Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 | 2 | /* |
d352ac68 | 3 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
4 | */ |
5 | ||
a6b6e75e | 6 | #include <linux/sched.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
bd989ba3 | 8 | #include <linux/rbtree.h> |
adf02123 | 9 | #include <linux/mm.h> |
e41d12f5 | 10 | #include <linux/error-injection.h> |
eb60ceac CM |
11 | #include "ctree.h" |
12 | #include "disk-io.h" | |
7f5c1516 | 13 | #include "transaction.h" |
5f39d397 | 14 | #include "print-tree.h" |
925baedd | 15 | #include "locking.h" |
de37aa51 | 16 | #include "volumes.h" |
f616f5cd | 17 | #include "qgroup.h" |
f3a84ccd | 18 | #include "tree-mod-log.h" |
88c602ab | 19 | #include "tree-checker.h" |
9a8dd150 | 20 | |
e089f05c CM |
21 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
22 | *root, struct btrfs_path *path, int level); | |
310712b2 OS |
23 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
24 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
25 | int data_size, int extend); | |
5f39d397 | 26 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 27 | struct extent_buffer *dst, |
971a1f66 | 28 | struct extent_buffer *src, int empty); |
5f39d397 | 29 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
30 | struct extent_buffer *dst_buf, |
31 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
32 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
33 | int level, int slot); | |
d97e63b6 | 34 | |
af024ed2 JT |
35 | static const struct btrfs_csums { |
36 | u16 size; | |
59a0fcdb DS |
37 | const char name[10]; |
38 | const char driver[12]; | |
af024ed2 JT |
39 | } btrfs_csums[] = { |
40 | [BTRFS_CSUM_TYPE_CRC32] = { .size = 4, .name = "crc32c" }, | |
3951e7f0 | 41 | [BTRFS_CSUM_TYPE_XXHASH] = { .size = 8, .name = "xxhash64" }, |
3831bf00 | 42 | [BTRFS_CSUM_TYPE_SHA256] = { .size = 32, .name = "sha256" }, |
352ae07b DS |
43 | [BTRFS_CSUM_TYPE_BLAKE2] = { .size = 32, .name = "blake2b", |
44 | .driver = "blake2b-256" }, | |
af024ed2 JT |
45 | }; |
46 | ||
47 | int btrfs_super_csum_size(const struct btrfs_super_block *s) | |
48 | { | |
49 | u16 t = btrfs_super_csum_type(s); | |
50 | /* | |
51 | * csum type is validated at mount time | |
52 | */ | |
53 | return btrfs_csums[t].size; | |
54 | } | |
55 | ||
56 | const char *btrfs_super_csum_name(u16 csum_type) | |
57 | { | |
58 | /* csum type is validated at mount time */ | |
59 | return btrfs_csums[csum_type].name; | |
60 | } | |
61 | ||
b4e967be DS |
62 | /* |
63 | * Return driver name if defined, otherwise the name that's also a valid driver | |
64 | * name | |
65 | */ | |
66 | const char *btrfs_super_csum_driver(u16 csum_type) | |
67 | { | |
68 | /* csum type is validated at mount time */ | |
59a0fcdb DS |
69 | return btrfs_csums[csum_type].driver[0] ? |
70 | btrfs_csums[csum_type].driver : | |
b4e967be DS |
71 | btrfs_csums[csum_type].name; |
72 | } | |
73 | ||
604997b4 | 74 | size_t __attribute_const__ btrfs_get_num_csums(void) |
f7cea56c DS |
75 | { |
76 | return ARRAY_SIZE(btrfs_csums); | |
77 | } | |
78 | ||
df24a2b9 | 79 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 80 | { |
e2c89907 | 81 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
82 | } |
83 | ||
d352ac68 | 84 | /* this also releases the path */ |
df24a2b9 | 85 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 86 | { |
ff175d57 JJ |
87 | if (!p) |
88 | return; | |
b3b4aa74 | 89 | btrfs_release_path(p); |
df24a2b9 | 90 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
91 | } |
92 | ||
d352ac68 CM |
93 | /* |
94 | * path release drops references on the extent buffers in the path | |
95 | * and it drops any locks held by this path | |
96 | * | |
97 | * It is safe to call this on paths that no locks or extent buffers held. | |
98 | */ | |
b3b4aa74 | 99 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
100 | { |
101 | int i; | |
a2135011 | 102 | |
234b63a0 | 103 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 104 | p->slots[i] = 0; |
eb60ceac | 105 | if (!p->nodes[i]) |
925baedd CM |
106 | continue; |
107 | if (p->locks[i]) { | |
bd681513 | 108 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
109 | p->locks[i] = 0; |
110 | } | |
5f39d397 | 111 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 112 | p->nodes[i] = NULL; |
eb60ceac CM |
113 | } |
114 | } | |
115 | ||
8bb808c6 DS |
116 | /* |
117 | * We want the transaction abort to print stack trace only for errors where the | |
118 | * cause could be a bug, eg. due to ENOSPC, and not for common errors that are | |
119 | * caused by external factors. | |
120 | */ | |
121 | bool __cold abort_should_print_stack(int errno) | |
122 | { | |
123 | switch (errno) { | |
124 | case -EIO: | |
125 | case -EROFS: | |
126 | case -ENOMEM: | |
127 | return false; | |
128 | } | |
129 | return true; | |
130 | } | |
131 | ||
d352ac68 CM |
132 | /* |
133 | * safely gets a reference on the root node of a tree. A lock | |
134 | * is not taken, so a concurrent writer may put a different node | |
135 | * at the root of the tree. See btrfs_lock_root_node for the | |
136 | * looping required. | |
137 | * | |
138 | * The extent buffer returned by this has a reference taken, so | |
139 | * it won't disappear. It may stop being the root of the tree | |
140 | * at any time because there are no locks held. | |
141 | */ | |
925baedd CM |
142 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
143 | { | |
144 | struct extent_buffer *eb; | |
240f62c8 | 145 | |
3083ee2e JB |
146 | while (1) { |
147 | rcu_read_lock(); | |
148 | eb = rcu_dereference(root->node); | |
149 | ||
150 | /* | |
151 | * RCU really hurts here, we could free up the root node because | |
01327610 | 152 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
153 | * the inc_not_zero dance and if it doesn't work then |
154 | * synchronize_rcu and try again. | |
155 | */ | |
156 | if (atomic_inc_not_zero(&eb->refs)) { | |
157 | rcu_read_unlock(); | |
158 | break; | |
159 | } | |
160 | rcu_read_unlock(); | |
161 | synchronize_rcu(); | |
162 | } | |
925baedd CM |
163 | return eb; |
164 | } | |
165 | ||
92a7cc42 QW |
166 | /* |
167 | * Cowonly root (not-shareable trees, everything not subvolume or reloc roots), | |
168 | * just get put onto a simple dirty list. Transaction walks this list to make | |
169 | * sure they get properly updated on disk. | |
d352ac68 | 170 | */ |
0b86a832 CM |
171 | static void add_root_to_dirty_list(struct btrfs_root *root) |
172 | { | |
0b246afa JM |
173 | struct btrfs_fs_info *fs_info = root->fs_info; |
174 | ||
e7070be1 JB |
175 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
176 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
177 | return; | |
178 | ||
0b246afa | 179 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
180 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
181 | /* Want the extent tree to be the last on the list */ | |
4fd786e6 | 182 | if (root->root_key.objectid == BTRFS_EXTENT_TREE_OBJECTID) |
e7070be1 | 183 | list_move_tail(&root->dirty_list, |
0b246afa | 184 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
185 | else |
186 | list_move(&root->dirty_list, | |
0b246afa | 187 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 188 | } |
0b246afa | 189 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
190 | } |
191 | ||
d352ac68 CM |
192 | /* |
193 | * used by snapshot creation to make a copy of a root for a tree with | |
194 | * a given objectid. The buffer with the new root node is returned in | |
195 | * cow_ret, and this func returns zero on success or a negative error code. | |
196 | */ | |
be20aa9d CM |
197 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
198 | struct btrfs_root *root, | |
199 | struct extent_buffer *buf, | |
200 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
201 | { | |
0b246afa | 202 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 203 | struct extent_buffer *cow; |
be20aa9d CM |
204 | int ret = 0; |
205 | int level; | |
5d4f98a2 | 206 | struct btrfs_disk_key disk_key; |
be20aa9d | 207 | |
92a7cc42 | 208 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 209 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 210 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 211 | trans->transid != root->last_trans); |
be20aa9d CM |
212 | |
213 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
214 | if (level == 0) |
215 | btrfs_item_key(buf, &disk_key, 0); | |
216 | else | |
217 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 218 | |
4d75f8a9 | 219 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
cf6f34aa JB |
220 | &disk_key, level, buf->start, 0, |
221 | BTRFS_NESTING_NEW_ROOT); | |
5d4f98a2 | 222 | if (IS_ERR(cow)) |
be20aa9d CM |
223 | return PTR_ERR(cow); |
224 | ||
58e8012c | 225 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
226 | btrfs_set_header_bytenr(cow, cow->start); |
227 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
228 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
229 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
230 | BTRFS_HEADER_FLAG_RELOC); | |
231 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
232 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
233 | else | |
234 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 235 | |
de37aa51 | 236 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 237 | |
be20aa9d | 238 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 239 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 240 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 241 | else |
e339a6b0 | 242 | ret = btrfs_inc_ref(trans, root, cow, 0); |
867ed321 | 243 | if (ret) { |
72c9925f FM |
244 | btrfs_tree_unlock(cow); |
245 | free_extent_buffer(cow); | |
867ed321 | 246 | btrfs_abort_transaction(trans, ret); |
be20aa9d | 247 | return ret; |
867ed321 | 248 | } |
be20aa9d CM |
249 | |
250 | btrfs_mark_buffer_dirty(cow); | |
251 | *cow_ret = cow; | |
252 | return 0; | |
253 | } | |
254 | ||
5d4f98a2 YZ |
255 | /* |
256 | * check if the tree block can be shared by multiple trees | |
257 | */ | |
258 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
259 | struct extent_buffer *buf) | |
260 | { | |
261 | /* | |
92a7cc42 QW |
262 | * Tree blocks not in shareable trees and tree roots are never shared. |
263 | * If a block was allocated after the last snapshot and the block was | |
264 | * not allocated by tree relocation, we know the block is not shared. | |
5d4f98a2 | 265 | */ |
92a7cc42 | 266 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
5d4f98a2 YZ |
267 | buf != root->node && buf != root->commit_root && |
268 | (btrfs_header_generation(buf) <= | |
269 | btrfs_root_last_snapshot(&root->root_item) || | |
270 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
271 | return 1; | |
a79865c6 | 272 | |
5d4f98a2 YZ |
273 | return 0; |
274 | } | |
275 | ||
276 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
277 | struct btrfs_root *root, | |
278 | struct extent_buffer *buf, | |
f0486c68 YZ |
279 | struct extent_buffer *cow, |
280 | int *last_ref) | |
5d4f98a2 | 281 | { |
0b246afa | 282 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
283 | u64 refs; |
284 | u64 owner; | |
285 | u64 flags; | |
286 | u64 new_flags = 0; | |
287 | int ret; | |
288 | ||
289 | /* | |
290 | * Backrefs update rules: | |
291 | * | |
292 | * Always use full backrefs for extent pointers in tree block | |
293 | * allocated by tree relocation. | |
294 | * | |
295 | * If a shared tree block is no longer referenced by its owner | |
296 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
297 | * use full backrefs for extent pointers in tree block. | |
298 | * | |
299 | * If a tree block is been relocating | |
300 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
301 | * use full backrefs for extent pointers in tree block. | |
302 | * The reason for this is some operations (such as drop tree) | |
303 | * are only allowed for blocks use full backrefs. | |
304 | */ | |
305 | ||
306 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 307 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
308 | btrfs_header_level(buf), 1, |
309 | &refs, &flags); | |
be1a5564 MF |
310 | if (ret) |
311 | return ret; | |
e5df9573 MF |
312 | if (refs == 0) { |
313 | ret = -EROFS; | |
0b246afa | 314 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
315 | return ret; |
316 | } | |
5d4f98a2 YZ |
317 | } else { |
318 | refs = 1; | |
319 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
320 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
321 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
322 | else | |
323 | flags = 0; | |
324 | } | |
325 | ||
326 | owner = btrfs_header_owner(buf); | |
327 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
328 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
329 | ||
330 | if (refs > 1) { | |
331 | if ((owner == root->root_key.objectid || | |
332 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
333 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 334 | ret = btrfs_inc_ref(trans, root, buf, 1); |
692826b2 JM |
335 | if (ret) |
336 | return ret; | |
5d4f98a2 YZ |
337 | |
338 | if (root->root_key.objectid == | |
339 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 340 | ret = btrfs_dec_ref(trans, root, buf, 0); |
692826b2 JM |
341 | if (ret) |
342 | return ret; | |
e339a6b0 | 343 | ret = btrfs_inc_ref(trans, root, cow, 1); |
692826b2 JM |
344 | if (ret) |
345 | return ret; | |
5d4f98a2 YZ |
346 | } |
347 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
348 | } else { | |
349 | ||
350 | if (root->root_key.objectid == | |
351 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 352 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 353 | else |
e339a6b0 | 354 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
355 | if (ret) |
356 | return ret; | |
5d4f98a2 YZ |
357 | } |
358 | if (new_flags != 0) { | |
b1c79e09 JB |
359 | int level = btrfs_header_level(buf); |
360 | ||
42c9d0b5 | 361 | ret = btrfs_set_disk_extent_flags(trans, buf, |
2fe6a5a1 | 362 | new_flags, level); |
be1a5564 MF |
363 | if (ret) |
364 | return ret; | |
5d4f98a2 YZ |
365 | } |
366 | } else { | |
367 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
368 | if (root->root_key.objectid == | |
369 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 370 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 371 | else |
e339a6b0 | 372 | ret = btrfs_inc_ref(trans, root, cow, 0); |
692826b2 JM |
373 | if (ret) |
374 | return ret; | |
e339a6b0 | 375 | ret = btrfs_dec_ref(trans, root, buf, 1); |
692826b2 JM |
376 | if (ret) |
377 | return ret; | |
5d4f98a2 | 378 | } |
6a884d7d | 379 | btrfs_clean_tree_block(buf); |
f0486c68 | 380 | *last_ref = 1; |
5d4f98a2 YZ |
381 | } |
382 | return 0; | |
383 | } | |
384 | ||
d352ac68 | 385 | /* |
d397712b CM |
386 | * does the dirty work in cow of a single block. The parent block (if |
387 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
388 | * dirty and returned locked. If you modify the block it needs to be marked | |
389 | * dirty again. | |
d352ac68 CM |
390 | * |
391 | * search_start -- an allocation hint for the new block | |
392 | * | |
d397712b CM |
393 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
394 | * bytes the allocator should try to find free next to the block it returns. | |
395 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 396 | */ |
d397712b | 397 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
398 | struct btrfs_root *root, |
399 | struct extent_buffer *buf, | |
400 | struct extent_buffer *parent, int parent_slot, | |
401 | struct extent_buffer **cow_ret, | |
9631e4cc JB |
402 | u64 search_start, u64 empty_size, |
403 | enum btrfs_lock_nesting nest) | |
02217ed2 | 404 | { |
0b246afa | 405 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 406 | struct btrfs_disk_key disk_key; |
5f39d397 | 407 | struct extent_buffer *cow; |
be1a5564 | 408 | int level, ret; |
f0486c68 | 409 | int last_ref = 0; |
925baedd | 410 | int unlock_orig = 0; |
0f5053eb | 411 | u64 parent_start = 0; |
7bb86316 | 412 | |
925baedd CM |
413 | if (*cow_ret == buf) |
414 | unlock_orig = 1; | |
415 | ||
49d0c642 | 416 | btrfs_assert_tree_write_locked(buf); |
925baedd | 417 | |
92a7cc42 | 418 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
0b246afa | 419 | trans->transid != fs_info->running_transaction->transid); |
92a7cc42 | 420 | WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
27cdeb70 | 421 | trans->transid != root->last_trans); |
5f39d397 | 422 | |
7bb86316 | 423 | level = btrfs_header_level(buf); |
31840ae1 | 424 | |
5d4f98a2 YZ |
425 | if (level == 0) |
426 | btrfs_item_key(buf, &disk_key, 0); | |
427 | else | |
428 | btrfs_node_key(buf, &disk_key, 0); | |
429 | ||
0f5053eb GR |
430 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
431 | parent_start = parent->start; | |
5d4f98a2 | 432 | |
79bd3712 FM |
433 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
434 | root->root_key.objectid, &disk_key, level, | |
435 | search_start, empty_size, nest); | |
54aa1f4d CM |
436 | if (IS_ERR(cow)) |
437 | return PTR_ERR(cow); | |
6702ed49 | 438 | |
b4ce94de CM |
439 | /* cow is set to blocking by btrfs_init_new_buffer */ |
440 | ||
58e8012c | 441 | copy_extent_buffer_full(cow, buf); |
db94535d | 442 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 443 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
444 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
445 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
446 | BTRFS_HEADER_FLAG_RELOC); | |
447 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
448 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
449 | else | |
450 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 451 | |
de37aa51 | 452 | write_extent_buffer_fsid(cow, fs_info->fs_devices->metadata_uuid); |
2b82032c | 453 | |
be1a5564 | 454 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 455 | if (ret) { |
572c83ac JB |
456 | btrfs_tree_unlock(cow); |
457 | free_extent_buffer(cow); | |
66642832 | 458 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
459 | return ret; |
460 | } | |
1a40e23b | 461 | |
92a7cc42 | 462 | if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) { |
83d4cfd4 | 463 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 464 | if (ret) { |
572c83ac JB |
465 | btrfs_tree_unlock(cow); |
466 | free_extent_buffer(cow); | |
66642832 | 467 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 468 | return ret; |
93314e3b | 469 | } |
83d4cfd4 | 470 | } |
3fd0a558 | 471 | |
02217ed2 | 472 | if (buf == root->node) { |
925baedd | 473 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
474 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
475 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
476 | parent_start = buf->start; | |
925baedd | 477 | |
67439dad | 478 | atomic_inc(&cow->refs); |
406808ab | 479 | ret = btrfs_tree_mod_log_insert_root(root->node, cow, true); |
d9d19a01 | 480 | BUG_ON(ret < 0); |
240f62c8 | 481 | rcu_assign_pointer(root->node, cow); |
925baedd | 482 | |
7a163608 FM |
483 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
484 | parent_start, last_ref); | |
5f39d397 | 485 | free_extent_buffer(buf); |
0b86a832 | 486 | add_root_to_dirty_list(root); |
02217ed2 | 487 | } else { |
5d4f98a2 | 488 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
f3a84ccd FM |
489 | btrfs_tree_mod_log_insert_key(parent, parent_slot, |
490 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
5f39d397 | 491 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 492 | cow->start); |
74493f7a CM |
493 | btrfs_set_node_ptr_generation(parent, parent_slot, |
494 | trans->transid); | |
d6025579 | 495 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 496 | if (last_ref) { |
f3a84ccd | 497 | ret = btrfs_tree_mod_log_free_eb(buf); |
5de865ee | 498 | if (ret) { |
572c83ac JB |
499 | btrfs_tree_unlock(cow); |
500 | free_extent_buffer(cow); | |
66642832 | 501 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
502 | return ret; |
503 | } | |
504 | } | |
7a163608 FM |
505 | btrfs_free_tree_block(trans, btrfs_root_id(root), buf, |
506 | parent_start, last_ref); | |
02217ed2 | 507 | } |
925baedd CM |
508 | if (unlock_orig) |
509 | btrfs_tree_unlock(buf); | |
3083ee2e | 510 | free_extent_buffer_stale(buf); |
ccd467d6 | 511 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 512 | *cow_ret = cow; |
02217ed2 CM |
513 | return 0; |
514 | } | |
515 | ||
5d4f98a2 YZ |
516 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
517 | struct btrfs_root *root, | |
518 | struct extent_buffer *buf) | |
519 | { | |
f5ee5c9a | 520 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 521 | return 0; |
fccb84c9 | 522 | |
d1980131 DS |
523 | /* Ensure we can see the FORCE_COW bit */ |
524 | smp_mb__before_atomic(); | |
f1ebcc74 LB |
525 | |
526 | /* | |
527 | * We do not need to cow a block if | |
528 | * 1) this block is not created or changed in this transaction; | |
529 | * 2) this block does not belong to TREE_RELOC tree; | |
530 | * 3) the root is not forced COW. | |
531 | * | |
532 | * What is forced COW: | |
01327610 | 533 | * when we create snapshot during committing the transaction, |
52042d8e | 534 | * after we've finished copying src root, we must COW the shared |
f1ebcc74 LB |
535 | * block to ensure the metadata consistency. |
536 | */ | |
5d4f98a2 YZ |
537 | if (btrfs_header_generation(buf) == trans->transid && |
538 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
539 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 540 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 541 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
542 | return 0; |
543 | return 1; | |
544 | } | |
545 | ||
d352ac68 CM |
546 | /* |
547 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 548 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
549 | * once per transaction, as long as it hasn't been written yet |
550 | */ | |
d397712b | 551 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
552 | struct btrfs_root *root, struct extent_buffer *buf, |
553 | struct extent_buffer *parent, int parent_slot, | |
9631e4cc JB |
554 | struct extent_buffer **cow_ret, |
555 | enum btrfs_lock_nesting nest) | |
6702ed49 | 556 | { |
0b246afa | 557 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 558 | u64 search_start; |
f510cfec | 559 | int ret; |
dc17ff8f | 560 | |
83354f07 JB |
561 | if (test_bit(BTRFS_ROOT_DELETING, &root->state)) |
562 | btrfs_err(fs_info, | |
563 | "COW'ing blocks on a fs root that's being dropped"); | |
564 | ||
0b246afa | 565 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 566 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 567 | trans->transid, |
0b246afa | 568 | fs_info->running_transaction->transid); |
31b1a2bd | 569 | |
0b246afa | 570 | if (trans->transid != fs_info->generation) |
31b1a2bd | 571 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 572 | trans->transid, fs_info->generation); |
dc17ff8f | 573 | |
5d4f98a2 | 574 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
575 | *cow_ret = buf; |
576 | return 0; | |
577 | } | |
c487685d | 578 | |
ee22184b | 579 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de | 580 | |
f616f5cd QW |
581 | /* |
582 | * Before CoWing this block for later modification, check if it's | |
583 | * the subtree root and do the delayed subtree trace if needed. | |
