
5f39d397 | 1129 | ret = btrfs_cow_block(trans, root, left, parent, |
9631e4cc | 1130 | pslot - 1, &left, |
bf59a5a2 | 1131 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
1132 | if (ret) |
1133 | wret = 1; | |
1134 | else { | |
d30a668f | 1135 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 1136 | } |
33ade1f8 | 1137 | } |
e66f709b CM |
1138 | if (wret < 0) |
1139 | ret = wret; | |
1140 | if (wret == 0) { | |
5f39d397 | 1141 | struct btrfs_disk_key disk_key; |
e66f709b | 1142 | orig_slot += left_nr; |
5f39d397 | 1143 | btrfs_node_key(mid, &disk_key, 0); |
f3a84ccd FM |
1144 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1145 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1146 | BUG_ON(ret < 0); |
5f39d397 CM |
1147 | btrfs_set_node_key(parent, &disk_key, pslot); |
1148 | btrfs_mark_buffer_dirty(parent); | |
1149 | if (btrfs_header_nritems(left) > orig_slot) { | |
1150 | path->nodes[level] = left; | |
e66f709b CM |
1151 | path->slots[level + 1] -= 1; |
1152 | path->slots[level] = orig_slot; | |
925baedd | 1153 | btrfs_tree_unlock(mid); |
5f39d397 | 1154 | free_extent_buffer(mid); |
e66f709b CM |
1155 | } else { |
1156 | orig_slot -= | |
5f39d397 | 1157 | btrfs_header_nritems(left); |
e66f709b | 1158 | path->slots[level] = orig_slot; |
925baedd | 1159 | btrfs_tree_unlock(left); |
5f39d397 | 1160 | free_extent_buffer(left); |
e66f709b | 1161 | } |
e66f709b CM |
1162 | return 0; |
1163 | } | |
925baedd | 1164 | btrfs_tree_unlock(left); |
5f39d397 | 1165 | free_extent_buffer(left); |
e66f709b | 1166 | } |
4b231ae4 | 1167 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1168 | if (IS_ERR(right)) |
1169 | right = NULL; | |
e66f709b CM |
1170 | |
1171 | /* | |
1172 | * then try to empty the right most buffer into the middle | |
1173 | */ | |
5f39d397 | 1174 | if (right) { |
33ade1f8 | 1175 | u32 right_nr; |
b4ce94de | 1176 | |
bf77467a | 1177 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
b4ce94de | 1178 | |
5f39d397 | 1179 | right_nr = btrfs_header_nritems(right); |
0b246afa | 1180 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1181 | wret = 1; |
1182 | } else { | |
5f39d397 CM |
1183 | ret = btrfs_cow_block(trans, root, right, |
1184 | parent, pslot + 1, | |
bf59a5a2 | 1185 | &right, BTRFS_NESTING_RIGHT_COW); |
54aa1f4d CM |
1186 | if (ret) |
1187 | wret = 1; | |
1188 | else { | |
55d32ed8 | 1189 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 1190 | } |
33ade1f8 | 1191 | } |
e66f709b CM |
1192 | if (wret < 0) |
1193 | ret = wret; | |
1194 | if (wret == 0) { | |
5f39d397 CM |
1195 | struct btrfs_disk_key disk_key; |
1196 | ||
1197 | btrfs_node_key(right, &disk_key, 0); | |
f3a84ccd FM |
1198 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
1199 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1200 | BUG_ON(ret < 0); |
5f39d397 CM |
1201 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
1202 | btrfs_mark_buffer_dirty(parent); | |
1203 | ||
1204 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1205 | path->nodes[level] = right; | |
e66f709b CM |
1206 | path->slots[level + 1] += 1; |
1207 | path->slots[level] = orig_slot - | |
5f39d397 | 1208 | btrfs_header_nritems(mid); |
925baedd | 1209 | btrfs_tree_unlock(mid); |
5f39d397 | 1210 | free_extent_buffer(mid); |
e66f709b | 1211 | } else { |
925baedd | 1212 | btrfs_tree_unlock(right); |
5f39d397 | 1213 | free_extent_buffer(right); |
e66f709b | 1214 | } |
e66f709b CM |
1215 | return 0; |
1216 | } | |
925baedd | 1217 | btrfs_tree_unlock(right); |
5f39d397 | 1218 | free_extent_buffer(right); |
e66f709b | 1219 | } |
e66f709b CM |
1220 | return 1; |
1221 | } | |
1222 | ||
3c69faec | 1223 | /* |
d352ac68 CM |
1224 | * readahead one full node of leaves, finding things that are close |
1225 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1226 | */ |
2ff7e61e | 1227 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
1228 | struct btrfs_path *path, |
1229 | int level, int slot, u64 objectid) | |
3c69faec | 1230 | { |
5f39d397 | 1231 | struct extent_buffer *node; |
01f46658 | 1232 | struct btrfs_disk_key disk_key; |
3c69faec | 1233 | u32 nritems; |
3c69faec | 1234 | u64 search; |
a7175319 | 1235 | u64 target; |
6b80053d | 1236 | u64 nread = 0; |
ace75066 | 1237 | u64 nread_max; |
6b80053d CM |
1238 | u32 nr; |
1239 | u32 blocksize; | |
1240 | u32 nscan = 0; | |
db94535d | 1241 | |
ace75066 | 1242 | if (level != 1 && path->reada != READA_FORWARD_ALWAYS) |
6702ed49 CM |
1243 | return; |
1244 | ||
1245 | if (!path->nodes[level]) | |
3c69faec CM |
1246 | return; |
1247 | ||
5f39d397 | 1248 | node = path->nodes[level]; |
925baedd | 1249 | |
ace75066 FM |
1250 | /* |
1251 | * Since the time between visiting leaves is much shorter than the time | |
1252 | * between visiting nodes, limit read ahead of nodes to 1, to avoid too | |
1253 | * much IO at once (possibly random). | |
1254 | */ | |
1255 | if (path->reada == READA_FORWARD_ALWAYS) { | |
1256 | if (level > 1) | |
1257 | nread_max = node->fs_info->nodesize; | |
1258 | else | |
1259 | nread_max = SZ_128K; | |
1260 | } else { | |
1261 | nread_max = SZ_64K; | |
1262 | } | |
1263 | ||
3c69faec | 1264 | search = btrfs_node_blockptr(node, slot); |
0b246afa | 1265 | blocksize = fs_info->nodesize; |
069a2e37 FM |
1266 | if (path->reada != READA_FORWARD_ALWAYS) { |
1267 | struct extent_buffer *eb; | |
1268 | ||
1269 | eb = find_extent_buffer(fs_info, search); | |
1270 | if (eb) { | |
1271 | free_extent_buffer(eb); | |
1272 | return; | |
1273 | } | |
3c69faec CM |
1274 | } |
1275 | ||
a7175319 | 1276 | target = search; |
6b80053d | 1277 | |
5f39d397 | 1278 | nritems = btrfs_header_nritems(node); |
6b80053d | 1279 | nr = slot; |
25b8b936 | 1280 | |
d397712b | 1281 | while (1) { |
e4058b54 | 1282 | if (path->reada == READA_BACK) { |
6b80053d CM |
1283 | if (nr == 0) |
1284 | break; | |
1285 | nr--; | |
ace75066 FM |
1286 | } else if (path->reada == READA_FORWARD || |
1287 | path->reada == READA_FORWARD_ALWAYS) { | |
6b80053d CM |
1288 | nr++; |
1289 | if (nr >= nritems) | |
1290 | break; | |
3c69faec | 1291 | } |
e4058b54 | 1292 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
1293 | btrfs_node_key(node, &disk_key, nr); |
1294 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1295 | break; | |
1296 | } | |
6b80053d | 1297 | search = btrfs_node_blockptr(node, nr); |
ace75066 FM |
1298 | if (path->reada == READA_FORWARD_ALWAYS || |
1299 | (search <= target && target - search <= 65536) || | |
a7175319 | 1300 | (search > target && search - target <= 65536)) { |
bfb484d9 | 1301 | btrfs_readahead_node_child(node, nr); |
6b80053d CM |
1302 | nread += blocksize; |
1303 | } | |
1304 | nscan++; | |
ace75066 | 1305 | if (nread > nread_max || nscan > 32) |
6b80053d | 1306 | break; |
3c69faec CM |
1307 | } |
1308 | } | |
925baedd | 1309 | |
bfb484d9 | 1310 | static noinline void reada_for_balance(struct btrfs_path *path, int level) |
b4ce94de | 1311 | { |
bfb484d9 | 1312 | struct extent_buffer *parent; |
b4ce94de CM |
1313 | int slot; |
1314 | int nritems; | |
b4ce94de | 1315 | |
8c594ea8 | 1316 | parent = path->nodes[level + 1]; |
b4ce94de | 1317 | if (!parent) |
0b08851f | 1318 | return; |
b4ce94de CM |
1319 | |
1320 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1321 | slot = path->slots[level + 1]; |
b4ce94de | 1322 | |
bfb484d9 JB |
1323 | if (slot > 0) |
1324 | btrfs_readahead_node_child(parent, slot - 1); | |
1325 | if (slot + 1 < nritems) | |
1326 | btrfs_readahead_node_child(parent, slot + 1); | |
b4ce94de CM |
1327 | } |
1328 | ||
1329 | ||
d352ac68 | 1330 | /* |
d397712b CM |
1331 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1332 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1333 | * operations on the tree might require changing key pointers higher up in the | |
1334 | * tree. | |
d352ac68 | 1335 | * |
d397712b CM |
1336 | * callers might also have set path->keep_locks, which tells this code to keep |
1337 | * the lock if the path points to the last slot in the block. This is part of | |
1338 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1339 | * |
d397712b CM |
1340 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1341 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1342 | */ |
e02119d5 | 1343 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1344 | int lowest_unlock, int min_write_lock_level, |
1345 | int *write_lock_level) | |
925baedd CM |
1346 | { |
1347 | int i; | |
1348 | int skip_level = level; | |
051e1b9f | 1349 | int no_skips = 0; |
925baedd CM |
1350 | struct extent_buffer *t; |
1351 | ||
1352 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1353 | if (!path->nodes[i]) | |
1354 | break; | |
1355 | if (!path->locks[i]) | |
1356 | break; | |
051e1b9f | 1357 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
1358 | skip_level = i + 1; |
1359 | continue; | |
1360 | } | |
051e1b9f | 1361 | if (!no_skips && path->keep_locks) { |
925baedd CM |
1362 | u32 nritems; |
1363 | t = path->nodes[i]; | |
1364 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 1365 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
1366 | skip_level = i + 1; |
1367 | continue; | |
1368 | } | |
1369 | } | |
051e1b9f CM |
1370 | if (skip_level < i && i >= lowest_unlock) |
1371 | no_skips = 1; | |
1372 | ||
925baedd | 1373 | t = path->nodes[i]; |
d80bb3f9 | 1374 | if (i >= lowest_unlock && i > skip_level) { |
bd681513 | 1375 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 1376 | path->locks[i] = 0; |
f7c79f30 CM |
1377 | if (write_lock_level && |
1378 | i > min_write_lock_level && | |
1379 | i <= *write_lock_level) { | |
1380 | *write_lock_level = i - 1; | |
1381 | } | |
925baedd CM |
1382 | } |
1383 | } | |
1384 | } | |
1385 | ||
c8c42864 CM |
1386 | /* |
1387 | * helper function for btrfs_search_slot. The goal is to find a block | |
1388 | * in cache without setting the path to blocking. If we find the block | |
1389 | * we return zero and the path is unchanged. | |
1390 | * | |
1391 | * If we can't find the block, we set the path blocking and do some | |
1392 | * reada. -EAGAIN is returned and the search must be repeated. | |
1393 | */ | |
1394 | static int | |
d07b8528 LB |
1395 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
1396 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 1397 | const struct btrfs_key *key) |
c8c42864 | 1398 | { |
0b246afa | 1399 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
1400 | u64 blocknr; |
1401 | u64 gen; | |
c8c42864 | 1402 | struct extent_buffer *tmp; |
581c1760 | 1403 | struct btrfs_key first_key; |
76a05b35 | 1404 | int ret; |
581c1760 | 1405 | int parent_level; |
c8c42864 | 1406 | |
213ff4b7 NB |
1407 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
1408 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
1409 | parent_level = btrfs_header_level(*eb_ret); | |
1410 | btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); | |
c8c42864 | 1411 | |
0b246afa | 1412 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 1413 | if (tmp) { |
ace75066 FM |
1414 | if (p->reada == READA_FORWARD_ALWAYS) |
1415 | reada_for_search(fs_info, p, level, slot, key->objectid); | |
1416 | ||
b9fab919 | 1417 | /* first we do an atomic uptodate check */ |
bdf7c00e | 1418 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
1419 | /* |
1420 | * Do extra check for first_key, eb can be stale due to | |
1421 | * being cached, read from scrub, or have multiple | |
1422 | * parents (shared tree blocks). | |
1423 | */ | |
e064d5e9 | 1424 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
1425 | parent_level - 1, &first_key, gen)) { |
1426 | free_extent_buffer(tmp); | |
1427 | return -EUCLEAN; | |
1428 | } | |
bdf7c00e JB |
1429 | *eb_ret = tmp; |
1430 | return 0; | |
1431 | } | |
1432 | ||
bdf7c00e | 1433 | /* now we're allowed to do a blocking uptodate check */ |
581c1760 | 1434 | ret = btrfs_read_buffer(tmp, gen, parent_level - 1, &first_key); |
bdf7c00e JB |
1435 | if (!ret) { |
1436 | *eb_ret = tmp; | |
1437 | return 0; | |
cb44921a | 1438 | } |
bdf7c00e JB |
1439 | free_extent_buffer(tmp); |
1440 | btrfs_release_path(p); | |
1441 | return -EIO; | |
c8c42864 CM |
1442 | } |
1443 | ||
1444 | /* | |
1445 | * reduce lock contention at high levels | |
1446 | * of the btree by dropping locks before | |
76a05b35 CM |
1447 | * we read. Don't release the lock on the current |
1448 | * level because we need to walk this node to figure | |
1449 | * out which blocks to read. | |
c8c42864 | 1450 | */ |
8c594ea8 | 1451 | btrfs_unlock_up_safe(p, level + 1); |
8c594ea8 | 1452 | |
e4058b54 | 1453 | if (p->reada != READA_NONE) |
2ff7e61e | 1454 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 1455 | |
76a05b35 | 1456 | ret = -EAGAIN; |
1b7ec85e JB |
1457 | tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, |
1458 | gen, parent_level - 1, &first_key); | |
64c043de | 1459 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
1460 | /* |
1461 | * If the read above didn't mark this buffer up to date, | |
1462 | * it will never end up being up to date. Set ret to EIO now | |
1463 | * and give up so that our caller doesn't loop forever | |
1464 | * on our EAGAINs. | |
1465 | */ | |
e6a1d6fd | 1466 | if (!extent_buffer_uptodate(tmp)) |
76a05b35 | 1467 | ret = -EIO; |
c8c42864 | 1468 | free_extent_buffer(tmp); |
c871b0f2 LB |
1469 | } else { |
1470 | ret = PTR_ERR(tmp); | |
76a05b35 | 1471 | } |
02a3307a LB |
1472 | |
1473 | btrfs_release_path(p); | |
76a05b35 | 1474 | return ret; |
c8c42864 CM |
1475 | } |
1476 | ||
1477 | /* | |
1478 | * helper function for btrfs_search_slot. This does all of the checks | |
1479 | * for node-level blocks and does any balancing required based on | |
1480 | * the ins_len. | |
1481 | * | |
1482 | * If no extra work was required, zero is returned. If we had to | |
1483 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1484 | * start over | |
1485 | */ | |
1486 | static int | |
1487 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1488 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1489 | struct extent_buffer *b, int level, int ins_len, |
1490 | int *write_lock_level) | |
c8c42864 | 1491 | { |
0b246afa | 1492 | struct btrfs_fs_info *fs_info = root->fs_info; |
95b982de | 1493 | int ret = 0; |
0b246afa | 1494 | |
c8c42864 | 1495 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 1496 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 | 1497 | |
bd681513 CM |
1498 | if (*write_lock_level < level + 1) { |
1499 | *write_lock_level = level + 1; | |
1500 | btrfs_release_path(p); | |
95b982de | 1501 | return -EAGAIN; |
bd681513 CM |
1502 | } |
1503 | ||
bfb484d9 | 1504 | reada_for_balance(p, level); |
95b982de | 1505 | ret = split_node(trans, root, p, level); |
c8c42864 | 1506 | |
c8c42864 CM |
1507 | b = p->nodes[level]; |
1508 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 1509 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 | 1510 | |
bd681513 CM |
1511 | if (*write_lock_level < level + 1) { |
1512 | *write_lock_level = level + 1; | |
1513 | btrfs_release_path(p); | |
95b982de | 1514 | return -EAGAIN; |
bd681513 CM |
1515 | } |
1516 | ||
bfb484d9 | 1517 | reada_for_balance(p, level); |
95b982de NB |
1518 | ret = balance_level(trans, root, p, level); |
1519 | if (ret) | |
1520 | return ret; | |
c8c42864 | 1521 | |
c8c42864 CM |
1522 | b = p->nodes[level]; |
1523 | if (!b) { | |
b3b4aa74 | 1524 | btrfs_release_path(p); |
95b982de | 1525 | return -EAGAIN; |
c8c42864 CM |
1526 | } |
1527 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1528 | } | |
c8c42864 CM |
1529 | return ret; |
1530 | } | |
1531 | ||
381cf658 | 1532 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
1533 | u64 iobjectid, u64 ioff, u8 key_type, |
1534 | struct btrfs_key *found_key) | |
1535 | { | |
1536 | int ret; | |
1537 | struct btrfs_key key; | |
1538 | struct extent_buffer *eb; | |
381cf658 DS |
1539 | |
1540 | ASSERT(path); | |
1d4c08e0 | 1541 | ASSERT(found_key); |
e33d5c3d KN |
1542 | |
1543 | key.type = key_type; | |
1544 | key.objectid = iobjectid; | |
1545 | key.offset = ioff; | |
1546 | ||
1547 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 1548 | if (ret < 0) |
e33d5c3d KN |
1549 | return ret; |
1550 | ||
1551 | eb = path->nodes[0]; | |
1552 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
1553 | ret = btrfs_next_leaf(fs_root, path); | |
1554 | if (ret) | |
1555 | return ret; | |
1556 | eb = path->nodes[0]; | |
1557 | } | |
1558 | ||
1559 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
1560 | if (found_key->type != key.type || | |
1561 | found_key->objectid != key.objectid) | |
1562 | return 1; | |
1563 | ||
1564 | return 0; | |
1565 | } | |
1566 | ||
1fc28d8e LB |
1567 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
1568 | struct btrfs_path *p, | |
1569 | int write_lock_level) | |
1570 | { | |