584 | * | |
585 | * Also We don't care about the error, as it's handled internally. | |
586 | */ | |
587 | btrfs_qgroup_trace_subtree_after_cow(trans, root, buf); | |
f510cfec | 588 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9631e4cc | 589 | parent_slot, cow_ret, search_start, 0, nest); |
1abe9b8a | 590 | |
591 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
592 | ||
f510cfec | 593 | return ret; |
6702ed49 | 594 | } |
f75e2b79 | 595 | ALLOW_ERROR_INJECTION(btrfs_cow_block, ERRNO); |
6702ed49 | 596 | |
d352ac68 CM |
597 | /* |
598 | * helper function for defrag to decide if two blocks pointed to by a | |
599 | * node are actually close by | |
600 | */ | |
6b80053d | 601 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 602 | { |
6b80053d | 603 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 604 | return 1; |
6b80053d | 605 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
606 | return 1; |
607 | return 0; | |
608 | } | |
609 | ||
ce6ef5ab DS |
610 | #ifdef __LITTLE_ENDIAN |
611 | ||
612 | /* | |
613 | * Compare two keys, on little-endian the disk order is same as CPU order and | |
614 | * we can avoid the conversion. | |
615 | */ | |
616 | static int comp_keys(const struct btrfs_disk_key *disk_key, | |
617 | const struct btrfs_key *k2) | |
618 | { | |
619 | const struct btrfs_key *k1 = (const struct btrfs_key *)disk_key; | |
620 | ||
621 | return btrfs_comp_cpu_keys(k1, k2); | |
622 | } | |
623 | ||
624 | #else | |
625 | ||
081e9573 CM |
626 | /* |
627 | * compare two keys in a memcmp fashion | |
628 | */ | |
310712b2 OS |
629 | static int comp_keys(const struct btrfs_disk_key *disk, |
630 | const struct btrfs_key *k2) | |
081e9573 CM |
631 | { |
632 | struct btrfs_key k1; | |
633 | ||
634 | btrfs_disk_key_to_cpu(&k1, disk); | |
635 | ||
20736aba | 636 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 | 637 | } |
ce6ef5ab | 638 | #endif |
081e9573 | 639 | |
f3465ca4 JB |
640 | /* |
641 | * same as comp_keys only with two btrfs_key's | |
642 | */ | |
e1f60a65 | 643 | int __pure btrfs_comp_cpu_keys(const struct btrfs_key *k1, const struct btrfs_key *k2) |
f3465ca4 JB |
644 | { |
645 | if (k1->objectid > k2->objectid) | |
646 | return 1; | |
647 | if (k1->objectid < k2->objectid) | |
648 | return -1; | |
649 | if (k1->type > k2->type) | |
650 | return 1; | |
651 | if (k1->type < k2->type) | |
652 | return -1; | |
653 | if (k1->offset > k2->offset) | |
654 | return 1; | |
655 | if (k1->offset < k2->offset) | |
656 | return -1; | |
657 | return 0; | |
658 | } | |
081e9573 | 659 | |
d352ac68 CM |
660 | /* |
661 | * this is used by the defrag code to go through all the | |
662 | * leaves pointed to by a node and reallocate them so that | |
663 | * disk order is close to key order | |
664 | */ | |
6702ed49 | 665 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 666 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 667 | int start_slot, u64 *last_ret, |
a6b6e75e | 668 | struct btrfs_key *progress) |
6702ed49 | 669 | { |
0b246afa | 670 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 671 | struct extent_buffer *cur; |
6702ed49 | 672 | u64 blocknr; |
e9d0b13b CM |
673 | u64 search_start = *last_ret; |
674 | u64 last_block = 0; | |
6702ed49 CM |
675 | u64 other; |
676 | u32 parent_nritems; | |
6702ed49 CM |
677 | int end_slot; |
678 | int i; | |
679 | int err = 0; | |
6b80053d | 680 | u32 blocksize; |
081e9573 CM |
681 | int progress_passed = 0; |
682 | struct btrfs_disk_key disk_key; | |
6702ed49 | 683 | |
0b246afa JM |
684 | WARN_ON(trans->transaction != fs_info->running_transaction); |
685 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 686 | |
6b80053d | 687 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 688 | blocksize = fs_info->nodesize; |
5dfe2be7 | 689 | end_slot = parent_nritems - 1; |
6702ed49 | 690 | |
5dfe2be7 | 691 | if (parent_nritems <= 1) |
6702ed49 CM |
692 | return 0; |
693 | ||
5dfe2be7 | 694 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 695 | int close = 1; |
a6b6e75e | 696 | |
081e9573 CM |
697 | btrfs_node_key(parent, &disk_key, i); |
698 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
699 | continue; | |
700 | ||
701 | progress_passed = 1; | |
6b80053d | 702 | blocknr = btrfs_node_blockptr(parent, i); |
e9d0b13b CM |
703 | if (last_block == 0) |
704 | last_block = blocknr; | |
5708b959 | 705 | |
6702ed49 | 706 | if (i > 0) { |
6b80053d CM |
707 | other = btrfs_node_blockptr(parent, i - 1); |
708 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 709 | } |
5dfe2be7 | 710 | if (!close && i < end_slot) { |
6b80053d CM |
711 | other = btrfs_node_blockptr(parent, i + 1); |
712 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 713 | } |
e9d0b13b CM |
714 | if (close) { |
715 | last_block = blocknr; | |
6702ed49 | 716 | continue; |
e9d0b13b | 717 | } |
6702ed49 | 718 | |
206983b7 JB |
719 | cur = btrfs_read_node_slot(parent, i); |
720 | if (IS_ERR(cur)) | |
721 | return PTR_ERR(cur); | |
e9d0b13b | 722 | if (search_start == 0) |
6b80053d | 723 | search_start = last_block; |
e9d0b13b | 724 | |
e7a84565 | 725 | btrfs_tree_lock(cur); |
6b80053d | 726 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 727 | &cur, search_start, |
6b80053d | 728 | min(16 * blocksize, |
9631e4cc JB |
729 | (end_slot - i) * blocksize), |
730 | BTRFS_NESTING_COW); | |
252c38f0 | 731 | if (err) { |
e7a84565 | 732 | btrfs_tree_unlock(cur); |
6b80053d | 733 | free_extent_buffer(cur); |
6702ed49 | 734 | break; |
252c38f0 | 735 | } |
e7a84565 CM |
736 | search_start = cur->start; |
737 | last_block = cur->start; | |
f2183bde | 738 | *last_ret = search_start; |
e7a84565 CM |
739 | btrfs_tree_unlock(cur); |
740 | free_extent_buffer(cur); | |
6702ed49 CM |
741 | } |
742 | return err; | |
743 | } | |
744 | ||
74123bd7 | 745 | /* |
fb81212c | 746 | * Search for a key in the given extent_buffer. |
5f39d397 | 747 | * |
fb81212c FM |
748 | * The lower boundary for the search is specified by the slot number @low. Use a |
749 | * value of 0 to search over the whole extent buffer. | |
74123bd7 | 750 | * |
fb81212c FM |
751 | * The slot in the extent buffer is returned via @slot. If the key exists in the |
752 | * extent buffer, then @slot will point to the slot where the key is, otherwise | |
753 | * it points to the slot where you would insert the key. | |
754 | * | |
755 | * Slot may point to the total number of items (i.e. one position beyond the last | |
756 | * key) if the key is bigger than the last key in the extent buffer. | |
74123bd7 | 757 | */ |
fb81212c | 758 | static noinline int generic_bin_search(struct extent_buffer *eb, int low, |
67d5e289 | 759 | const struct btrfs_key *key, int *slot) |
be0e5c09 | 760 | { |
fb81212c FM |
761 | unsigned long p; |
762 | int item_size; | |
67d5e289 | 763 | int high = btrfs_header_nritems(eb); |
be0e5c09 | 764 | int ret; |
5cd17f34 | 765 | const int key_size = sizeof(struct btrfs_disk_key); |
be0e5c09 | 766 | |
5e24e9af LB |
767 | if (low > high) { |
768 | btrfs_err(eb->fs_info, | |
769 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
770 | __func__, low, high, eb->start, | |
771 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
772 | return -EINVAL; | |
773 | } | |
774 | ||
fb81212c FM |
775 | if (btrfs_header_level(eb) == 0) { |
776 | p = offsetof(struct btrfs_leaf, items); | |
777 | item_size = sizeof(struct btrfs_item); | |
778 | } else { | |
779 | p = offsetof(struct btrfs_node, ptrs); | |
780 | item_size = sizeof(struct btrfs_key_ptr); | |
781 | } | |
782 | ||
d397712b | 783 | while (low < high) { |
5cd17f34 DS |
784 | unsigned long oip; |
785 | unsigned long offset; | |
786 | struct btrfs_disk_key *tmp; | |
787 | struct btrfs_disk_key unaligned; | |
788 | int mid; | |
789 | ||
be0e5c09 | 790 | mid = (low + high) / 2; |
5f39d397 | 791 | offset = p + mid * item_size; |
5cd17f34 | 792 | oip = offset_in_page(offset); |
5f39d397 | 793 | |
5cd17f34 | 794 | if (oip + key_size <= PAGE_SIZE) { |
884b07d0 | 795 | const unsigned long idx = get_eb_page_index(offset); |
5cd17f34 | 796 | char *kaddr = page_address(eb->pages[idx]); |
5f39d397 | 797 | |
884b07d0 | 798 | oip = get_eb_offset_in_page(eb, offset); |
5cd17f34 | 799 | tmp = (struct btrfs_disk_key *)(kaddr + oip); |
5f39d397 | 800 | } else { |
5cd17f34 DS |
801 | read_extent_buffer(eb, &unaligned, offset, key_size); |
802 | tmp = &unaligned; | |
5f39d397 | 803 | } |
5cd17f34 | 804 | |
be0e5c09 CM |
805 | ret = comp_keys(tmp, key); |
806 | ||
807 | if (ret < 0) | |
808 | low = mid + 1; | |
809 | else if (ret > 0) | |
810 | high = mid; | |
811 | else { | |
812 | *slot = mid; | |
813 | return 0; | |
814 | } | |
815 | } | |
816 | *slot = low; | |
817 | return 1; | |
818 | } | |
819 | ||
97571fd0 | 820 | /* |
fb81212c FM |
821 | * Simple binary search on an extent buffer. Works for both leaves and nodes, and |
822 | * always searches over the whole range of keys (slot 0 to slot 'nritems - 1'). | |
97571fd0 | 823 | */ |
a74b35ec | 824 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
e3b83361 | 825 | int *slot) |
be0e5c09 | 826 | { |
fb81212c | 827 | return generic_bin_search(eb, 0, key, slot); |
be0e5c09 CM |
828 | } |
829 | ||
f0486c68 YZ |
830 | static void root_add_used(struct btrfs_root *root, u32 size) |
831 | { | |
832 | spin_lock(&root->accounting_lock); | |
833 | btrfs_set_root_used(&root->root_item, | |
834 | btrfs_root_used(&root->root_item) + size); | |
835 | spin_unlock(&root->accounting_lock); | |
836 | } | |
837 | ||
838 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
839 | { | |
840 | spin_lock(&root->accounting_lock); | |
841 | btrfs_set_root_used(&root->root_item, | |
842 | btrfs_root_used(&root->root_item) - size); | |
843 | spin_unlock(&root->accounting_lock); | |
844 | } | |
845 | ||
d352ac68 CM |
846 | /* given a node and slot number, this reads the blocks it points to. The |
847 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 848 | */ |
4b231ae4 DS |
849 | struct extent_buffer *btrfs_read_node_slot(struct extent_buffer *parent, |
850 | int slot) | |
bb803951 | 851 | { |
ca7a79ad | 852 | int level = btrfs_header_level(parent); |
416bc658 | 853 | struct extent_buffer *eb; |
581c1760 | 854 | struct btrfs_key first_key; |
416bc658 | 855 | |
fb770ae4 LB |
856 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
857 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
858 | |
859 | BUG_ON(level == 0); | |
860 | ||
581c1760 | 861 | btrfs_node_key_to_cpu(parent, &first_key, slot); |
d0d20b0f | 862 | eb = read_tree_block(parent->fs_info, btrfs_node_blockptr(parent, slot), |
1b7ec85e | 863 | btrfs_header_owner(parent), |
581c1760 QW |
864 | btrfs_node_ptr_generation(parent, slot), |
865 | level - 1, &first_key); | |
4eb150d6 QW |
866 | if (IS_ERR(eb)) |
867 | return eb; | |
868 | if (!extent_buffer_uptodate(eb)) { | |
fb770ae4 | 869 | free_extent_buffer(eb); |
4eb150d6 | 870 | return ERR_PTR(-EIO); |
416bc658 JB |
871 | } |
872 | ||
873 | return eb; | |
bb803951 CM |
874 | } |
875 | ||
d352ac68 CM |
876 | /* |
877 | * node level balancing, used to make sure nodes are in proper order for | |
878 | * item deletion. We balance from the top down, so we have to make sure | |
879 | * that a deletion won't leave an node completely empty later on. | |
880 | */ | |
e02119d5 | 881 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
882 | struct btrfs_root *root, |
883 | struct btrfs_path *path, int level) | |
bb803951 | 884 | { |
0b246afa | 885 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
886 | struct extent_buffer *right = NULL; |
887 | struct extent_buffer *mid; | |
888 | struct extent_buffer *left = NULL; | |
889 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
890 | int ret = 0; |
891 | int wret; | |
892 | int pslot; | |
bb803951 | 893 | int orig_slot = path->slots[level]; |
79f95c82 | 894 | u64 orig_ptr; |
bb803951 | 895 | |
98e6b1eb | 896 | ASSERT(level > 0); |
bb803951 | 897 | |
5f39d397 | 898 | mid = path->nodes[level]; |
b4ce94de | 899 | |
ac5887c8 | 900 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK); |
7bb86316 CM |
901 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
902 | ||
1d4f8a0c | 903 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 904 | |
a05a9bb1 | 905 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 906 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
907 | pslot = path->slots[level + 1]; |
908 | } | |
bb803951 | 909 | |
40689478 CM |
910 | /* |
911 | * deal with the case where there is only one pointer in the root | |
912 | * by promoting the node below to a root | |
913 | */ | |
5f39d397 CM |
914 | if (!parent) { |
915 | struct extent_buffer *child; | |
bb803951 | 916 | |
5f39d397 | 917 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
918 | return 0; |
919 | ||
920 | /* promote the child to a root */ | |
4b231ae4 | 921 | child = btrfs_read_node_slot(mid, 0); |
fb770ae4 LB |
922 | if (IS_ERR(child)) { |
923 | ret = PTR_ERR(child); | |
0b246afa | 924 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
925 | goto enospc; |
926 | } | |
927 | ||
925baedd | 928 | btrfs_tree_lock(child); |
9631e4cc JB |
929 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child, |
930 | BTRFS_NESTING_COW); | |
f0486c68 YZ |
931 | if (ret) { |
932 | btrfs_tree_unlock(child); | |
933 | free_extent_buffer(child); | |
934 | goto enospc; | |
935 | } | |
2f375ab9 | 936 | |
406808ab | 937 | ret = btrfs_tree_mod_log_insert_root(root->node, child, true); |
d9d19a01 | 938 | BUG_ON(ret < 0); |
240f62c8 | 939 | rcu_assign_pointer(root->node, child); |
925baedd | 940 | |
0b86a832 | 941 | add_root_to_dirty_list(root); |
925baedd | 942 | btrfs_tree_unlock(child); |
b4ce94de | 943 | |
925baedd | 944 | path->locks[level] = 0; |
bb803951 | 945 | path->nodes[level] = NULL; |
6a884d7d | 946 | btrfs_clean_tree_block(mid); |
925baedd | 947 | btrfs_tree_unlock(mid); |
bb803951 | 948 | /* once for the path */ |
5f39d397 | 949 | free_extent_buffer(mid); |
f0486c68 YZ |
950 | |
951 | root_sub_used(root, mid->len); | |
7a163608 | 952 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
bb803951 | 953 | /* once for the root ptr */ |
3083ee2e | 954 | free_extent_buffer_stale(mid); |
f0486c68 | 955 | return 0; |
bb803951 | 956 | } |
5f39d397 | 957 | if (btrfs_header_nritems(mid) > |
0b246afa | 958 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
959 | return 0; |
960 | ||
4b231ae4 | 961 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
962 | if (IS_ERR(left)) |
963 | left = NULL; | |
964 | ||
5f39d397 | 965 | if (left) { |
bf77467a | 966 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
5f39d397 | 967 | wret = btrfs_cow_block(trans, root, left, |
9631e4cc | 968 | parent, pslot - 1, &left, |
bf59a5a2 | 969 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
970 | if (wret) { |
971 | ret = wret; | |
972 | goto enospc; | |
973 | } | |
2cc58cf2 | 974 | } |
fb770ae4 | 975 | |
4b231ae4 | 976 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
977 | if (IS_ERR(right)) |
978 | right = NULL; | |
979 | ||
5f39d397 | 980 | if (right) { |
bf77467a | 981 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
5f39d397 | 982 | wret = btrfs_cow_block(trans, root, right, |
9631e4cc | 983 | parent, pslot + 1, &right, |
bf59a5a2 | 984 | BTRFS_NESTING_RIGHT_COW); |
2cc58cf2 CM |
985 | if (wret) { |
986 | ret = wret; | |
987 | goto enospc; | |
988 | } | |
989 | } | |
990 | ||
991 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
992 | if (left) { |
993 | orig_slot += btrfs_header_nritems(left); | |
d30a668f | 994 | wret = push_node_left(trans, left, mid, 1); |
79f95c82 CM |
995 | if (wret < 0) |
996 | ret = wret; | |
bb803951 | 997 | } |
79f95c82 CM |
998 | |
999 | /* | |
1000 | * then try to empty the right most buffer into the middle | |
1001 | */ | |
5f39d397 | 1002 | if (right) { |
d30a668f | 1003 | wret = push_node_left(trans, mid, right, 1); |
54aa1f4d | 1004 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1005 | ret = wret; |
5f39d397 | 1006 | if (btrfs_header_nritems(right) == 0) { |
6a884d7d | 1007 | btrfs_clean_tree_block(right); |
925baedd | 1008 | btrfs_tree_unlock(right); |
afe5fea7 | 1009 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1010 | root_sub_used(root, right->len); |
7a163608 FM |
1011 | btrfs_free_tree_block(trans, btrfs_root_id(root), right, |
1012 | 0, 1); | |
3083ee2e | 1013 | free_extent_buffer_stale(right); |
f0486c68 | 1014 | right = NULL; |
bb803951 | 1015 | } else { |
5f39d397 CM |
1016 | struct btrfs_disk_key right_key; |
1017 | btrfs_node_key(right, &right_key, 0); | |
f3a84ccd FM |
1018 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
1019 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1020 | BUG_ON(ret < 0); |
5f39d397 CM |
1021 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1022 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1023 | } |
1024 | } | |
5f39d397 | 1025 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1026 | /* |
1027 | * we're not allowed to leave a node with one item in the | |
1028 | * tree during a delete. A deletion from lower in the tree | |
1029 | * could try to delete the only pointer in this node. | |
1030 | * So, pull some keys from the left. | |
1031 | * There has to be a left pointer at this point because | |
1032 | * otherwise we would have pulled some pointers from the | |
1033 | * right | |
1034 | */ | |
305a26af MF |
1035 | if (!left) { |
1036 | ret = -EROFS; | |
0b246afa | 1037 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1038 | goto enospc; |
1039 | } | |
55d32ed8 | 1040 | wret = balance_node_right(trans, mid, left); |
54aa1f4d | 1041 | if (wret < 0) { |
79f95c82 | 1042 | ret = wret; |
54aa1f4d CM |
1043 | goto enospc; |
1044 | } | |
bce4eae9 | 1045 | if (wret == 1) { |
d30a668f | 1046 | wret = push_node_left(trans, left, mid, 1); |
bce4eae9 CM |
1047 | if (wret < 0) |
1048 | ret = wret; | |
1049 | } | |
79f95c82 CM |
1050 | BUG_ON(wret == 1); |
1051 | } | |
5f39d397 | 1052 | if (btrfs_header_nritems(mid) == 0) { |
6a884d7d | 1053 | btrfs_clean_tree_block(mid); |
925baedd | 1054 | btrfs_tree_unlock(mid); |
afe5fea7 | 1055 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1056 | root_sub_used(root, mid->len); |
7a163608 | 1057 | btrfs_free_tree_block(trans, btrfs_root_id(root), mid, 0, 1); |
3083ee2e | 1058 | free_extent_buffer_stale(mid); |
f0486c68 | 1059 | mid = NULL; |
79f95c82 CM |
1060 | } else { |
1061 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1062 | struct btrfs_disk_key mid_key; |
1063 | btrfs_node_key(mid, &mid_key, 0); | |
f3a84ccd FM |
1064 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1065 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1066 | BUG_ON(ret < 0); |
5f39d397 CM |
1067 | btrfs_set_node_key(parent, &mid_key, pslot); |
1068 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1069 | } |
bb803951 | 1070 | |
79f95c82 | 1071 | /* update the path */ |
5f39d397 CM |
1072 | if (left) { |
1073 | if (btrfs_header_nritems(left) > orig_slot) { | |
67439dad | 1074 | atomic_inc(&left->refs); |
925baedd | 1075 | /* left was locked after cow */ |
5f39d397 | 1076 | path->nodes[level] = left; |
bb803951 CM |
1077 | path->slots[level + 1] -= 1; |
1078 | path->slots[level] = orig_slot; | |
925baedd CM |
1079 | if (mid) { |
1080 | btrfs_tree_unlock(mid); | |
5f39d397 | 1081 | free_extent_buffer(mid); |
925baedd | 1082 | } |
bb803951 | 1083 | } else { |
5f39d397 | 1084 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1085 | path->slots[level] = orig_slot; |
1086 | } | |
1087 | } | |
79f95c82 | 1088 | /* double check we haven't messed things up */ |
e20d96d6 | 1089 | if (orig_ptr != |
5f39d397 | 1090 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1091 | BUG(); |
54aa1f4d | 1092 | enospc: |
925baedd CM |
1093 | if (right) { |
1094 | btrfs_tree_unlock(right); | |
5f39d397 | 1095 | free_extent_buffer(right); |
925baedd CM |
1096 | } |
1097 | if (left) { | |
1098 | if (path->nodes[level] != left) | |
1099 | btrfs_tree_unlock(left); | |
5f39d397 | 1100 | free_extent_buffer(left); |
925baedd | 1101 | } |
bb803951 CM |
1102 | return ret; |
1103 | } | |
1104 | ||
d352ac68 CM |
1105 | /* Node balancing for insertion. Here we only split or push nodes around |
1106 | * when they are completely full. This is also done top down, so we | |
1107 | * have to be pessimistic. | |
1108 | */ | |
d397712b | 1109 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1110 | struct btrfs_root *root, |
1111 | struct btrfs_path *path, int level) | |
e66f709b | 1112 | { |
0b246afa | 1113 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1114 | struct extent_buffer *right = NULL; |
1115 | struct extent_buffer *mid; | |
1116 | struct extent_buffer *left = NULL; | |
1117 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1118 | int ret = 0; |
1119 | int wret; | |
1120 | int pslot; | |
1121 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1122 | |
1123 | if (level == 0) | |
1124 | return 1; | |
1125 | ||
5f39d397 | 1126 | mid = path->nodes[level]; |
7bb86316 | 1127 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1128 | |
a05a9bb1 | 1129 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1130 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1131 | pslot = path->slots[level + 1]; |
1132 | } | |
e66f709b | 1133 | |
5f39d397 | 1134 | if (!parent) |
e66f709b | 1135 | return 1; |
e66f709b | 1136 | |
4b231ae4 | 1137 | left = btrfs_read_node_slot(parent, pslot - 1); |
fb770ae4 LB |
1138 | if (IS_ERR(left)) |
1139 | left = NULL; | |
e66f709b CM |
1140 | |
1141 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1142 | if (left) { |
e66f709b | 1143 | u32 left_nr; |
925baedd | 1144 | |
bf77467a | 1145 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
b4ce94de | 1146 | |
5f39d397 | 1147 | left_nr = btrfs_header_nritems(left); |
0b246afa | 1148 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1149 | wret = 1; |
1150 | } else { | |
5f39d397 | 1151 | ret = btrfs_cow_block(trans, root, left, parent, |
9631e4cc | 1152 | pslot - 1, &left, |
bf59a5a2 | 1153 | BTRFS_NESTING_LEFT_COW); |
54aa1f4d CM |
1154 | if (ret) |
1155 | wret = 1; | |
1156 | else { | |
d30a668f | 1157 | wret = push_node_left(trans, left, mid, 0); |
54aa1f4d | 1158 | } |
33ade1f8 | 1159 | } |
e66f709b CM |
1160 | if (wret < 0) |
1161 | ret = wret; | |
1162 | if (wret == 0) { | |
5f39d397 | 1163 | struct btrfs_disk_key disk_key; |
e66f709b | 1164 | orig_slot += left_nr; |
5f39d397 | 1165 | btrfs_node_key(mid, &disk_key, 0); |
f3a84ccd FM |
1166 | ret = btrfs_tree_mod_log_insert_key(parent, pslot, |
1167 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1168 | BUG_ON(ret < 0); |
5f39d397 CM |
1169 | btrfs_set_node_key(parent, &disk_key, pslot); |
1170 | btrfs_mark_buffer_dirty(parent); | |
1171 | if (btrfs_header_nritems(left) > orig_slot) { | |
1172 | path->nodes[level] = left; | |
e66f709b CM |
1173 | path->slots[level + 1] -= 1; |
1174 | path->slots[level] = orig_slot; | |
925baedd | 1175 | btrfs_tree_unlock(mid); |
5f39d397 | 1176 | free_extent_buffer(mid); |
e66f709b CM |
1177 | } else { |
1178 | orig_slot -= | |
5f39d397 | 1179 | btrfs_header_nritems(left); |
e66f709b | 1180 | path->slots[level] = orig_slot; |
925baedd | 1181 | btrfs_tree_unlock(left); |
5f39d397 | 1182 | free_extent_buffer(left); |
e66f709b | 1183 | } |
e66f709b CM |
1184 | return 0; |
1185 | } | |
925baedd | 1186 | btrfs_tree_unlock(left); |
5f39d397 | 1187 | free_extent_buffer(left); |