1571 | struct btrfs_fs_info *fs_info = root->fs_info; | |
1572 | struct extent_buffer *b; | |
1573 | int root_lock; | |
1574 | int level = 0; | |
1575 | ||
1576 | /* We try very hard to do read locks on the root */ | |
1577 | root_lock = BTRFS_READ_LOCK; | |
1578 | ||
1579 | if (p->search_commit_root) { | |
be6821f8 FM |
1580 | /* |
1581 | * The commit roots are read only so we always do read locks, | |
1582 | * and we always must hold the commit_root_sem when doing | |
1583 | * searches on them, the only exception is send where we don't | |
1584 | * want to block transaction commits for a long time, so | |
1585 | * we need to clone the commit root in order to avoid races | |
1586 | * with transaction commits that create a snapshot of one of | |
1587 | * the roots used by a send operation. | |
1588 | */ | |
1589 | if (p->need_commit_sem) { | |
1fc28d8e | 1590 | down_read(&fs_info->commit_root_sem); |
be6821f8 | 1591 | b = btrfs_clone_extent_buffer(root->commit_root); |
1fc28d8e | 1592 | up_read(&fs_info->commit_root_sem); |
be6821f8 FM |
1593 | if (!b) |
1594 | return ERR_PTR(-ENOMEM); | |
1595 | ||
1596 | } else { | |
1597 | b = root->commit_root; | |
67439dad | 1598 | atomic_inc(&b->refs); |
be6821f8 FM |
1599 | } |
1600 | level = btrfs_header_level(b); | |
f9ddfd05 LB |
1601 | /* |
1602 | * Ensure that all callers have set skip_locking when | |
1603 | * p->search_commit_root = 1. | |
1604 | */ | |
1605 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
1606 | |
1607 | goto out; | |
1608 | } | |
1609 | ||
1610 | if (p->skip_locking) { | |
1611 | b = btrfs_root_node(root); | |
1612 | level = btrfs_header_level(b); | |
1613 | goto out; | |
1614 | } | |
1615 | ||
1616 | /* | |
662c653b LB |
1617 | * If the level is set to maximum, we can skip trying to get the read |
1618 | * lock. | |
1fc28d8e | 1619 | */ |
662c653b LB |
1620 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
1621 | /* | |
1622 | * We don't know the level of the root node until we actually | |
1623 | * have it read locked | |
1624 | */ | |
1bb96598 | 1625 | b = btrfs_read_lock_root_node(root); |
662c653b LB |
1626 | level = btrfs_header_level(b); |
1627 | if (level > write_lock_level) | |
1628 | goto out; | |
1629 | ||
1630 | /* Whoops, must trade for write lock */ | |
1631 | btrfs_tree_read_unlock(b); | |
1632 | free_extent_buffer(b); | |
1633 | } | |
1fc28d8e | 1634 | |
1fc28d8e LB |
1635 | b = btrfs_lock_root_node(root); |
1636 | root_lock = BTRFS_WRITE_LOCK; | |
1637 | ||
1638 | /* The level might have changed, check again */ | |
1639 | level = btrfs_header_level(b); | |
1640 | ||
1641 | out: | |
1642 | p->nodes[level] = b; | |
1643 | if (!p->skip_locking) | |
1644 | p->locks[level] = root_lock; | |
1645 | /* | |
1646 | * Callers are responsible for dropping b's references. | |
1647 | */ | |
1648 | return b; | |
1649 | } | |
1650 | ||
1651 | ||
74123bd7 | 1652 | /* |
4271ecea NB |
1653 | * btrfs_search_slot - look for a key in a tree and perform necessary |
1654 | * modifications to preserve tree invariants. | |
74123bd7 | 1655 | * |
4271ecea NB |
1656 | * @trans: Handle of transaction, used when modifying the tree |
1657 | * @p: Holds all btree nodes along the search path | |
1658 | * @root: The root node of the tree | |
1659 | * @key: The key we are looking for | |
9a664971 | 1660 | * @ins_len: Indicates purpose of search: |
1661 | * >0 for inserts it's size of item inserted (*) | |
1662 | * <0 for deletions | |
1663 | * 0 for plain searches, not modifying the tree | |
1664 | * | |
1665 | * (*) If size of item inserted doesn't include | |
1666 | * sizeof(struct btrfs_item), then p->search_for_extension must | |
1667 | * be set. | |
4271ecea NB |
1668 | * @cow: boolean should CoW operations be performed. Must always be 1 |
1669 | * when modifying the tree. | |
97571fd0 | 1670 | * |
4271ecea NB |
1671 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
1672 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
1673 | * | |
1674 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
1675 | * of the path (level 0) | |
1676 | * | |
1677 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
1678 | * points to the slot where it should be inserted | |
1679 | * | |
1680 | * If an error is encountered while searching the tree a negative error number | |
1681 | * is returned | |
74123bd7 | 1682 | */ |
310712b2 OS |
1683 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1684 | const struct btrfs_key *key, struct btrfs_path *p, | |
1685 | int ins_len, int cow) | |
be0e5c09 | 1686 | { |
5f39d397 | 1687 | struct extent_buffer *b; |
be0e5c09 CM |
1688 | int slot; |
1689 | int ret; | |
33c66f43 | 1690 | int err; |
be0e5c09 | 1691 | int level; |
925baedd | 1692 | int lowest_unlock = 1; |
bd681513 CM |
1693 | /* everything at write_lock_level or lower must be write locked */ |
1694 | int write_lock_level = 0; | |
9f3a7427 | 1695 | u8 lowest_level = 0; |
f7c79f30 | 1696 | int min_write_lock_level; |
d7396f07 | 1697 | int prev_cmp; |
9f3a7427 | 1698 | |
6702ed49 | 1699 | lowest_level = p->lowest_level; |
323ac95b | 1700 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1701 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 1702 | BUG_ON(!cow && ins_len); |
25179201 | 1703 | |
bd681513 | 1704 | if (ins_len < 0) { |
925baedd | 1705 | lowest_unlock = 2; |
65b51a00 | 1706 | |
bd681513 CM |
1707 | /* when we are removing items, we might have to go up to level |
1708 | * two as we update tree pointers Make sure we keep write | |
1709 | * for those levels as well | |
1710 | */ | |
1711 | write_lock_level = 2; | |
1712 | } else if (ins_len > 0) { | |
1713 | /* | |
1714 | * for inserting items, make sure we have a write lock on | |
1715 | * level 1 so we can update keys | |
1716 | */ | |
1717 | write_lock_level = 1; | |
1718 | } | |
1719 | ||
1720 | if (!cow) | |
1721 | write_lock_level = -1; | |
1722 | ||
09a2a8f9 | 1723 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
1724 | write_lock_level = BTRFS_MAX_LEVEL; |
1725 | ||
f7c79f30 CM |
1726 | min_write_lock_level = write_lock_level; |
1727 | ||
bb803951 | 1728 | again: |
d7396f07 | 1729 | prev_cmp = -1; |
1fc28d8e | 1730 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
1731 | if (IS_ERR(b)) { |
1732 | ret = PTR_ERR(b); | |
1733 | goto done; | |
1734 | } | |
925baedd | 1735 | |
eb60ceac | 1736 | while (b) { |
f624d976 QW |
1737 | int dec = 0; |
1738 | ||
5f39d397 | 1739 | level = btrfs_header_level(b); |
65b51a00 | 1740 | |
02217ed2 | 1741 | if (cow) { |
9ea2c7c9 NB |
1742 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
1743 | ||
c8c42864 CM |
1744 | /* |
1745 | * if we don't really need to cow this block | |
1746 | * then we don't want to set the path blocking, | |
1747 | * so we test it here | |
1748 | */ | |
5963ffca | 1749 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 1750 | goto cow_done; |
5d4f98a2 | 1751 | |
bd681513 CM |
1752 | /* |
1753 | * must have write locks on this node and the | |
1754 | * parent | |
1755 | */ | |
5124e00e JB |
1756 | if (level > write_lock_level || |
1757 | (level + 1 > write_lock_level && | |
1758 | level + 1 < BTRFS_MAX_LEVEL && | |
1759 | p->nodes[level + 1])) { | |
bd681513 CM |
1760 | write_lock_level = level + 1; |
1761 | btrfs_release_path(p); | |
1762 | goto again; | |
1763 | } | |
1764 | ||
9ea2c7c9 NB |
1765 | if (last_level) |
1766 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
9631e4cc JB |
1767 | &b, |
1768 | BTRFS_NESTING_COW); | |
9ea2c7c9 NB |
1769 | else |
1770 | err = btrfs_cow_block(trans, root, b, | |
1771 | p->nodes[level + 1], | |
9631e4cc JB |
1772 | p->slots[level + 1], &b, |
1773 | BTRFS_NESTING_COW); | |
33c66f43 | 1774 | if (err) { |
33c66f43 | 1775 | ret = err; |
65b51a00 | 1776 | goto done; |
54aa1f4d | 1777 | } |
02217ed2 | 1778 | } |
65b51a00 | 1779 | cow_done: |
eb60ceac | 1780 | p->nodes[level] = b; |
52398340 LB |
1781 | /* |
1782 | * Leave path with blocking locks to avoid massive | |
1783 | * lock context switch, this is made on purpose. | |
1784 | */ | |
b4ce94de CM |
1785 | |
1786 | /* | |
1787 | * we have a lock on b and as long as we aren't changing | |
1788 | * the tree, there is no way to for the items in b to change. | |
1789 | * It is safe to drop the lock on our parent before we | |
1790 | * go through the expensive btree search on b. | |
1791 | * | |
eb653de1 FDBM |
1792 | * If we're inserting or deleting (ins_len != 0), then we might |
1793 | * be changing slot zero, which may require changing the parent. | |
1794 | * So, we can't drop the lock until after we know which slot | |
1795 | * we're operating on. | |
b4ce94de | 1796 | */ |
eb653de1 FDBM |
1797 | if (!ins_len && !p->keep_locks) { |
1798 | int u = level + 1; | |
1799 | ||
1800 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
1801 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
1802 | p->locks[u] = 0; | |
1803 | } | |
1804 | } | |
b4ce94de | 1805 | |
995e9a16 NB |
1806 | /* |
1807 | * If btrfs_bin_search returns an exact match (prev_cmp == 0) | |
1808 | * we can safely assume the target key will always be in slot 0 | |
1809 | * on lower levels due to the invariants BTRFS' btree provides, | |
1810 | * namely that a btrfs_key_ptr entry always points to the | |
1811 | * lowest key in the child node, thus we can skip searching | |
1812 | * lower levels | |
1813 | */ | |
1814 | if (prev_cmp == 0) { | |
1815 | slot = 0; | |
1816 | ret = 0; | |
1817 | } else { | |
1818 | ret = btrfs_bin_search(b, key, &slot); | |
1819 | prev_cmp = ret; | |
1820 | if (ret < 0) | |
1821 | goto done; | |
1822 | } | |
b4ce94de | 1823 | |
f624d976 | 1824 | if (level == 0) { |
be0e5c09 | 1825 | p->slots[level] = slot; |
9a664971 | 1826 | /* |
1827 | * Item key already exists. In this case, if we are | |
1828 | * allowed to insert the item (for example, in dir_item | |
1829 | * case, item key collision is allowed), it will be | |
1830 | * merged with the original item. Only the item size | |
1831 | * grows, no new btrfs item will be added. If | |
1832 | * search_for_extension is not set, ins_len already | |
1833 | * accounts the size btrfs_item, deduct it here so leaf | |
1834 | * space check will be correct. | |
1835 | */ | |
1836 | if (ret == 0 && ins_len > 0 && !p->search_for_extension) { | |
1837 | ASSERT(ins_len >= sizeof(struct btrfs_item)); | |
1838 | ins_len -= sizeof(struct btrfs_item); | |
1839 | } | |
87b29b20 | 1840 | if (ins_len > 0 && |
e902baac | 1841 | btrfs_leaf_free_space(b) < ins_len) { |
bd681513 CM |
1842 | if (write_lock_level < 1) { |
1843 | write_lock_level = 1; | |
1844 | btrfs_release_path(p); | |
1845 | goto again; | |
1846 | } | |
1847 | ||
33c66f43 YZ |
1848 | err = split_leaf(trans, root, key, |
1849 | p, ins_len, ret == 0); | |
b4ce94de | 1850 | |
33c66f43 YZ |
1851 | BUG_ON(err > 0); |
1852 | if (err) { | |
1853 | ret = err; | |
65b51a00 CM |
1854 | goto done; |
1855 | } | |
5c680ed6 | 1856 | } |
459931ec | 1857 | if (!p->search_for_split) |
f7c79f30 | 1858 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 1859 | min_write_lock_level, NULL); |
65b51a00 | 1860 | goto done; |
be0e5c09 | 1861 | } |
f624d976 QW |
1862 | if (ret && slot > 0) { |
1863 | dec = 1; | |
1864 | slot--; | |
1865 | } | |
1866 | p->slots[level] = slot; | |
1867 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
1868 | &write_lock_level); | |
1869 | if (err == -EAGAIN) | |
1870 | goto again; | |
1871 | if (err) { | |
1872 | ret = err; | |
1873 | goto done; | |
1874 | } | |
1875 | b = p->nodes[level]; | |
1876 | slot = p->slots[level]; | |
1877 | ||
1878 | /* | |
1879 | * Slot 0 is special, if we change the key we have to update | |
1880 | * the parent pointer which means we must have a write lock on | |
1881 | * the parent | |
1882 | */ | |
1883 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
1884 | write_lock_level = level + 1; | |
1885 | btrfs_release_path(p); | |
1886 | goto again; | |
1887 | } | |
1888 | ||
1889 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
1890 | &write_lock_level); | |
1891 | ||
1892 | if (level == lowest_level) { | |
1893 | if (dec) | |
1894 | p->slots[level]++; | |
1895 | goto done; | |
1896 | } | |
1897 | ||
1898 | err = read_block_for_search(root, p, &b, level, slot, key); | |
1899 | if (err == -EAGAIN) | |
1900 | goto again; | |
1901 | if (err) { | |
1902 | ret = err; | |
1903 | goto done; | |
1904 | } | |
1905 | ||
1906 | if (!p->skip_locking) { | |
1907 | level = btrfs_header_level(b); | |
1908 | if (level <= write_lock_level) { | |
ac5887c8 | 1909 | btrfs_tree_lock(b); |
f624d976 QW |
1910 | p->locks[level] = BTRFS_WRITE_LOCK; |
1911 | } else { | |
fe596ca3 | 1912 | btrfs_tree_read_lock(b); |
f624d976 QW |
1913 | p->locks[level] = BTRFS_READ_LOCK; |
1914 | } | |
1915 | p->nodes[level] = b; | |
1916 | } | |
be0e5c09 | 1917 | } |
65b51a00 CM |
1918 | ret = 1; |
1919 | done: | |
5f5bc6b1 | 1920 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 1921 | btrfs_release_path(p); |
65b51a00 | 1922 | return ret; |
be0e5c09 | 1923 | } |
f75e2b79 | 1924 | ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); |
be0e5c09 | 1925 | |
5d9e75c4 JS |
1926 | /* |
1927 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
1928 | * current state of the tree together with the operations recorded in the tree | |
1929 | * modification log to search for the key in a previous version of this tree, as | |
1930 | * denoted by the time_seq parameter. | |
1931 | * | |
1932 | * Naturally, there is no support for insert, delete or cow operations. | |
1933 | * | |
1934 | * The resulting path and return value will be set up as if we called | |
1935 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
1936 | */ | |
310712b2 | 1937 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
1938 | struct btrfs_path *p, u64 time_seq) |
1939 | { | |
0b246afa | 1940 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
1941 | struct extent_buffer *b; |
1942 | int slot; | |
1943 | int ret; | |
1944 | int err; | |
1945 | int level; | |
1946 | int lowest_unlock = 1; | |
1947 | u8 lowest_level = 0; | |
1948 | ||
1949 | lowest_level = p->lowest_level; | |
1950 | WARN_ON(p->nodes[0] != NULL); | |
1951 | ||
1952 | if (p->search_commit_root) { | |
1953 | BUG_ON(time_seq); | |
1954 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
1955 | } | |
1956 | ||
1957 | again: | |
f3a84ccd | 1958 | b = btrfs_get_old_root(root, time_seq); |
315bed43 NB |
1959 | if (!b) { |
1960 | ret = -EIO; | |
1961 | goto done; | |
1962 | } | |
5d9e75c4 | 1963 | level = btrfs_header_level(b); |
5d9e75c4 JS |
1964 | p->locks[level] = BTRFS_READ_LOCK; |
1965 | ||
1966 | while (b) { | |
abe9339d QW |
1967 | int dec = 0; |
1968 | ||
5d9e75c4 JS |
1969 | level = btrfs_header_level(b); |
1970 | p->nodes[level] = b; | |
5d9e75c4 JS |
1971 | |
1972 | /* | |
1973 | * we have a lock on b and as long as we aren't changing | |
1974 | * the tree, there is no way to for the items in b to change. | |
1975 | * It is safe to drop the lock on our parent before we | |
1976 | * go through the expensive btree search on b. | |
1977 | */ | |
1978 | btrfs_unlock_up_safe(p, level + 1); | |
1979 | ||
995e9a16 | 1980 | ret = btrfs_bin_search(b, key, &slot); |
cbca7d59 FM |
1981 | if (ret < 0) |
1982 | goto done; | |
5d9e75c4 | 1983 | |
abe9339d | 1984 | if (level == 0) { |
5d9e75c4 JS |
1985 | p->slots[level] = slot; |
1986 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
1987 | goto done; |
1988 | } | |
5d9e75c4 | 1989 | |
abe9339d QW |
1990 | if (ret && slot > 0) { |
1991 | dec = 1; | |
1992 | slot--; | |
1993 | } | |
1994 | p->slots[level] = slot; | |
1995 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 1996 | |
abe9339d QW |
1997 | if (level == lowest_level) { |
1998 | if (dec) | |
1999 | p->slots[level]++; | |
2000 | goto done; | |
2001 | } | |
5d9e75c4 | 2002 | |
abe9339d QW |
2003 | err = read_block_for_search(root, p, &b, level, slot, key); |
2004 | if (err == -EAGAIN) | |
2005 | goto again; | |
2006 | if (err) { | |
2007 | ret = err; | |
5d9e75c4 JS |
2008 | goto done; |
2009 | } | |
abe9339d QW |
2010 | |
2011 | level = btrfs_header_level(b); | |
ac5887c8 | 2012 | btrfs_tree_read_lock(b); |
f3a84ccd | 2013 | b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); |
abe9339d QW |
2014 | if (!b) { |
2015 | ret = -ENOMEM; | |
2016 | goto done; | |
2017 | } | |
2018 | p->locks[level] = BTRFS_READ_LOCK; | |
2019 | p->nodes[level] = b; | |
5d9e75c4 JS |
2020 | } |
2021 | ret = 1; | |
2022 | done: | |
5d9e75c4 JS |
2023 | if (ret < 0) |
2024 | btrfs_release_path(p); | |
2025 | ||
2026 | return ret; | |
2027 | } | |
2028 | ||
2f38b3e1 AJ |
2029 | /* |
2030 | * helper to use instead of search slot if no exact match is needed but | |
2031 | * instead the next or previous item should be returned. | |
2032 | * When find_higher is true, the next higher item is returned, the next lower | |