e66f709b | 1188 | } |
4b231ae4 | 1189 | right = btrfs_read_node_slot(parent, pslot + 1); |
fb770ae4 LB |
1190 | if (IS_ERR(right)) |
1191 | right = NULL; | |
e66f709b CM |
1192 | |
1193 | /* | |
1194 | * then try to empty the right most buffer into the middle | |
1195 | */ | |
5f39d397 | 1196 | if (right) { |
33ade1f8 | 1197 | u32 right_nr; |
b4ce94de | 1198 | |
bf77467a | 1199 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
b4ce94de | 1200 | |
5f39d397 | 1201 | right_nr = btrfs_header_nritems(right); |
0b246afa | 1202 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
1203 | wret = 1; |
1204 | } else { | |
5f39d397 CM |
1205 | ret = btrfs_cow_block(trans, root, right, |
1206 | parent, pslot + 1, | |
bf59a5a2 | 1207 | &right, BTRFS_NESTING_RIGHT_COW); |
54aa1f4d CM |
1208 | if (ret) |
1209 | wret = 1; | |
1210 | else { | |
55d32ed8 | 1211 | wret = balance_node_right(trans, right, mid); |
54aa1f4d | 1212 | } |
33ade1f8 | 1213 | } |
e66f709b CM |
1214 | if (wret < 0) |
1215 | ret = wret; | |
1216 | if (wret == 0) { | |
5f39d397 CM |
1217 | struct btrfs_disk_key disk_key; |
1218 | ||
1219 | btrfs_node_key(right, &disk_key, 0); | |
f3a84ccd FM |
1220 | ret = btrfs_tree_mod_log_insert_key(parent, pslot + 1, |
1221 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_NOFS); | |
0e82bcfe | 1222 | BUG_ON(ret < 0); |
5f39d397 CM |
1223 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
1224 | btrfs_mark_buffer_dirty(parent); | |
1225 | ||
1226 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
1227 | path->nodes[level] = right; | |
e66f709b CM |
1228 | path->slots[level + 1] += 1; |
1229 | path->slots[level] = orig_slot - | |
5f39d397 | 1230 | btrfs_header_nritems(mid); |
925baedd | 1231 | btrfs_tree_unlock(mid); |
5f39d397 | 1232 | free_extent_buffer(mid); |
e66f709b | 1233 | } else { |
925baedd | 1234 | btrfs_tree_unlock(right); |
5f39d397 | 1235 | free_extent_buffer(right); |
e66f709b | 1236 | } |
e66f709b CM |
1237 | return 0; |
1238 | } | |
925baedd | 1239 | btrfs_tree_unlock(right); |
5f39d397 | 1240 | free_extent_buffer(right); |
e66f709b | 1241 | } |
e66f709b CM |
1242 | return 1; |
1243 | } | |
1244 | ||
3c69faec | 1245 | /* |
d352ac68 CM |
1246 | * readahead one full node of leaves, finding things that are close |
1247 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 1248 | */ |
2ff7e61e | 1249 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
1250 | struct btrfs_path *path, |
1251 | int level, int slot, u64 objectid) | |
3c69faec | 1252 | { |
5f39d397 | 1253 | struct extent_buffer *node; |
01f46658 | 1254 | struct btrfs_disk_key disk_key; |
3c69faec | 1255 | u32 nritems; |
3c69faec | 1256 | u64 search; |
a7175319 | 1257 | u64 target; |
6b80053d | 1258 | u64 nread = 0; |
ace75066 | 1259 | u64 nread_max; |
6b80053d CM |
1260 | u32 nr; |
1261 | u32 blocksize; | |
1262 | u32 nscan = 0; | |
db94535d | 1263 | |
ace75066 | 1264 | if (level != 1 && path->reada != READA_FORWARD_ALWAYS) |
6702ed49 CM |
1265 | return; |
1266 | ||
1267 | if (!path->nodes[level]) | |
3c69faec CM |
1268 | return; |
1269 | ||
5f39d397 | 1270 | node = path->nodes[level]; |
925baedd | 1271 | |
ace75066 FM |
1272 | /* |
1273 | * Since the time between visiting leaves is much shorter than the time | |
1274 | * between visiting nodes, limit read ahead of nodes to 1, to avoid too | |
1275 | * much IO at once (possibly random). | |
1276 | */ | |
1277 | if (path->reada == READA_FORWARD_ALWAYS) { | |
1278 | if (level > 1) | |
1279 | nread_max = node->fs_info->nodesize; | |
1280 | else | |
1281 | nread_max = SZ_128K; | |
1282 | } else { | |
1283 | nread_max = SZ_64K; | |
1284 | } | |
1285 | ||
3c69faec | 1286 | search = btrfs_node_blockptr(node, slot); |
0b246afa | 1287 | blocksize = fs_info->nodesize; |
069a2e37 FM |
1288 | if (path->reada != READA_FORWARD_ALWAYS) { |
1289 | struct extent_buffer *eb; | |
1290 | ||
1291 | eb = find_extent_buffer(fs_info, search); | |
1292 | if (eb) { | |
1293 | free_extent_buffer(eb); | |
1294 | return; | |
1295 | } | |
3c69faec CM |
1296 | } |
1297 | ||
a7175319 | 1298 | target = search; |
6b80053d | 1299 | |
5f39d397 | 1300 | nritems = btrfs_header_nritems(node); |
6b80053d | 1301 | nr = slot; |
25b8b936 | 1302 | |
d397712b | 1303 | while (1) { |
e4058b54 | 1304 | if (path->reada == READA_BACK) { |
6b80053d CM |
1305 | if (nr == 0) |
1306 | break; | |
1307 | nr--; | |
ace75066 FM |
1308 | } else if (path->reada == READA_FORWARD || |
1309 | path->reada == READA_FORWARD_ALWAYS) { | |
6b80053d CM |
1310 | nr++; |
1311 | if (nr >= nritems) | |
1312 | break; | |
3c69faec | 1313 | } |
e4058b54 | 1314 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
1315 | btrfs_node_key(node, &disk_key, nr); |
1316 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
1317 | break; | |
1318 | } | |
6b80053d | 1319 | search = btrfs_node_blockptr(node, nr); |
ace75066 FM |
1320 | if (path->reada == READA_FORWARD_ALWAYS || |
1321 | (search <= target && target - search <= 65536) || | |
a7175319 | 1322 | (search > target && search - target <= 65536)) { |
bfb484d9 | 1323 | btrfs_readahead_node_child(node, nr); |
6b80053d CM |
1324 | nread += blocksize; |
1325 | } | |
1326 | nscan++; | |
ace75066 | 1327 | if (nread > nread_max || nscan > 32) |
6b80053d | 1328 | break; |
3c69faec CM |
1329 | } |
1330 | } | |
925baedd | 1331 | |
bfb484d9 | 1332 | static noinline void reada_for_balance(struct btrfs_path *path, int level) |
b4ce94de | 1333 | { |
bfb484d9 | 1334 | struct extent_buffer *parent; |
b4ce94de CM |
1335 | int slot; |
1336 | int nritems; | |
b4ce94de | 1337 | |
8c594ea8 | 1338 | parent = path->nodes[level + 1]; |
b4ce94de | 1339 | if (!parent) |
0b08851f | 1340 | return; |
b4ce94de CM |
1341 | |
1342 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 1343 | slot = path->slots[level + 1]; |
b4ce94de | 1344 | |
bfb484d9 JB |
1345 | if (slot > 0) |
1346 | btrfs_readahead_node_child(parent, slot - 1); | |
1347 | if (slot + 1 < nritems) | |
1348 | btrfs_readahead_node_child(parent, slot + 1); | |
b4ce94de CM |
1349 | } |
1350 | ||
1351 | ||
d352ac68 | 1352 | /* |
d397712b CM |
1353 | * when we walk down the tree, it is usually safe to unlock the higher layers |
1354 | * in the tree. The exceptions are when our path goes through slot 0, because | |
1355 | * operations on the tree might require changing key pointers higher up in the | |
1356 | * tree. | |
d352ac68 | 1357 | * |
d397712b CM |
1358 | * callers might also have set path->keep_locks, which tells this code to keep |
1359 | * the lock if the path points to the last slot in the block. This is part of | |
1360 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 1361 | * |
d397712b CM |
1362 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
1363 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 1364 | */ |
e02119d5 | 1365 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
1366 | int lowest_unlock, int min_write_lock_level, |
1367 | int *write_lock_level) | |
925baedd CM |
1368 | { |
1369 | int i; | |
1370 | int skip_level = level; | |
c1227996 | 1371 | bool check_skip = true; |
925baedd CM |
1372 | |
1373 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
1374 | if (!path->nodes[i]) | |
1375 | break; | |
1376 | if (!path->locks[i]) | |
1377 | break; | |
c1227996 NB |
1378 | |
1379 | if (check_skip) { | |
1380 | if (path->slots[i] == 0) { | |
925baedd CM |
1381 | skip_level = i + 1; |
1382 | continue; | |
1383 | } | |
c1227996 NB |
1384 | |
1385 | if (path->keep_locks) { | |
1386 | u32 nritems; | |
1387 | ||
1388 | nritems = btrfs_header_nritems(path->nodes[i]); | |
1389 | if (nritems < 1 || path->slots[i] >= nritems - 1) { | |
1390 | skip_level = i + 1; | |
1391 | continue; | |
1392 | } | |
1393 | } | |
925baedd | 1394 | } |
051e1b9f | 1395 | |
d80bb3f9 | 1396 | if (i >= lowest_unlock && i > skip_level) { |
c1227996 NB |
1397 | check_skip = false; |
1398 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); | |
925baedd | 1399 | path->locks[i] = 0; |
f7c79f30 CM |
1400 | if (write_lock_level && |
1401 | i > min_write_lock_level && | |
1402 | i <= *write_lock_level) { | |
1403 | *write_lock_level = i - 1; | |
1404 | } | |
925baedd CM |
1405 | } |
1406 | } | |
1407 | } | |
1408 | ||
c8c42864 | 1409 | /* |
376a21d7 FM |
1410 | * Helper function for btrfs_search_slot() and other functions that do a search |
1411 | * on a btree. The goal is to find a tree block in the cache (the radix tree at | |
1412 | * fs_info->buffer_radix), but if we can't find it, or it's not up to date, read | |
1413 | * its pages from disk. | |
c8c42864 | 1414 | * |
376a21d7 FM |
1415 | * Returns -EAGAIN, with the path unlocked, if the caller needs to repeat the |
1416 | * whole btree search, starting again from the current root node. | |
c8c42864 CM |
1417 | */ |
1418 | static int | |
d07b8528 LB |
1419 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
1420 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 1421 | const struct btrfs_key *key) |
c8c42864 | 1422 | { |
0b246afa | 1423 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
1424 | u64 blocknr; |
1425 | u64 gen; | |
c8c42864 | 1426 | struct extent_buffer *tmp; |
581c1760 | 1427 | struct btrfs_key first_key; |
76a05b35 | 1428 | int ret; |
581c1760 | 1429 | int parent_level; |
b246666e | 1430 | bool unlock_up; |
c8c42864 | 1431 | |
b246666e | 1432 | unlock_up = ((level + 1 < BTRFS_MAX_LEVEL) && p->locks[level + 1]); |
213ff4b7 NB |
1433 | blocknr = btrfs_node_blockptr(*eb_ret, slot); |
1434 | gen = btrfs_node_ptr_generation(*eb_ret, slot); | |
1435 | parent_level = btrfs_header_level(*eb_ret); | |
1436 | btrfs_node_key_to_cpu(*eb_ret, &first_key, slot); | |
c8c42864 | 1437 | |
b246666e FM |
1438 | /* |
1439 | * If we need to read an extent buffer from disk and we are holding locks | |
1440 | * on upper level nodes, we unlock all the upper nodes before reading the | |
1441 | * extent buffer, and then return -EAGAIN to the caller as it needs to | |
1442 | * restart the search. We don't release the lock on the current level | |
1443 | * because we need to walk this node to figure out which blocks to read. | |
1444 | */ | |
0b246afa | 1445 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 1446 | if (tmp) { |
ace75066 FM |
1447 | if (p->reada == READA_FORWARD_ALWAYS) |
1448 | reada_for_search(fs_info, p, level, slot, key->objectid); | |
1449 | ||
b9fab919 | 1450 | /* first we do an atomic uptodate check */ |
bdf7c00e | 1451 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
448de471 QW |
1452 | /* |
1453 | * Do extra check for first_key, eb can be stale due to | |
1454 | * being cached, read from scrub, or have multiple | |
1455 | * parents (shared tree blocks). | |
1456 | */ | |
e064d5e9 | 1457 | if (btrfs_verify_level_key(tmp, |
448de471 QW |
1458 | parent_level - 1, &first_key, gen)) { |
1459 | free_extent_buffer(tmp); | |
1460 | return -EUCLEAN; | |
1461 | } | |
bdf7c00e JB |
1462 | *eb_ret = tmp; |
1463 | return 0; | |
1464 | } | |
1465 | ||
857bc13f JB |
1466 | if (p->nowait) { |
1467 | free_extent_buffer(tmp); | |
1468 | return -EAGAIN; | |
1469 | } | |
1470 | ||
b246666e FM |
1471 | if (unlock_up) |
1472 | btrfs_unlock_up_safe(p, level + 1); | |
1473 | ||
bdf7c00e | 1474 | /* now we're allowed to do a blocking uptodate check */ |
6a2e9dc4 | 1475 | ret = btrfs_read_extent_buffer(tmp, gen, parent_level - 1, &first_key); |
9a4ffa1b QW |
1476 | if (ret) { |
1477 | free_extent_buffer(tmp); | |
1478 | btrfs_release_path(p); | |
1479 | return -EIO; | |
cb44921a | 1480 | } |
88c602ab QW |
1481 | if (btrfs_check_eb_owner(tmp, root->root_key.objectid)) { |
1482 | free_extent_buffer(tmp); | |
1483 | btrfs_release_path(p); | |
1484 | return -EUCLEAN; | |
1485 | } | |
b246666e FM |
1486 | |
1487 | if (unlock_up) | |
1488 | ret = -EAGAIN; | |
1489 | ||
1490 | goto out; | |
857bc13f JB |
1491 | } else if (p->nowait) { |
1492 | return -EAGAIN; | |
c8c42864 CM |
1493 | } |
1494 | ||
b246666e | 1495 | if (unlock_up) { |
4bb59055 FM |
1496 | btrfs_unlock_up_safe(p, level + 1); |
1497 | ret = -EAGAIN; | |
1498 | } else { | |
1499 | ret = 0; | |
1500 | } | |
8c594ea8 | 1501 | |
e4058b54 | 1502 | if (p->reada != READA_NONE) |
2ff7e61e | 1503 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 1504 | |
1b7ec85e JB |
1505 | tmp = read_tree_block(fs_info, blocknr, root->root_key.objectid, |
1506 | gen, parent_level - 1, &first_key); | |
4eb150d6 QW |
1507 | if (IS_ERR(tmp)) { |
1508 | btrfs_release_path(p); | |
1509 | return PTR_ERR(tmp); | |
76a05b35 | 1510 | } |
4eb150d6 QW |
1511 | /* |
1512 | * If the read above didn't mark this buffer up to date, | |
1513 | * it will never end up being up to date. Set ret to EIO now | |
1514 | * and give up so that our caller doesn't loop forever | |
1515 | * on our EAGAINs. | |
1516 | */ | |
1517 | if (!extent_buffer_uptodate(tmp)) | |
1518 | ret = -EIO; | |
02a3307a | 1519 | |
b246666e | 1520 | out: |
4bb59055 FM |
1521 | if (ret == 0) { |
1522 | *eb_ret = tmp; | |
1523 | } else { | |
1524 | free_extent_buffer(tmp); | |
1525 | btrfs_release_path(p); | |
1526 | } | |
1527 | ||
76a05b35 | 1528 | return ret; |
c8c42864 CM |
1529 | } |
1530 | ||
1531 | /* | |
1532 | * helper function for btrfs_search_slot. This does all of the checks | |
1533 | * for node-level blocks and does any balancing required based on | |
1534 | * the ins_len. | |
1535 | * | |
1536 | * If no extra work was required, zero is returned. If we had to | |
1537 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
1538 | * start over | |
1539 | */ | |
1540 | static int | |
1541 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
1542 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
1543 | struct extent_buffer *b, int level, int ins_len, |
1544 | int *write_lock_level) | |
c8c42864 | 1545 | { |
0b246afa | 1546 | struct btrfs_fs_info *fs_info = root->fs_info; |
95b982de | 1547 | int ret = 0; |
0b246afa | 1548 | |
c8c42864 | 1549 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 1550 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 | 1551 | |
bd681513 CM |
1552 | if (*write_lock_level < level + 1) { |
1553 | *write_lock_level = level + 1; | |
1554 | btrfs_release_path(p); | |
95b982de | 1555 | return -EAGAIN; |
bd681513 CM |
1556 | } |
1557 | ||
bfb484d9 | 1558 | reada_for_balance(p, level); |
95b982de | 1559 | ret = split_node(trans, root, p, level); |
c8c42864 | 1560 | |
c8c42864 CM |
1561 | b = p->nodes[level]; |
1562 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 1563 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 | 1564 | |
bd681513 CM |
1565 | if (*write_lock_level < level + 1) { |
1566 | *write_lock_level = level + 1; | |
1567 | btrfs_release_path(p); | |
95b982de | 1568 | return -EAGAIN; |
bd681513 CM |
1569 | } |
1570 | ||
bfb484d9 | 1571 | reada_for_balance(p, level); |
95b982de NB |
1572 | ret = balance_level(trans, root, p, level); |
1573 | if (ret) | |
1574 | return ret; | |
c8c42864 | 1575 | |
c8c42864 CM |
1576 | b = p->nodes[level]; |
1577 | if (!b) { | |
b3b4aa74 | 1578 | btrfs_release_path(p); |
95b982de | 1579 | return -EAGAIN; |
c8c42864 CM |
1580 | } |
1581 | BUG_ON(btrfs_header_nritems(b) == 1); | |
1582 | } | |
c8c42864 CM |
1583 | return ret; |
1584 | } | |
1585 | ||
381cf658 | 1586 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
1587 | u64 iobjectid, u64 ioff, u8 key_type, |
1588 | struct btrfs_key *found_key) | |
1589 | { | |
1590 | int ret; | |
1591 | struct btrfs_key key; | |
1592 | struct extent_buffer *eb; | |
381cf658 DS |
1593 | |
1594 | ASSERT(path); | |
1d4c08e0 | 1595 | ASSERT(found_key); |
e33d5c3d KN |
1596 | |
1597 | key.type = key_type; | |
1598 | key.objectid = iobjectid; | |
1599 | key.offset = ioff; | |
1600 | ||
1601 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 1602 | if (ret < 0) |
e33d5c3d KN |
1603 | return ret; |
1604 | ||
1605 | eb = path->nodes[0]; | |
1606 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
1607 | ret = btrfs_next_leaf(fs_root, path); | |
1608 | if (ret) | |
1609 | return ret; | |
1610 | eb = path->nodes[0]; | |
1611 | } | |
1612 | ||
1613 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
1614 | if (found_key->type != key.type || | |
1615 | found_key->objectid != key.objectid) | |
1616 | return 1; | |
1617 | ||
1618 | return 0; | |
1619 | } | |
1620 | ||
1fc28d8e LB |
1621 | static struct extent_buffer *btrfs_search_slot_get_root(struct btrfs_root *root, |
1622 | struct btrfs_path *p, | |
1623 | int write_lock_level) | |
1624 | { | |
1fc28d8e | 1625 | struct extent_buffer *b; |
120de408 | 1626 | int root_lock = 0; |
1fc28d8e LB |
1627 | int level = 0; |
1628 | ||
1fc28d8e | 1629 | if (p->search_commit_root) { |
d96b3424 FM |
1630 | b = root->commit_root; |
1631 | atomic_inc(&b->refs); | |
be6821f8 | 1632 | level = btrfs_header_level(b); |
f9ddfd05 LB |
1633 | /* |
1634 | * Ensure that all callers have set skip_locking when | |
1635 | * p->search_commit_root = 1. | |
1636 | */ | |
1637 | ASSERT(p->skip_locking == 1); | |
1fc28d8e LB |
1638 | |
1639 | goto out; | |
1640 | } | |
1641 | ||
1642 | if (p->skip_locking) { | |
1643 | b = btrfs_root_node(root); | |
1644 | level = btrfs_header_level(b); | |
1645 | goto out; | |
1646 | } | |
1647 | ||
120de408 JB |
1648 | /* We try very hard to do read locks on the root */ |
1649 | root_lock = BTRFS_READ_LOCK; | |
1650 | ||
1fc28d8e | 1651 | /* |
662c653b LB |
1652 | * If the level is set to maximum, we can skip trying to get the read |
1653 | * lock. | |
1fc28d8e | 1654 | */ |
662c653b LB |
1655 | if (write_lock_level < BTRFS_MAX_LEVEL) { |
1656 | /* | |
1657 | * We don't know the level of the root node until we actually | |
1658 | * have it read locked | |
1659 | */ | |
857bc13f JB |
1660 | if (p->nowait) { |
1661 | b = btrfs_try_read_lock_root_node(root); | |
1662 | if (IS_ERR(b)) | |
1663 | return b; | |
1664 | } else { | |
1665 | b = btrfs_read_lock_root_node(root); | |
1666 | } | |
662c653b LB |
1667 | level = btrfs_header_level(b); |
1668 | if (level > write_lock_level) | |
1669 | goto out; | |
1670 | ||
1671 | /* Whoops, must trade for write lock */ | |
1672 | btrfs_tree_read_unlock(b); | |
1673 | free_extent_buffer(b); | |
1674 | } | |
1fc28d8e | 1675 | |
1fc28d8e LB |
1676 | b = btrfs_lock_root_node(root); |
1677 | root_lock = BTRFS_WRITE_LOCK; | |
1678 | ||
1679 | /* The level might have changed, check again */ | |
1680 | level = btrfs_header_level(b); | |
1681 | ||
1682 | out: | |
120de408 JB |
1683 | /* |
1684 | * The root may have failed to write out at some point, and thus is no | |
1685 | * longer valid, return an error in this case. | |
1686 | */ | |
1687 | if (!extent_buffer_uptodate(b)) { | |
1688 | if (root_lock) | |
1689 | btrfs_tree_unlock_rw(b, root_lock); | |
1690 | free_extent_buffer(b); | |
1691 | return ERR_PTR(-EIO); | |
1692 | } | |
1693 | ||
1fc28d8e LB |
1694 | p->nodes[level] = b; |
1695 | if (!p->skip_locking) | |
1696 | p->locks[level] = root_lock; | |
1697 | /* | |
1698 | * Callers are responsible for dropping b's references. | |
1699 | */ | |
1700 | return b; | |
1701 | } | |
1702 | ||
d96b3424 FM |
1703 | /* |
1704 | * Replace the extent buffer at the lowest level of the path with a cloned | |
1705 | * version. The purpose is to be able to use it safely, after releasing the | |
1706 | * commit root semaphore, even if relocation is happening in parallel, the | |
1707 | * transaction used for relocation is committed and the extent buffer is | |
1708 | * reallocated in the next transaction. | |
1709 | * | |
1710 | * This is used in a context where the caller does not prevent transaction | |
1711 | * commits from happening, either by holding a transaction handle or holding | |
1712 | * some lock, while it's doing searches through a commit root. | |
1713 | * At the moment it's only used for send operations. | |
1714 | */ | |
1715 | static int finish_need_commit_sem_search(struct btrfs_path *path) | |
1716 | { | |
1717 | const int i = path->lowest_level; | |
1718 | const int slot = path->slots[i]; | |
1719 | struct extent_buffer *lowest = path->nodes[i]; | |
1720 | struct extent_buffer *clone; | |
1721 | ||
1722 | ASSERT(path->need_commit_sem); | |
1723 | ||
1724 | if (!lowest) | |
1725 | return 0; | |
1726 | ||
1727 | lockdep_assert_held_read(&lowest->fs_info->commit_root_sem); | |
1728 | ||
1729 | clone = btrfs_clone_extent_buffer(lowest); | |
1730 | if (!clone) | |
1731 | return -ENOMEM; | |
1732 | ||
1733 | btrfs_release_path(path); | |
1734 | path->nodes[i] = clone; | |
1735 | path->slots[i] = slot; | |
1736 | ||
1737 | return 0; | |
1738 | } | |
1fc28d8e | 1739 | |
e2e58d0f FM |
1740 | static inline int search_for_key_slot(struct extent_buffer *eb, |
1741 | int search_low_slot, | |
1742 | const struct btrfs_key *key, | |
1743 | int prev_cmp, | |
1744 | int *slot) | |
1745 | { | |
1746 | /* | |
1747 | * If a previous call to btrfs_bin_search() on a parent node returned an | |
1748 | * exact match (prev_cmp == 0), we can safely assume the target key will | |
1749 | * always be at slot 0 on lower levels, since each key pointer | |
1750 | * (struct btrfs_key_ptr) refers to the lowest key accessible from the | |
1751 | * subtree it points to. Thus we can skip searching lower levels. | |
1752 | */ | |
1753 | if (prev_cmp == 0) { | |
1754 | *slot = 0; | |
1755 | return 0; | |
1756 | } | |
1757 | ||
1758 | return generic_bin_search(eb, search_low_slot, key, slot); | |
1759 | } | |
1760 | ||
109324cf FM |
1761 | static int search_leaf(struct btrfs_trans_handle *trans, |
1762 | struct btrfs_root *root, | |
1763 | const struct btrfs_key *key, | |
1764 | struct btrfs_path *path, | |
1765 | int ins_len, | |
1766 | int prev_cmp) | |
1767 | { | |
1768 | struct extent_buffer *leaf = path->nodes[0]; | |
1769 | int leaf_free_space = -1; | |
1770 | int search_low_slot = 0; | |
1771 | int ret; | |
1772 | bool do_bin_search = true; | |
1773 | ||
1774 | /* | |
1775 | * If we are doing an insertion, the leaf has enough free space and the | |
1776 | * destination slot for the key is not slot 0, then we can unlock our | |
1777 | * write lock on the parent, and any other upper nodes, before doing the | |
1778 | * binary search on the leaf (with search_for_key_slot()), allowing other | |
1779 | * tasks to lock the parent and any other upper nodes. | |
1780 | */ | |
1781 | if (ins_len > 0) { | |
1782 | /* | |
1783 | * Cache the leaf free space, since we will need it later and it | |
1784 | * will not change until then. | |
1785 | */ | |
1786 | leaf_free_space = btrfs_leaf_free_space(leaf); | |
1787 | ||
1788 | /* | |
1789 | * !path->locks[1] means we have a single node tree, the leaf is | |
1790 | * the root of the tree. | |
1791 | */ | |
1792 | if (path->locks[1] && leaf_free_space >= ins_len) { | |
1793 | struct btrfs_disk_key first_key; | |
1794 | ||
1795 | ASSERT(btrfs_header_nritems(leaf) > 0); | |
1796 | btrfs_item_key(leaf, &first_key, 0); | |
1797 | ||
1798 | /* | |
1799 | * Doing the extra comparison with the first key is cheap, | |
1800 | * taking into account that the first key is very likely | |
1801 | * already in a cache line because it immediately follows | |
1802 | * the extent buffer's header and we have recently accessed | |
1803 | * the header's level field. | |
1804 | */ | |
1805 | ret = comp_keys(&first_key, key); | |
1806 | if (ret < 0) { | |
1807 | /* | |
1808 | * The first key is smaller than the key we want | |