2033 | * otherwise. | |
2034 | * When return_any and find_higher are both true, and no higher item is found, | |
2035 | * return the next lower instead. | |
2036 | * When return_any is true and find_higher is false, and no lower item is found, | |
2037 | * return the next higher instead. | |
2038 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2039 | * < 0 on error | |
2040 | */ | |
2041 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2042 | const struct btrfs_key *key, |
2043 | struct btrfs_path *p, int find_higher, | |
2044 | int return_any) | |
2f38b3e1 AJ |
2045 | { |
2046 | int ret; | |
2047 | struct extent_buffer *leaf; | |
2048 | ||
2049 | again: | |
2050 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2051 | if (ret <= 0) | |
2052 | return ret; | |
2053 | /* | |
2054 | * a return value of 1 means the path is at the position where the | |
2055 | * item should be inserted. Normally this is the next bigger item, | |
2056 | * but in case the previous item is the last in a leaf, path points | |
2057 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2058 | * item. | |
2059 | */ | |
2060 | leaf = p->nodes[0]; | |
2061 | ||
2062 | if (find_higher) { | |
2063 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2064 | ret = btrfs_next_leaf(root, p); | |
2065 | if (ret <= 0) | |
2066 | return ret; | |
2067 | if (!return_any) | |
2068 | return 1; | |
2069 | /* | |
2070 | * no higher item found, return the next | |
2071 | * lower instead | |
2072 | */ | |
2073 | return_any = 0; | |
2074 | find_higher = 0; | |
2075 | btrfs_release_path(p); | |
2076 | goto again; | |
2077 | } | |
2078 | } else { | |
e6793769 AJ |
2079 | if (p->slots[0] == 0) { |
2080 | ret = btrfs_prev_leaf(root, p); | |
2081 | if (ret < 0) | |
2082 | return ret; | |
2083 | if (!ret) { | |
23c6bf6a FDBM |
2084 | leaf = p->nodes[0]; |
2085 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
2086 | p->slots[0]--; | |
e6793769 | 2087 | return 0; |
2f38b3e1 | 2088 | } |
e6793769 AJ |
2089 | if (!return_any) |
2090 | return 1; | |
2091 | /* | |
2092 | * no lower item found, return the next | |
2093 | * higher instead | |
2094 | */ | |
2095 | return_any = 0; | |
2096 | find_higher = 1; | |
2097 | btrfs_release_path(p); | |
2098 | goto again; | |
2099 | } else { | |
2f38b3e1 AJ |
2100 | --p->slots[0]; |
2101 | } | |
2102 | } | |
2103 | return 0; | |
2104 | } | |
2105 | ||
0ff40a91 MPS |
2106 | /* |
2107 | * Execute search and call btrfs_previous_item to traverse backwards if the item | |
2108 | * was not found. | |
2109 | * | |
2110 | * Return 0 if found, 1 if not found and < 0 if error. | |
2111 | */ | |
2112 | int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, | |
2113 | struct btrfs_path *path) | |
2114 | { | |
2115 | int ret; | |
2116 | ||
2117 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
2118 | if (ret > 0) | |
2119 | ret = btrfs_previous_item(root, path, key->objectid, key->type); | |
2120 | ||
2121 | if (ret == 0) | |
2122 | btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); | |
2123 | ||
2124 | return ret; | |
2125 | } | |
2126 | ||
74123bd7 CM |
2127 | /* |
2128 | * adjust the pointers going up the tree, starting at level | |
2129 | * making sure the right key of each node is points to 'key'. | |
2130 | * This is used after shifting pointers to the left, so it stops | |
2131 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2132 | * higher levels | |
aa5d6bed | 2133 | * |
74123bd7 | 2134 | */ |
b167fa91 | 2135 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 2136 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2137 | { |
2138 | int i; | |
5f39d397 | 2139 | struct extent_buffer *t; |
0e82bcfe | 2140 | int ret; |
5f39d397 | 2141 | |
234b63a0 | 2142 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2143 | int tslot = path->slots[i]; |
0e82bcfe | 2144 | |
eb60ceac | 2145 | if (!path->nodes[i]) |
be0e5c09 | 2146 | break; |
5f39d397 | 2147 | t = path->nodes[i]; |
f3a84ccd FM |
2148 | ret = btrfs_tree_mod_log_insert_key(t, tslot, |
2149 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_ATOMIC); | |
0e82bcfe | 2150 | BUG_ON(ret < 0); |
5f39d397 | 2151 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2152 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2153 | if (tslot != 0) |
2154 | break; | |
2155 | } | |
2156 | } | |
2157 | ||
31840ae1 ZY |
2158 | /* |
2159 | * update item key. | |
2160 | * | |
2161 | * This function isn't completely safe. It's the caller's responsibility | |
2162 | * that the new key won't break the order | |
2163 | */ | |
b7a0365e DD |
2164 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
2165 | struct btrfs_path *path, | |
310712b2 | 2166 | const struct btrfs_key *new_key) |
31840ae1 ZY |
2167 | { |
2168 | struct btrfs_disk_key disk_key; | |
2169 | struct extent_buffer *eb; | |
2170 | int slot; | |
2171 | ||
2172 | eb = path->nodes[0]; | |
2173 | slot = path->slots[0]; | |
2174 | if (slot > 0) { | |
2175 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
2176 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
2177 | btrfs_crit(fs_info, | |
2178 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2179 | slot, btrfs_disk_key_objectid(&disk_key), | |
2180 | btrfs_disk_key_type(&disk_key), | |
2181 | btrfs_disk_key_offset(&disk_key), | |
2182 | new_key->objectid, new_key->type, | |
2183 | new_key->offset); | |
2184 | btrfs_print_leaf(eb); | |
2185 | BUG(); | |
2186 | } | |
31840ae1 ZY |
2187 | } |
2188 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2189 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
2190 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
2191 | btrfs_crit(fs_info, | |
2192 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2193 | slot, btrfs_disk_key_objectid(&disk_key), | |
2194 | btrfs_disk_key_type(&disk_key), | |
2195 | btrfs_disk_key_offset(&disk_key), | |
2196 | new_key->objectid, new_key->type, | |
2197 | new_key->offset); | |
2198 | btrfs_print_leaf(eb); | |
2199 | BUG(); | |
2200 | } | |
31840ae1 ZY |
2201 | } |
2202 | ||
2203 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2204 | btrfs_set_item_key(eb, &disk_key, slot); | |
2205 | btrfs_mark_buffer_dirty(eb); | |
2206 | if (slot == 0) | |
b167fa91 | 2207 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
2208 | } |
2209 | ||
d16c702f QW |
2210 | /* |
2211 | * Check key order of two sibling extent buffers. | |
2212 | * | |
2213 | * Return true if something is wrong. | |
2214 | * Return false if everything is fine. | |
2215 | * | |
2216 | * Tree-checker only works inside one tree block, thus the following | |
2217 | * corruption can not be detected by tree-checker: | |
2218 | * | |
2219 | * Leaf @left | Leaf @right | |
2220 | * -------------------------------------------------------------- | |
2221 | * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | | |
2222 | * | |
2223 | * Key f6 in leaf @left itself is valid, but not valid when the next | |
2224 | * key in leaf @right is 7. | |
2225 | * This can only be checked at tree block merge time. | |
2226 | * And since tree checker has ensured all key order in each tree block | |
2227 | * is correct, we only need to bother the last key of @left and the first | |
2228 | * key of @right. | |
2229 | */ | |
2230 | static bool check_sibling_keys(struct extent_buffer *left, | |
2231 | struct extent_buffer *right) | |
2232 | { | |
2233 | struct btrfs_key left_last; | |
2234 | struct btrfs_key right_first; | |
2235 | int level = btrfs_header_level(left); | |
2236 | int nr_left = btrfs_header_nritems(left); | |
2237 | int nr_right = btrfs_header_nritems(right); | |
2238 | ||
2239 | /* No key to check in one of the tree blocks */ | |
2240 | if (!nr_left || !nr_right) | |
2241 | return false; | |
2242 | ||
2243 | if (level) { | |
2244 | btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); | |
2245 | btrfs_node_key_to_cpu(right, &right_first, 0); | |
2246 | } else { | |
2247 | btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); | |
2248 | btrfs_item_key_to_cpu(right, &right_first, 0); | |
2249 | } | |
2250 | ||
2251 | if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { | |
2252 | btrfs_crit(left->fs_info, | |
2253 | "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", | |
2254 | left_last.objectid, left_last.type, | |
2255 | left_last.offset, right_first.objectid, | |
2256 | right_first.type, right_first.offset); | |
2257 | return true; | |
2258 | } | |
2259 | return false; | |
2260 | } | |
2261 | ||
74123bd7 CM |
2262 | /* |
2263 | * try to push data from one node into the next node left in the | |
79f95c82 | 2264 | * tree. |
aa5d6bed CM |
2265 | * |
2266 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2267 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2268 | */ |
98ed5174 | 2269 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 2270 | struct extent_buffer *dst, |
971a1f66 | 2271 | struct extent_buffer *src, int empty) |
be0e5c09 | 2272 | { |
d30a668f | 2273 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 2274 | int push_items = 0; |
bb803951 CM |
2275 | int src_nritems; |
2276 | int dst_nritems; | |
aa5d6bed | 2277 | int ret = 0; |
be0e5c09 | 2278 | |
5f39d397 CM |
2279 | src_nritems = btrfs_header_nritems(src); |
2280 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2281 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
2282 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2283 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2284 | |
bce4eae9 | 2285 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2286 | return 1; |
2287 | ||
d397712b | 2288 | if (push_items <= 0) |
be0e5c09 CM |
2289 | return 1; |
2290 | ||
bce4eae9 | 2291 | if (empty) { |
971a1f66 | 2292 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2293 | if (push_items < src_nritems) { |
2294 | /* leave at least 8 pointers in the node if | |
2295 | * we aren't going to empty it | |
2296 | */ | |
2297 | if (src_nritems - push_items < 8) { | |
2298 | if (push_items <= 8) | |
2299 | return 1; | |
2300 | push_items -= 8; | |
2301 | } | |
2302 | } | |
2303 | } else | |
2304 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2305 | |
d16c702f QW |
2306 | /* dst is the left eb, src is the middle eb */ |
2307 | if (check_sibling_keys(dst, src)) { | |
2308 | ret = -EUCLEAN; | |
2309 | btrfs_abort_transaction(trans, ret); | |
2310 | return ret; | |
2311 | } | |
f3a84ccd | 2312 | ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 2313 | if (ret) { |
66642832 | 2314 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2315 | return ret; |
2316 | } | |
5f39d397 CM |
2317 | copy_extent_buffer(dst, src, |
2318 | btrfs_node_key_ptr_offset(dst_nritems), | |
2319 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2320 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2321 | |
bb803951 | 2322 | if (push_items < src_nritems) { |
57911b8b | 2323 | /* |
f3a84ccd FM |
2324 | * Don't call btrfs_tree_mod_log_insert_move() here, key removal |
2325 | * was already fully logged by btrfs_tree_mod_log_eb_copy() above. | |
57911b8b | 2326 | */ |
5f39d397 CM |
2327 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2328 | btrfs_node_key_ptr_offset(push_items), | |
2329 | (src_nritems - push_items) * | |
2330 | sizeof(struct btrfs_key_ptr)); | |
2331 | } | |
2332 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2333 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2334 | btrfs_mark_buffer_dirty(src); | |
2335 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2336 | |
79f95c82 CM |
2337 | return ret; |
2338 | } | |
2339 | ||
2340 | /* | |
2341 | * try to push data from one node into the next node right in the | |
2342 | * tree. | |
2343 | * | |
2344 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2345 | * error, and > 0 if there was no room in the right hand block. | |
2346 | * | |
2347 | * this will only push up to 1/2 the contents of the left node over | |
2348 | */ | |
5f39d397 | 2349 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2350 | struct extent_buffer *dst, |
2351 | struct extent_buffer *src) | |
79f95c82 | 2352 | { |
55d32ed8 | 2353 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
2354 | int push_items = 0; |
2355 | int max_push; | |
2356 | int src_nritems; | |
2357 | int dst_nritems; | |
2358 | int ret = 0; | |
79f95c82 | 2359 | |
7bb86316 CM |
2360 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2361 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2362 | ||
5f39d397 CM |
2363 | src_nritems = btrfs_header_nritems(src); |
2364 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2365 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 2366 | if (push_items <= 0) |
79f95c82 | 2367 | return 1; |
bce4eae9 | 2368 | |
d397712b | 2369 | if (src_nritems < 4) |
bce4eae9 | 2370 | return 1; |
79f95c82 CM |
2371 | |
2372 | max_push = src_nritems / 2 + 1; | |
2373 | /* don't try to empty the node */ | |
d397712b | 2374 | if (max_push >= src_nritems) |
79f95c82 | 2375 | return 1; |
252c38f0 | 2376 | |
79f95c82 CM |
2377 | if (max_push < push_items) |
2378 | push_items = max_push; | |
2379 | ||
d16c702f QW |
2380 | /* dst is the right eb, src is the middle eb */ |
2381 | if (check_sibling_keys(src, dst)) { | |
2382 | ret = -EUCLEAN; | |
2383 | btrfs_abort_transaction(trans, ret); | |
2384 | return ret; | |
2385 | } | |
f3a84ccd | 2386 | ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
bf1d3425 | 2387 | BUG_ON(ret < 0); |
5f39d397 CM |
2388 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2389 | btrfs_node_key_ptr_offset(0), | |
2390 | (dst_nritems) * | |
2391 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2392 | |
f3a84ccd FM |
2393 | ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
2394 | push_items); | |
5de865ee | 2395 | if (ret) { |
66642832 | 2396 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2397 | return ret; |
2398 | } | |
5f39d397 CM |
2399 | copy_extent_buffer(dst, src, |
2400 | btrfs_node_key_ptr_offset(0), | |
2401 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2402 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2403 | |
5f39d397 CM |
2404 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2405 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2406 | |
5f39d397 CM |
2407 | btrfs_mark_buffer_dirty(src); |
2408 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2409 | |
aa5d6bed | 2410 | return ret; |
be0e5c09 CM |
2411 | } |
2412 | ||
97571fd0 CM |
2413 | /* |
2414 | * helper function to insert a new root level in the tree. | |
2415 | * A new node is allocated, and a single item is inserted to | |
2416 | * point to the existing root | |
aa5d6bed CM |
2417 | * |
2418 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2419 | */ |
d397712b | 2420 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 2421 | struct btrfs_root *root, |
fdd99c72 | 2422 | struct btrfs_path *path, int level) |
5c680ed6 | 2423 | { |
0b246afa | 2424 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 2425 | u64 lower_gen; |
5f39d397 CM |
2426 | struct extent_buffer *lower; |
2427 | struct extent_buffer *c; | |
925baedd | 2428 | struct extent_buffer *old; |
5f39d397 | 2429 | struct btrfs_disk_key lower_key; |
d9d19a01 | 2430 | int ret; |
5c680ed6 CM |
2431 | |
2432 | BUG_ON(path->nodes[level]); | |
2433 | BUG_ON(path->nodes[level-1] != root->node); | |
2434 | ||
7bb86316 CM |
2435 | lower = path->nodes[level-1]; |
2436 | if (level == 1) | |
2437 | btrfs_item_key(lower, &lower_key, 0); | |
2438 | else | |
2439 | btrfs_node_key(lower, &lower_key, 0); | |
2440 | ||
79bd3712 FM |
2441 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2442 | &lower_key, level, root->node->start, 0, | |
2443 | BTRFS_NESTING_NEW_ROOT); | |
5f39d397 CM |
2444 | if (IS_ERR(c)) |
2445 | return PTR_ERR(c); | |
925baedd | 2446 | |
0b246afa | 2447 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 2448 | |
5f39d397 | 2449 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 2450 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2451 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2452 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2453 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2454 | |
2455 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2456 | |
5f39d397 | 2457 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2458 | |
925baedd | 2459 | old = root->node; |
406808ab | 2460 | ret = btrfs_tree_mod_log_insert_root(root->node, c, false); |
d9d19a01 | 2461 | BUG_ON(ret < 0); |
240f62c8 | 2462 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2463 | |
2464 | /* the super has an extra ref to root->node */ | |
2465 | free_extent_buffer(old); | |
2466 | ||
0b86a832 | 2467 | add_root_to_dirty_list(root); |
67439dad | 2468 | atomic_inc(&c->refs); |
5f39d397 | 2469 | path->nodes[level] = c; |
ac5887c8 | 2470 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2471 | path->slots[level] = 0; |
2472 | return 0; | |
2473 | } | |
2474 | ||
74123bd7 CM |
2475 | /* |
2476 | * worker function to insert a single pointer in a node. | |
2477 | * the node should have enough room for the pointer already | |
97571fd0 | 2478 | * |
74123bd7 CM |
2479 | * slot and level indicate where you want the key to go, and |
2480 | * blocknr is the block the key points to. | |
2481 | */ | |
143bede5 | 2482 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 2483 | struct btrfs_path *path, |
143bede5 | 2484 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 2485 | int slot, int level) |
74123bd7 | 2486 | { |
5f39d397 | 2487 | struct extent_buffer *lower; |
74123bd7 | 2488 | int nritems; |
f3ea38da | 2489 | int ret; |
5c680ed6 CM |
2490 | |
2491 | BUG_ON(!path->nodes[level]); | |
49d0c642 | 2492 | btrfs_assert_tree_write_locked(path->nodes[level]); |
5f39d397 CM |
2493 | lower = path->nodes[level]; |
2494 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2495 | BUG_ON(slot > nritems); |
6ad3cf6d | 2496 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 2497 | if (slot != nritems) { |
bf1d3425 | 2498 | if (level) { |
f3a84ccd FM |
2499 | ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, |
2500 | slot, nritems - slot); | |
bf1d3425 DS |
2501 | BUG_ON(ret < 0); |
2502 | } | |
5f39d397 CM |
2503 | memmove_extent_buffer(lower, |
2504 | btrfs_node_key_ptr_offset(slot + 1), | |
2505 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2506 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2507 | } |
c3e06965 | 2508 | if (level) { |
f3a84ccd FM |
2509 | ret = btrfs_tree_mod_log_insert_key(lower, slot, |
2510 | BTRFS_MOD_LOG_KEY_ADD, GFP_NOFS); | |
f3ea38da JS |
2511 | BUG_ON(ret < 0); |
2512 | } | |
5f39d397 | 2513 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2514 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2515 | WARN_ON(trans->transid == 0); |
2516 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2517 | btrfs_set_header_nritems(lower, nritems + 1); |
2518 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2519 | } |
2520 | ||
97571fd0 CM |
2521 | /* |
2522 | * split the node at the specified level in path in two. | |
2523 | * The path is corrected to point to the appropriate node after the split | |
2524 | * | |
2525 | * Before splitting this tries to make some room in the node by pushing | |
2526 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2527 | * |
2528 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2529 | */ |
e02119d5 CM |
2530 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2531 | struct btrfs_root *root, | |
2532 | struct btrfs_path *path, int level) | |
be0e5c09 | 2533 | { |
0b246afa | 2534 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2535 | struct extent_buffer *c; |
2536 | struct extent_buffer *split; | |
2537 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2538 | int mid; |
5c680ed6 | 2539 | int ret; |
7518a238 | 2540 | u32 c_nritems; |
eb60ceac | 2541 | |
5f39d397 | 2542 | c = path->nodes[level]; |
7bb86316 | 2543 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2544 | if (c == root->node) { |
d9abbf1c | 2545 | /* |
90f8d62e JS |
2546 | * trying to split the root, lets make a new one |
2547 | * | |
fdd99c72 | 2548 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
2549 | * insert_new_root, because that root buffer will be kept as a |
2550 | * normal node. We are going to log removal of half of the | |
f3a84ccd FM |
2551 | * elements below with btrfs_tree_mod_log_eb_copy(). We're |
2552 | * holding a tree lock on the buffer, which is why we cannot | |
2553 | * race with other tree_mod_log users. | |
d9abbf1c | 2554 | */ |
fdd99c72 | 2555 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2556 | if (ret) |
2557 | return ret; | |
b3612421 | 2558 | } else { |
e66f709b | 2559 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2560 | c = path->nodes[level]; |
2561 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 2562 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 2563 | return 0; |
54aa1f4d CM |
2564 | if (ret < 0) |
2565 | return ret; | |
be0e5c09 | 2566 | } |
e66f709b | 2567 | |
5f39d397 | 2568 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2569 | mid = (c_nritems + 1) / 2; |
2570 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2571 | |
79bd3712 FM |
2572 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2573 | &disk_key, level, c->start, 0, | |
2574 | BTRFS_NESTING_SPLIT); | |
5f39d397 CM |
2575 | if (IS_ERR(split)) |
2576 | return PTR_ERR(split); | |
2577 | ||
0b246afa | 2578 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 2579 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 2580 | |
f3a84ccd | 2581 | ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 2582 | if (ret) { |
66642832 | 2583 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2584 | return ret; |
2585 | } | |
5f39d397 CM |
2586 | copy_extent_buffer(split, c, |
2587 | btrfs_node_key_ptr_offset(0), | |
2588 | btrfs_node_key_ptr_offset(mid), | |
2589 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2590 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2591 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed | 2592 | |
5f39d397 CM |
2593 | btrfs_mark_buffer_dirty(c); |
2594 | btrfs_mark_buffer_dirty(split); | |
2595 | ||
6ad3cf6d | 2596 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 2597 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 2598 | |
5de08d7d | 2599 | if (path->slots[level] >= mid) { |
5c680ed6 | 2600 | path->slots[level] -= mid; |
925baedd | 2601 | btrfs_tree_unlock(c); |
5f39d397 CM |
2602 | free_extent_buffer(c); |
2603 | path->nodes[level] = split; | |
5c680ed6 CM |
2604 | path->slots[level + 1] += 1; |
2605 | } else { | |
925baedd | 2606 | btrfs_tree_unlock(split); |
5f39d397 | 2607 | free_extent_buffer(split); |
be0e5c09 | 2608 | } |
d5286a92 | 2609 | return 0; |
be0e5c09 CM |
2610 | } |
2611 | ||
74123bd7 CM |
2612 | /* |
2613 | * how many bytes are required to store the items in a leaf. start | |
2614 | * and nr indicate which items in the leaf to check. This totals up the | |
2615 | * space used both by the item structs and the item data | |
2616 | */ | |
5f39d397 | 2617 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2618 | { |
2619 | int data_len; | |
5f39d397 | 2620 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2621 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2622 | |
2623 | if (!nr) | |
2624 | return 0; | |
227f3cd0 JB |
2625 | data_len = btrfs_item_offset_nr(l, start) + btrfs_item_size_nr(l, start); |
2626 | data_len = data_len - btrfs_item_offset_nr(l, end); | |
0783fcfc | 2627 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2628 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2629 | return data_len; |
2630 | } | |
2631 | ||
d4dbff95 CM |
2632 | /* |
2633 | * The space between the end of the leaf items and | |
2634 | * the start of the leaf data. IOW, how much room | |
2635 | * the leaf has left for both items and data | |
2636 | */ | |
e902baac | 2637 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 2638 | { |
e902baac | 2639 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
2640 | int nritems = btrfs_header_nritems(leaf); |
2641 | int ret; | |
0b246afa JM |
2642 | |
2643 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 2644 | if (ret < 0) { |
0b246afa JM |
2645 | btrfs_crit(fs_info, |
2646 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
2647 | ret, | |
2648 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
2649 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
2650 | } |
2651 | return ret; | |
d4dbff95 CM |
2652 | } |
2653 | ||
99d8f83c CM |
2654 | /* |
2655 | * min slot controls the lowest index we're willing to push to the | |
2656 | * right. We'll push up to and including min_slot, but no lower | |
2657 | */ | |
f72f0010 | 2658 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
2659 | int data_size, int empty, |
2660 | struct extent_buffer *right, | |
99d8f83c CM |
2661 | int free_space, u32 left_nritems, |
2662 | u32 min_slot) | |
00ec4c51 | 2663 | { |
f72f0010 | 2664 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 2665 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2666 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2667 | struct btrfs_map_token token; |
5f39d397 | 2668 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2669 | int slot; |
34a38218 | 2670 | u32 i; |
00ec4c51 CM |
2671 | int push_space = 0; |
2672 | int push_items = 0; | |
0783fcfc | 2673 | struct btrfs_item *item; |
34a38218 | 2674 | u32 nr; |
7518a238 | 2675 | u32 right_nritems; |
5f39d397 | 2676 | u32 data_end; |
db94535d | 2677 | u32 this_item_size; |
00ec4c51 | 2678 | |
34a38218 CM |
2679 | if (empty) |
2680 | nr = 0; | |
2681 | else | |
99d8f83c | 2682 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2683 | |
31840ae1 | 2684 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2685 | push_space += data_size; |
31840ae1 | 2686 | |
44871b1b | 2687 | slot = path->slots[1]; |
34a38218 CM |
2688 | i = left_nritems - 1; |
2689 | while (i >= nr) { | |
31840ae1 ZY |
2690 | if (!empty && push_items > 0) { |
2691 | if (path->slots[0] > i) | |
2692 | break; | |
2693 | if (path->slots[0] == i) { | |
e902baac DS |
2694 | int space = btrfs_leaf_free_space(left); |
2695 | ||
31840ae1 ZY |
2696 | if (space + push_space * 2 > free_space) |
2697 | break; | |
2698 | } | |
2699 | } | |
2700 | ||
00ec4c51 | 2701 | if (path->slots[0] == i) |
87b29b20 | 2702 | push_space += data_size; |
db94535d | 2703 | |
227f3cd0 | 2704 | this_item_size = btrfs_item_size_nr(left, i); |
db94535d | 2705 | if (this_item_size + sizeof(*item) + push_space > free_space) |
00ec4c51 | 2706 | break; |
31840ae1 | 2707 | |
00ec4c51 | 2708 | push_items++; |
db94535d | 2709 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
2710 | if (i == 0) |
2711 | break; | |
2712 | i--; | |
db94535d | 2713 | } |
5f39d397 | 2714 | |
925baedd CM |
2715 | if (push_items == 0) |
2716 | goto out_unlock; | |
5f39d397 | 2717 | |
6c1500f2 | 2718 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 2719 | |
00ec4c51 | 2720 | /* push left to right */ |
5f39d397 | 2721 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2722 | |
5f39d397 | 2723 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
8f881e8c | 2724 | push_space -= leaf_data_end(left); |
5f39d397 | 2725 | |
00ec4c51 | 2726 | /* make room in the right data area */ |
8f881e8c | 2727 | data_end = leaf_data_end(right); |
5f39d397 | 2728 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
2729 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
2730 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 2731 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 2732 | |
00ec4c51 | 2733 | /* copy from the left data area */ |
3d9ec8c4 | 2734 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 2735 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 2736 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 2737 | push_space); |
5f39d397 CM |
2738 | |
2739 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
2740 | btrfs_item_nr_offset(0), | |
2741 | right_nritems * sizeof(struct btrfs_item)); | |
2742 | ||
00ec4c51 | 2743 | /* copy the items from left to right */ |
5f39d397 CM |
2744 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
2745 | btrfs_item_nr_offset(left_nritems - push_items), | |
2746 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
2747 | |
2748 | /* update the item pointers */ | |
c82f823c | 2749 | btrfs_init_map_token(&token, right); |
7518a238 | 2750 | right_nritems += push_items; |
5f39d397 | 2751 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 2752 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 2753 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 2754 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
2755 | push_space -= btrfs_token_item_size(&token, item); |
2756 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d CM |
2757 | } |
2758 | ||
7518a238 | 2759 | left_nritems -= push_items; |
5f39d397 | 2760 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 2761 | |
34a38218 CM |
2762 | if (left_nritems) |
2763 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 2764 | else |
6a884d7d | 2765 | btrfs_clean_tree_block(left); |
f0486c68 | 2766 | |
5f39d397 | 2767 | btrfs_mark_buffer_dirty(right); |
a429e513 | 2768 | |
5f39d397 CM |
2769 | btrfs_item_key(right, &disk_key, 0); |
2770 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 2771 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 2772 | |
00ec4c51 | 2773 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
2774 | if (path->slots[0] >= left_nritems) { |
2775 | path->slots[0] -= left_nritems; | |
925baedd | 2776 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 2777 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 2778 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
2779 | free_extent_buffer(path->nodes[0]); |
2780 | path->nodes[0] = right; | |
00ec4c51 CM |
2781 | path->slots[1] += 1; |
2782 | } else { | |
925baedd | 2783 | btrfs_tree_unlock(right); |
5f39d397 | 2784 | free_extent_buffer(right); |
00ec4c51 CM |
2785 | } |
2786 | return 0; | |
925baedd CM |
2787 | |
2788 | out_unlock: | |
2789 | btrfs_tree_unlock(right); | |
2790 | free_extent_buffer(right); | |
2791 | return 1; | |
00ec4c51 | 2792 | } |
925baedd | 2793 | |
44871b1b CM |
2794 | /* |
2795 | * push some data in the path leaf to the right, trying to free up at | |
2796 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
2797 | * | |
2798 | * returns 1 if the push failed because the other node didn't have enough | |
2799 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
2800 | * |
2801 | * this will push starting from min_slot to the end of the leaf. It won't | |
2802 | * push any slot lower than min_slot | |
44871b1b CM |
2803 | */ |
2804 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
2805 | *root, struct btrfs_path *path, |
2806 | int min_data_size, int data_size, | |
2807 | int empty, u32 min_slot) | |
44871b1b CM |
2808 | { |
2809 | struct extent_buffer *left = path->nodes[0]; | |
2810 | struct extent_buffer *right; | |
2811 | struct extent_buffer *upper; | |
2812 | int slot; | |
2813 | int free_space; | |
2814 | u32 left_nritems; | |
2815 | int ret; | |
2816 | ||
2817 | if (!path->nodes[1]) | |
2818 | return 1; | |
2819 | ||
2820 | slot = path->slots[1]; | |
2821 | upper = path->nodes[1]; | |
2822 | if (slot >= btrfs_header_nritems(upper) - 1) | |
2823 | return 1; | |
2824 | ||
49d0c642 | 2825 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 2826 | |
4b231ae4 | 2827 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
2828 | /* |
2829 | * slot + 1 is not valid or we fail to read the right node, | |
2830 | * no big deal, just return. | |
2831 | */ | |
2832 | if (IS_ERR(right)) | |
91ca338d TI |
2833 | return 1; |
2834 | ||
bf77467a | 2835 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
44871b1b | 2836 | |
e902baac | 2837 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
2838 | if (free_space < data_size) |
2839 | goto out_unlock; | |
2840 | ||
2841 | /* cow and double check */ | |
2842 | ret = btrfs_cow_block(trans, root, right, upper, | |
bf59a5a2 | 2843 | slot + 1, &right, BTRFS_NESTING_RIGHT_COW); |
44871b1b CM |
2844 | if (ret) |
2845 | goto out_unlock; | |
2846 | ||
e902baac | 2847 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
2848 | if (free_space < data_size) |
2849 | goto out_unlock; | |
2850 | ||
2851 | left_nritems = btrfs_header_nritems(left); | |
2852 | if (left_nritems == 0) | |
2853 | goto out_unlock; | |
2854 | ||
d16c702f QW |
2855 | if (check_sibling_keys(left, right)) { |
2856 | ret = -EUCLEAN; | |
2857 | btrfs_tree_unlock(right); | |
2858 | free_extent_buffer(right); | |
2859 | return ret; | |
2860 | } | |
2ef1fed2 FDBM |
2861 | if (path->slots[0] == left_nritems && !empty) { |
2862 | /* Key greater than all keys in the leaf, right neighbor has | |
2863 | * enough room for it and we're not emptying our leaf to delete | |
2864 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 2865 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
2866 | btrfs_tree_unlock(left); |
2867 | free_extent_buffer(left); | |
2868 | path->nodes[0] = right; | |
2869 | path->slots[0] = 0; | |
2870 | path->slots[1]++; | |
2871 | return 0; | |
2872 | } | |
2873 | ||
f72f0010 | 2874 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 2875 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
2876 | out_unlock: |
2877 | btrfs_tree_unlock(right); | |
2878 | free_extent_buffer(right); | |
2879 | return 1; | |
2880 | } | |
2881 | ||
74123bd7 CM |
2882 | /* |