1809 | * to insert, so we are safe to unlock all upper | |
1810 | * nodes and we have to do the binary search. | |
1811 | * | |
1812 | * We do use btrfs_unlock_up_safe() and not | |
1813 | * unlock_up() because the later does not unlock | |
1814 | * nodes with a slot of 0 - we can safely unlock | |
1815 | * any node even if its slot is 0 since in this | |
1816 | * case the key does not end up at slot 0 of the | |
1817 | * leaf and there's no need to split the leaf. | |
1818 | */ | |
1819 | btrfs_unlock_up_safe(path, 1); | |
1820 | search_low_slot = 1; | |
1821 | } else { | |
1822 | /* | |
1823 | * The first key is >= then the key we want to | |
1824 | * insert, so we can skip the binary search as | |
1825 | * the target key will be at slot 0. | |
1826 | * | |
1827 | * We can not unlock upper nodes when the key is | |
1828 | * less than the first key, because we will need | |
1829 | * to update the key at slot 0 of the parent node | |
1830 | * and possibly of other upper nodes too. | |
1831 | * If the key matches the first key, then we can | |
1832 | * unlock all the upper nodes, using | |
1833 | * btrfs_unlock_up_safe() instead of unlock_up() | |
1834 | * as stated above. | |
1835 | */ | |
1836 | if (ret == 0) | |
1837 | btrfs_unlock_up_safe(path, 1); | |
1838 | /* | |
1839 | * ret is already 0 or 1, matching the result of | |
1840 | * a btrfs_bin_search() call, so there is no need | |
1841 | * to adjust it. | |
1842 | */ | |
1843 | do_bin_search = false; | |
1844 | path->slots[0] = 0; | |
1845 | } | |
1846 | } | |
1847 | } | |
1848 | ||
1849 | if (do_bin_search) { | |
1850 | ret = search_for_key_slot(leaf, search_low_slot, key, | |
1851 | prev_cmp, &path->slots[0]); | |
1852 | if (ret < 0) | |
1853 | return ret; | |
1854 | } | |
1855 | ||
1856 | if (ins_len > 0) { | |
1857 | /* | |
1858 | * Item key already exists. In this case, if we are allowed to | |
1859 | * insert the item (for example, in dir_item case, item key | |
1860 | * collision is allowed), it will be merged with the original | |
1861 | * item. Only the item size grows, no new btrfs item will be | |
1862 | * added. If search_for_extension is not set, ins_len already | |
1863 | * accounts the size btrfs_item, deduct it here so leaf space | |
1864 | * check will be correct. | |
1865 | */ | |
1866 | if (ret == 0 && !path->search_for_extension) { | |
1867 | ASSERT(ins_len >= sizeof(struct btrfs_item)); | |
1868 | ins_len -= sizeof(struct btrfs_item); | |
1869 | } | |
1870 | ||
1871 | ASSERT(leaf_free_space >= 0); | |
1872 | ||
1873 | if (leaf_free_space < ins_len) { | |
1874 | int err; | |
1875 | ||
1876 | err = split_leaf(trans, root, key, path, ins_len, | |
1877 | (ret == 0)); | |
bb8e9a60 FM |
1878 | ASSERT(err <= 0); |
1879 | if (WARN_ON(err > 0)) | |
1880 | err = -EUCLEAN; | |
109324cf FM |
1881 | if (err) |
1882 | ret = err; | |
1883 | } | |
1884 | } | |
1885 | ||
1886 | return ret; | |
1887 | } | |
1888 | ||
74123bd7 | 1889 | /* |
4271ecea NB |
1890 | * btrfs_search_slot - look for a key in a tree and perform necessary |
1891 | * modifications to preserve tree invariants. | |
74123bd7 | 1892 | * |
4271ecea NB |
1893 | * @trans: Handle of transaction, used when modifying the tree |
1894 | * @p: Holds all btree nodes along the search path | |
1895 | * @root: The root node of the tree | |
1896 | * @key: The key we are looking for | |
9a664971 | 1897 | * @ins_len: Indicates purpose of search: |
1898 | * >0 for inserts it's size of item inserted (*) | |
1899 | * <0 for deletions | |
1900 | * 0 for plain searches, not modifying the tree | |
1901 | * | |
1902 | * (*) If size of item inserted doesn't include | |
1903 | * sizeof(struct btrfs_item), then p->search_for_extension must | |
1904 | * be set. | |
4271ecea NB |
1905 | * @cow: boolean should CoW operations be performed. Must always be 1 |
1906 | * when modifying the tree. | |
97571fd0 | 1907 | * |
4271ecea NB |
1908 | * If @ins_len > 0, nodes and leaves will be split as we walk down the tree. |
1909 | * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible) | |
1910 | * | |
1911 | * If @key is found, 0 is returned and you can find the item in the leaf level | |
1912 | * of the path (level 0) | |
1913 | * | |
1914 | * If @key isn't found, 1 is returned and the leaf level of the path (level 0) | |
1915 | * points to the slot where it should be inserted | |
1916 | * | |
1917 | * If an error is encountered while searching the tree a negative error number | |
1918 | * is returned | |
74123bd7 | 1919 | */ |
310712b2 OS |
1920 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
1921 | const struct btrfs_key *key, struct btrfs_path *p, | |
1922 | int ins_len, int cow) | |
be0e5c09 | 1923 | { |
d96b3424 | 1924 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 1925 | struct extent_buffer *b; |
be0e5c09 CM |
1926 | int slot; |
1927 | int ret; | |
33c66f43 | 1928 | int err; |
be0e5c09 | 1929 | int level; |
925baedd | 1930 | int lowest_unlock = 1; |
bd681513 CM |
1931 | /* everything at write_lock_level or lower must be write locked */ |
1932 | int write_lock_level = 0; | |
9f3a7427 | 1933 | u8 lowest_level = 0; |
f7c79f30 | 1934 | int min_write_lock_level; |
d7396f07 | 1935 | int prev_cmp; |
9f3a7427 | 1936 | |
6702ed49 | 1937 | lowest_level = p->lowest_level; |
323ac95b | 1938 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 1939 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 1940 | BUG_ON(!cow && ins_len); |
25179201 | 1941 | |
857bc13f JB |
1942 | /* |
1943 | * For now only allow nowait for read only operations. There's no | |
1944 | * strict reason why we can't, we just only need it for reads so it's | |
1945 | * only implemented for reads. | |
1946 | */ | |
1947 | ASSERT(!p->nowait || !cow); | |
1948 | ||
bd681513 | 1949 | if (ins_len < 0) { |
925baedd | 1950 | lowest_unlock = 2; |
65b51a00 | 1951 | |
bd681513 CM |
1952 | /* when we are removing items, we might have to go up to level |
1953 | * two as we update tree pointers Make sure we keep write | |
1954 | * for those levels as well | |
1955 | */ | |
1956 | write_lock_level = 2; | |
1957 | } else if (ins_len > 0) { | |
1958 | /* | |
1959 | * for inserting items, make sure we have a write lock on | |
1960 | * level 1 so we can update keys | |
1961 | */ | |
1962 | write_lock_level = 1; | |
1963 | } | |
1964 | ||
1965 | if (!cow) | |
1966 | write_lock_level = -1; | |
1967 | ||
09a2a8f9 | 1968 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
1969 | write_lock_level = BTRFS_MAX_LEVEL; |
1970 | ||
f7c79f30 CM |
1971 | min_write_lock_level = write_lock_level; |
1972 | ||
d96b3424 FM |
1973 | if (p->need_commit_sem) { |
1974 | ASSERT(p->search_commit_root); | |
857bc13f JB |
1975 | if (p->nowait) { |
1976 | if (!down_read_trylock(&fs_info->commit_root_sem)) | |
1977 | return -EAGAIN; | |
1978 | } else { | |
1979 | down_read(&fs_info->commit_root_sem); | |
1980 | } | |
d96b3424 FM |
1981 | } |
1982 | ||
bb803951 | 1983 | again: |
d7396f07 | 1984 | prev_cmp = -1; |
1fc28d8e | 1985 | b = btrfs_search_slot_get_root(root, p, write_lock_level); |
be6821f8 FM |
1986 | if (IS_ERR(b)) { |
1987 | ret = PTR_ERR(b); | |
1988 | goto done; | |
1989 | } | |
925baedd | 1990 | |
eb60ceac | 1991 | while (b) { |
f624d976 QW |
1992 | int dec = 0; |
1993 | ||
5f39d397 | 1994 | level = btrfs_header_level(b); |
65b51a00 | 1995 | |
02217ed2 | 1996 | if (cow) { |
9ea2c7c9 NB |
1997 | bool last_level = (level == (BTRFS_MAX_LEVEL - 1)); |
1998 | ||
c8c42864 CM |
1999 | /* |
2000 | * if we don't really need to cow this block | |
2001 | * then we don't want to set the path blocking, | |
2002 | * so we test it here | |
2003 | */ | |
5963ffca | 2004 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2005 | goto cow_done; |
5d4f98a2 | 2006 | |
bd681513 CM |
2007 | /* |
2008 | * must have write locks on this node and the | |
2009 | * parent | |
2010 | */ | |
5124e00e JB |
2011 | if (level > write_lock_level || |
2012 | (level + 1 > write_lock_level && | |
2013 | level + 1 < BTRFS_MAX_LEVEL && | |
2014 | p->nodes[level + 1])) { | |
bd681513 CM |
2015 | write_lock_level = level + 1; |
2016 | btrfs_release_path(p); | |
2017 | goto again; | |
2018 | } | |
2019 | ||
9ea2c7c9 NB |
2020 | if (last_level) |
2021 | err = btrfs_cow_block(trans, root, b, NULL, 0, | |
9631e4cc JB |
2022 | &b, |
2023 | BTRFS_NESTING_COW); | |
9ea2c7c9 NB |
2024 | else |
2025 | err = btrfs_cow_block(trans, root, b, | |
2026 | p->nodes[level + 1], | |
9631e4cc JB |
2027 | p->slots[level + 1], &b, |
2028 | BTRFS_NESTING_COW); | |
33c66f43 | 2029 | if (err) { |
33c66f43 | 2030 | ret = err; |
65b51a00 | 2031 | goto done; |
54aa1f4d | 2032 | } |
02217ed2 | 2033 | } |
65b51a00 | 2034 | cow_done: |
eb60ceac | 2035 | p->nodes[level] = b; |
b4ce94de CM |
2036 | |
2037 | /* | |
2038 | * we have a lock on b and as long as we aren't changing | |
2039 | * the tree, there is no way to for the items in b to change. | |
2040 | * It is safe to drop the lock on our parent before we | |
2041 | * go through the expensive btree search on b. | |
2042 | * | |
eb653de1 FDBM |
2043 | * If we're inserting or deleting (ins_len != 0), then we might |
2044 | * be changing slot zero, which may require changing the parent. | |
2045 | * So, we can't drop the lock until after we know which slot | |
2046 | * we're operating on. | |
b4ce94de | 2047 | */ |
eb653de1 FDBM |
2048 | if (!ins_len && !p->keep_locks) { |
2049 | int u = level + 1; | |
2050 | ||
2051 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2052 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2053 | p->locks[u] = 0; | |
2054 | } | |
2055 | } | |
b4ce94de | 2056 | |
e2e58d0f | 2057 | if (level == 0) { |
109324cf | 2058 | if (ins_len > 0) |
e5e1c174 | 2059 | ASSERT(write_lock_level >= 1); |
bd681513 | 2060 | |
109324cf | 2061 | ret = search_leaf(trans, root, key, p, ins_len, prev_cmp); |
459931ec | 2062 | if (!p->search_for_split) |
f7c79f30 | 2063 | unlock_up(p, level, lowest_unlock, |
4b6f8e96 | 2064 | min_write_lock_level, NULL); |
65b51a00 | 2065 | goto done; |
be0e5c09 | 2066 | } |
e2e58d0f FM |
2067 | |
2068 | ret = search_for_key_slot(b, 0, key, prev_cmp, &slot); | |
2069 | if (ret < 0) | |
2070 | goto done; | |
2071 | prev_cmp = ret; | |
2072 | ||
f624d976 QW |
2073 | if (ret && slot > 0) { |
2074 | dec = 1; | |
2075 | slot--; | |
2076 | } | |
2077 | p->slots[level] = slot; | |
2078 | err = setup_nodes_for_search(trans, root, p, b, level, ins_len, | |
2079 | &write_lock_level); | |
2080 | if (err == -EAGAIN) | |
2081 | goto again; | |
2082 | if (err) { | |
2083 | ret = err; | |
2084 | goto done; | |
2085 | } | |
2086 | b = p->nodes[level]; | |
2087 | slot = p->slots[level]; | |
2088 | ||
2089 | /* | |
2090 | * Slot 0 is special, if we change the key we have to update | |
2091 | * the parent pointer which means we must have a write lock on | |
2092 | * the parent | |
2093 | */ | |
2094 | if (slot == 0 && ins_len && write_lock_level < level + 1) { | |
2095 | write_lock_level = level + 1; | |
2096 | btrfs_release_path(p); | |
2097 | goto again; | |
2098 | } | |
2099 | ||
2100 | unlock_up(p, level, lowest_unlock, min_write_lock_level, | |
2101 | &write_lock_level); | |
2102 | ||
2103 | if (level == lowest_level) { | |
2104 | if (dec) | |
2105 | p->slots[level]++; | |
2106 | goto done; | |
2107 | } | |
2108 | ||
2109 | err = read_block_for_search(root, p, &b, level, slot, key); | |
2110 | if (err == -EAGAIN) | |
2111 | goto again; | |
2112 | if (err) { | |
2113 | ret = err; | |
2114 | goto done; | |
2115 | } | |
2116 | ||
2117 | if (!p->skip_locking) { | |
2118 | level = btrfs_header_level(b); | |
b40130b2 JB |
2119 | |
2120 | btrfs_maybe_reset_lockdep_class(root, b); | |
2121 | ||
f624d976 | 2122 | if (level <= write_lock_level) { |
ac5887c8 | 2123 | btrfs_tree_lock(b); |
f624d976 QW |
2124 | p->locks[level] = BTRFS_WRITE_LOCK; |
2125 | } else { | |
857bc13f JB |
2126 | if (p->nowait) { |
2127 | if (!btrfs_try_tree_read_lock(b)) { | |
2128 | free_extent_buffer(b); | |
2129 | ret = -EAGAIN; | |
2130 | goto done; | |
2131 | } | |
2132 | } else { | |
2133 | btrfs_tree_read_lock(b); | |
2134 | } | |
f624d976 QW |
2135 | p->locks[level] = BTRFS_READ_LOCK; |
2136 | } | |
2137 | p->nodes[level] = b; | |
2138 | } | |
be0e5c09 | 2139 | } |
65b51a00 CM |
2140 | ret = 1; |
2141 | done: | |
5f5bc6b1 | 2142 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2143 | btrfs_release_path(p); |
d96b3424 FM |
2144 | |
2145 | if (p->need_commit_sem) { | |
2146 | int ret2; | |
2147 | ||
2148 | ret2 = finish_need_commit_sem_search(p); | |
2149 | up_read(&fs_info->commit_root_sem); | |
2150 | if (ret2) | |
2151 | ret = ret2; | |
2152 | } | |
2153 | ||
65b51a00 | 2154 | return ret; |
be0e5c09 | 2155 | } |
f75e2b79 | 2156 | ALLOW_ERROR_INJECTION(btrfs_search_slot, ERRNO); |
be0e5c09 | 2157 | |
5d9e75c4 JS |
2158 | /* |
2159 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2160 | * current state of the tree together with the operations recorded in the tree | |
2161 | * modification log to search for the key in a previous version of this tree, as | |
2162 | * denoted by the time_seq parameter. | |
2163 | * | |
2164 | * Naturally, there is no support for insert, delete or cow operations. | |
2165 | * | |
2166 | * The resulting path and return value will be set up as if we called | |
2167 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2168 | */ | |
310712b2 | 2169 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2170 | struct btrfs_path *p, u64 time_seq) |
2171 | { | |
0b246afa | 2172 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2173 | struct extent_buffer *b; |
2174 | int slot; | |
2175 | int ret; | |
2176 | int err; | |
2177 | int level; | |
2178 | int lowest_unlock = 1; | |
2179 | u8 lowest_level = 0; | |
2180 | ||
2181 | lowest_level = p->lowest_level; | |
2182 | WARN_ON(p->nodes[0] != NULL); | |
c922b016 | 2183 | ASSERT(!p->nowait); |
5d9e75c4 JS |
2184 | |
2185 | if (p->search_commit_root) { | |
2186 | BUG_ON(time_seq); | |
2187 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2188 | } | |
2189 | ||
2190 | again: | |
f3a84ccd | 2191 | b = btrfs_get_old_root(root, time_seq); |
315bed43 NB |
2192 | if (!b) { |
2193 | ret = -EIO; | |
2194 | goto done; | |
2195 | } | |
5d9e75c4 | 2196 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2197 | p->locks[level] = BTRFS_READ_LOCK; |
2198 | ||
2199 | while (b) { | |
abe9339d QW |
2200 | int dec = 0; |
2201 | ||
5d9e75c4 JS |
2202 | level = btrfs_header_level(b); |
2203 | p->nodes[level] = b; | |
5d9e75c4 JS |
2204 | |
2205 | /* | |
2206 | * we have a lock on b and as long as we aren't changing | |
2207 | * the tree, there is no way to for the items in b to change. | |
2208 | * It is safe to drop the lock on our parent before we | |
2209 | * go through the expensive btree search on b. | |
2210 | */ | |
2211 | btrfs_unlock_up_safe(p, level + 1); | |
2212 | ||
995e9a16 | 2213 | ret = btrfs_bin_search(b, key, &slot); |
cbca7d59 FM |
2214 | if (ret < 0) |
2215 | goto done; | |
5d9e75c4 | 2216 | |
abe9339d | 2217 | if (level == 0) { |
5d9e75c4 JS |
2218 | p->slots[level] = slot; |
2219 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
abe9339d QW |
2220 | goto done; |
2221 | } | |
5d9e75c4 | 2222 | |
abe9339d QW |
2223 | if (ret && slot > 0) { |
2224 | dec = 1; | |
2225 | slot--; | |
2226 | } | |
2227 | p->slots[level] = slot; | |
2228 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
5d9e75c4 | 2229 | |
abe9339d QW |
2230 | if (level == lowest_level) { |
2231 | if (dec) | |
2232 | p->slots[level]++; | |
2233 | goto done; | |
2234 | } | |
5d9e75c4 | 2235 | |
abe9339d QW |
2236 | err = read_block_for_search(root, p, &b, level, slot, key); |
2237 | if (err == -EAGAIN) | |
2238 | goto again; | |
2239 | if (err) { | |
2240 | ret = err; | |
5d9e75c4 JS |
2241 | goto done; |
2242 | } | |
abe9339d QW |
2243 | |
2244 | level = btrfs_header_level(b); | |
ac5887c8 | 2245 | btrfs_tree_read_lock(b); |
f3a84ccd | 2246 | b = btrfs_tree_mod_log_rewind(fs_info, p, b, time_seq); |
abe9339d QW |
2247 | if (!b) { |
2248 | ret = -ENOMEM; | |
2249 | goto done; | |
2250 | } | |
2251 | p->locks[level] = BTRFS_READ_LOCK; | |
2252 | p->nodes[level] = b; | |
5d9e75c4 JS |
2253 | } |
2254 | ret = 1; | |
2255 | done: | |
5d9e75c4 JS |
2256 | if (ret < 0) |
2257 | btrfs_release_path(p); | |
2258 | ||
2259 | return ret; | |
2260 | } | |
2261 | ||
2f38b3e1 AJ |
2262 | /* |
2263 | * helper to use instead of search slot if no exact match is needed but | |
2264 | * instead the next or previous item should be returned. | |
2265 | * When find_higher is true, the next higher item is returned, the next lower | |
2266 | * otherwise. | |
2267 | * When return_any and find_higher are both true, and no higher item is found, | |
2268 | * return the next lower instead. | |
2269 | * When return_any is true and find_higher is false, and no lower item is found, | |
2270 | * return the next higher instead. | |
2271 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2272 | * < 0 on error | |
2273 | */ | |
2274 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
2275 | const struct btrfs_key *key, |
2276 | struct btrfs_path *p, int find_higher, | |
2277 | int return_any) | |
2f38b3e1 AJ |
2278 | { |
2279 | int ret; | |
2280 | struct extent_buffer *leaf; | |
2281 | ||
2282 | again: | |
2283 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2284 | if (ret <= 0) | |
2285 | return ret; | |
2286 | /* | |
2287 | * a return value of 1 means the path is at the position where the | |
2288 | * item should be inserted. Normally this is the next bigger item, | |
2289 | * but in case the previous item is the last in a leaf, path points | |
2290 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2291 | * item. | |
2292 | */ | |
2293 | leaf = p->nodes[0]; | |
2294 | ||
2295 | if (find_higher) { | |
2296 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2297 | ret = btrfs_next_leaf(root, p); | |
2298 | if (ret <= 0) | |
2299 | return ret; | |
2300 | if (!return_any) | |
2301 | return 1; | |
2302 | /* | |
2303 | * no higher item found, return the next | |
2304 | * lower instead | |
2305 | */ | |
2306 | return_any = 0; | |
2307 | find_higher = 0; | |
2308 | btrfs_release_path(p); | |
2309 | goto again; | |
2310 | } | |
2311 | } else { | |
e6793769 AJ |
2312 | if (p->slots[0] == 0) { |
2313 | ret = btrfs_prev_leaf(root, p); | |
2314 | if (ret < 0) | |
2315 | return ret; | |
2316 | if (!ret) { | |
23c6bf6a FDBM |
2317 | leaf = p->nodes[0]; |
2318 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
2319 | p->slots[0]--; | |
e6793769 | 2320 | return 0; |
2f38b3e1 | 2321 | } |
e6793769 AJ |
2322 | if (!return_any) |
2323 | return 1; | |
2324 | /* | |
2325 | * no lower item found, return the next | |
2326 | * higher instead | |
2327 | */ | |
2328 | return_any = 0; | |
2329 | find_higher = 1; | |
2330 | btrfs_release_path(p); | |
2331 | goto again; | |
2332 | } else { | |
2f38b3e1 AJ |
2333 | --p->slots[0]; |
2334 | } | |
2335 | } | |
2336 | return 0; | |
2337 | } | |
2338 | ||
0ff40a91 MPS |
2339 | /* |
2340 | * Execute search and call btrfs_previous_item to traverse backwards if the item | |
2341 | * was not found. | |
2342 | * | |
2343 | * Return 0 if found, 1 if not found and < 0 if error. | |
2344 | */ | |
2345 | int btrfs_search_backwards(struct btrfs_root *root, struct btrfs_key *key, | |
2346 | struct btrfs_path *path) | |
2347 | { | |
2348 | int ret; | |
2349 | ||
2350 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); | |
2351 | if (ret > 0) | |
2352 | ret = btrfs_previous_item(root, path, key->objectid, key->type); | |
2353 | ||
2354 | if (ret == 0) | |
2355 | btrfs_item_key_to_cpu(path->nodes[0], key, path->slots[0]); | |
2356 | ||
2357 | return ret; | |
2358 | } | |
2359 | ||
62142be3 GN |
2360 | /** |
2361 | * Search for a valid slot for the given path. | |
2362 | * | |
2363 | * @root: The root node of the tree. | |
2364 | * @key: Will contain a valid item if found. | |
2365 | * @path: The starting point to validate the slot. | |
2366 | * | |
2367 | * Return: 0 if the item is valid | |
2368 | * 1 if not found | |
2369 | * <0 if error. | |
2370 | */ | |
2371 | int btrfs_get_next_valid_item(struct btrfs_root *root, struct btrfs_key *key, | |
2372 | struct btrfs_path *path) | |
2373 | { | |
2374 | while (1) { | |
2375 | int ret; | |
2376 | const int slot = path->slots[0]; | |
2377 | const struct extent_buffer *leaf = path->nodes[0]; | |
2378 | ||
2379 | /* This is where we start walking the path. */ | |
2380 | if (slot >= btrfs_header_nritems(leaf)) { | |
2381 | /* | |
2382 | * If we've reached the last slot in this leaf we need | |
2383 | * to go to the next leaf and reset the path. | |
2384 | */ | |
2385 | ret = btrfs_next_leaf(root, path); | |
2386 | if (ret) | |
2387 | return ret; | |
2388 | continue; | |
2389 | } | |
2390 | /* Store the found, valid item in @key. */ | |
2391 | btrfs_item_key_to_cpu(leaf, key, slot); | |
2392 | break; | |
2393 | } | |
2394 | return 0; | |
2395 | } | |
2396 | ||
74123bd7 CM |
2397 | /* |
2398 | * adjust the pointers going up the tree, starting at level | |
2399 | * making sure the right key of each node is points to 'key'. | |
2400 | * This is used after shifting pointers to the left, so it stops | |
2401 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2402 | * higher levels | |
aa5d6bed | 2403 | * |
74123bd7 | 2404 | */ |
b167fa91 | 2405 | static void fixup_low_keys(struct btrfs_path *path, |
143bede5 | 2406 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2407 | { |
2408 | int i; | |
5f39d397 | 2409 | struct extent_buffer *t; |
0e82bcfe | 2410 | int ret; |
5f39d397 | 2411 | |
234b63a0 | 2412 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2413 | int tslot = path->slots[i]; |
0e82bcfe | 2414 | |
eb60ceac | 2415 | if (!path->nodes[i]) |
be0e5c09 | 2416 | break; |
5f39d397 | 2417 | t = path->nodes[i]; |
f3a84ccd FM |
2418 | ret = btrfs_tree_mod_log_insert_key(t, tslot, |
2419 | BTRFS_MOD_LOG_KEY_REPLACE, GFP_ATOMIC); | |
0e82bcfe | 2420 | BUG_ON(ret < 0); |
5f39d397 | 2421 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2422 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2423 | if (tslot != 0) |
2424 | break; | |
2425 | } | |
2426 | } | |
2427 | ||
31840ae1 ZY |
2428 | /* |
2429 | * update item key. | |
2430 | * | |
2431 | * This function isn't completely safe. It's the caller's responsibility | |
2432 | * that the new key won't break the order | |
2433 | */ | |
b7a0365e DD |
2434 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
2435 | struct btrfs_path *path, | |
310712b2 | 2436 | const struct btrfs_key *new_key) |
31840ae1 ZY |
2437 | { |
2438 | struct btrfs_disk_key disk_key; | |
2439 | struct extent_buffer *eb; | |
2440 | int slot; | |
2441 | ||
2442 | eb = path->nodes[0]; | |
2443 | slot = path->slots[0]; | |
2444 | if (slot > 0) { | |
2445 | btrfs_item_key(eb, &disk_key, slot - 1); | |
7c15d410 QW |
2446 | if (unlikely(comp_keys(&disk_key, new_key) >= 0)) { |
2447 | btrfs_crit(fs_info, | |
2448 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2449 | slot, btrfs_disk_key_objectid(&disk_key), | |
2450 | btrfs_disk_key_type(&disk_key), | |
2451 | btrfs_disk_key_offset(&disk_key), | |
2452 | new_key->objectid, new_key->type, | |
2453 | new_key->offset); | |
2454 | btrfs_print_leaf(eb); | |
2455 | BUG(); | |
2456 | } | |
31840ae1 ZY |
2457 | } |