2883 | * push some data in the path leaf to the left, trying to free up at | |
2884 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
2885 | * |
2886 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
2887 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
2888 | * items | |
74123bd7 | 2889 | */ |
8087c193 | 2890 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 2891 | int empty, struct extent_buffer *left, |
99d8f83c CM |
2892 | int free_space, u32 right_nritems, |
2893 | u32 max_slot) | |
be0e5c09 | 2894 | { |
8087c193 | 2895 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
2896 | struct btrfs_disk_key disk_key; |
2897 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 2898 | int i; |
be0e5c09 CM |
2899 | int push_space = 0; |
2900 | int push_items = 0; | |
0783fcfc | 2901 | struct btrfs_item *item; |
7518a238 | 2902 | u32 old_left_nritems; |
34a38218 | 2903 | u32 nr; |
aa5d6bed | 2904 | int ret = 0; |
db94535d CM |
2905 | u32 this_item_size; |
2906 | u32 old_left_item_size; | |
cfed81a0 CM |
2907 | struct btrfs_map_token token; |
2908 | ||
34a38218 | 2909 | if (empty) |
99d8f83c | 2910 | nr = min(right_nritems, max_slot); |
34a38218 | 2911 | else |
99d8f83c | 2912 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
2913 | |
2914 | for (i = 0; i < nr; i++) { | |
31840ae1 ZY |
2915 | if (!empty && push_items > 0) { |
2916 | if (path->slots[0] < i) | |
2917 | break; | |
2918 | if (path->slots[0] == i) { | |
e902baac DS |
2919 | int space = btrfs_leaf_free_space(right); |
2920 | ||
31840ae1 ZY |
2921 | if (space + push_space * 2 > free_space) |
2922 | break; | |
2923 | } | |
2924 | } | |
2925 | ||
be0e5c09 | 2926 | if (path->slots[0] == i) |
87b29b20 | 2927 | push_space += data_size; |
db94535d | 2928 | |
227f3cd0 | 2929 | this_item_size = btrfs_item_size_nr(right, i); |
db94535d | 2930 | if (this_item_size + sizeof(*item) + push_space > free_space) |
be0e5c09 | 2931 | break; |
db94535d | 2932 | |
be0e5c09 | 2933 | push_items++; |
db94535d CM |
2934 | push_space += this_item_size + sizeof(*item); |
2935 | } | |
2936 | ||
be0e5c09 | 2937 | if (push_items == 0) { |
925baedd CM |
2938 | ret = 1; |
2939 | goto out; | |
be0e5c09 | 2940 | } |
fae7f21c | 2941 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 2942 | |
be0e5c09 | 2943 | /* push data from right to left */ |
5f39d397 CM |
2944 | copy_extent_buffer(left, right, |
2945 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
2946 | btrfs_item_nr_offset(0), | |
2947 | push_items * sizeof(struct btrfs_item)); | |
2948 | ||
0b246afa | 2949 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 2950 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 2951 | |
3d9ec8c4 | 2952 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 2953 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 2954 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 2955 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 2956 | push_space); |
5f39d397 | 2957 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 2958 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 2959 | |
c82f823c | 2960 | btrfs_init_map_token(&token, left); |
db94535d | 2961 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 2962 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 2963 | u32 ioff; |
db94535d | 2964 | |
dd3cc16b | 2965 | item = btrfs_item_nr(i); |
db94535d | 2966 | |
cc4c13d5 DS |
2967 | ioff = btrfs_token_item_offset(&token, item); |
2968 | btrfs_set_token_item_offset(&token, item, | |
2969 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); | |
be0e5c09 | 2970 | } |
5f39d397 | 2971 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
2972 | |
2973 | /* fixup right node */ | |
31b1a2bd JL |
2974 | if (push_items > right_nritems) |
2975 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 2976 | right_nritems); |
34a38218 CM |
2977 | |
2978 | if (push_items < right_nritems) { | |
2979 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
8f881e8c | 2980 | leaf_data_end(right); |
3d9ec8c4 | 2981 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 2982 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 2983 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 2984 | leaf_data_end(right), push_space); |
34a38218 CM |
2985 | |
2986 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
2987 | btrfs_item_nr_offset(push_items), |
2988 | (btrfs_header_nritems(right) - push_items) * | |
2989 | sizeof(struct btrfs_item)); | |
34a38218 | 2990 | } |
c82f823c DS |
2991 | |
2992 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
2993 | right_nritems -= push_items; |
2994 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 2995 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 2996 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 2997 | item = btrfs_item_nr(i); |
db94535d | 2998 | |
cc4c13d5 DS |
2999 | push_space = push_space - btrfs_token_item_size(&token, item); |
3000 | btrfs_set_token_item_offset(&token, item, push_space); | |
db94535d | 3001 | } |
eb60ceac | 3002 | |
5f39d397 | 3003 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3004 | if (right_nritems) |
3005 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3006 | else |
6a884d7d | 3007 | btrfs_clean_tree_block(right); |
098f59c2 | 3008 | |
5f39d397 | 3009 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3010 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3011 | |
3012 | /* then fixup the leaf pointer in the path */ | |
3013 | if (path->slots[0] < push_items) { | |
3014 | path->slots[0] += old_left_nritems; | |
925baedd | 3015 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3016 | free_extent_buffer(path->nodes[0]); |
3017 | path->nodes[0] = left; | |
be0e5c09 CM |
3018 | path->slots[1] -= 1; |
3019 | } else { | |
925baedd | 3020 | btrfs_tree_unlock(left); |
5f39d397 | 3021 | free_extent_buffer(left); |
be0e5c09 CM |
3022 | path->slots[0] -= push_items; |
3023 | } | |
eb60ceac | 3024 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3025 | return ret; |
925baedd CM |
3026 | out: |
3027 | btrfs_tree_unlock(left); | |
3028 | free_extent_buffer(left); | |
3029 | return ret; | |
be0e5c09 CM |
3030 | } |
3031 | ||
44871b1b CM |
3032 | /* |
3033 | * push some data in the path leaf to the left, trying to free up at | |
3034 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3035 | * |
3036 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3037 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3038 | * items | |
44871b1b CM |
3039 | */ |
3040 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3041 | *root, struct btrfs_path *path, int min_data_size, |
3042 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3043 | { |
3044 | struct extent_buffer *right = path->nodes[0]; | |
3045 | struct extent_buffer *left; | |
3046 | int slot; | |
3047 | int free_space; | |
3048 | u32 right_nritems; | |
3049 | int ret = 0; | |
3050 | ||
3051 | slot = path->slots[1]; | |
3052 | if (slot == 0) | |
3053 | return 1; | |
3054 | if (!path->nodes[1]) | |
3055 | return 1; | |
3056 | ||
3057 | right_nritems = btrfs_header_nritems(right); | |
3058 | if (right_nritems == 0) | |
3059 | return 1; | |
3060 | ||
49d0c642 | 3061 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3062 | |
4b231ae4 | 3063 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3064 | /* |
3065 | * slot - 1 is not valid or we fail to read the left node, | |
3066 | * no big deal, just return. | |
3067 | */ | |
3068 | if (IS_ERR(left)) | |
91ca338d TI |
3069 | return 1; |
3070 | ||
bf77467a | 3071 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
44871b1b | 3072 | |
e902baac | 3073 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3074 | if (free_space < data_size) { |
3075 | ret = 1; | |
3076 | goto out; | |
3077 | } | |
3078 | ||
3079 | /* cow and double check */ | |
3080 | ret = btrfs_cow_block(trans, root, left, | |
9631e4cc | 3081 | path->nodes[1], slot - 1, &left, |
bf59a5a2 | 3082 | BTRFS_NESTING_LEFT_COW); |
44871b1b CM |
3083 | if (ret) { |
3084 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3085 | if (ret == -ENOSPC) |
3086 | ret = 1; | |
44871b1b CM |
3087 | goto out; |
3088 | } | |
3089 | ||
e902baac | 3090 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3091 | if (free_space < data_size) { |
3092 | ret = 1; | |
3093 | goto out; | |
3094 | } | |
3095 | ||
d16c702f QW |
3096 | if (check_sibling_keys(left, right)) { |
3097 | ret = -EUCLEAN; | |
3098 | goto out; | |
3099 | } | |
8087c193 | 3100 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
3101 | empty, left, free_space, right_nritems, |
3102 | max_slot); | |
44871b1b CM |
3103 | out: |
3104 | btrfs_tree_unlock(left); | |
3105 | free_extent_buffer(left); | |
3106 | return ret; | |
3107 | } | |
3108 | ||
3109 | /* | |
3110 | * split the path's leaf in two, making sure there is at least data_size | |
3111 | * available for the resulting leaf level of the path. | |
44871b1b | 3112 | */ |
143bede5 | 3113 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
3114 | struct btrfs_path *path, |
3115 | struct extent_buffer *l, | |
3116 | struct extent_buffer *right, | |
3117 | int slot, int mid, int nritems) | |
44871b1b | 3118 | { |
94f94ad9 | 3119 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
3120 | int data_copy_size; |
3121 | int rt_data_off; | |
3122 | int i; | |
44871b1b | 3123 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3124 | struct btrfs_map_token token; |
3125 | ||
44871b1b CM |
3126 | nritems = nritems - mid; |
3127 | btrfs_set_header_nritems(right, nritems); | |
8f881e8c | 3128 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(l); |
44871b1b CM |
3129 | |
3130 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3131 | btrfs_item_nr_offset(mid), | |
3132 | nritems * sizeof(struct btrfs_item)); | |
3133 | ||
3134 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
3135 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
3136 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 3137 | leaf_data_end(l), data_copy_size); |
44871b1b | 3138 | |
0b246afa | 3139 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b | 3140 | |
c82f823c | 3141 | btrfs_init_map_token(&token, right); |
44871b1b | 3142 | for (i = 0; i < nritems; i++) { |
dd3cc16b | 3143 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
3144 | u32 ioff; |
3145 | ||
cc4c13d5 DS |
3146 | ioff = btrfs_token_item_offset(&token, item); |
3147 | btrfs_set_token_item_offset(&token, item, ioff + rt_data_off); | |
44871b1b CM |
3148 | } |
3149 | ||
44871b1b | 3150 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3151 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 3152 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
3153 | |
3154 | btrfs_mark_buffer_dirty(right); | |
3155 | btrfs_mark_buffer_dirty(l); | |
3156 | BUG_ON(path->slots[0] != slot); | |
3157 | ||
44871b1b CM |
3158 | if (mid <= slot) { |
3159 | btrfs_tree_unlock(path->nodes[0]); | |
3160 | free_extent_buffer(path->nodes[0]); | |
3161 | path->nodes[0] = right; | |
3162 | path->slots[0] -= mid; | |
3163 | path->slots[1] += 1; | |
3164 | } else { | |
3165 | btrfs_tree_unlock(right); | |
3166 | free_extent_buffer(right); | |
3167 | } | |
3168 | ||
3169 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3170 | } |
3171 | ||
99d8f83c CM |
3172 | /* |
3173 | * double splits happen when we need to insert a big item in the middle | |
3174 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3175 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3176 | * A B C | |
3177 | * | |
3178 | * We avoid this by trying to push the items on either side of our target | |
3179 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3180 | * completely. | |
3181 | */ | |
3182 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3183 | struct btrfs_root *root, | |
3184 | struct btrfs_path *path, | |
3185 | int data_size) | |
3186 | { | |
3187 | int ret; | |
3188 | int progress = 0; | |
3189 | int slot; | |
3190 | u32 nritems; | |
5a4267ca | 3191 | int space_needed = data_size; |
99d8f83c CM |
3192 | |
3193 | slot = path->slots[0]; | |
5a4267ca | 3194 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 3195 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
3196 | |
3197 | /* | |
3198 | * try to push all the items after our slot into the | |
3199 | * right leaf | |
3200 | */ | |
5a4267ca | 3201 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3202 | if (ret < 0) |
3203 | return ret; | |
3204 | ||
3205 | if (ret == 0) | |
3206 | progress++; | |
3207 | ||
3208 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3209 | /* | |
3210 | * our goal is to get our slot at the start or end of a leaf. If | |
3211 | * we've done so we're done | |
3212 | */ | |
3213 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3214 | return 0; | |
3215 | ||
e902baac | 3216 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3217 | return 0; |
3218 | ||
3219 | /* try to push all the items before our slot into the next leaf */ | |
3220 | slot = path->slots[0]; | |
263d3995 FM |
3221 | space_needed = data_size; |
3222 | if (slot > 0) | |
e902baac | 3223 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 3224 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3225 | if (ret < 0) |
3226 | return ret; | |
3227 | ||
3228 | if (ret == 0) | |
3229 | progress++; | |
3230 | ||
3231 | if (progress) | |
3232 | return 0; | |
3233 | return 1; | |
3234 | } | |
3235 | ||
74123bd7 CM |
3236 | /* |
3237 | * split the path's leaf in two, making sure there is at least data_size | |
3238 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3239 | * |
3240 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3241 | */ |
e02119d5 CM |
3242 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3243 | struct btrfs_root *root, | |
310712b2 | 3244 | const struct btrfs_key *ins_key, |
e02119d5 CM |
3245 | struct btrfs_path *path, int data_size, |
3246 | int extend) | |
be0e5c09 | 3247 | { |
5d4f98a2 | 3248 | struct btrfs_disk_key disk_key; |
5f39d397 | 3249 | struct extent_buffer *l; |
7518a238 | 3250 | u32 nritems; |
eb60ceac CM |
3251 | int mid; |
3252 | int slot; | |
5f39d397 | 3253 | struct extent_buffer *right; |
b7a0365e | 3254 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 3255 | int ret = 0; |
aa5d6bed | 3256 | int wret; |
5d4f98a2 | 3257 | int split; |
cc0c5538 | 3258 | int num_doubles = 0; |
99d8f83c | 3259 | int tried_avoid_double = 0; |
aa5d6bed | 3260 | |
a5719521 YZ |
3261 | l = path->nodes[0]; |
3262 | slot = path->slots[0]; | |
3263 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 3264 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
3265 | return -EOVERFLOW; |
3266 | ||
40689478 | 3267 | /* first try to make some room by pushing left and right */ |
33157e05 | 3268 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
3269 | int space_needed = data_size; |
3270 | ||
3271 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 3272 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3273 | |
3274 | wret = push_leaf_right(trans, root, path, space_needed, | |
3275 | space_needed, 0, 0); | |
d397712b | 3276 | if (wret < 0) |
eaee50e8 | 3277 | return wret; |
3685f791 | 3278 | if (wret) { |
263d3995 FM |
3279 | space_needed = data_size; |
3280 | if (slot > 0) | |
e902baac | 3281 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3282 | wret = push_leaf_left(trans, root, path, space_needed, |
3283 | space_needed, 0, (u32)-1); | |
3685f791 CM |
3284 | if (wret < 0) |
3285 | return wret; | |
3286 | } | |
3287 | l = path->nodes[0]; | |
aa5d6bed | 3288 | |
3685f791 | 3289 | /* did the pushes work? */ |
e902baac | 3290 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 3291 | return 0; |
3326d1b0 | 3292 | } |
aa5d6bed | 3293 | |
5c680ed6 | 3294 | if (!path->nodes[1]) { |
fdd99c72 | 3295 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3296 | if (ret) |
3297 | return ret; | |
3298 | } | |
cc0c5538 | 3299 | again: |
5d4f98a2 | 3300 | split = 1; |
cc0c5538 | 3301 | l = path->nodes[0]; |
eb60ceac | 3302 | slot = path->slots[0]; |
5f39d397 | 3303 | nritems = btrfs_header_nritems(l); |
d397712b | 3304 | mid = (nritems + 1) / 2; |
54aa1f4d | 3305 | |
5d4f98a2 YZ |
3306 | if (mid <= slot) { |
3307 | if (nritems == 1 || | |
3308 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 3309 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3310 | if (slot >= nritems) { |
3311 | split = 0; | |
3312 | } else { | |
3313 | mid = slot; | |
3314 | if (mid != nritems && | |
3315 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3316 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3317 | if (data_size && !tried_avoid_double) |
3318 | goto push_for_double; | |
5d4f98a2 YZ |
3319 | split = 2; |
3320 | } | |
3321 | } | |
3322 | } | |
3323 | } else { | |
3324 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 3325 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3326 | if (!extend && data_size && slot == 0) { |
3327 | split = 0; | |
3328 | } else if ((extend || !data_size) && slot == 0) { | |
3329 | mid = 1; | |
3330 | } else { | |
3331 | mid = slot; | |
3332 | if (mid != nritems && | |
3333 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3334 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3335 | if (data_size && !tried_avoid_double) |
3336 | goto push_for_double; | |
67871254 | 3337 | split = 2; |
5d4f98a2 YZ |
3338 | } |
3339 | } | |
3340 | } | |
3341 | } | |
3342 | ||
3343 | if (split == 0) | |
3344 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3345 | else | |
3346 | btrfs_item_key(l, &disk_key, mid); | |
3347 | ||
ca9d473a JB |
3348 | /* |
3349 | * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double | |
3350 | * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES | |
3351 | * subclasses, which is 8 at the time of this patch, and we've maxed it | |
3352 | * out. In the future we could add a | |
3353 | * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just | |
3354 | * use BTRFS_NESTING_NEW_ROOT. | |
3355 | */ | |
79bd3712 FM |
3356 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3357 | &disk_key, 0, l->start, 0, | |
3358 | num_doubles ? BTRFS_NESTING_NEW_ROOT : | |
3359 | BTRFS_NESTING_SPLIT); | |
f0486c68 | 3360 | if (IS_ERR(right)) |
5f39d397 | 3361 | return PTR_ERR(right); |
f0486c68 | 3362 | |
0b246afa | 3363 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 3364 | |
5d4f98a2 YZ |
3365 | if (split == 0) { |
3366 | if (mid <= slot) { | |
3367 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3368 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3369 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
3370 | btrfs_tree_unlock(path->nodes[0]); |
3371 | free_extent_buffer(path->nodes[0]); | |
3372 | path->nodes[0] = right; | |
3373 | path->slots[0] = 0; | |
3374 | path->slots[1] += 1; | |
3375 | } else { | |
3376 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3377 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3378 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
3379 | btrfs_tree_unlock(path->nodes[0]); |
3380 | free_extent_buffer(path->nodes[0]); | |
3381 | path->nodes[0] = right; | |
3382 | path->slots[0] = 0; | |
143bede5 | 3383 | if (path->slots[1] == 0) |
b167fa91 | 3384 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 3385 | } |
196e0249 LB |
3386 | /* |
3387 | * We create a new leaf 'right' for the required ins_len and | |
3388 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
3389 | * the content of ins_len to 'right'. | |
3390 | */ | |
5d4f98a2 | 3391 | return ret; |
d4dbff95 | 3392 | } |
74123bd7 | 3393 | |
94f94ad9 | 3394 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 3395 | |
5d4f98a2 | 3396 | if (split == 2) { |
cc0c5538 CM |
3397 | BUG_ON(num_doubles != 0); |
3398 | num_doubles++; | |
3399 | goto again; | |
a429e513 | 3400 | } |
44871b1b | 3401 | |
143bede5 | 3402 | return 0; |
99d8f83c CM |
3403 | |
3404 | push_for_double: | |
3405 | push_for_double_split(trans, root, path, data_size); | |
3406 | tried_avoid_double = 1; | |
e902baac | 3407 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3408 | return 0; |
3409 | goto again; | |
be0e5c09 CM |
3410 | } |
3411 | ||
ad48fd75 YZ |
3412 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3413 | struct btrfs_root *root, | |
3414 | struct btrfs_path *path, int ins_len) | |
459931ec | 3415 | { |
ad48fd75 | 3416 | struct btrfs_key key; |
459931ec | 3417 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3418 | struct btrfs_file_extent_item *fi; |
3419 | u64 extent_len = 0; | |
3420 | u32 item_size; | |
3421 | int ret; | |
459931ec CM |
3422 | |
3423 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3424 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3425 | ||
3426 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3427 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3428 | ||
e902baac | 3429 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 3430 | return 0; |
459931ec CM |
3431 | |
3432 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
3433 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3434 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3435 | struct btrfs_file_extent_item); | |
3436 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3437 | } | |
b3b4aa74 | 3438 | btrfs_release_path(path); |
459931ec | 3439 | |
459931ec | 3440 | path->keep_locks = 1; |
ad48fd75 YZ |
3441 | path->search_for_split = 1; |
3442 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3443 | path->search_for_split = 0; |
a8df6fe6 FM |
3444 | if (ret > 0) |
3445 | ret = -EAGAIN; | |
ad48fd75 YZ |
3446 | if (ret < 0) |
3447 | goto err; | |
459931ec | 3448 | |
ad48fd75 YZ |
3449 | ret = -EAGAIN; |
3450 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
3451 | /* if our item isn't there, return now */ |
3452 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
3453 | goto err; |
3454 | ||
109f6aef | 3455 | /* the leaf has changed, it now has room. return now */ |
e902baac | 3456 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
3457 | goto err; |
3458 | ||
ad48fd75 YZ |
3459 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3460 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3461 | struct btrfs_file_extent_item); | |
3462 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3463 | goto err; | |
459931ec CM |
3464 | } |
3465 | ||
ad48fd75 | 3466 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3467 | if (ret) |
3468 | goto err; | |
459931ec | 3469 | |
ad48fd75 | 3470 | path->keep_locks = 0; |
b9473439 | 3471 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3472 | return 0; |
3473 | err: | |
3474 | path->keep_locks = 0; | |
3475 | return ret; | |
3476 | } | |
3477 | ||
25263cd7 | 3478 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 3479 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3480 | unsigned long split_offset) |
3481 | { | |
3482 | struct extent_buffer *leaf; | |
3483 | struct btrfs_item *item; | |
3484 | struct btrfs_item *new_item; | |
3485 | int slot; | |
3486 | char *buf; | |
3487 | u32 nritems; | |
3488 | u32 item_size; | |
3489 | u32 orig_offset; | |
3490 | struct btrfs_disk_key disk_key; | |
3491 | ||
b9473439 | 3492 | leaf = path->nodes[0]; |
e902baac | 3493 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 3494 | |
dd3cc16b | 3495 | item = btrfs_item_nr(path->slots[0]); |
227f3cd0 JB |
3496 | orig_offset = btrfs_item_offset_nr(leaf, path->slots[0]); |
3497 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
459931ec | 3498 | |
459931ec | 3499 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3500 | if (!buf) |
3501 | return -ENOMEM; | |
3502 | ||
459931ec CM |
3503 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3504 | path->slots[0]), item_size); | |
459931ec | 3505 | |
ad48fd75 | 3506 | slot = path->slots[0] + 1; |
459931ec | 3507 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3508 | if (slot != nritems) { |
3509 | /* shift the items */ | |
3510 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3511 | btrfs_item_nr_offset(slot), |
3512 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3513 | } |
3514 | ||
3515 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3516 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3517 | ||
dd3cc16b | 3518 | new_item = btrfs_item_nr(slot); |
459931ec CM |
3519 | |
3520 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
3521 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
3522 | ||
3523 | btrfs_set_item_offset(leaf, item, | |
3524 | orig_offset + item_size - split_offset); | |
3525 | btrfs_set_item_size(leaf, item, split_offset); | |
3526 | ||
3527 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3528 | ||
3529 | /* write the data for the start of the original item */ | |
3530 | write_extent_buffer(leaf, buf, | |
3531 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3532 | split_offset); | |
3533 | ||
3534 | /* write the data for the new item */ | |
3535 | write_extent_buffer(leaf, buf + split_offset, | |
3536 | btrfs_item_ptr_offset(leaf, slot), | |
3537 | item_size - split_offset); | |
3538 | btrfs_mark_buffer_dirty(leaf); | |
3539 | ||
e902baac | 3540 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 3541 | kfree(buf); |
ad48fd75 YZ |
3542 | return 0; |
3543 | } | |
3544 | ||
3545 | /* | |
3546 | * This function splits a single item into two items, | |
3547 | * giving 'new_key' to the new item and splitting the | |
3548 | * old one at split_offset (from the start of the item). | |
3549 | * | |
3550 | * The path may be released by this operation. After | |
3551 | * the split, the path is pointing to the old item. The | |
3552 | * new item is going to be in the same node as the old one. | |
3553 | * | |
3554 | * Note, the item being split must be smaller enough to live alone on | |
3555 | * a tree block with room for one extra struct btrfs_item | |
3556 | * | |
3557 | * This allows us to split the item in place, keeping a lock on the | |
3558 | * leaf the entire time. | |
3559 | */ | |
3560 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3561 | struct btrfs_root *root, | |
3562 | struct btrfs_path *path, | |
310712b2 | 3563 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3564 | unsigned long split_offset) |
3565 | { | |
3566 | int ret; | |
3567 | ret = setup_leaf_for_split(trans, root, path, | |
3568 | sizeof(struct btrfs_item)); | |
3569 | if (ret) | |
3570 | return ret; | |
3571 | ||
25263cd7 | 3572 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
3573 | return ret; |
3574 | } | |
3575 | ||
d352ac68 CM |
3576 | /* |
3577 | * make the item pointed to by the path smaller. new_size indicates | |
3578 | * how small to make it, and from_end tells us if we just chop bytes | |
3579 | * off the end of the item or if we shift the item to chop bytes off | |
3580 | * the front. | |
3581 | */ | |
78ac4f9e | 3582 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 3583 | { |
b18c6685 | 3584 | int slot; |
5f39d397 CM |
3585 | struct extent_buffer *leaf; |
3586 | struct btrfs_item *item; | |
b18c6685 CM |
3587 | u32 nritems; |
3588 | unsigned int data_end; | |
3589 | unsigned int old_data_start; | |
3590 | unsigned int old_size; | |
3591 | unsigned int size_diff; | |
3592 | int i; | |
cfed81a0 CM |
3593 | struct btrfs_map_token token; |
3594 | ||
5f39d397 | 3595 | leaf = path->nodes[0]; |
179e29e4 CM |
3596 | slot = path->slots[0]; |
3597 | ||
3598 | old_size = btrfs_item_size_nr(leaf, slot); | |
3599 | if (old_size == new_size) | |
143bede5 | 3600 | return; |
b18c6685 | 3601 | |
5f39d397 | 3602 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3603 | data_end = leaf_data_end(leaf); |
b18c6685 | 3604 | |
5f39d397 | 3605 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 3606 | |
b18c6685 CM |
3607 | size_diff = old_size - new_size; |
3608 | ||
3609 | BUG_ON(slot < 0); | |
3610 | BUG_ON(slot >= nritems); | |
3611 | ||
3612 | /* | |
3613 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3614 | */ | |
3615 | /* first correct the data pointers */ | |
c82f823c | 3616 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 3617 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3618 | u32 ioff; |
dd3cc16b | 3619 | item = btrfs_item_nr(i); |
db94535d | 3620 | |
cc4c13d5 DS |
3621 | ioff = btrfs_token_item_offset(&token, item); |
3622 | btrfs_set_token_item_offset(&token, item, ioff + size_diff); | |
b18c6685 | 3623 | } |
db94535d | 3624 | |
b18c6685 | 3625 | /* shift the data */ |
179e29e4 | 3626 | if (from_end) { |
3d9ec8c4 NB |
3627 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3628 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3629 | data_end, old_data_start + new_size - data_end); |
3630 | } else { | |
3631 | struct btrfs_disk_key disk_key; | |
3632 | u64 offset; | |
3633 | ||
3634 | btrfs_item_key(leaf, &disk_key, slot); | |
3635 | ||
3636 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3637 | unsigned long ptr; | |
3638 | struct btrfs_file_extent_item *fi; | |
3639 | ||
3640 | fi = btrfs_item_ptr(leaf, slot, | |
3641 | struct btrfs_file_extent_item); | |
3642 | fi = (struct btrfs_file_extent_item *)( | |
3643 | (unsigned long)fi - size_diff); | |
3644 | ||
3645 | if (btrfs_file_extent_type(leaf, fi) == | |
3646 | BTRFS_FILE_EXTENT_INLINE) { | |
3647 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3648 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 3649 | (unsigned long)fi, |
7ec20afb | 3650 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
3651 | } |
3652 | } | |
3653 | ||
3d9ec8c4 NB |
3654 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3655 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3656 | data_end, old_data_start - data_end); |
3657 | ||
3658 | offset = btrfs_disk_key_offset(&disk_key); | |
3659 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3660 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3661 | if (slot == 0) | |
b167fa91 | 3662 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 3663 | } |
5f39d397 | 3664 | |
dd3cc16b | 3665 | item = btrfs_item_nr(slot); |
5f39d397 CM |
3666 | btrfs_set_item_size(leaf, item, new_size); |
3667 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 3668 | |
e902baac | 3669 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3670 | btrfs_print_leaf(leaf); |
b18c6685 | 3671 | BUG(); |
5f39d397 | 3672 | } |
b18c6685 CM |
3673 | } |
3674 | ||
d352ac68 | 3675 | /* |
8f69dbd2 | 3676 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 3677 | */ |
c71dd880 | 3678 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 3679 | { |
6567e837 | 3680 | int slot; |
5f39d397 CM |
3681 | struct extent_buffer *leaf; |
3682 | struct btrfs_item *item; | |
6567e837 CM |
3683 | u32 nritems; |
3684 | unsigned int data_end; | |
3685 | unsigned int old_data; | |
3686 | unsigned int old_size; | |
3687 | int i; | |
cfed81a0 CM |
3688 | struct btrfs_map_token token; |
3689 | ||
5f39d397 | 3690 | leaf = path->nodes[0]; |
6567e837 | 3691 | |
5f39d397 | 3692 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3693 | data_end = leaf_data_end(leaf); |
6567e837 | 3694 | |
e902baac | 3695 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 3696 | btrfs_print_leaf(leaf); |
6567e837 | 3697 | BUG(); |
5f39d397 | 3698 | } |
6567e837 | 3699 | slot = path->slots[0]; |
5f39d397 | 3700 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
3701 | |
3702 | BUG_ON(slot < 0); | |
3326d1b0 | 3703 | if (slot >= nritems) { |
a4f78750 | 3704 | btrfs_print_leaf(leaf); |
c71dd880 | 3705 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 3706 | slot, nritems); |
290342f6 | 3707 | BUG(); |
3326d1b0 | 3708 | } |
6567e837 CM |
3709 | |
3710 | /* | |
3711 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3712 | */ | |
3713 | /* first correct the data pointers */ | |
c82f823c | 3714 | btrfs_init_map_token(&token, leaf); |
6567e837 | 3715 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3716 | u32 ioff; |
dd3cc16b | 3717 | item = btrfs_item_nr(i); |
db94535d | 3718 | |
cc4c13d5 DS |
3719 | ioff = btrfs_token_item_offset(&token, item); |
3720 | btrfs_set_token_item_offset(&token, item, ioff - data_size); | |
6567e837 | 3721 | } |
5f39d397 | 3722 | |
6567e837 | 3723 | /* shift the data */ |
3d9ec8c4 NB |
3724 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3725 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 3726 | data_end, old_data - data_end); |
5f39d397 | 3727 | |
6567e837 | 3728 | data_end = old_data; |
5f39d397 | 3729 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 3730 | item = btrfs_item_nr(slot); |
5f39d397 CM |
3731 | btrfs_set_item_size(leaf, item, old_size + data_size); |
3732 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 3733 | |
e902baac | 3734 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3735 | btrfs_print_leaf(leaf); |
6567e837 | 3736 | BUG(); |
5f39d397 | 3737 | } |
6567e837 CM |
3738 | } |
3739 | ||
da9ffb24 NB |
3740 | /** |
3741 | * setup_items_for_insert - Helper called before inserting one or more items | |
3742 | * to a leaf. Main purpose is to save stack depth by doing the bulk of the work | |
3743 | * in a function that doesn't call btrfs_search_slot | |
3744 | * | |
3745 | * @root: root we are inserting items to | |
3746 | * @path: points to the leaf/slot where we are going to insert new items | |
b7ef5f3a | 3747 | * @batch: information about the batch of items to insert |
74123bd7 | 3748 | */ |
f0641656 FM |
3749 | static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
3750 | const struct btrfs_item_batch *batch) | |
be0e5c09 | 3751 | { |
0b246afa | 3752 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 3753 | struct btrfs_item *item; |