2458 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2459 | btrfs_item_key(eb, &disk_key, slot + 1); | |
7c15d410 QW |
2460 | if (unlikely(comp_keys(&disk_key, new_key) <= 0)) { |
2461 | btrfs_crit(fs_info, | |
2462 | "slot %u key (%llu %u %llu) new key (%llu %u %llu)", | |
2463 | slot, btrfs_disk_key_objectid(&disk_key), | |
2464 | btrfs_disk_key_type(&disk_key), | |
2465 | btrfs_disk_key_offset(&disk_key), | |
2466 | new_key->objectid, new_key->type, | |
2467 | new_key->offset); | |
2468 | btrfs_print_leaf(eb); | |
2469 | BUG(); | |
2470 | } | |
31840ae1 ZY |
2471 | } |
2472 | ||
2473 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2474 | btrfs_set_item_key(eb, &disk_key, slot); | |
2475 | btrfs_mark_buffer_dirty(eb); | |
2476 | if (slot == 0) | |
b167fa91 | 2477 | fixup_low_keys(path, &disk_key, 1); |
31840ae1 ZY |
2478 | } |
2479 | ||
d16c702f QW |
2480 | /* |
2481 | * Check key order of two sibling extent buffers. | |
2482 | * | |
2483 | * Return true if something is wrong. | |
2484 | * Return false if everything is fine. | |
2485 | * | |
2486 | * Tree-checker only works inside one tree block, thus the following | |
2487 | * corruption can not be detected by tree-checker: | |
2488 | * | |
2489 | * Leaf @left | Leaf @right | |
2490 | * -------------------------------------------------------------- | |
2491 | * | 1 | 2 | 3 | 4 | 5 | f6 | | 7 | 8 | | |
2492 | * | |
2493 | * Key f6 in leaf @left itself is valid, but not valid when the next | |
2494 | * key in leaf @right is 7. | |
2495 | * This can only be checked at tree block merge time. | |
2496 | * And since tree checker has ensured all key order in each tree block | |
2497 | * is correct, we only need to bother the last key of @left and the first | |
2498 | * key of @right. | |
2499 | */ | |
2500 | static bool check_sibling_keys(struct extent_buffer *left, | |
2501 | struct extent_buffer *right) | |
2502 | { | |
2503 | struct btrfs_key left_last; | |
2504 | struct btrfs_key right_first; | |
2505 | int level = btrfs_header_level(left); | |
2506 | int nr_left = btrfs_header_nritems(left); | |
2507 | int nr_right = btrfs_header_nritems(right); | |
2508 | ||
2509 | /* No key to check in one of the tree blocks */ | |
2510 | if (!nr_left || !nr_right) | |
2511 | return false; | |
2512 | ||
2513 | if (level) { | |
2514 | btrfs_node_key_to_cpu(left, &left_last, nr_left - 1); | |
2515 | btrfs_node_key_to_cpu(right, &right_first, 0); | |
2516 | } else { | |
2517 | btrfs_item_key_to_cpu(left, &left_last, nr_left - 1); | |
2518 | btrfs_item_key_to_cpu(right, &right_first, 0); | |
2519 | } | |
2520 | ||
2521 | if (btrfs_comp_cpu_keys(&left_last, &right_first) >= 0) { | |
2522 | btrfs_crit(left->fs_info, | |
2523 | "bad key order, sibling blocks, left last (%llu %u %llu) right first (%llu %u %llu)", | |
2524 | left_last.objectid, left_last.type, | |
2525 | left_last.offset, right_first.objectid, | |
2526 | right_first.type, right_first.offset); | |
2527 | return true; | |
2528 | } | |
2529 | return false; | |
2530 | } | |
2531 | ||
74123bd7 CM |
2532 | /* |
2533 | * try to push data from one node into the next node left in the | |
79f95c82 | 2534 | * tree. |
aa5d6bed CM |
2535 | * |
2536 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2537 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2538 | */ |
98ed5174 | 2539 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e | 2540 | struct extent_buffer *dst, |
971a1f66 | 2541 | struct extent_buffer *src, int empty) |
be0e5c09 | 2542 | { |
d30a668f | 2543 | struct btrfs_fs_info *fs_info = trans->fs_info; |
be0e5c09 | 2544 | int push_items = 0; |
bb803951 CM |
2545 | int src_nritems; |
2546 | int dst_nritems; | |
aa5d6bed | 2547 | int ret = 0; |
be0e5c09 | 2548 | |
5f39d397 CM |
2549 | src_nritems = btrfs_header_nritems(src); |
2550 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2551 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
2552 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2553 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2554 | |
bce4eae9 | 2555 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2556 | return 1; |
2557 | ||
d397712b | 2558 | if (push_items <= 0) |
be0e5c09 CM |
2559 | return 1; |
2560 | ||
bce4eae9 | 2561 | if (empty) { |
971a1f66 | 2562 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2563 | if (push_items < src_nritems) { |
2564 | /* leave at least 8 pointers in the node if | |
2565 | * we aren't going to empty it | |
2566 | */ | |
2567 | if (src_nritems - push_items < 8) { | |
2568 | if (push_items <= 8) | |
2569 | return 1; | |
2570 | push_items -= 8; | |
2571 | } | |
2572 | } | |
2573 | } else | |
2574 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2575 | |
d16c702f QW |
2576 | /* dst is the left eb, src is the middle eb */ |
2577 | if (check_sibling_keys(dst, src)) { | |
2578 | ret = -EUCLEAN; | |
2579 | btrfs_abort_transaction(trans, ret); | |
2580 | return ret; | |
2581 | } | |
f3a84ccd | 2582 | ret = btrfs_tree_mod_log_eb_copy(dst, src, dst_nritems, 0, push_items); |
5de865ee | 2583 | if (ret) { |
66642832 | 2584 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2585 | return ret; |
2586 | } | |
5f39d397 CM |
2587 | copy_extent_buffer(dst, src, |
2588 | btrfs_node_key_ptr_offset(dst_nritems), | |
2589 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2590 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2591 | |
bb803951 | 2592 | if (push_items < src_nritems) { |
57911b8b | 2593 | /* |
f3a84ccd FM |
2594 | * Don't call btrfs_tree_mod_log_insert_move() here, key removal |
2595 | * was already fully logged by btrfs_tree_mod_log_eb_copy() above. | |
57911b8b | 2596 | */ |
5f39d397 CM |
2597 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2598 | btrfs_node_key_ptr_offset(push_items), | |
2599 | (src_nritems - push_items) * | |
2600 | sizeof(struct btrfs_key_ptr)); | |
2601 | } | |
2602 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2603 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
2604 | btrfs_mark_buffer_dirty(src); | |
2605 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2606 | |
79f95c82 CM |
2607 | return ret; |
2608 | } | |
2609 | ||
2610 | /* | |
2611 | * try to push data from one node into the next node right in the | |
2612 | * tree. | |
2613 | * | |
2614 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
2615 | * error, and > 0 if there was no room in the right hand block. | |
2616 | * | |
2617 | * this will only push up to 1/2 the contents of the left node over | |
2618 | */ | |
5f39d397 | 2619 | static int balance_node_right(struct btrfs_trans_handle *trans, |
5f39d397 CM |
2620 | struct extent_buffer *dst, |
2621 | struct extent_buffer *src) | |
79f95c82 | 2622 | { |
55d32ed8 | 2623 | struct btrfs_fs_info *fs_info = trans->fs_info; |
79f95c82 CM |
2624 | int push_items = 0; |
2625 | int max_push; | |
2626 | int src_nritems; | |
2627 | int dst_nritems; | |
2628 | int ret = 0; | |
79f95c82 | 2629 | |
7bb86316 CM |
2630 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2631 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
2632 | ||
5f39d397 CM |
2633 | src_nritems = btrfs_header_nritems(src); |
2634 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 2635 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 2636 | if (push_items <= 0) |
79f95c82 | 2637 | return 1; |
bce4eae9 | 2638 | |
d397712b | 2639 | if (src_nritems < 4) |
bce4eae9 | 2640 | return 1; |
79f95c82 CM |
2641 | |
2642 | max_push = src_nritems / 2 + 1; | |
2643 | /* don't try to empty the node */ | |
d397712b | 2644 | if (max_push >= src_nritems) |
79f95c82 | 2645 | return 1; |
252c38f0 | 2646 | |
79f95c82 CM |
2647 | if (max_push < push_items) |
2648 | push_items = max_push; | |
2649 | ||
d16c702f QW |
2650 | /* dst is the right eb, src is the middle eb */ |
2651 | if (check_sibling_keys(src, dst)) { | |
2652 | ret = -EUCLEAN; | |
2653 | btrfs_abort_transaction(trans, ret); | |
2654 | return ret; | |
2655 | } | |
f3a84ccd | 2656 | ret = btrfs_tree_mod_log_insert_move(dst, push_items, 0, dst_nritems); |
bf1d3425 | 2657 | BUG_ON(ret < 0); |
5f39d397 CM |
2658 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
2659 | btrfs_node_key_ptr_offset(0), | |
2660 | (dst_nritems) * | |
2661 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 2662 | |
f3a84ccd FM |
2663 | ret = btrfs_tree_mod_log_eb_copy(dst, src, 0, src_nritems - push_items, |
2664 | push_items); | |
5de865ee | 2665 | if (ret) { |
66642832 | 2666 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2667 | return ret; |
2668 | } | |
5f39d397 CM |
2669 | copy_extent_buffer(dst, src, |
2670 | btrfs_node_key_ptr_offset(0), | |
2671 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 2672 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 2673 | |
5f39d397 CM |
2674 | btrfs_set_header_nritems(src, src_nritems - push_items); |
2675 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 2676 | |
5f39d397 CM |
2677 | btrfs_mark_buffer_dirty(src); |
2678 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 2679 | |
aa5d6bed | 2680 | return ret; |
be0e5c09 CM |
2681 | } |
2682 | ||
97571fd0 CM |
2683 | /* |
2684 | * helper function to insert a new root level in the tree. | |
2685 | * A new node is allocated, and a single item is inserted to | |
2686 | * point to the existing root | |
aa5d6bed CM |
2687 | * |
2688 | * returns zero on success or < 0 on failure. | |
97571fd0 | 2689 | */ |
d397712b | 2690 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 2691 | struct btrfs_root *root, |
fdd99c72 | 2692 | struct btrfs_path *path, int level) |
5c680ed6 | 2693 | { |
0b246afa | 2694 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 2695 | u64 lower_gen; |
5f39d397 CM |
2696 | struct extent_buffer *lower; |
2697 | struct extent_buffer *c; | |
925baedd | 2698 | struct extent_buffer *old; |
5f39d397 | 2699 | struct btrfs_disk_key lower_key; |
d9d19a01 | 2700 | int ret; |
5c680ed6 CM |
2701 | |
2702 | BUG_ON(path->nodes[level]); | |
2703 | BUG_ON(path->nodes[level-1] != root->node); | |
2704 | ||
7bb86316 CM |
2705 | lower = path->nodes[level-1]; |
2706 | if (level == 1) | |
2707 | btrfs_item_key(lower, &lower_key, 0); | |
2708 | else | |
2709 | btrfs_node_key(lower, &lower_key, 0); | |
2710 | ||
79bd3712 FM |
2711 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2712 | &lower_key, level, root->node->start, 0, | |
2713 | BTRFS_NESTING_NEW_ROOT); | |
5f39d397 CM |
2714 | if (IS_ERR(c)) |
2715 | return PTR_ERR(c); | |
925baedd | 2716 | |
0b246afa | 2717 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 2718 | |
5f39d397 | 2719 | btrfs_set_header_nritems(c, 1); |
5f39d397 | 2720 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 2721 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 2722 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 2723 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
2724 | |
2725 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 2726 | |
5f39d397 | 2727 | btrfs_mark_buffer_dirty(c); |
d5719762 | 2728 | |
925baedd | 2729 | old = root->node; |
406808ab | 2730 | ret = btrfs_tree_mod_log_insert_root(root->node, c, false); |
d9d19a01 | 2731 | BUG_ON(ret < 0); |
240f62c8 | 2732 | rcu_assign_pointer(root->node, c); |
925baedd CM |
2733 | |
2734 | /* the super has an extra ref to root->node */ | |
2735 | free_extent_buffer(old); | |
2736 | ||
0b86a832 | 2737 | add_root_to_dirty_list(root); |
67439dad | 2738 | atomic_inc(&c->refs); |
5f39d397 | 2739 | path->nodes[level] = c; |
ac5887c8 | 2740 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
2741 | path->slots[level] = 0; |
2742 | return 0; | |
2743 | } | |
2744 | ||
74123bd7 CM |
2745 | /* |
2746 | * worker function to insert a single pointer in a node. | |
2747 | * the node should have enough room for the pointer already | |
97571fd0 | 2748 | * |
74123bd7 CM |
2749 | * slot and level indicate where you want the key to go, and |
2750 | * blocknr is the block the key points to. | |
2751 | */ | |
143bede5 | 2752 | static void insert_ptr(struct btrfs_trans_handle *trans, |
6ad3cf6d | 2753 | struct btrfs_path *path, |
143bede5 | 2754 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 2755 | int slot, int level) |
74123bd7 | 2756 | { |
5f39d397 | 2757 | struct extent_buffer *lower; |
74123bd7 | 2758 | int nritems; |
f3ea38da | 2759 | int ret; |
5c680ed6 CM |
2760 | |
2761 | BUG_ON(!path->nodes[level]); | |
49d0c642 | 2762 | btrfs_assert_tree_write_locked(path->nodes[level]); |
5f39d397 CM |
2763 | lower = path->nodes[level]; |
2764 | nritems = btrfs_header_nritems(lower); | |
c293498b | 2765 | BUG_ON(slot > nritems); |
6ad3cf6d | 2766 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(trans->fs_info)); |
74123bd7 | 2767 | if (slot != nritems) { |
bf1d3425 | 2768 | if (level) { |
f3a84ccd FM |
2769 | ret = btrfs_tree_mod_log_insert_move(lower, slot + 1, |
2770 | slot, nritems - slot); | |
bf1d3425 DS |
2771 | BUG_ON(ret < 0); |
2772 | } | |
5f39d397 CM |
2773 | memmove_extent_buffer(lower, |
2774 | btrfs_node_key_ptr_offset(slot + 1), | |
2775 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 2776 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 2777 | } |
c3e06965 | 2778 | if (level) { |
f3a84ccd FM |
2779 | ret = btrfs_tree_mod_log_insert_key(lower, slot, |
2780 | BTRFS_MOD_LOG_KEY_ADD, GFP_NOFS); | |
f3ea38da JS |
2781 | BUG_ON(ret < 0); |
2782 | } | |
5f39d397 | 2783 | btrfs_set_node_key(lower, key, slot); |
db94535d | 2784 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
2785 | WARN_ON(trans->transid == 0); |
2786 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
2787 | btrfs_set_header_nritems(lower, nritems + 1); |
2788 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
2789 | } |
2790 | ||
97571fd0 CM |
2791 | /* |
2792 | * split the node at the specified level in path in two. | |
2793 | * The path is corrected to point to the appropriate node after the split | |
2794 | * | |
2795 | * Before splitting this tries to make some room in the node by pushing | |
2796 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
2797 | * |
2798 | * returns 0 on success and < 0 on failure | |
97571fd0 | 2799 | */ |
e02119d5 CM |
2800 | static noinline int split_node(struct btrfs_trans_handle *trans, |
2801 | struct btrfs_root *root, | |
2802 | struct btrfs_path *path, int level) | |
be0e5c09 | 2803 | { |
0b246afa | 2804 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2805 | struct extent_buffer *c; |
2806 | struct extent_buffer *split; | |
2807 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 2808 | int mid; |
5c680ed6 | 2809 | int ret; |
7518a238 | 2810 | u32 c_nritems; |
eb60ceac | 2811 | |
5f39d397 | 2812 | c = path->nodes[level]; |
7bb86316 | 2813 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 2814 | if (c == root->node) { |
d9abbf1c | 2815 | /* |
90f8d62e JS |
2816 | * trying to split the root, lets make a new one |
2817 | * | |
fdd99c72 | 2818 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
2819 | * insert_new_root, because that root buffer will be kept as a |
2820 | * normal node. We are going to log removal of half of the | |
f3a84ccd FM |
2821 | * elements below with btrfs_tree_mod_log_eb_copy(). We're |
2822 | * holding a tree lock on the buffer, which is why we cannot | |
2823 | * race with other tree_mod_log users. | |
d9abbf1c | 2824 | */ |
fdd99c72 | 2825 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
2826 | if (ret) |
2827 | return ret; | |
b3612421 | 2828 | } else { |
e66f709b | 2829 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
2830 | c = path->nodes[level]; |
2831 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 2832 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 2833 | return 0; |
54aa1f4d CM |
2834 | if (ret < 0) |
2835 | return ret; | |
be0e5c09 | 2836 | } |
e66f709b | 2837 | |
5f39d397 | 2838 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
2839 | mid = (c_nritems + 1) / 2; |
2840 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 2841 | |
79bd3712 FM |
2842 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
2843 | &disk_key, level, c->start, 0, | |
2844 | BTRFS_NESTING_SPLIT); | |
5f39d397 CM |
2845 | if (IS_ERR(split)) |
2846 | return PTR_ERR(split); | |
2847 | ||
0b246afa | 2848 | root_add_used(root, fs_info->nodesize); |
bc877d28 | 2849 | ASSERT(btrfs_header_level(c) == level); |
54aa1f4d | 2850 | |
f3a84ccd | 2851 | ret = btrfs_tree_mod_log_eb_copy(split, c, 0, mid, c_nritems - mid); |
5de865ee | 2852 | if (ret) { |
66642832 | 2853 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
2854 | return ret; |
2855 | } | |
5f39d397 CM |
2856 | copy_extent_buffer(split, c, |
2857 | btrfs_node_key_ptr_offset(0), | |
2858 | btrfs_node_key_ptr_offset(mid), | |
2859 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
2860 | btrfs_set_header_nritems(split, c_nritems - mid); | |
2861 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed | 2862 | |
5f39d397 CM |
2863 | btrfs_mark_buffer_dirty(c); |
2864 | btrfs_mark_buffer_dirty(split); | |
2865 | ||
6ad3cf6d | 2866 | insert_ptr(trans, path, &disk_key, split->start, |
c3e06965 | 2867 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 2868 | |
5de08d7d | 2869 | if (path->slots[level] >= mid) { |
5c680ed6 | 2870 | path->slots[level] -= mid; |
925baedd | 2871 | btrfs_tree_unlock(c); |
5f39d397 CM |
2872 | free_extent_buffer(c); |
2873 | path->nodes[level] = split; | |
5c680ed6 CM |
2874 | path->slots[level + 1] += 1; |
2875 | } else { | |
925baedd | 2876 | btrfs_tree_unlock(split); |
5f39d397 | 2877 | free_extent_buffer(split); |
be0e5c09 | 2878 | } |
d5286a92 | 2879 | return 0; |
be0e5c09 CM |
2880 | } |
2881 | ||
74123bd7 CM |
2882 | /* |
2883 | * how many bytes are required to store the items in a leaf. start | |
2884 | * and nr indicate which items in the leaf to check. This totals up the | |
2885 | * space used both by the item structs and the item data | |
2886 | */ | |
5f39d397 | 2887 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 CM |
2888 | { |
2889 | int data_len; | |
5f39d397 | 2890 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 2891 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
2892 | |
2893 | if (!nr) | |
2894 | return 0; | |
3212fa14 JB |
2895 | data_len = btrfs_item_offset(l, start) + btrfs_item_size(l, start); |
2896 | data_len = data_len - btrfs_item_offset(l, end); | |
0783fcfc | 2897 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 2898 | WARN_ON(data_len < 0); |
be0e5c09 CM |
2899 | return data_len; |
2900 | } | |
2901 | ||
d4dbff95 CM |
2902 | /* |
2903 | * The space between the end of the leaf items and | |
2904 | * the start of the leaf data. IOW, how much room | |
2905 | * the leaf has left for both items and data | |
2906 | */ | |
e902baac | 2907 | noinline int btrfs_leaf_free_space(struct extent_buffer *leaf) |
d4dbff95 | 2908 | { |
e902baac | 2909 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
5f39d397 CM |
2910 | int nritems = btrfs_header_nritems(leaf); |
2911 | int ret; | |
0b246afa JM |
2912 | |
2913 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 2914 | if (ret < 0) { |
0b246afa JM |
2915 | btrfs_crit(fs_info, |
2916 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
2917 | ret, | |
2918 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
2919 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
2920 | } |
2921 | return ret; | |
d4dbff95 CM |
2922 | } |
2923 | ||
99d8f83c CM |
2924 | /* |
2925 | * min slot controls the lowest index we're willing to push to the | |
2926 | * right. We'll push up to and including min_slot, but no lower | |
2927 | */ | |
f72f0010 | 2928 | static noinline int __push_leaf_right(struct btrfs_path *path, |
44871b1b CM |
2929 | int data_size, int empty, |
2930 | struct extent_buffer *right, | |
99d8f83c CM |
2931 | int free_space, u32 left_nritems, |
2932 | u32 min_slot) | |
00ec4c51 | 2933 | { |
f72f0010 | 2934 | struct btrfs_fs_info *fs_info = right->fs_info; |
5f39d397 | 2935 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 2936 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 2937 | struct btrfs_map_token token; |
5f39d397 | 2938 | struct btrfs_disk_key disk_key; |
00ec4c51 | 2939 | int slot; |
34a38218 | 2940 | u32 i; |
00ec4c51 CM |
2941 | int push_space = 0; |
2942 | int push_items = 0; | |
34a38218 | 2943 | u32 nr; |
7518a238 | 2944 | u32 right_nritems; |
5f39d397 | 2945 | u32 data_end; |
db94535d | 2946 | u32 this_item_size; |
00ec4c51 | 2947 | |
34a38218 CM |
2948 | if (empty) |
2949 | nr = 0; | |
2950 | else | |
99d8f83c | 2951 | nr = max_t(u32, 1, min_slot); |
34a38218 | 2952 | |
31840ae1 | 2953 | if (path->slots[0] >= left_nritems) |
87b29b20 | 2954 | push_space += data_size; |
31840ae1 | 2955 | |
44871b1b | 2956 | slot = path->slots[1]; |
34a38218 CM |
2957 | i = left_nritems - 1; |
2958 | while (i >= nr) { | |
31840ae1 ZY |
2959 | if (!empty && push_items > 0) { |
2960 | if (path->slots[0] > i) | |
2961 | break; | |
2962 | if (path->slots[0] == i) { | |
e902baac DS |
2963 | int space = btrfs_leaf_free_space(left); |
2964 | ||
31840ae1 ZY |
2965 | if (space + push_space * 2 > free_space) |
2966 | break; | |
2967 | } | |
2968 | } | |
2969 | ||
00ec4c51 | 2970 | if (path->slots[0] == i) |
87b29b20 | 2971 | push_space += data_size; |
db94535d | 2972 | |
3212fa14 | 2973 | this_item_size = btrfs_item_size(left, i); |
74794207 JB |
2974 | if (this_item_size + sizeof(struct btrfs_item) + |
2975 | push_space > free_space) | |
00ec4c51 | 2976 | break; |
31840ae1 | 2977 | |
00ec4c51 | 2978 | push_items++; |
74794207 | 2979 | push_space += this_item_size + sizeof(struct btrfs_item); |
34a38218 CM |
2980 | if (i == 0) |
2981 | break; | |
2982 | i--; | |
db94535d | 2983 | } |
5f39d397 | 2984 | |
925baedd CM |
2985 | if (push_items == 0) |
2986 | goto out_unlock; | |
5f39d397 | 2987 | |
6c1500f2 | 2988 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 2989 | |
00ec4c51 | 2990 | /* push left to right */ |
5f39d397 | 2991 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 2992 | |
dc2e724e | 2993 | push_space = btrfs_item_data_end(left, left_nritems - push_items); |
8f881e8c | 2994 | push_space -= leaf_data_end(left); |
5f39d397 | 2995 | |
00ec4c51 | 2996 | /* make room in the right data area */ |
8f881e8c | 2997 | data_end = leaf_data_end(right); |
5f39d397 | 2998 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
2999 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3000 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3001 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3002 | |
00ec4c51 | 3003 | /* copy from the left data area */ |
3d9ec8c4 | 3004 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3005 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
8f881e8c | 3006 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(left), |
d6025579 | 3007 | push_space); |
5f39d397 CM |
3008 | |
3009 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3010 | btrfs_item_nr_offset(0), | |
3011 | right_nritems * sizeof(struct btrfs_item)); | |
3012 | ||
00ec4c51 | 3013 | /* copy the items from left to right */ |
5f39d397 CM |
3014 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3015 | btrfs_item_nr_offset(left_nritems - push_items), | |
3016 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3017 | |
3018 | /* update the item pointers */ | |
c82f823c | 3019 | btrfs_init_map_token(&token, right); |
7518a238 | 3020 | right_nritems += push_items; |
5f39d397 | 3021 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3022 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3023 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
3024 | push_space -= btrfs_token_item_size(&token, i); |
3025 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d CM |
3026 | } |
3027 | ||
7518a238 | 3028 | left_nritems -= push_items; |
5f39d397 | 3029 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3030 | |
34a38218 CM |
3031 | if (left_nritems) |
3032 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3033 | else |
6a884d7d | 3034 | btrfs_clean_tree_block(left); |
f0486c68 | 3035 | |
5f39d397 | 3036 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3037 | |
5f39d397 CM |
3038 | btrfs_item_key(right, &disk_key, 0); |
3039 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3040 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3041 | |
00ec4c51 | 3042 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3043 | if (path->slots[0] >= left_nritems) { |
3044 | path->slots[0] -= left_nritems; | |
925baedd | 3045 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
6a884d7d | 3046 | btrfs_clean_tree_block(path->nodes[0]); |
925baedd | 3047 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3048 | free_extent_buffer(path->nodes[0]); |
3049 | path->nodes[0] = right; | |
00ec4c51 CM |
3050 | path->slots[1] += 1; |
3051 | } else { | |
925baedd | 3052 | btrfs_tree_unlock(right); |
5f39d397 | 3053 | free_extent_buffer(right); |
00ec4c51 CM |
3054 | } |
3055 | return 0; | |
925baedd CM |
3056 | |
3057 | out_unlock: | |
3058 | btrfs_tree_unlock(right); | |
3059 | free_extent_buffer(right); | |
3060 | return 1; | |
00ec4c51 | 3061 | } |
925baedd | 3062 | |
44871b1b CM |
3063 | /* |
3064 | * push some data in the path leaf to the right, trying to free up at | |
3065 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3066 | * | |
3067 | * returns 1 if the push failed because the other node didn't have enough | |
3068 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3069 | * |
3070 | * this will push starting from min_slot to the end of the leaf. It won't | |
3071 | * push any slot lower than min_slot | |
44871b1b CM |
3072 | */ |
3073 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3074 | *root, struct btrfs_path *path, |
3075 | int min_data_size, int data_size, | |
3076 | int empty, u32 min_slot) | |
44871b1b CM |
3077 | { |
3078 | struct extent_buffer *left = path->nodes[0]; | |
3079 | struct extent_buffer *right; | |
3080 | struct extent_buffer *upper; | |
3081 | int slot; | |
3082 | int free_space; | |
3083 | u32 left_nritems; | |
3084 | int ret; | |
3085 | ||
3086 | if (!path->nodes[1]) | |
3087 | return 1; | |
3088 | ||
3089 | slot = path->slots[1]; | |
3090 | upper = path->nodes[1]; | |
3091 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3092 | return 1; | |
3093 | ||
49d0c642 | 3094 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3095 | |
4b231ae4 | 3096 | right = btrfs_read_node_slot(upper, slot + 1); |
fb770ae4 LB |
3097 | /* |
3098 | * slot + 1 is not valid or we fail to read the right node, | |
3099 | * no big deal, just return. | |
3100 | */ | |
3101 | if (IS_ERR(right)) | |
91ca338d TI |
3102 | return 1; |
3103 | ||
bf77467a | 3104 | __btrfs_tree_lock(right, BTRFS_NESTING_RIGHT); |
44871b1b | 3105 | |
e902baac | 3106 | free_space = btrfs_leaf_free_space(right); |
44871b1b CM |
3107 | if (free_space < data_size) |
3108 | goto out_unlock; | |
3109 | ||
44871b1b | 3110 | ret = btrfs_cow_block(trans, root, right, upper, |
bf59a5a2 | 3111 | slot + 1, &right, BTRFS_NESTING_RIGHT_COW); |
44871b1b CM |
3112 | if (ret) |
3113 | goto out_unlock; | |
3114 | ||
44871b1b CM |
3115 | left_nritems = btrfs_header_nritems(left); |
3116 | if (left_nritems == 0) | |
3117 | goto out_unlock; | |
3118 | ||
d16c702f QW |
3119 | if (check_sibling_keys(left, right)) { |
3120 | ret = -EUCLEAN; | |
3121 | btrfs_tree_unlock(right); | |
3122 | free_extent_buffer(right); | |
3123 | return ret; | |
3124 | } | |
2ef1fed2 FDBM |
3125 | if (path->slots[0] == left_nritems && !empty) { |
3126 | /* Key greater than all keys in the leaf, right neighbor has | |
3127 | * enough room for it and we're not emptying our leaf to delete | |
3128 | * it, therefore use right neighbor to insert the new item and | |
52042d8e | 3129 | * no need to touch/dirty our left leaf. */ |
2ef1fed2 FDBM |
3130 | btrfs_tree_unlock(left); |
3131 | free_extent_buffer(left); | |
3132 | path->nodes[0] = right; | |
3133 | path->slots[0] = 0; | |
3134 | path->slots[1]++; | |
3135 | return 0; | |
3136 | } | |
3137 | ||
f72f0010 | 3138 | return __push_leaf_right(path, min_data_size, empty, |
99d8f83c | 3139 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3140 | out_unlock: |
3141 | btrfs_tree_unlock(right); | |
3142 | free_extent_buffer(right); | |
3143 | return 1; | |
3144 | } | |
3145 | ||
74123bd7 CM |
3146 | /* |
3147 | * push some data in the path leaf to the left, trying to free up at | |
3148 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3149 | * |
3150 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3151 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3152 | * items | |
74123bd7 | 3153 | */ |
8087c193 | 3154 | static noinline int __push_leaf_left(struct btrfs_path *path, int data_size, |
44871b1b | 3155 | int empty, struct extent_buffer *left, |
99d8f83c CM |
3156 | int free_space, u32 right_nritems, |
3157 | u32 max_slot) | |
be0e5c09 | 3158 | { |
8087c193 | 3159 | struct btrfs_fs_info *fs_info = left->fs_info; |
5f39d397 CM |
3160 | struct btrfs_disk_key disk_key; |
3161 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3162 | int i; |
be0e5c09 CM |
3163 | int push_space = 0; |
3164 | int push_items = 0; | |
7518a238 | 3165 | u32 old_left_nritems; |
34a38218 | 3166 | u32 nr; |
aa5d6bed | 3167 | int ret = 0; |
db94535d CM |
3168 | u32 this_item_size; |
3169 | u32 old_left_item_size; | |
cfed81a0 CM |
3170 | struct btrfs_map_token token; |
3171 | ||
34a38218 | 3172 | if (empty) |
99d8f83c | 3173 | nr = min(right_nritems, max_slot); |
34a38218 | 3174 | else |
99d8f83c | 3175 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3176 | |
3177 | for (i = 0; i < nr; i++) { | |
31840ae1 ZY |
3178 | if (!empty && push_items > 0) { |
3179 | if (path->slots[0] < i) | |
3180 | break; | |
3181 | if (path->slots[0] == i) { | |
e902baac DS |
3182 | int space = btrfs_leaf_free_space(right); |
3183 | ||
31840ae1 ZY |
3184 | if (space + push_space * 2 > free_space) |
3185 | break; | |
3186 | } | |
3187 | } | |
3188 | ||
be0e5c09 | 3189 | if (path->slots[0] == i) |
87b29b20 | 3190 | push_space += data_size; |
db94535d | 3191 | |
3212fa14 | 3192 | this_item_size = btrfs_item_size(right, i); |
74794207 JB |
3193 | if (this_item_size + sizeof(struct btrfs_item) + push_space > |
3194 | free_space) | |
be0e5c09 | 3195 | break; |
db94535d | 3196 | |
be0e5c09 | 3197 | push_items++; |
74794207 | 3198 | push_space += this_item_size + sizeof(struct btrfs_item); |
db94535d CM |
3199 | } |
3200 | ||
be0e5c09 | 3201 | if (push_items == 0) { |
925baedd CM |
3202 | ret = 1; |
3203 | goto out; | |
be0e5c09 | 3204 | } |
fae7f21c | 3205 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3206 | |
be0e5c09 | 3207 | /* push data from right to left */ |
5f39d397 CM |
3208 | copy_extent_buffer(left, right, |
3209 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3210 | btrfs_item_nr_offset(0), | |
3211 | push_items * sizeof(struct btrfs_item)); | |
3212 | ||
0b246afa | 3213 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
3212fa14 | 3214 | btrfs_item_offset(right, push_items - 1); |
5f39d397 | 3215 | |
3d9ec8c4 | 3216 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3217 | leaf_data_end(left) - push_space, |
3d9ec8c4 | 3218 | BTRFS_LEAF_DATA_OFFSET + |
3212fa14 | 3219 | btrfs_item_offset(right, push_items - 1), |
d6025579 | 3220 | push_space); |
5f39d397 | 3221 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3222 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3223 | |
c82f823c | 3224 | btrfs_init_map_token(&token, left); |
3212fa14 | 3225 | old_left_item_size = btrfs_item_offset(left, old_left_nritems - 1); |
0783fcfc | 3226 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3227 | u32 ioff; |
db94535d | 3228 | |
3212fa14 JB |
3229 | ioff = btrfs_token_item_offset(&token, i); |
3230 | btrfs_set_token_item_offset(&token, i, | |
cc4c13d5 | 3231 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size)); |
be0e5c09 | 3232 | } |
5f39d397 | 3233 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3234 | |
3235 | /* fixup right node */ | |
31b1a2bd JL |
3236 | if (push_items > right_nritems) |
3237 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3238 | right_nritems); |
34a38218 CM |
3239 | |
3240 | if (push_items < right_nritems) { | |
3212fa14 | 3241 | push_space = btrfs_item_offset(right, push_items - 1) - |
8f881e8c | 3242 | leaf_data_end(right); |
3d9ec8c4 | 3243 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3244 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3245 | BTRFS_LEAF_DATA_OFFSET + |
8f881e8c | 3246 | leaf_data_end(right), push_space); |
34a38218 CM |
3247 | |
3248 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3249 | btrfs_item_nr_offset(push_items), |
3250 | (btrfs_header_nritems(right) - push_items) * | |
3251 | sizeof(struct btrfs_item)); | |
34a38218 | 3252 | } |
c82f823c DS |
3253 | |
3254 | btrfs_init_map_token(&token, right); | |
eef1c494 Y |
3255 | right_nritems -= push_items; |
3256 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3257 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3258 | for (i = 0; i < right_nritems; i++) { |
3212fa14 JB |
3259 | push_space = push_space - btrfs_token_item_size(&token, i); |
3260 | btrfs_set_token_item_offset(&token, i, push_space); | |
db94535d | 3261 | } |
eb60ceac | 3262 | |
5f39d397 | 3263 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3264 | if (right_nritems) |
3265 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3266 | else |
6a884d7d | 3267 | btrfs_clean_tree_block(right); |
098f59c2 | 3268 | |
5f39d397 | 3269 | btrfs_item_key(right, &disk_key, 0); |
b167fa91 | 3270 | fixup_low_keys(path, &disk_key, 1); |
be0e5c09 CM |
3271 | |
3272 | /* then fixup the leaf pointer in the path */ | |
3273 | if (path->slots[0] < push_items) { | |
3274 | path->slots[0] += old_left_nritems; | |
925baedd | 3275 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3276 | free_extent_buffer(path->nodes[0]); |
3277 | path->nodes[0] = left; | |
be0e5c09 CM |
3278 | path->slots[1] -= 1; |
3279 | } else { | |
925baedd | 3280 | btrfs_tree_unlock(left); |
5f39d397 | 3281 | free_extent_buffer(left); |
be0e5c09 CM |
3282 | path->slots[0] -= push_items; |
3283 | } | |
eb60ceac | 3284 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3285 | return ret; |
925baedd CM |
3286 | out: |
3287 | btrfs_tree_unlock(left); | |
3288 | free_extent_buffer(left); | |
3289 | return ret; | |
be0e5c09 CM |
3290 | } |
3291 | ||
44871b1b CM |
3292 | /* |
3293 | * push some data in the path leaf to the left, trying to free up at | |
3294 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3295 | * |
3296 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3297 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3298 | * items | |
44871b1b CM |
3299 | */ |
3300 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3301 | *root, struct btrfs_path *path, int min_data_size, |
3302 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3303 | { |
3304 | struct extent_buffer *right = path->nodes[0]; | |
3305 | struct extent_buffer *left; | |
3306 | int slot; | |
3307 | int free_space; | |
3308 | u32 right_nritems; | |
3309 | int ret = 0; | |
3310 | ||
3311 | slot = path->slots[1]; | |
3312 | if (slot == 0) | |
3313 | return 1; | |
3314 | if (!path->nodes[1]) | |
3315 | return 1; | |
3316 | ||
3317 | right_nritems = btrfs_header_nritems(right); | |
3318 | if (right_nritems == 0) | |
3319 | return 1; | |
3320 | ||
49d0c642 | 3321 | btrfs_assert_tree_write_locked(path->nodes[1]); |
44871b1b | 3322 | |
4b231ae4 | 3323 | left = btrfs_read_node_slot(path->nodes[1], slot - 1); |
fb770ae4 LB |
3324 | /* |
3325 | * slot - 1 is not valid or we fail to read the left node, | |
3326 | * no big deal, just return. | |
3327 | */ | |
3328 | if (IS_ERR(left)) | |
91ca338d TI |
3329 | return 1; |
3330 | ||
bf77467a | 3331 | __btrfs_tree_lock(left, BTRFS_NESTING_LEFT); |
44871b1b | 3332 | |
e902baac | 3333 | free_space = btrfs_leaf_free_space(left); |
44871b1b CM |
3334 | if (free_space < data_size) { |
3335 | ret = 1; | |
3336 | goto out; | |
3337 | } | |
3338 | ||
44871b1b | 3339 | ret = btrfs_cow_block(trans, root, left, |
9631e4cc | 3340 | path->nodes[1], slot - 1, &left, |
bf59a5a2 | 3341 | BTRFS_NESTING_LEFT_COW); |
44871b1b CM |
3342 | if (ret) { |
3343 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3344 | if (ret == -ENOSPC) |
3345 | ret = 1; | |
44871b1b CM |
3346 | goto out; |
3347 | } | |
3348 | ||
d16c702f QW |
3349 | if (check_sibling_keys(left, right)) { |
3350 | ret = -EUCLEAN; | |
3351 | goto out; | |
3352 | } | |
8087c193 | 3353 | return __push_leaf_left(path, min_data_size, |
99d8f83c CM |
3354 | empty, left, free_space, right_nritems, |
3355 | max_slot); | |
44871b1b CM |
3356 | out: |
3357 | btrfs_tree_unlock(left); | |
3358 | free_extent_buffer(left); | |
3359 | return ret; | |
3360 | } | |
3361 | ||
3362 | /* | |
3363 | * split the path's leaf in two, making sure there is at least data_size | |
3364 | * available for the resulting leaf level of the path. | |
44871b1b | 3365 | */ |
143bede5 | 3366 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
143bede5 JM |
3367 | struct btrfs_path *path, |
3368 | struct extent_buffer *l, | |
3369 | struct extent_buffer *right, | |
3370 | int slot, int mid, int nritems) | |
44871b1b | 3371 | { |
94f94ad9 | 3372 | struct btrfs_fs_info *fs_info = trans->fs_info; |
44871b1b CM |
3373 | int data_copy_size; |
3374 | int rt_data_off; | |
3375 | int i; | |
44871b1b | 3376 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3377 | struct btrfs_map_token token; |
3378 | ||
44871b1b CM |
3379 | nritems = nritems - mid; |
3380 | btrfs_set_header_nritems(right, nritems); | |
dc2e724e | 3381 | data_copy_size = btrfs_item_data_end(l, mid) - leaf_data_end(l); |
44871b1b CM |
3382 | |
3383 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3384 | btrfs_item_nr_offset(mid), | |
3385 | nritems * sizeof(struct btrfs_item)); | |
3386 | ||
3387 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
3388 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
3389 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
8f881e8c | 3390 | leaf_data_end(l), data_copy_size); |
44871b1b | 3391 | |
dc2e724e | 3392 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_data_end(l, mid); |
44871b1b | 3393 | |
c82f823c | 3394 | btrfs_init_map_token(&token, right); |
44871b1b | 3395 | for (i = 0; i < nritems; i++) { |
44871b1b CM |
3396 | u32 ioff; |
3397 | ||
3212fa14 JB |
3398 | ioff = btrfs_token_item_offset(&token, i); |
3399 | btrfs_set_token_item_offset(&token, i, ioff + rt_data_off); | |
44871b1b CM |
3400 | } |
3401 | ||
44871b1b | 3402 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3403 | btrfs_item_key(right, &disk_key, 0); |
6ad3cf6d | 3404 | insert_ptr(trans, path, &disk_key, right->start, path->slots[1] + 1, 1); |
44871b1b CM |
3405 | |
3406 | btrfs_mark_buffer_dirty(right); | |
3407 | btrfs_mark_buffer_dirty(l); | |
3408 | BUG_ON(path->slots[0] != slot); | |
3409 | ||
44871b1b CM |
3410 | if (mid <= slot) { |
3411 | btrfs_tree_unlock(path->nodes[0]); | |
3412 | free_extent_buffer(path->nodes[0]); | |
3413 | path->nodes[0] = right; | |
3414 | path->slots[0] -= mid; | |
3415 | path->slots[1] += 1; | |
3416 | } else { | |
3417 | btrfs_tree_unlock(right); | |
3418 | free_extent_buffer(right); | |
3419 | } | |
3420 | ||
3421 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3422 | } |
3423 | ||
99d8f83c CM |
3424 | /* |
3425 | * double splits happen when we need to insert a big item in the middle | |
3426 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3427 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3428 | * A B C | |
3429 | * | |
3430 | * We avoid this by trying to push the items on either side of our target | |
3431 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3432 | * completely. | |
3433 | */ | |
3434 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3435 | struct btrfs_root *root, | |
3436 | struct btrfs_path *path, | |
3437 | int data_size) | |
3438 | { | |
3439 | int ret; | |
3440 | int progress = 0; | |
3441 | int slot; | |
3442 | u32 nritems; | |
5a4267ca | 3443 | int space_needed = data_size; |
99d8f83c CM |
3444 | |
3445 | slot = path->slots[0]; | |
5a4267ca | 3446 | if (slot < btrfs_header_nritems(path->nodes[0])) |
e902baac | 3447 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
99d8f83c CM |
3448 | |
3449 | /* | |
3450 | * try to push all the items after our slot into the | |
3451 | * right leaf | |
3452 | */ | |
5a4267ca | 3453 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3454 | if (ret < 0) |
3455 | return ret; | |
3456 | ||
3457 | if (ret == 0) | |
3458 | progress++; | |
3459 | ||
3460 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3461 | /* | |
3462 | * our goal is to get our slot at the start or end of a leaf. If | |
3463 | * we've done so we're done | |
3464 | */ | |
3465 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3466 | return 0; | |
3467 | ||
e902baac | 3468 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3469 | return 0; |
3470 | ||
3471 | /* try to push all the items before our slot into the next leaf */ | |
3472 | slot = path->slots[0]; | |
263d3995 FM |
3473 | space_needed = data_size; |
3474 | if (slot > 0) | |
e902baac | 3475 | space_needed -= btrfs_leaf_free_space(path->nodes[0]); |
5a4267ca | 3476 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3477 | if (ret < 0) |
3478 | return ret; | |
3479 | ||
3480 | if (ret == 0) | |
3481 | progress++; | |
3482 | ||
3483 | if (progress) | |
3484 | return 0; | |
3485 | return 1; | |
3486 | } | |
3487 | ||
74123bd7 CM |
3488 | /* |
3489 | * split the path's leaf in two, making sure there is at least data_size | |
3490 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3491 | * |
3492 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3493 | */ |
e02119d5 CM |
3494 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3495 | struct btrfs_root *root, | |
310712b2 | 3496 | const struct btrfs_key *ins_key, |
e02119d5 CM |
3497 | struct btrfs_path *path, int data_size, |
3498 | int extend) | |
be0e5c09 | 3499 | { |
5d4f98a2 | 3500 | struct btrfs_disk_key disk_key; |
5f39d397 | 3501 | struct extent_buffer *l; |
7518a238 | 3502 | u32 nritems; |
eb60ceac CM |
3503 | int mid; |
3504 | int slot; | |
5f39d397 | 3505 | struct extent_buffer *right; |
b7a0365e | 3506 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 3507 | int ret = 0; |
aa5d6bed | 3508 | int wret; |
5d4f98a2 | 3509 | int split; |
cc0c5538 | 3510 | int num_doubles = 0; |
99d8f83c | 3511 | int tried_avoid_double = 0; |
aa5d6bed | 3512 | |
a5719521 YZ |
3513 | l = path->nodes[0]; |
3514 | slot = path->slots[0]; | |
3212fa14 | 3515 | if (extend && data_size + btrfs_item_size(l, slot) + |
0b246afa | 3516 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
3517 | return -EOVERFLOW; |
3518 | ||
40689478 | 3519 | /* first try to make some room by pushing left and right */ |
33157e05 | 3520 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
3521 | int space_needed = data_size; |
3522 | ||
3523 | if (slot < btrfs_header_nritems(l)) | |
e902baac | 3524 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3525 | |
3526 | wret = push_leaf_right(trans, root, path, space_needed, | |
3527 | space_needed, 0, 0); | |
d397712b | 3528 | if (wret < 0) |
eaee50e8 | 3529 | return wret; |
3685f791 | 3530 | if (wret) { |
263d3995 FM |
3531 | space_needed = data_size; |
3532 | if (slot > 0) | |
e902baac | 3533 | space_needed -= btrfs_leaf_free_space(l); |
5a4267ca FDBM |
3534 | wret = push_leaf_left(trans, root, path, space_needed, |
3535 | space_needed, 0, (u32)-1); | |
3685f791 CM |
3536 | if (wret < 0) |
3537 | return wret; | |
3538 | } | |
3539 | l = path->nodes[0]; | |
aa5d6bed | 3540 | |
3685f791 | 3541 | /* did the pushes work? */ |
e902baac | 3542 | if (btrfs_leaf_free_space(l) >= data_size) |
3685f791 | 3543 | return 0; |
3326d1b0 | 3544 | } |
aa5d6bed | 3545 | |
5c680ed6 | 3546 | if (!path->nodes[1]) { |
fdd99c72 | 3547 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3548 | if (ret) |
3549 | return ret; | |
3550 | } | |
cc0c5538 | 3551 | again: |
5d4f98a2 | 3552 | split = 1; |
cc0c5538 | 3553 | l = path->nodes[0]; |
eb60ceac | 3554 | slot = path->slots[0]; |
5f39d397 | 3555 | nritems = btrfs_header_nritems(l); |
d397712b | 3556 | mid = (nritems + 1) / 2; |
54aa1f4d | 3557 | |
5d4f98a2 YZ |
3558 | if (mid <= slot) { |
3559 | if (nritems == 1 || | |
3560 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 3561 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3562 | if (slot >= nritems) { |
3563 | split = 0; | |
3564 | } else { | |
3565 | mid = slot; | |
3566 | if (mid != nritems && | |
3567 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3568 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3569 | if (data_size && !tried_avoid_double) |
3570 | goto push_for_double; | |
5d4f98a2 YZ |
3571 | split = 2; |
3572 | } | |
3573 | } | |
3574 | } | |
3575 | } else { | |
3576 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 3577 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
3578 | if (!extend && data_size && slot == 0) { |
3579 | split = 0; | |
3580 | } else if ((extend || !data_size) && slot == 0) { | |
3581 | mid = 1; | |
3582 | } else { | |
3583 | mid = slot; | |
3584 | if (mid != nritems && | |
3585 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 3586 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
3587 | if (data_size && !tried_avoid_double) |
3588 | goto push_for_double; | |
67871254 | 3589 | split = 2; |
5d4f98a2 YZ |
3590 | } |
3591 | } | |
3592 | } | |
3593 | } | |
3594 | ||
3595 | if (split == 0) | |
3596 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3597 | else | |
3598 | btrfs_item_key(l, &disk_key, mid); | |
3599 | ||
ca9d473a JB |
3600 | /* |
3601 | * We have to about BTRFS_NESTING_NEW_ROOT here if we've done a double | |
3602 | * split, because we're only allowed to have MAX_LOCKDEP_SUBCLASSES | |
3603 | * subclasses, which is 8 at the time of this patch, and we've maxed it | |
3604 | * out. In the future we could add a | |
3605 | * BTRFS_NESTING_SPLIT_THE_SPLITTENING if we need to, but for now just | |
3606 | * use BTRFS_NESTING_NEW_ROOT. | |
3607 | */ | |
79bd3712 FM |
3608 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3609 | &disk_key, 0, l->start, 0, | |
3610 | num_doubles ? BTRFS_NESTING_NEW_ROOT : | |
3611 | BTRFS_NESTING_SPLIT); | |
f0486c68 | 3612 | if (IS_ERR(right)) |
5f39d397 | 3613 | return PTR_ERR(right); |
f0486c68 | 3614 | |
0b246afa | 3615 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 3616 | |
5d4f98a2 YZ |
3617 | if (split == 0) { |
3618 | if (mid <= slot) { | |
3619 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3620 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3621 | right->start, path->slots[1] + 1, 1); |