9c58309d | 3754 | int i; |
7518a238 | 3755 | u32 nritems; |
be0e5c09 | 3756 | unsigned int data_end; |
e2fa7227 | 3757 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3758 | struct extent_buffer *leaf; |
3759 | int slot; | |
cfed81a0 | 3760 | struct btrfs_map_token token; |
fc0d82e1 | 3761 | u32 total_size; |
cfed81a0 | 3762 | |
b7ef5f3a FM |
3763 | /* |
3764 | * Before anything else, update keys in the parent and other ancestors | |
3765 | * if needed, then release the write locks on them, so that other tasks | |
3766 | * can use them while we modify the leaf. | |
3767 | */ | |
24cdc847 | 3768 | if (path->slots[0] == 0) { |
b7ef5f3a | 3769 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); |
b167fa91 | 3770 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
3771 | } |
3772 | btrfs_unlock_up_safe(path, 1); | |
3773 | ||
5f39d397 | 3774 | leaf = path->nodes[0]; |
44871b1b | 3775 | slot = path->slots[0]; |
74123bd7 | 3776 | |
5f39d397 | 3777 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3778 | data_end = leaf_data_end(leaf); |
b7ef5f3a | 3779 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
eb60ceac | 3780 | |
e902baac | 3781 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 3782 | btrfs_print_leaf(leaf); |
0b246afa | 3783 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 3784 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 3785 | BUG(); |
d4dbff95 | 3786 | } |
5f39d397 | 3787 | |
c82f823c | 3788 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 3789 | if (slot != nritems) { |
5f39d397 | 3790 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 3791 | |
5f39d397 | 3792 | if (old_data < data_end) { |
a4f78750 | 3793 | btrfs_print_leaf(leaf); |
7269ddd2 NB |
3794 | btrfs_crit(fs_info, |
3795 | "item at slot %d with data offset %u beyond data end of leaf %u", | |
5d163e0e | 3796 | slot, old_data, data_end); |
290342f6 | 3797 | BUG(); |
5f39d397 | 3798 | } |
be0e5c09 CM |
3799 | /* |
3800 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3801 | */ | |
3802 | /* first correct the data pointers */ | |
0783fcfc | 3803 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3804 | u32 ioff; |
db94535d | 3805 | |
62e85577 | 3806 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
3807 | ioff = btrfs_token_item_offset(&token, item); |
3808 | btrfs_set_token_item_offset(&token, item, | |
b7ef5f3a | 3809 | ioff - batch->total_data_size); |
0783fcfc | 3810 | } |
be0e5c09 | 3811 | /* shift the items */ |
b7ef5f3a | 3812 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + batch->nr), |
5f39d397 | 3813 | btrfs_item_nr_offset(slot), |
d6025579 | 3814 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
3815 | |
3816 | /* shift the data */ | |
3d9ec8c4 | 3817 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
b7ef5f3a FM |
3818 | data_end - batch->total_data_size, |
3819 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
3820 | old_data - data_end); | |
be0e5c09 CM |
3821 | data_end = old_data; |
3822 | } | |
5f39d397 | 3823 | |
62e2749e | 3824 | /* setup the item for the new data */ |
b7ef5f3a FM |
3825 | for (i = 0; i < batch->nr; i++) { |
3826 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); | |
9c58309d | 3827 | btrfs_set_item_key(leaf, &disk_key, slot + i); |
dd3cc16b | 3828 | item = btrfs_item_nr(slot + i); |
b7ef5f3a | 3829 | data_end -= batch->data_sizes[i]; |
fc0716c2 | 3830 | btrfs_set_token_item_offset(&token, item, data_end); |
b7ef5f3a | 3831 | btrfs_set_token_item_size(&token, item, batch->data_sizes[i]); |
9c58309d | 3832 | } |
44871b1b | 3833 | |
b7ef5f3a | 3834 | btrfs_set_header_nritems(leaf, nritems + batch->nr); |
b9473439 | 3835 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 3836 | |
e902baac | 3837 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3838 | btrfs_print_leaf(leaf); |
be0e5c09 | 3839 | BUG(); |
5f39d397 | 3840 | } |
44871b1b CM |
3841 | } |
3842 | ||
f0641656 FM |
3843 | /* |
3844 | * Insert a new item into a leaf. | |
3845 | * | |
3846 | * @root: The root of the btree. | |
3847 | * @path: A path pointing to the target leaf and slot. | |
3848 | * @key: The key of the new item. | |
3849 | * @data_size: The size of the data associated with the new key. | |
3850 | */ | |
3851 | void btrfs_setup_item_for_insert(struct btrfs_root *root, | |
3852 | struct btrfs_path *path, | |
3853 | const struct btrfs_key *key, | |
3854 | u32 data_size) | |
3855 | { | |
3856 | struct btrfs_item_batch batch; | |
3857 | ||
3858 | batch.keys = key; | |
3859 | batch.data_sizes = &data_size; | |
3860 | batch.total_data_size = data_size; | |
3861 | batch.nr = 1; | |
3862 | ||
3863 | setup_items_for_insert(root, path, &batch); | |
3864 | } | |
3865 | ||
44871b1b CM |
3866 | /* |
3867 | * Given a key and some data, insert items into the tree. | |
3868 | * This does all the path init required, making room in the tree if needed. | |
3869 | */ | |
3870 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
3871 | struct btrfs_root *root, | |
3872 | struct btrfs_path *path, | |
b7ef5f3a | 3873 | const struct btrfs_item_batch *batch) |
44871b1b | 3874 | { |
44871b1b CM |
3875 | int ret = 0; |
3876 | int slot; | |
b7ef5f3a | 3877 | u32 total_size; |
44871b1b | 3878 | |
b7ef5f3a FM |
3879 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
3880 | ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); | |
44871b1b CM |
3881 | if (ret == 0) |
3882 | return -EEXIST; | |
3883 | if (ret < 0) | |
143bede5 | 3884 | return ret; |
44871b1b | 3885 | |
44871b1b CM |
3886 | slot = path->slots[0]; |
3887 | BUG_ON(slot < 0); | |
3888 | ||
b7ef5f3a | 3889 | setup_items_for_insert(root, path, batch); |
143bede5 | 3890 | return 0; |
62e2749e CM |
3891 | } |
3892 | ||
3893 | /* | |
3894 | * Given a key and some data, insert an item into the tree. | |
3895 | * This does all the path init required, making room in the tree if needed. | |
3896 | */ | |
310712b2 OS |
3897 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
3898 | const struct btrfs_key *cpu_key, void *data, | |
3899 | u32 data_size) | |
62e2749e CM |
3900 | { |
3901 | int ret = 0; | |
2c90e5d6 | 3902 | struct btrfs_path *path; |
5f39d397 CM |
3903 | struct extent_buffer *leaf; |
3904 | unsigned long ptr; | |
62e2749e | 3905 | |
2c90e5d6 | 3906 | path = btrfs_alloc_path(); |
db5b493a TI |
3907 | if (!path) |
3908 | return -ENOMEM; | |
2c90e5d6 | 3909 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 3910 | if (!ret) { |
5f39d397 CM |
3911 | leaf = path->nodes[0]; |
3912 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3913 | write_extent_buffer(leaf, data, ptr, data_size); | |
3914 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 3915 | } |
2c90e5d6 | 3916 | btrfs_free_path(path); |
aa5d6bed | 3917 | return ret; |
be0e5c09 CM |
3918 | } |
3919 | ||
f0641656 FM |
3920 | /* |
3921 | * This function duplicates an item, giving 'new_key' to the new item. | |
3922 | * It guarantees both items live in the same tree leaf and the new item is | |
3923 | * contiguous with the original item. | |
3924 | * | |
3925 | * This allows us to split a file extent in place, keeping a lock on the leaf | |
3926 | * the entire time. | |
3927 | */ | |
3928 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
3929 | struct btrfs_root *root, | |
3930 | struct btrfs_path *path, | |
3931 | const struct btrfs_key *new_key) | |
3932 | { | |
3933 | struct extent_buffer *leaf; | |
3934 | int ret; | |
3935 | u32 item_size; | |
3936 | ||
3937 | leaf = path->nodes[0]; | |
3938 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3939 | ret = setup_leaf_for_split(trans, root, path, | |
3940 | item_size + sizeof(struct btrfs_item)); | |
3941 | if (ret) | |
3942 | return ret; | |
3943 | ||
3944 | path->slots[0]++; | |
3945 | btrfs_setup_item_for_insert(root, path, new_key, item_size); | |
3946 | leaf = path->nodes[0]; | |
3947 | memcpy_extent_buffer(leaf, | |
3948 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3949 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
3950 | item_size); | |
3951 | return 0; | |
3952 | } | |
3953 | ||
74123bd7 | 3954 | /* |
5de08d7d | 3955 | * delete the pointer from a given node. |
74123bd7 | 3956 | * |
d352ac68 CM |
3957 | * the tree should have been previously balanced so the deletion does not |
3958 | * empty a node. | |
74123bd7 | 3959 | */ |
afe5fea7 TI |
3960 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
3961 | int level, int slot) | |
be0e5c09 | 3962 | { |
5f39d397 | 3963 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 3964 | u32 nritems; |
f3ea38da | 3965 | int ret; |
be0e5c09 | 3966 | |
5f39d397 | 3967 | nritems = btrfs_header_nritems(parent); |
d397712b | 3968 | if (slot != nritems - 1) { |
bf1d3425 | 3969 | if (level) { |
f3a84ccd FM |
3970 | ret = btrfs_tree_mod_log_insert_move(parent, slot, |
3971 | slot + 1, nritems - slot - 1); | |
bf1d3425 DS |
3972 | BUG_ON(ret < 0); |
3973 | } | |
5f39d397 CM |
3974 | memmove_extent_buffer(parent, |
3975 | btrfs_node_key_ptr_offset(slot), | |
3976 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
3977 | sizeof(struct btrfs_key_ptr) * |
3978 | (nritems - slot - 1)); | |
57ba86c0 | 3979 | } else if (level) { |
f3a84ccd FM |
3980 | ret = btrfs_tree_mod_log_insert_key(parent, slot, |
3981 | BTRFS_MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
57ba86c0 | 3982 | BUG_ON(ret < 0); |
bb803951 | 3983 | } |
f3ea38da | 3984 | |
7518a238 | 3985 | nritems--; |
5f39d397 | 3986 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 3987 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 3988 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 3989 | /* just turn the root into a leaf and break */ |
5f39d397 | 3990 | btrfs_set_header_level(root->node, 0); |
bb803951 | 3991 | } else if (slot == 0) { |
5f39d397 CM |
3992 | struct btrfs_disk_key disk_key; |
3993 | ||
3994 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 3995 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 3996 | } |
d6025579 | 3997 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
3998 | } |
3999 | ||
323ac95b CM |
4000 | /* |
4001 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4002 | * path->nodes[1]. |
323ac95b CM |
4003 | * |
4004 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4005 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4006 | * | |
4007 | * The path must have already been setup for deleting the leaf, including | |
4008 | * all the proper balancing. path->nodes[1] must be locked. | |
4009 | */ | |
143bede5 JM |
4010 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4011 | struct btrfs_root *root, | |
4012 | struct btrfs_path *path, | |
4013 | struct extent_buffer *leaf) | |
323ac95b | 4014 | { |
5d4f98a2 | 4015 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4016 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4017 | |
4d081c41 CM |
4018 | /* |
4019 | * btrfs_free_extent is expensive, we want to make sure we | |
4020 | * aren't holding any locks when we call it | |
4021 | */ | |
4022 | btrfs_unlock_up_safe(path, 0); | |
4023 | ||
f0486c68 YZ |
4024 | root_sub_used(root, leaf->len); |
4025 | ||
67439dad | 4026 | atomic_inc(&leaf->refs); |
7a163608 | 4027 | btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); |
3083ee2e | 4028 | free_extent_buffer_stale(leaf); |
323ac95b | 4029 | } |
74123bd7 CM |
4030 | /* |
4031 | * delete the item at the leaf level in path. If that empties | |
4032 | * the leaf, remove it from the tree | |
4033 | */ | |
85e21bac CM |
4034 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4035 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4036 | { |
0b246afa | 4037 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4038 | struct extent_buffer *leaf; |
4039 | struct btrfs_item *item; | |
ce0eac2a AM |
4040 | u32 last_off; |
4041 | u32 dsize = 0; | |
aa5d6bed CM |
4042 | int ret = 0; |
4043 | int wret; | |
85e21bac | 4044 | int i; |
7518a238 | 4045 | u32 nritems; |
be0e5c09 | 4046 | |
5f39d397 | 4047 | leaf = path->nodes[0]; |
85e21bac CM |
4048 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4049 | ||
4050 | for (i = 0; i < nr; i++) | |
4051 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4052 | ||
5f39d397 | 4053 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4054 | |
85e21bac | 4055 | if (slot + nr != nritems) { |
8f881e8c | 4056 | int data_end = leaf_data_end(leaf); |
c82f823c | 4057 | struct btrfs_map_token token; |
5f39d397 | 4058 | |
3d9ec8c4 | 4059 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4060 | data_end + dsize, |
3d9ec8c4 | 4061 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4062 | last_off - data_end); |
5f39d397 | 4063 | |
c82f823c | 4064 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4065 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4066 | u32 ioff; |
db94535d | 4067 | |
dd3cc16b | 4068 | item = btrfs_item_nr(i); |
cc4c13d5 DS |
4069 | ioff = btrfs_token_item_offset(&token, item); |
4070 | btrfs_set_token_item_offset(&token, item, ioff + dsize); | |
0783fcfc | 4071 | } |
db94535d | 4072 | |
5f39d397 | 4073 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4074 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4075 | sizeof(struct btrfs_item) * |
85e21bac | 4076 | (nritems - slot - nr)); |
be0e5c09 | 4077 | } |
85e21bac CM |
4078 | btrfs_set_header_nritems(leaf, nritems - nr); |
4079 | nritems -= nr; | |
5f39d397 | 4080 | |
74123bd7 | 4081 | /* delete the leaf if we've emptied it */ |
7518a238 | 4082 | if (nritems == 0) { |
5f39d397 CM |
4083 | if (leaf == root->node) { |
4084 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4085 | } else { |
6a884d7d | 4086 | btrfs_clean_tree_block(leaf); |
143bede5 | 4087 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4088 | } |
be0e5c09 | 4089 | } else { |
7518a238 | 4090 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4091 | if (slot == 0) { |
5f39d397 CM |
4092 | struct btrfs_disk_key disk_key; |
4093 | ||
4094 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4095 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4096 | } |
aa5d6bed | 4097 | |
74123bd7 | 4098 | /* delete the leaf if it is mostly empty */ |
0b246afa | 4099 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
4100 | /* push_leaf_left fixes the path. |
4101 | * make sure the path still points to our leaf | |
4102 | * for possible call to del_ptr below | |
4103 | */ | |
4920c9ac | 4104 | slot = path->slots[1]; |
67439dad | 4105 | atomic_inc(&leaf->refs); |
5f39d397 | 4106 | |
99d8f83c CM |
4107 | wret = push_leaf_left(trans, root, path, 1, 1, |
4108 | 1, (u32)-1); | |
54aa1f4d | 4109 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4110 | ret = wret; |
5f39d397 CM |
4111 | |
4112 | if (path->nodes[0] == leaf && | |
4113 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4114 | wret = push_leaf_right(trans, root, path, 1, |
4115 | 1, 1, 0); | |
54aa1f4d | 4116 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4117 | ret = wret; |
4118 | } | |
5f39d397 CM |
4119 | |
4120 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4121 | path->slots[1] = slot; |
143bede5 | 4122 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4123 | free_extent_buffer(leaf); |
143bede5 | 4124 | ret = 0; |
5de08d7d | 4125 | } else { |
925baedd CM |
4126 | /* if we're still in the path, make sure |
4127 | * we're dirty. Otherwise, one of the | |
4128 | * push_leaf functions must have already | |
4129 | * dirtied this buffer | |
4130 | */ | |
4131 | if (path->nodes[0] == leaf) | |
4132 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4133 | free_extent_buffer(leaf); |
be0e5c09 | 4134 | } |
d5719762 | 4135 | } else { |
5f39d397 | 4136 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4137 | } |
4138 | } | |
aa5d6bed | 4139 | return ret; |
be0e5c09 CM |
4140 | } |
4141 | ||
7bb86316 | 4142 | /* |
925baedd | 4143 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4144 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4145 | * returns < 0 on io errors. | |
d352ac68 CM |
4146 | * |
4147 | * This may release the path, and so you may lose any locks held at the | |
4148 | * time you call it. | |
7bb86316 | 4149 | */ |
16e7549f | 4150 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4151 | { |
925baedd CM |
4152 | struct btrfs_key key; |
4153 | struct btrfs_disk_key found_key; | |
4154 | int ret; | |
7bb86316 | 4155 | |
925baedd | 4156 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4157 | |
e8b0d724 | 4158 | if (key.offset > 0) { |
925baedd | 4159 | key.offset--; |
e8b0d724 | 4160 | } else if (key.type > 0) { |
925baedd | 4161 | key.type--; |
e8b0d724 FDBM |
4162 | key.offset = (u64)-1; |
4163 | } else if (key.objectid > 0) { | |
925baedd | 4164 | key.objectid--; |
e8b0d724 FDBM |
4165 | key.type = (u8)-1; |
4166 | key.offset = (u64)-1; | |