5d4f98a2 YZ |
3622 | btrfs_tree_unlock(path->nodes[0]); |
3623 | free_extent_buffer(path->nodes[0]); | |
3624 | path->nodes[0] = right; | |
3625 | path->slots[0] = 0; | |
3626 | path->slots[1] += 1; | |
3627 | } else { | |
3628 | btrfs_set_header_nritems(right, 0); | |
6ad3cf6d | 3629 | insert_ptr(trans, path, &disk_key, |
2ff7e61e | 3630 | right->start, path->slots[1], 1); |
5d4f98a2 YZ |
3631 | btrfs_tree_unlock(path->nodes[0]); |
3632 | free_extent_buffer(path->nodes[0]); | |
3633 | path->nodes[0] = right; | |
3634 | path->slots[0] = 0; | |
143bede5 | 3635 | if (path->slots[1] == 0) |
b167fa91 | 3636 | fixup_low_keys(path, &disk_key, 1); |
d4dbff95 | 3637 | } |
196e0249 LB |
3638 | /* |
3639 | * We create a new leaf 'right' for the required ins_len and | |
3640 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
3641 | * the content of ins_len to 'right'. | |
3642 | */ | |
5d4f98a2 | 3643 | return ret; |
d4dbff95 | 3644 | } |
74123bd7 | 3645 | |
94f94ad9 | 3646 | copy_for_split(trans, path, l, right, slot, mid, nritems); |
31840ae1 | 3647 | |
5d4f98a2 | 3648 | if (split == 2) { |
cc0c5538 CM |
3649 | BUG_ON(num_doubles != 0); |
3650 | num_doubles++; | |
3651 | goto again; | |
a429e513 | 3652 | } |
44871b1b | 3653 | |
143bede5 | 3654 | return 0; |
99d8f83c CM |
3655 | |
3656 | push_for_double: | |
3657 | push_for_double_split(trans, root, path, data_size); | |
3658 | tried_avoid_double = 1; | |
e902baac | 3659 | if (btrfs_leaf_free_space(path->nodes[0]) >= data_size) |
99d8f83c CM |
3660 | return 0; |
3661 | goto again; | |
be0e5c09 CM |
3662 | } |
3663 | ||
ad48fd75 YZ |
3664 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
3665 | struct btrfs_root *root, | |
3666 | struct btrfs_path *path, int ins_len) | |
459931ec | 3667 | { |
ad48fd75 | 3668 | struct btrfs_key key; |
459931ec | 3669 | struct extent_buffer *leaf; |
ad48fd75 YZ |
3670 | struct btrfs_file_extent_item *fi; |
3671 | u64 extent_len = 0; | |
3672 | u32 item_size; | |
3673 | int ret; | |
459931ec CM |
3674 | |
3675 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
3676 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
3677 | ||
3678 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
3679 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
3680 | ||
e902baac | 3681 | if (btrfs_leaf_free_space(leaf) >= ins_len) |
ad48fd75 | 3682 | return 0; |
459931ec | 3683 | |
3212fa14 | 3684 | item_size = btrfs_item_size(leaf, path->slots[0]); |
ad48fd75 YZ |
3685 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3686 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3687 | struct btrfs_file_extent_item); | |
3688 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
3689 | } | |
b3b4aa74 | 3690 | btrfs_release_path(path); |
459931ec | 3691 | |
459931ec | 3692 | path->keep_locks = 1; |
ad48fd75 YZ |
3693 | path->search_for_split = 1; |
3694 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 3695 | path->search_for_split = 0; |
a8df6fe6 FM |
3696 | if (ret > 0) |
3697 | ret = -EAGAIN; | |
ad48fd75 YZ |
3698 | if (ret < 0) |
3699 | goto err; | |
459931ec | 3700 | |
ad48fd75 YZ |
3701 | ret = -EAGAIN; |
3702 | leaf = path->nodes[0]; | |
a8df6fe6 | 3703 | /* if our item isn't there, return now */ |
3212fa14 | 3704 | if (item_size != btrfs_item_size(leaf, path->slots[0])) |
ad48fd75 YZ |
3705 | goto err; |
3706 | ||
109f6aef | 3707 | /* the leaf has changed, it now has room. return now */ |
e902baac | 3708 | if (btrfs_leaf_free_space(path->nodes[0]) >= ins_len) |
109f6aef CM |
3709 | goto err; |
3710 | ||
ad48fd75 YZ |
3711 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
3712 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
3713 | struct btrfs_file_extent_item); | |
3714 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
3715 | goto err; | |
459931ec CM |
3716 | } |
3717 | ||
ad48fd75 | 3718 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
3719 | if (ret) |
3720 | goto err; | |
459931ec | 3721 | |
ad48fd75 | 3722 | path->keep_locks = 0; |
b9473439 | 3723 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
3724 | return 0; |
3725 | err: | |
3726 | path->keep_locks = 0; | |
3727 | return ret; | |
3728 | } | |
3729 | ||
25263cd7 | 3730 | static noinline int split_item(struct btrfs_path *path, |
310712b2 | 3731 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3732 | unsigned long split_offset) |
3733 | { | |
3734 | struct extent_buffer *leaf; | |
c91666b1 | 3735 | int orig_slot, slot; |
ad48fd75 YZ |
3736 | char *buf; |
3737 | u32 nritems; | |
3738 | u32 item_size; | |
3739 | u32 orig_offset; | |
3740 | struct btrfs_disk_key disk_key; | |
3741 | ||
b9473439 | 3742 | leaf = path->nodes[0]; |
e902baac | 3743 | BUG_ON(btrfs_leaf_free_space(leaf) < sizeof(struct btrfs_item)); |
b9473439 | 3744 | |
c91666b1 | 3745 | orig_slot = path->slots[0]; |
3212fa14 JB |
3746 | orig_offset = btrfs_item_offset(leaf, path->slots[0]); |
3747 | item_size = btrfs_item_size(leaf, path->slots[0]); | |
459931ec | 3748 | |
459931ec | 3749 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
3750 | if (!buf) |
3751 | return -ENOMEM; | |
3752 | ||
459931ec CM |
3753 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
3754 | path->slots[0]), item_size); | |
459931ec | 3755 | |
ad48fd75 | 3756 | slot = path->slots[0] + 1; |
459931ec | 3757 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
3758 | if (slot != nritems) { |
3759 | /* shift the items */ | |
3760 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
3761 | btrfs_item_nr_offset(slot), |
3762 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
3763 | } |
3764 | ||
3765 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3766 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3767 | ||
3212fa14 JB |
3768 | btrfs_set_item_offset(leaf, slot, orig_offset); |
3769 | btrfs_set_item_size(leaf, slot, item_size - split_offset); | |
459931ec | 3770 | |
3212fa14 | 3771 | btrfs_set_item_offset(leaf, orig_slot, |
c91666b1 | 3772 | orig_offset + item_size - split_offset); |
3212fa14 | 3773 | btrfs_set_item_size(leaf, orig_slot, split_offset); |
459931ec CM |
3774 | |
3775 | btrfs_set_header_nritems(leaf, nritems + 1); | |
3776 | ||
3777 | /* write the data for the start of the original item */ | |
3778 | write_extent_buffer(leaf, buf, | |
3779 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
3780 | split_offset); | |
3781 | ||
3782 | /* write the data for the new item */ | |
3783 | write_extent_buffer(leaf, buf + split_offset, | |
3784 | btrfs_item_ptr_offset(leaf, slot), | |
3785 | item_size - split_offset); | |
3786 | btrfs_mark_buffer_dirty(leaf); | |
3787 | ||
e902baac | 3788 | BUG_ON(btrfs_leaf_free_space(leaf) < 0); |
459931ec | 3789 | kfree(buf); |
ad48fd75 YZ |
3790 | return 0; |
3791 | } | |
3792 | ||
3793 | /* | |
3794 | * This function splits a single item into two items, | |
3795 | * giving 'new_key' to the new item and splitting the | |
3796 | * old one at split_offset (from the start of the item). | |
3797 | * | |
3798 | * The path may be released by this operation. After | |
3799 | * the split, the path is pointing to the old item. The | |
3800 | * new item is going to be in the same node as the old one. | |
3801 | * | |
3802 | * Note, the item being split must be smaller enough to live alone on | |
3803 | * a tree block with room for one extra struct btrfs_item | |
3804 | * | |
3805 | * This allows us to split the item in place, keeping a lock on the | |
3806 | * leaf the entire time. | |
3807 | */ | |
3808 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
3809 | struct btrfs_root *root, | |
3810 | struct btrfs_path *path, | |
310712b2 | 3811 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
3812 | unsigned long split_offset) |
3813 | { | |
3814 | int ret; | |
3815 | ret = setup_leaf_for_split(trans, root, path, | |
3816 | sizeof(struct btrfs_item)); | |
3817 | if (ret) | |
3818 | return ret; | |
3819 | ||
25263cd7 | 3820 | ret = split_item(path, new_key, split_offset); |
459931ec CM |
3821 | return ret; |
3822 | } | |
3823 | ||
d352ac68 CM |
3824 | /* |
3825 | * make the item pointed to by the path smaller. new_size indicates | |
3826 | * how small to make it, and from_end tells us if we just chop bytes | |
3827 | * off the end of the item or if we shift the item to chop bytes off | |
3828 | * the front. | |
3829 | */ | |
78ac4f9e | 3830 | void btrfs_truncate_item(struct btrfs_path *path, u32 new_size, int from_end) |
b18c6685 | 3831 | { |
b18c6685 | 3832 | int slot; |
5f39d397 | 3833 | struct extent_buffer *leaf; |
b18c6685 CM |
3834 | u32 nritems; |
3835 | unsigned int data_end; | |
3836 | unsigned int old_data_start; | |
3837 | unsigned int old_size; | |
3838 | unsigned int size_diff; | |
3839 | int i; | |
cfed81a0 CM |
3840 | struct btrfs_map_token token; |
3841 | ||
5f39d397 | 3842 | leaf = path->nodes[0]; |
179e29e4 CM |
3843 | slot = path->slots[0]; |
3844 | ||
3212fa14 | 3845 | old_size = btrfs_item_size(leaf, slot); |
179e29e4 | 3846 | if (old_size == new_size) |
143bede5 | 3847 | return; |
b18c6685 | 3848 | |
5f39d397 | 3849 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3850 | data_end = leaf_data_end(leaf); |
b18c6685 | 3851 | |
3212fa14 | 3852 | old_data_start = btrfs_item_offset(leaf, slot); |
179e29e4 | 3853 | |
b18c6685 CM |
3854 | size_diff = old_size - new_size; |
3855 | ||
3856 | BUG_ON(slot < 0); | |
3857 | BUG_ON(slot >= nritems); | |
3858 | ||
3859 | /* | |
3860 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3861 | */ | |
3862 | /* first correct the data pointers */ | |
c82f823c | 3863 | btrfs_init_map_token(&token, leaf); |
b18c6685 | 3864 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3865 | u32 ioff; |
db94535d | 3866 | |
3212fa14 JB |
3867 | ioff = btrfs_token_item_offset(&token, i); |
3868 | btrfs_set_token_item_offset(&token, i, ioff + size_diff); | |
b18c6685 | 3869 | } |
db94535d | 3870 | |
b18c6685 | 3871 | /* shift the data */ |
179e29e4 | 3872 | if (from_end) { |
3d9ec8c4 NB |
3873 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3874 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3875 | data_end, old_data_start + new_size - data_end); |
3876 | } else { | |
3877 | struct btrfs_disk_key disk_key; | |
3878 | u64 offset; | |
3879 | ||
3880 | btrfs_item_key(leaf, &disk_key, slot); | |
3881 | ||
3882 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
3883 | unsigned long ptr; | |
3884 | struct btrfs_file_extent_item *fi; | |
3885 | ||
3886 | fi = btrfs_item_ptr(leaf, slot, | |
3887 | struct btrfs_file_extent_item); | |
3888 | fi = (struct btrfs_file_extent_item *)( | |
3889 | (unsigned long)fi - size_diff); | |
3890 | ||
3891 | if (btrfs_file_extent_type(leaf, fi) == | |
3892 | BTRFS_FILE_EXTENT_INLINE) { | |
3893 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
3894 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 3895 | (unsigned long)fi, |
7ec20afb | 3896 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
3897 | } |
3898 | } | |
3899 | ||
3d9ec8c4 NB |
3900 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3901 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
3902 | data_end, old_data_start - data_end); |
3903 | ||
3904 | offset = btrfs_disk_key_offset(&disk_key); | |
3905 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
3906 | btrfs_set_item_key(leaf, &disk_key, slot); | |
3907 | if (slot == 0) | |
b167fa91 | 3908 | fixup_low_keys(path, &disk_key, 1); |
179e29e4 | 3909 | } |
5f39d397 | 3910 | |
3212fa14 | 3911 | btrfs_set_item_size(leaf, slot, new_size); |
5f39d397 | 3912 | btrfs_mark_buffer_dirty(leaf); |
b18c6685 | 3913 | |
e902baac | 3914 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3915 | btrfs_print_leaf(leaf); |
b18c6685 | 3916 | BUG(); |
5f39d397 | 3917 | } |
b18c6685 CM |
3918 | } |
3919 | ||
d352ac68 | 3920 | /* |
8f69dbd2 | 3921 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 3922 | */ |
c71dd880 | 3923 | void btrfs_extend_item(struct btrfs_path *path, u32 data_size) |
6567e837 | 3924 | { |
6567e837 | 3925 | int slot; |
5f39d397 | 3926 | struct extent_buffer *leaf; |
6567e837 CM |
3927 | u32 nritems; |
3928 | unsigned int data_end; | |
3929 | unsigned int old_data; | |
3930 | unsigned int old_size; | |
3931 | int i; | |
cfed81a0 CM |
3932 | struct btrfs_map_token token; |
3933 | ||
5f39d397 | 3934 | leaf = path->nodes[0]; |
6567e837 | 3935 | |
5f39d397 | 3936 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 3937 | data_end = leaf_data_end(leaf); |
6567e837 | 3938 | |
e902baac | 3939 | if (btrfs_leaf_free_space(leaf) < data_size) { |
a4f78750 | 3940 | btrfs_print_leaf(leaf); |
6567e837 | 3941 | BUG(); |
5f39d397 | 3942 | } |
6567e837 | 3943 | slot = path->slots[0]; |
dc2e724e | 3944 | old_data = btrfs_item_data_end(leaf, slot); |
6567e837 CM |
3945 | |
3946 | BUG_ON(slot < 0); | |
3326d1b0 | 3947 | if (slot >= nritems) { |
a4f78750 | 3948 | btrfs_print_leaf(leaf); |
c71dd880 | 3949 | btrfs_crit(leaf->fs_info, "slot %d too large, nritems %d", |
0b246afa | 3950 | slot, nritems); |
290342f6 | 3951 | BUG(); |
3326d1b0 | 3952 | } |
6567e837 CM |
3953 | |
3954 | /* | |
3955 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
3956 | */ | |
3957 | /* first correct the data pointers */ | |
c82f823c | 3958 | btrfs_init_map_token(&token, leaf); |
6567e837 | 3959 | for (i = slot; i < nritems; i++) { |
5f39d397 | 3960 | u32 ioff; |
db94535d | 3961 | |
3212fa14 JB |
3962 | ioff = btrfs_token_item_offset(&token, i); |
3963 | btrfs_set_token_item_offset(&token, i, ioff - data_size); | |
6567e837 | 3964 | } |
5f39d397 | 3965 | |
6567e837 | 3966 | /* shift the data */ |
3d9ec8c4 NB |
3967 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
3968 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 3969 | data_end, old_data - data_end); |
5f39d397 | 3970 | |
6567e837 | 3971 | data_end = old_data; |
3212fa14 JB |
3972 | old_size = btrfs_item_size(leaf, slot); |
3973 | btrfs_set_item_size(leaf, slot, old_size + data_size); | |
5f39d397 | 3974 | btrfs_mark_buffer_dirty(leaf); |
6567e837 | 3975 | |
e902baac | 3976 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 3977 | btrfs_print_leaf(leaf); |
6567e837 | 3978 | BUG(); |
5f39d397 | 3979 | } |
6567e837 CM |
3980 | } |
3981 | ||
da9ffb24 NB |
3982 | /** |
3983 | * setup_items_for_insert - Helper called before inserting one or more items | |
3984 | * to a leaf. Main purpose is to save stack depth by doing the bulk of the work | |
3985 | * in a function that doesn't call btrfs_search_slot | |
3986 | * | |
3987 | * @root: root we are inserting items to | |
3988 | * @path: points to the leaf/slot where we are going to insert new items | |
b7ef5f3a | 3989 | * @batch: information about the batch of items to insert |
74123bd7 | 3990 | */ |
f0641656 FM |
3991 | static void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
3992 | const struct btrfs_item_batch *batch) | |
be0e5c09 | 3993 | { |
0b246afa | 3994 | struct btrfs_fs_info *fs_info = root->fs_info; |
9c58309d | 3995 | int i; |
7518a238 | 3996 | u32 nritems; |
be0e5c09 | 3997 | unsigned int data_end; |
e2fa7227 | 3998 | struct btrfs_disk_key disk_key; |
44871b1b CM |
3999 | struct extent_buffer *leaf; |
4000 | int slot; | |
cfed81a0 | 4001 | struct btrfs_map_token token; |
fc0d82e1 | 4002 | u32 total_size; |
cfed81a0 | 4003 | |
b7ef5f3a FM |
4004 | /* |
4005 | * Before anything else, update keys in the parent and other ancestors | |
4006 | * if needed, then release the write locks on them, so that other tasks | |
4007 | * can use them while we modify the leaf. | |
4008 | */ | |
24cdc847 | 4009 | if (path->slots[0] == 0) { |
b7ef5f3a | 4010 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[0]); |
b167fa91 | 4011 | fixup_low_keys(path, &disk_key, 1); |
24cdc847 FM |
4012 | } |
4013 | btrfs_unlock_up_safe(path, 1); | |
4014 | ||
5f39d397 | 4015 | leaf = path->nodes[0]; |
44871b1b | 4016 | slot = path->slots[0]; |
74123bd7 | 4017 | |
5f39d397 | 4018 | nritems = btrfs_header_nritems(leaf); |
8f881e8c | 4019 | data_end = leaf_data_end(leaf); |
b7ef5f3a | 4020 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
eb60ceac | 4021 | |
e902baac | 4022 | if (btrfs_leaf_free_space(leaf) < total_size) { |
a4f78750 | 4023 | btrfs_print_leaf(leaf); |
0b246afa | 4024 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
e902baac | 4025 | total_size, btrfs_leaf_free_space(leaf)); |
be0e5c09 | 4026 | BUG(); |
d4dbff95 | 4027 | } |
5f39d397 | 4028 | |
c82f823c | 4029 | btrfs_init_map_token(&token, leaf); |
be0e5c09 | 4030 | if (slot != nritems) { |
dc2e724e | 4031 | unsigned int old_data = btrfs_item_data_end(leaf, slot); |
be0e5c09 | 4032 | |
5f39d397 | 4033 | if (old_data < data_end) { |
a4f78750 | 4034 | btrfs_print_leaf(leaf); |
7269ddd2 NB |
4035 | btrfs_crit(fs_info, |
4036 | "item at slot %d with data offset %u beyond data end of leaf %u", | |
5d163e0e | 4037 | slot, old_data, data_end); |
290342f6 | 4038 | BUG(); |
5f39d397 | 4039 | } |
be0e5c09 CM |
4040 | /* |
4041 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4042 | */ | |
4043 | /* first correct the data pointers */ | |
0783fcfc | 4044 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4045 | u32 ioff; |
db94535d | 4046 | |
3212fa14 JB |
4047 | ioff = btrfs_token_item_offset(&token, i); |
4048 | btrfs_set_token_item_offset(&token, i, | |
74794207 | 4049 | ioff - batch->total_data_size); |
0783fcfc | 4050 | } |
be0e5c09 | 4051 | /* shift the items */ |
b7ef5f3a | 4052 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + batch->nr), |
5f39d397 | 4053 | btrfs_item_nr_offset(slot), |
d6025579 | 4054 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4055 | |
4056 | /* shift the data */ | |
3d9ec8c4 | 4057 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
b7ef5f3a FM |
4058 | data_end - batch->total_data_size, |
4059 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
4060 | old_data - data_end); | |
be0e5c09 CM |
4061 | data_end = old_data; |
4062 | } | |
5f39d397 | 4063 | |
62e2749e | 4064 | /* setup the item for the new data */ |
b7ef5f3a FM |
4065 | for (i = 0; i < batch->nr; i++) { |
4066 | btrfs_cpu_key_to_disk(&disk_key, &batch->keys[i]); | |
9c58309d | 4067 | btrfs_set_item_key(leaf, &disk_key, slot + i); |
b7ef5f3a | 4068 | data_end -= batch->data_sizes[i]; |
3212fa14 JB |
4069 | btrfs_set_token_item_offset(&token, slot + i, data_end); |
4070 | btrfs_set_token_item_size(&token, slot + i, batch->data_sizes[i]); | |
9c58309d | 4071 | } |
44871b1b | 4072 | |
b7ef5f3a | 4073 | btrfs_set_header_nritems(leaf, nritems + batch->nr); |
b9473439 | 4074 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4075 | |
e902baac | 4076 | if (btrfs_leaf_free_space(leaf) < 0) { |
a4f78750 | 4077 | btrfs_print_leaf(leaf); |
be0e5c09 | 4078 | BUG(); |
5f39d397 | 4079 | } |
44871b1b CM |
4080 | } |
4081 | ||
f0641656 FM |
4082 | /* |
4083 | * Insert a new item into a leaf. | |
4084 | * | |
4085 | * @root: The root of the btree. | |
4086 | * @path: A path pointing to the target leaf and slot. | |
4087 | * @key: The key of the new item. | |
4088 | * @data_size: The size of the data associated with the new key. | |
4089 | */ | |
4090 | void btrfs_setup_item_for_insert(struct btrfs_root *root, | |
4091 | struct btrfs_path *path, | |
4092 | const struct btrfs_key *key, | |
4093 | u32 data_size) | |
4094 | { | |
4095 | struct btrfs_item_batch batch; | |
4096 | ||
4097 | batch.keys = key; | |
4098 | batch.data_sizes = &data_size; | |
4099 | batch.total_data_size = data_size; | |
4100 | batch.nr = 1; | |
4101 | ||
4102 | setup_items_for_insert(root, path, &batch); | |
4103 | } | |
4104 | ||
44871b1b CM |
4105 | /* |
4106 | * Given a key and some data, insert items into the tree. | |
4107 | * This does all the path init required, making room in the tree if needed. | |
4108 | */ | |
4109 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4110 | struct btrfs_root *root, | |
4111 | struct btrfs_path *path, | |
b7ef5f3a | 4112 | const struct btrfs_item_batch *batch) |
44871b1b | 4113 | { |
44871b1b CM |
4114 | int ret = 0; |
4115 | int slot; | |
b7ef5f3a | 4116 | u32 total_size; |
44871b1b | 4117 | |
b7ef5f3a FM |
4118 | total_size = batch->total_data_size + (batch->nr * sizeof(struct btrfs_item)); |
4119 | ret = btrfs_search_slot(trans, root, &batch->keys[0], path, total_size, 1); | |
44871b1b CM |
4120 | if (ret == 0) |
4121 | return -EEXIST; | |
4122 | if (ret < 0) | |
143bede5 | 4123 | return ret; |
44871b1b | 4124 | |
44871b1b CM |
4125 | slot = path->slots[0]; |
4126 | BUG_ON(slot < 0); | |
4127 | ||
b7ef5f3a | 4128 | setup_items_for_insert(root, path, batch); |
143bede5 | 4129 | return 0; |
62e2749e CM |
4130 | } |
4131 | ||
4132 | /* | |
4133 | * Given a key and some data, insert an item into the tree. | |
4134 | * This does all the path init required, making room in the tree if needed. | |
4135 | */ | |
310712b2 OS |
4136 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4137 | const struct btrfs_key *cpu_key, void *data, | |
4138 | u32 data_size) | |
62e2749e CM |
4139 | { |
4140 | int ret = 0; | |
2c90e5d6 | 4141 | struct btrfs_path *path; |
5f39d397 CM |
4142 | struct extent_buffer *leaf; |
4143 | unsigned long ptr; | |
62e2749e | 4144 | |
2c90e5d6 | 4145 | path = btrfs_alloc_path(); |
db5b493a TI |
4146 | if (!path) |
4147 | return -ENOMEM; | |
2c90e5d6 | 4148 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4149 | if (!ret) { |
5f39d397 CM |
4150 | leaf = path->nodes[0]; |
4151 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4152 | write_extent_buffer(leaf, data, ptr, data_size); | |
4153 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4154 | } |
2c90e5d6 | 4155 | btrfs_free_path(path); |
aa5d6bed | 4156 | return ret; |
be0e5c09 CM |
4157 | } |
4158 | ||
f0641656 FM |
4159 | /* |
4160 | * This function duplicates an item, giving 'new_key' to the new item. | |
4161 | * It guarantees both items live in the same tree leaf and the new item is | |
4162 | * contiguous with the original item. | |
4163 | * | |
4164 | * This allows us to split a file extent in place, keeping a lock on the leaf | |
4165 | * the entire time. | |
4166 | */ | |
4167 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4168 | struct btrfs_root *root, | |
4169 | struct btrfs_path *path, | |
4170 | const struct btrfs_key *new_key) | |
4171 | { | |
4172 | struct extent_buffer *leaf; | |
4173 | int ret; | |
4174 | u32 item_size; | |
4175 | ||
4176 | leaf = path->nodes[0]; | |
3212fa14 | 4177 | item_size = btrfs_item_size(leaf, path->slots[0]); |
f0641656 FM |
4178 | ret = setup_leaf_for_split(trans, root, path, |
4179 | item_size + sizeof(struct btrfs_item)); | |
4180 | if (ret) | |
4181 | return ret; | |
4182 | ||
4183 | path->slots[0]++; | |
4184 | btrfs_setup_item_for_insert(root, path, new_key, item_size); | |
4185 | leaf = path->nodes[0]; | |
4186 | memcpy_extent_buffer(leaf, | |
4187 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4188 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4189 | item_size); | |
4190 | return 0; | |
4191 | } | |
4192 | ||
74123bd7 | 4193 | /* |
5de08d7d | 4194 | * delete the pointer from a given node. |
74123bd7 | 4195 | * |
d352ac68 CM |
4196 | * the tree should have been previously balanced so the deletion does not |
4197 | * empty a node. | |
74123bd7 | 4198 | */ |
afe5fea7 TI |
4199 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4200 | int level, int slot) | |