4167 | } else { | |
925baedd | 4168 | return 1; |
e8b0d724 | 4169 | } |
7bb86316 | 4170 | |
b3b4aa74 | 4171 | btrfs_release_path(path); |
925baedd CM |
4172 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4173 | if (ret < 0) | |
4174 | return ret; | |
4175 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4176 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
4177 | /* |
4178 | * We might have had an item with the previous key in the tree right | |
4179 | * before we released our path. And after we released our path, that | |
4180 | * item might have been pushed to the first slot (0) of the leaf we | |
4181 | * were holding due to a tree balance. Alternatively, an item with the | |
4182 | * previous key can exist as the only element of a leaf (big fat item). | |
4183 | * Therefore account for these 2 cases, so that our callers (like | |
4184 | * btrfs_previous_item) don't miss an existing item with a key matching | |
4185 | * the previous key we computed above. | |
4186 | */ | |
4187 | if (ret <= 0) | |
925baedd CM |
4188 | return 0; |
4189 | return 1; | |
7bb86316 CM |
4190 | } |
4191 | ||
3f157a2f CM |
4192 | /* |
4193 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4194 | * for nodes or leaves that are have a minimum transaction id. |
4195 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4196 | * |
4197 | * This does not cow, but it does stuff the starting key it finds back | |
4198 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4199 | * key and get a writable path. | |
4200 | * | |
3f157a2f CM |
4201 | * This honors path->lowest_level to prevent descent past a given level |
4202 | * of the tree. | |
4203 | * | |
d352ac68 CM |
4204 | * min_trans indicates the oldest transaction that you are interested |
4205 | * in walking through. Any nodes or leaves older than min_trans are | |
4206 | * skipped over (without reading them). | |
4207 | * | |
3f157a2f CM |
4208 | * returns zero if something useful was found, < 0 on error and 1 if there |
4209 | * was nothing in the tree that matched the search criteria. | |
4210 | */ | |
4211 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 4212 | struct btrfs_path *path, |
3f157a2f CM |
4213 | u64 min_trans) |
4214 | { | |
4215 | struct extent_buffer *cur; | |
4216 | struct btrfs_key found_key; | |
4217 | int slot; | |
9652480b | 4218 | int sret; |
3f157a2f CM |
4219 | u32 nritems; |
4220 | int level; | |
4221 | int ret = 1; | |
f98de9b9 | 4222 | int keep_locks = path->keep_locks; |
3f157a2f | 4223 | |
f98de9b9 | 4224 | path->keep_locks = 1; |
3f157a2f | 4225 | again: |
bd681513 | 4226 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4227 | level = btrfs_header_level(cur); |
e02119d5 | 4228 | WARN_ON(path->nodes[level]); |
3f157a2f | 4229 | path->nodes[level] = cur; |
bd681513 | 4230 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4231 | |
4232 | if (btrfs_header_generation(cur) < min_trans) { | |
4233 | ret = 1; | |
4234 | goto out; | |
4235 | } | |
d397712b | 4236 | while (1) { |
3f157a2f CM |
4237 | nritems = btrfs_header_nritems(cur); |
4238 | level = btrfs_header_level(cur); | |
e3b83361 | 4239 | sret = btrfs_bin_search(cur, min_key, &slot); |
cbca7d59 FM |
4240 | if (sret < 0) { |
4241 | ret = sret; | |
4242 | goto out; | |
4243 | } | |
3f157a2f | 4244 | |
323ac95b CM |
4245 | /* at the lowest level, we're done, setup the path and exit */ |
4246 | if (level == path->lowest_level) { | |
e02119d5 CM |
4247 | if (slot >= nritems) |
4248 | goto find_next_key; | |
3f157a2f CM |
4249 | ret = 0; |
4250 | path->slots[level] = slot; | |
4251 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4252 | goto out; | |
4253 | } | |
9652480b Y |
4254 | if (sret && slot > 0) |
4255 | slot--; | |
3f157a2f | 4256 | /* |
de78b51a | 4257 | * check this node pointer against the min_trans parameters. |
260db43c | 4258 | * If it is too old, skip to the next one. |
3f157a2f | 4259 | */ |
d397712b | 4260 | while (slot < nritems) { |
3f157a2f | 4261 | u64 gen; |
e02119d5 | 4262 | |
3f157a2f CM |
4263 | gen = btrfs_node_ptr_generation(cur, slot); |
4264 | if (gen < min_trans) { | |
4265 | slot++; | |
4266 | continue; | |
4267 | } | |
de78b51a | 4268 | break; |
3f157a2f | 4269 | } |
e02119d5 | 4270 | find_next_key: |
3f157a2f CM |
4271 | /* |
4272 | * we didn't find a candidate key in this node, walk forward | |
4273 | * and find another one | |
4274 | */ | |
4275 | if (slot >= nritems) { | |
e02119d5 CM |
4276 | path->slots[level] = slot; |
4277 | sret = btrfs_find_next_key(root, path, min_key, level, | |
de78b51a | 4278 | min_trans); |
e02119d5 | 4279 | if (sret == 0) { |
b3b4aa74 | 4280 | btrfs_release_path(path); |
3f157a2f CM |
4281 | goto again; |
4282 | } else { | |
4283 | goto out; | |
4284 | } | |
4285 | } | |
4286 | /* save our key for returning back */ | |
4287 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4288 | path->slots[level] = slot; | |
4289 | if (level == path->lowest_level) { | |
4290 | ret = 0; | |
3f157a2f CM |
4291 | goto out; |
4292 | } | |
4b231ae4 | 4293 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
4294 | if (IS_ERR(cur)) { |
4295 | ret = PTR_ERR(cur); | |
4296 | goto out; | |
4297 | } | |
3f157a2f | 4298 | |
bd681513 | 4299 | btrfs_tree_read_lock(cur); |
b4ce94de | 4300 | |
bd681513 | 4301 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4302 | path->nodes[level - 1] = cur; |
f7c79f30 | 4303 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4304 | } |
4305 | out: | |
f98de9b9 FM |
4306 | path->keep_locks = keep_locks; |
4307 | if (ret == 0) { | |
4308 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
3f157a2f | 4309 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 4310 | } |
3f157a2f CM |
4311 | return ret; |
4312 | } | |
4313 | ||
4314 | /* | |
4315 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4316 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 4317 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
4318 | * |
4319 | * 0 is returned if another key is found, < 0 if there are any errors | |
4320 | * and 1 is returned if there are no higher keys in the tree | |
4321 | * | |
4322 | * path->keep_locks should be set to 1 on the search made before | |
4323 | * calling this function. | |
4324 | */ | |
e7a84565 | 4325 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 4326 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 4327 | { |
e7a84565 CM |
4328 | int slot; |
4329 | struct extent_buffer *c; | |
4330 | ||
6a9fb468 | 4331 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 4332 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4333 | if (!path->nodes[level]) |
4334 | return 1; | |
4335 | ||
4336 | slot = path->slots[level] + 1; | |
4337 | c = path->nodes[level]; | |
3f157a2f | 4338 | next: |
e7a84565 | 4339 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4340 | int ret; |
4341 | int orig_lowest; | |
4342 | struct btrfs_key cur_key; | |
4343 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4344 | !path->nodes[level + 1]) | |
e7a84565 | 4345 | return 1; |
33c66f43 | 4346 | |
6a9fb468 | 4347 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
4348 | level++; |
4349 | continue; | |
4350 | } | |
4351 | ||
4352 | slot = btrfs_header_nritems(c) - 1; | |
4353 | if (level == 0) | |
4354 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4355 | else | |
4356 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4357 | ||
4358 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4359 | btrfs_release_path(path); |
33c66f43 YZ |
4360 | path->lowest_level = level; |
4361 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4362 | 0, 0); | |
4363 | path->lowest_level = orig_lowest; | |
4364 | if (ret < 0) | |
4365 | return ret; | |
4366 | ||
4367 | c = path->nodes[level]; | |
4368 | slot = path->slots[level]; | |
4369 | if (ret == 0) | |
4370 | slot++; | |
4371 | goto next; | |
e7a84565 | 4372 | } |
33c66f43 | 4373 | |
e7a84565 CM |
4374 | if (level == 0) |
4375 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 4376 | else { |
3f157a2f CM |
4377 | u64 gen = btrfs_node_ptr_generation(c, slot); |
4378 | ||
3f157a2f CM |
4379 | if (gen < min_trans) { |
4380 | slot++; | |
4381 | goto next; | |
4382 | } | |
e7a84565 | 4383 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4384 | } |
e7a84565 CM |
4385 | return 0; |
4386 | } | |
4387 | return 1; | |
4388 | } | |
4389 | ||
3d7806ec JS |
4390 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, |
4391 | u64 time_seq) | |
d97e63b6 CM |
4392 | { |
4393 | int slot; | |
8e73f275 | 4394 | int level; |
5f39d397 | 4395 | struct extent_buffer *c; |
8e73f275 | 4396 | struct extent_buffer *next; |
925baedd CM |
4397 | struct btrfs_key key; |
4398 | u32 nritems; | |
4399 | int ret; | |
0e46318d | 4400 | int i; |
925baedd CM |
4401 | |
4402 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 4403 | if (nritems == 0) |
925baedd | 4404 | return 1; |
925baedd | 4405 | |
8e73f275 CM |
4406 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4407 | again: | |
4408 | level = 1; | |
4409 | next = NULL; | |
b3b4aa74 | 4410 | btrfs_release_path(path); |
8e73f275 | 4411 | |
a2135011 | 4412 | path->keep_locks = 1; |
8e73f275 | 4413 | |
3d7806ec JS |
4414 | if (time_seq) |
4415 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
4416 | else | |
4417 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
4418 | path->keep_locks = 0; |
4419 | ||
4420 | if (ret < 0) | |
4421 | return ret; | |
4422 | ||
a2135011 | 4423 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4424 | /* |
4425 | * by releasing the path above we dropped all our locks. A balance | |
4426 | * could have added more items next to the key that used to be | |
4427 | * at the very end of the block. So, check again here and | |
4428 | * advance the path if there are now more items available. | |
4429 | */ | |
a2135011 | 4430 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4431 | if (ret == 0) |
4432 | path->slots[0]++; | |
8e73f275 | 4433 | ret = 0; |
925baedd CM |
4434 | goto done; |
4435 | } | |
0b43e04f LB |
4436 | /* |
4437 | * So the above check misses one case: | |
4438 | * - after releasing the path above, someone has removed the item that | |
4439 | * used to be at the very end of the block, and balance between leafs | |
4440 | * gets another one with bigger key.offset to replace it. | |
4441 | * | |
4442 | * This one should be returned as well, or we can get leaf corruption | |
4443 | * later(esp. in __btrfs_drop_extents()). | |
4444 | * | |
4445 | * And a bit more explanation about this check, | |
4446 | * with ret > 0, the key isn't found, the path points to the slot | |
4447 | * where it should be inserted, so the path->slots[0] item must be the | |
4448 | * bigger one. | |
4449 | */ | |
4450 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
4451 | ret = 0; | |
4452 | goto done; | |
4453 | } | |
d97e63b6 | 4454 | |
d397712b | 4455 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4456 | if (!path->nodes[level]) { |
4457 | ret = 1; | |
4458 | goto done; | |
4459 | } | |
5f39d397 | 4460 | |
d97e63b6 CM |
4461 | slot = path->slots[level] + 1; |
4462 | c = path->nodes[level]; | |
5f39d397 | 4463 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4464 | level++; |
8e73f275 CM |
4465 | if (level == BTRFS_MAX_LEVEL) { |
4466 | ret = 1; | |
4467 | goto done; | |
4468 | } | |
d97e63b6 CM |
4469 | continue; |
4470 | } | |
5f39d397 | 4471 | |
0e46318d JB |
4472 | |
4473 | /* | |
4474 | * Our current level is where we're going to start from, and to | |
4475 | * make sure lockdep doesn't complain we need to drop our locks | |
4476 | * and nodes from 0 to our current level. | |
4477 | */ | |
4478 | for (i = 0; i < level; i++) { | |
4479 | if (path->locks[level]) { | |
4480 | btrfs_tree_read_unlock(path->nodes[i]); | |
4481 | path->locks[i] = 0; | |
4482 | } | |
4483 | free_extent_buffer(path->nodes[i]); | |
4484 | path->nodes[i] = NULL; | |
925baedd | 4485 | } |
5f39d397 | 4486 | |
8e73f275 | 4487 | next = c; |
d07b8528 | 4488 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4489 | slot, &key); |
8e73f275 CM |
4490 | if (ret == -EAGAIN) |
4491 | goto again; | |
5f39d397 | 4492 | |
76a05b35 | 4493 | if (ret < 0) { |
b3b4aa74 | 4494 | btrfs_release_path(path); |
76a05b35 CM |
4495 | goto done; |
4496 | } | |
4497 | ||
5cd57b2c | 4498 | if (!path->skip_locking) { |
bd681513 | 4499 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
4500 | if (!ret && time_seq) { |
4501 | /* | |
4502 | * If we don't get the lock, we may be racing | |
4503 | * with push_leaf_left, holding that lock while | |
4504 | * itself waiting for the leaf we've currently | |
4505 | * locked. To solve this situation, we give up | |
4506 | * on our lock and cycle. | |
4507 | */ | |
cf538830 | 4508 | free_extent_buffer(next); |
d42244a0 JS |
4509 | btrfs_release_path(path); |
4510 | cond_resched(); | |
4511 | goto again; | |
4512 | } | |
0e46318d JB |
4513 | if (!ret) |
4514 | btrfs_tree_read_lock(next); | |
5cd57b2c | 4515 | } |
d97e63b6 CM |
4516 | break; |
4517 | } | |
4518 | path->slots[level] = slot; | |
d397712b | 4519 | while (1) { |
d97e63b6 | 4520 | level--; |
d97e63b6 CM |
4521 | path->nodes[level] = next; |
4522 | path->slots[level] = 0; | |
a74a4b97 | 4523 | if (!path->skip_locking) |
ffeb03cf | 4524 | path->locks[level] = BTRFS_READ_LOCK; |
d97e63b6 CM |
4525 | if (!level) |
4526 | break; | |
b4ce94de | 4527 | |
d07b8528 | 4528 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4529 | 0, &key); |
8e73f275 CM |
4530 | if (ret == -EAGAIN) |
4531 | goto again; | |
4532 | ||
76a05b35 | 4533 | if (ret < 0) { |
b3b4aa74 | 4534 | btrfs_release_path(path); |
76a05b35 CM |
4535 | goto done; |
4536 | } | |
4537 | ||
ffeb03cf | 4538 | if (!path->skip_locking) |
0e46318d | 4539 | btrfs_tree_read_lock(next); |
d97e63b6 | 4540 | } |
8e73f275 | 4541 | ret = 0; |
925baedd | 4542 | done: |
f7c79f30 | 4543 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
4544 | |
4545 | return ret; | |
d97e63b6 | 4546 | } |
0b86a832 | 4547 | |
3f157a2f CM |
4548 | /* |
4549 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4550 | * searching until it gets past min_objectid or finds an item of 'type' | |
4551 | * | |
4552 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4553 | */ | |
0b86a832 CM |
4554 | int btrfs_previous_item(struct btrfs_root *root, |
4555 | struct btrfs_path *path, u64 min_objectid, | |
4556 | int type) | |
4557 | { | |
4558 | struct btrfs_key found_key; | |
4559 | struct extent_buffer *leaf; | |
e02119d5 | 4560 | u32 nritems; |
0b86a832 CM |
4561 | int ret; |
4562 | ||
d397712b | 4563 | while (1) { |
0b86a832 CM |
4564 | if (path->slots[0] == 0) { |
4565 | ret = btrfs_prev_leaf(root, path); | |
4566 | if (ret != 0) | |
4567 | return ret; | |
4568 | } else { | |
4569 | path->slots[0]--; | |
4570 | } | |
4571 | leaf = path->nodes[0]; | |
e02119d5 CM |
4572 | nritems = btrfs_header_nritems(leaf); |
4573 | if (nritems == 0) | |
4574 | return 1; | |
4575 | if (path->slots[0] == nritems) | |
4576 | path->slots[0]--; | |
4577 | ||
0b86a832 | 4578 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4579 | if (found_key.objectid < min_objectid) |
4580 | break; | |
0a4eefbb YZ |
4581 | if (found_key.type == type) |
4582 | return 0; | |
e02119d5 CM |
4583 | if (found_key.objectid == min_objectid && |
4584 | found_key.type < type) | |
4585 | break; | |
0b86a832 CM |
4586 | } |
4587 | return 1; | |
4588 | } | |
ade2e0b3 WS |
4589 | |
4590 | /* | |
4591 | * search in extent tree to find a previous Metadata/Data extent item with | |
4592 | * min objecitd. | |
4593 | * | |
4594 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4595 | */ | |
4596 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
4597 | struct btrfs_path *path, u64 min_objectid) | |
4598 | { | |
4599 | struct btrfs_key found_key; | |
4600 | struct extent_buffer *leaf; | |
4601 | u32 nritems; | |
4602 | int ret; | |
4603 | ||
4604 | while (1) { | |
4605 | if (path->slots[0] == 0) { | |
ade2e0b3 WS |
4606 | ret = btrfs_prev_leaf(root, path); |
4607 | if (ret != 0) | |
4608 | return ret; | |
4609 | } else { | |
4610 | path->slots[0]--; | |
4611 | } | |
4612 | leaf = path->nodes[0]; | |
4613 | nritems = btrfs_header_nritems(leaf); | |
4614 | if (nritems == 0) | |
4615 | return 1; | |
4616 | if (path->slots[0] == nritems) | |
4617 | path->slots[0]--; | |
4618 | ||
4619 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4620 | if (found_key.objectid < min_objectid) | |
4621 | break; | |
4622 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
4623 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
4624 | return 0; | |
4625 | if (found_key.objectid == min_objectid && | |
4626 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
4627 | break; | |
4628 | } | |
4629 | return 1; | |
4630 | } |