be0e5c09 | 4201 | { |
5f39d397 | 4202 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4203 | u32 nritems; |
f3ea38da | 4204 | int ret; |
be0e5c09 | 4205 | |
5f39d397 | 4206 | nritems = btrfs_header_nritems(parent); |
d397712b | 4207 | if (slot != nritems - 1) { |
bf1d3425 | 4208 | if (level) { |
f3a84ccd FM |
4209 | ret = btrfs_tree_mod_log_insert_move(parent, slot, |
4210 | slot + 1, nritems - slot - 1); | |
bf1d3425 DS |
4211 | BUG_ON(ret < 0); |
4212 | } | |
5f39d397 CM |
4213 | memmove_extent_buffer(parent, |
4214 | btrfs_node_key_ptr_offset(slot), | |
4215 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4216 | sizeof(struct btrfs_key_ptr) * |
4217 | (nritems - slot - 1)); | |
57ba86c0 | 4218 | } else if (level) { |
f3a84ccd FM |
4219 | ret = btrfs_tree_mod_log_insert_key(parent, slot, |
4220 | BTRFS_MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
57ba86c0 | 4221 | BUG_ON(ret < 0); |
bb803951 | 4222 | } |
f3ea38da | 4223 | |
7518a238 | 4224 | nritems--; |
5f39d397 | 4225 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4226 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4227 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4228 | /* just turn the root into a leaf and break */ |
5f39d397 | 4229 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4230 | } else if (slot == 0) { |
5f39d397 CM |
4231 | struct btrfs_disk_key disk_key; |
4232 | ||
4233 | btrfs_node_key(parent, &disk_key, 0); | |
b167fa91 | 4234 | fixup_low_keys(path, &disk_key, level + 1); |
be0e5c09 | 4235 | } |
d6025579 | 4236 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4237 | } |
4238 | ||
323ac95b CM |
4239 | /* |
4240 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4241 | * path->nodes[1]. |
323ac95b CM |
4242 | * |
4243 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4244 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4245 | * | |
4246 | * The path must have already been setup for deleting the leaf, including | |
4247 | * all the proper balancing. path->nodes[1] must be locked. | |
4248 | */ | |
143bede5 JM |
4249 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4250 | struct btrfs_root *root, | |
4251 | struct btrfs_path *path, | |
4252 | struct extent_buffer *leaf) | |
323ac95b | 4253 | { |
5d4f98a2 | 4254 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4255 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4256 | |
4d081c41 CM |
4257 | /* |
4258 | * btrfs_free_extent is expensive, we want to make sure we | |
4259 | * aren't holding any locks when we call it | |
4260 | */ | |
4261 | btrfs_unlock_up_safe(path, 0); | |
4262 | ||
f0486c68 YZ |
4263 | root_sub_used(root, leaf->len); |
4264 | ||
67439dad | 4265 | atomic_inc(&leaf->refs); |
7a163608 | 4266 | btrfs_free_tree_block(trans, btrfs_root_id(root), leaf, 0, 1); |
3083ee2e | 4267 | free_extent_buffer_stale(leaf); |
323ac95b | 4268 | } |
74123bd7 CM |
4269 | /* |
4270 | * delete the item at the leaf level in path. If that empties | |
4271 | * the leaf, remove it from the tree | |
4272 | */ | |
85e21bac CM |
4273 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4274 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4275 | { |
0b246afa | 4276 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4277 | struct extent_buffer *leaf; |
aa5d6bed CM |
4278 | int ret = 0; |
4279 | int wret; | |
7518a238 | 4280 | u32 nritems; |
be0e5c09 | 4281 | |
5f39d397 | 4282 | leaf = path->nodes[0]; |
5f39d397 | 4283 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4284 | |
85e21bac | 4285 | if (slot + nr != nritems) { |
0cae23b6 FM |
4286 | const u32 last_off = btrfs_item_offset(leaf, slot + nr - 1); |
4287 | const int data_end = leaf_data_end(leaf); | |
c82f823c | 4288 | struct btrfs_map_token token; |
0cae23b6 FM |
4289 | u32 dsize = 0; |
4290 | int i; | |
4291 | ||
4292 | for (i = 0; i < nr; i++) | |
4293 | dsize += btrfs_item_size(leaf, slot + i); | |
5f39d397 | 4294 | |
3d9ec8c4 | 4295 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4296 | data_end + dsize, |
3d9ec8c4 | 4297 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4298 | last_off - data_end); |
5f39d397 | 4299 | |
c82f823c | 4300 | btrfs_init_map_token(&token, leaf); |
85e21bac | 4301 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4302 | u32 ioff; |
db94535d | 4303 | |
3212fa14 JB |
4304 | ioff = btrfs_token_item_offset(&token, i); |
4305 | btrfs_set_token_item_offset(&token, i, ioff + dsize); | |
0783fcfc | 4306 | } |
db94535d | 4307 | |
5f39d397 | 4308 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4309 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4310 | sizeof(struct btrfs_item) * |
85e21bac | 4311 | (nritems - slot - nr)); |
be0e5c09 | 4312 | } |
85e21bac CM |
4313 | btrfs_set_header_nritems(leaf, nritems - nr); |
4314 | nritems -= nr; | |
5f39d397 | 4315 | |
74123bd7 | 4316 | /* delete the leaf if we've emptied it */ |
7518a238 | 4317 | if (nritems == 0) { |
5f39d397 CM |
4318 | if (leaf == root->node) { |
4319 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4320 | } else { |
6a884d7d | 4321 | btrfs_clean_tree_block(leaf); |
143bede5 | 4322 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4323 | } |
be0e5c09 | 4324 | } else { |
7518a238 | 4325 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4326 | if (slot == 0) { |
5f39d397 CM |
4327 | struct btrfs_disk_key disk_key; |
4328 | ||
4329 | btrfs_item_key(leaf, &disk_key, 0); | |
b167fa91 | 4330 | fixup_low_keys(path, &disk_key, 1); |
aa5d6bed | 4331 | } |
aa5d6bed | 4332 | |
7c4063d1 FM |
4333 | /* |
4334 | * Try to delete the leaf if it is mostly empty. We do this by | |
4335 | * trying to move all its items into its left and right neighbours. | |
4336 | * If we can't move all the items, then we don't delete it - it's | |
4337 | * not ideal, but future insertions might fill the leaf with more | |
4338 | * items, or items from other leaves might be moved later into our | |
4339 | * leaf due to deletions on those leaves. | |
4340 | */ | |
0b246afa | 4341 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
7c4063d1 FM |
4342 | u32 min_push_space; |
4343 | ||
be0e5c09 CM |
4344 | /* push_leaf_left fixes the path. |
4345 | * make sure the path still points to our leaf | |
4346 | * for possible call to del_ptr below | |
4347 | */ | |
4920c9ac | 4348 | slot = path->slots[1]; |
67439dad | 4349 | atomic_inc(&leaf->refs); |
7c4063d1 FM |
4350 | /* |
4351 | * We want to be able to at least push one item to the | |
4352 | * left neighbour leaf, and that's the first item. | |
4353 | */ | |
4354 | min_push_space = sizeof(struct btrfs_item) + | |
4355 | btrfs_item_size(leaf, 0); | |
4356 | wret = push_leaf_left(trans, root, path, 0, | |
4357 | min_push_space, 1, (u32)-1); | |
54aa1f4d | 4358 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4359 | ret = wret; |
5f39d397 CM |
4360 | |
4361 | if (path->nodes[0] == leaf && | |
4362 | btrfs_header_nritems(leaf)) { | |
7c4063d1 FM |
4363 | /* |
4364 | * If we were not able to push all items from our | |
4365 | * leaf to its left neighbour, then attempt to | |
4366 | * either push all the remaining items to the | |
4367 | * right neighbour or none. There's no advantage | |
4368 | * in pushing only some items, instead of all, as | |
4369 | * it's pointless to end up with a leaf having | |
4370 | * too few items while the neighbours can be full | |
4371 | * or nearly full. | |
4372 | */ | |
4373 | nritems = btrfs_header_nritems(leaf); | |
4374 | min_push_space = leaf_space_used(leaf, 0, nritems); | |
4375 | wret = push_leaf_right(trans, root, path, 0, | |
4376 | min_push_space, 1, 0); | |
54aa1f4d | 4377 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4378 | ret = wret; |
4379 | } | |
5f39d397 CM |
4380 | |
4381 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4382 | path->slots[1] = slot; |
143bede5 | 4383 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4384 | free_extent_buffer(leaf); |
143bede5 | 4385 | ret = 0; |
5de08d7d | 4386 | } else { |
925baedd CM |
4387 | /* if we're still in the path, make sure |
4388 | * we're dirty. Otherwise, one of the | |
4389 | * push_leaf functions must have already | |
4390 | * dirtied this buffer | |
4391 | */ | |
4392 | if (path->nodes[0] == leaf) | |
4393 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4394 | free_extent_buffer(leaf); |
be0e5c09 | 4395 | } |
d5719762 | 4396 | } else { |
5f39d397 | 4397 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4398 | } |
4399 | } | |
aa5d6bed | 4400 | return ret; |
be0e5c09 CM |
4401 | } |
4402 | ||
7bb86316 | 4403 | /* |
925baedd | 4404 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4405 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4406 | * returns < 0 on io errors. | |
d352ac68 CM |
4407 | * |
4408 | * This may release the path, and so you may lose any locks held at the | |
4409 | * time you call it. | |
7bb86316 | 4410 | */ |
16e7549f | 4411 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4412 | { |
925baedd CM |
4413 | struct btrfs_key key; |
4414 | struct btrfs_disk_key found_key; | |
4415 | int ret; | |
7bb86316 | 4416 | |
925baedd | 4417 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4418 | |
e8b0d724 | 4419 | if (key.offset > 0) { |
925baedd | 4420 | key.offset--; |
e8b0d724 | 4421 | } else if (key.type > 0) { |
925baedd | 4422 | key.type--; |
e8b0d724 FDBM |
4423 | key.offset = (u64)-1; |
4424 | } else if (key.objectid > 0) { | |
925baedd | 4425 | key.objectid--; |
e8b0d724 FDBM |
4426 | key.type = (u8)-1; |
4427 | key.offset = (u64)-1; | |
4428 | } else { | |
925baedd | 4429 | return 1; |
e8b0d724 | 4430 | } |
7bb86316 | 4431 | |
b3b4aa74 | 4432 | btrfs_release_path(path); |
925baedd CM |
4433 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4434 | if (ret < 0) | |
4435 | return ret; | |
4436 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4437 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
4438 | /* |
4439 | * We might have had an item with the previous key in the tree right | |
4440 | * before we released our path. And after we released our path, that | |
4441 | * item might have been pushed to the first slot (0) of the leaf we | |
4442 | * were holding due to a tree balance. Alternatively, an item with the | |
4443 | * previous key can exist as the only element of a leaf (big fat item). | |
4444 | * Therefore account for these 2 cases, so that our callers (like | |
4445 | * btrfs_previous_item) don't miss an existing item with a key matching | |
4446 | * the previous key we computed above. | |
4447 | */ | |
4448 | if (ret <= 0) | |
925baedd CM |
4449 | return 0; |
4450 | return 1; | |
7bb86316 CM |
4451 | } |
4452 | ||
3f157a2f CM |
4453 | /* |
4454 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4455 | * for nodes or leaves that are have a minimum transaction id. |
4456 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4457 | * |
4458 | * This does not cow, but it does stuff the starting key it finds back | |
4459 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4460 | * key and get a writable path. | |
4461 | * | |
3f157a2f CM |
4462 | * This honors path->lowest_level to prevent descent past a given level |
4463 | * of the tree. | |
4464 | * | |
d352ac68 CM |
4465 | * min_trans indicates the oldest transaction that you are interested |
4466 | * in walking through. Any nodes or leaves older than min_trans are | |
4467 | * skipped over (without reading them). | |
4468 | * | |
3f157a2f CM |
4469 | * returns zero if something useful was found, < 0 on error and 1 if there |
4470 | * was nothing in the tree that matched the search criteria. | |
4471 | */ | |
4472 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 4473 | struct btrfs_path *path, |
3f157a2f CM |
4474 | u64 min_trans) |
4475 | { | |
4476 | struct extent_buffer *cur; | |
4477 | struct btrfs_key found_key; | |
4478 | int slot; | |
9652480b | 4479 | int sret; |
3f157a2f CM |
4480 | u32 nritems; |
4481 | int level; | |
4482 | int ret = 1; | |
f98de9b9 | 4483 | int keep_locks = path->keep_locks; |
3f157a2f | 4484 | |
c922b016 | 4485 | ASSERT(!path->nowait); |
f98de9b9 | 4486 | path->keep_locks = 1; |
3f157a2f | 4487 | again: |
bd681513 | 4488 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4489 | level = btrfs_header_level(cur); |
e02119d5 | 4490 | WARN_ON(path->nodes[level]); |
3f157a2f | 4491 | path->nodes[level] = cur; |
bd681513 | 4492 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4493 | |
4494 | if (btrfs_header_generation(cur) < min_trans) { | |
4495 | ret = 1; | |
4496 | goto out; | |
4497 | } | |
d397712b | 4498 | while (1) { |
3f157a2f CM |
4499 | nritems = btrfs_header_nritems(cur); |
4500 | level = btrfs_header_level(cur); | |
e3b83361 | 4501 | sret = btrfs_bin_search(cur, min_key, &slot); |
cbca7d59 FM |
4502 | if (sret < 0) { |
4503 | ret = sret; | |
4504 | goto out; | |
4505 | } | |
3f157a2f | 4506 | |
323ac95b CM |
4507 | /* at the lowest level, we're done, setup the path and exit */ |
4508 | if (level == path->lowest_level) { | |
e02119d5 CM |
4509 | if (slot >= nritems) |
4510 | goto find_next_key; | |
3f157a2f CM |
4511 | ret = 0; |
4512 | path->slots[level] = slot; | |
4513 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4514 | goto out; | |
4515 | } | |
9652480b Y |
4516 | if (sret && slot > 0) |
4517 | slot--; | |
3f157a2f | 4518 | /* |
de78b51a | 4519 | * check this node pointer against the min_trans parameters. |
260db43c | 4520 | * If it is too old, skip to the next one. |
3f157a2f | 4521 | */ |
d397712b | 4522 | while (slot < nritems) { |
3f157a2f | 4523 | u64 gen; |
e02119d5 | 4524 | |
3f157a2f CM |
4525 | gen = btrfs_node_ptr_generation(cur, slot); |
4526 | if (gen < min_trans) { | |
4527 | slot++; | |
4528 | continue; | |
4529 | } | |
de78b51a | 4530 | break; |
3f157a2f | 4531 | } |
e02119d5 | 4532 | find_next_key: |
3f157a2f CM |
4533 | /* |
4534 | * we didn't find a candidate key in this node, walk forward | |
4535 | * and find another one | |
4536 | */ | |
4537 | if (slot >= nritems) { | |
e02119d5 CM |
4538 | path->slots[level] = slot; |
4539 | sret = btrfs_find_next_key(root, path, min_key, level, | |
de78b51a | 4540 | min_trans); |
e02119d5 | 4541 | if (sret == 0) { |
b3b4aa74 | 4542 | btrfs_release_path(path); |
3f157a2f CM |
4543 | goto again; |
4544 | } else { | |
4545 | goto out; | |
4546 | } | |
4547 | } | |
4548 | /* save our key for returning back */ | |
4549 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4550 | path->slots[level] = slot; | |
4551 | if (level == path->lowest_level) { | |
4552 | ret = 0; | |
3f157a2f CM |
4553 | goto out; |
4554 | } | |
4b231ae4 | 4555 | cur = btrfs_read_node_slot(cur, slot); |
fb770ae4 LB |
4556 | if (IS_ERR(cur)) { |
4557 | ret = PTR_ERR(cur); | |
4558 | goto out; | |
4559 | } | |
3f157a2f | 4560 | |
bd681513 | 4561 | btrfs_tree_read_lock(cur); |
b4ce94de | 4562 | |
bd681513 | 4563 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4564 | path->nodes[level - 1] = cur; |
f7c79f30 | 4565 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4566 | } |
4567 | out: | |
f98de9b9 FM |
4568 | path->keep_locks = keep_locks; |
4569 | if (ret == 0) { | |
4570 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
3f157a2f | 4571 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 4572 | } |
3f157a2f CM |
4573 | return ret; |
4574 | } | |
4575 | ||
4576 | /* | |
4577 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
4578 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 4579 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
4580 | * |
4581 | * 0 is returned if another key is found, < 0 if there are any errors | |
4582 | * and 1 is returned if there are no higher keys in the tree | |
4583 | * | |
4584 | * path->keep_locks should be set to 1 on the search made before | |
4585 | * calling this function. | |
4586 | */ | |
e7a84565 | 4587 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 4588 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 4589 | { |
e7a84565 CM |
4590 | int slot; |
4591 | struct extent_buffer *c; | |
4592 | ||
6a9fb468 | 4593 | WARN_ON(!path->keep_locks && !path->skip_locking); |
d397712b | 4594 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
4595 | if (!path->nodes[level]) |
4596 | return 1; | |
4597 | ||
4598 | slot = path->slots[level] + 1; | |
4599 | c = path->nodes[level]; | |
3f157a2f | 4600 | next: |
e7a84565 | 4601 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
4602 | int ret; |
4603 | int orig_lowest; | |
4604 | struct btrfs_key cur_key; | |
4605 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
4606 | !path->nodes[level + 1]) | |
e7a84565 | 4607 | return 1; |
33c66f43 | 4608 | |
6a9fb468 | 4609 | if (path->locks[level + 1] || path->skip_locking) { |
33c66f43 YZ |
4610 | level++; |
4611 | continue; | |
4612 | } | |
4613 | ||
4614 | slot = btrfs_header_nritems(c) - 1; | |
4615 | if (level == 0) | |
4616 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
4617 | else | |
4618 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
4619 | ||
4620 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 4621 | btrfs_release_path(path); |
33c66f43 YZ |
4622 | path->lowest_level = level; |
4623 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
4624 | 0, 0); | |
4625 | path->lowest_level = orig_lowest; | |
4626 | if (ret < 0) | |
4627 | return ret; | |
4628 | ||
4629 | c = path->nodes[level]; | |
4630 | slot = path->slots[level]; | |
4631 | if (ret == 0) | |
4632 | slot++; | |
4633 | goto next; | |
e7a84565 | 4634 | } |
33c66f43 | 4635 | |
e7a84565 CM |
4636 | if (level == 0) |
4637 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 4638 | else { |
3f157a2f CM |
4639 | u64 gen = btrfs_node_ptr_generation(c, slot); |
4640 | ||
3f157a2f CM |
4641 | if (gen < min_trans) { |
4642 | slot++; | |
4643 | goto next; | |
4644 | } | |
e7a84565 | 4645 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 4646 | } |
e7a84565 CM |
4647 | return 0; |
4648 | } | |
4649 | return 1; | |
4650 | } | |
4651 | ||
3d7806ec JS |
4652 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, |
4653 | u64 time_seq) | |
d97e63b6 CM |
4654 | { |
4655 | int slot; | |
8e73f275 | 4656 | int level; |
5f39d397 | 4657 | struct extent_buffer *c; |
8e73f275 | 4658 | struct extent_buffer *next; |
d96b3424 | 4659 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 4660 | struct btrfs_key key; |
d96b3424 | 4661 | bool need_commit_sem = false; |
925baedd CM |
4662 | u32 nritems; |
4663 | int ret; | |
0e46318d | 4664 | int i; |
925baedd | 4665 | |
bdcdd86c FM |
4666 | /* |
4667 | * The nowait semantics are used only for write paths, where we don't | |
4668 | * use the tree mod log and sequence numbers. | |
4669 | */ | |
4670 | if (time_seq) | |
4671 | ASSERT(!path->nowait); | |
c922b016 | 4672 | |
925baedd | 4673 | nritems = btrfs_header_nritems(path->nodes[0]); |
d397712b | 4674 | if (nritems == 0) |
925baedd | 4675 | return 1; |
925baedd | 4676 | |
8e73f275 CM |
4677 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
4678 | again: | |
4679 | level = 1; | |
4680 | next = NULL; | |
b3b4aa74 | 4681 | btrfs_release_path(path); |
8e73f275 | 4682 | |
a2135011 | 4683 | path->keep_locks = 1; |
8e73f275 | 4684 | |
d96b3424 | 4685 | if (time_seq) { |
3d7806ec | 4686 | ret = btrfs_search_old_slot(root, &key, path, time_seq); |
d96b3424 FM |
4687 | } else { |
4688 | if (path->need_commit_sem) { | |
4689 | path->need_commit_sem = 0; | |
4690 | need_commit_sem = true; | |
bdcdd86c FM |
4691 | if (path->nowait) { |
4692 | if (!down_read_trylock(&fs_info->commit_root_sem)) { | |
4693 | ret = -EAGAIN; | |
4694 | goto done; | |
4695 | } | |
4696 | } else { | |
4697 | down_read(&fs_info->commit_root_sem); | |
4698 | } | |
d96b3424 | 4699 | } |
3d7806ec | 4700 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
d96b3424 | 4701 | } |
925baedd CM |
4702 | path->keep_locks = 0; |
4703 | ||
4704 | if (ret < 0) | |
d96b3424 | 4705 | goto done; |
925baedd | 4706 | |
a2135011 | 4707 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
4708 | /* |
4709 | * by releasing the path above we dropped all our locks. A balance | |
4710 | * could have added more items next to the key that used to be | |
4711 | * at the very end of the block. So, check again here and | |
4712 | * advance the path if there are now more items available. | |
4713 | */ | |
a2135011 | 4714 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
4715 | if (ret == 0) |
4716 | path->slots[0]++; | |
8e73f275 | 4717 | ret = 0; |
925baedd CM |
4718 | goto done; |
4719 | } | |
0b43e04f LB |
4720 | /* |
4721 | * So the above check misses one case: | |
4722 | * - after releasing the path above, someone has removed the item that | |
4723 | * used to be at the very end of the block, and balance between leafs | |
4724 | * gets another one with bigger key.offset to replace it. | |
4725 | * | |
4726 | * This one should be returned as well, or we can get leaf corruption | |
4727 | * later(esp. in __btrfs_drop_extents()). | |
4728 | * | |
4729 | * And a bit more explanation about this check, | |
4730 | * with ret > 0, the key isn't found, the path points to the slot | |
4731 | * where it should be inserted, so the path->slots[0] item must be the | |
4732 | * bigger one. | |
4733 | */ | |
4734 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
4735 | ret = 0; | |
4736 | goto done; | |
4737 | } | |
d97e63b6 | 4738 | |
d397712b | 4739 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
4740 | if (!path->nodes[level]) { |
4741 | ret = 1; | |
4742 | goto done; | |
4743 | } | |
5f39d397 | 4744 | |
d97e63b6 CM |
4745 | slot = path->slots[level] + 1; |
4746 | c = path->nodes[level]; | |
5f39d397 | 4747 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 4748 | level++; |
8e73f275 CM |
4749 | if (level == BTRFS_MAX_LEVEL) { |
4750 | ret = 1; | |
4751 | goto done; | |
4752 | } | |
d97e63b6 CM |
4753 | continue; |
4754 | } | |
5f39d397 | 4755 | |
0e46318d JB |
4756 | |
4757 | /* | |
4758 | * Our current level is where we're going to start from, and to | |
4759 | * make sure lockdep doesn't complain we need to drop our locks | |
4760 | * and nodes from 0 to our current level. | |
4761 | */ | |
4762 | for (i = 0; i < level; i++) { | |
4763 | if (path->locks[level]) { | |
4764 | btrfs_tree_read_unlock(path->nodes[i]); | |
4765 | path->locks[i] = 0; | |
4766 | } | |
4767 | free_extent_buffer(path->nodes[i]); | |
4768 | path->nodes[i] = NULL; | |
925baedd | 4769 | } |
5f39d397 | 4770 | |
8e73f275 | 4771 | next = c; |
d07b8528 | 4772 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4773 | slot, &key); |
bdcdd86c | 4774 | if (ret == -EAGAIN && !path->nowait) |
8e73f275 | 4775 | goto again; |
5f39d397 | 4776 | |
76a05b35 | 4777 | if (ret < 0) { |
b3b4aa74 | 4778 | btrfs_release_path(path); |
76a05b35 CM |
4779 | goto done; |
4780 | } | |
4781 | ||
5cd57b2c | 4782 | if (!path->skip_locking) { |
bd681513 | 4783 | ret = btrfs_try_tree_read_lock(next); |
bdcdd86c FM |
4784 | if (!ret && path->nowait) { |
4785 | ret = -EAGAIN; | |
4786 | goto done; | |
4787 | } | |
d42244a0 JS |
4788 | if (!ret && time_seq) { |
4789 | /* | |
4790 | * If we don't get the lock, we may be racing | |
4791 | * with push_leaf_left, holding that lock while | |
4792 | * itself waiting for the leaf we've currently | |
4793 | * locked. To solve this situation, we give up | |
4794 | * on our lock and cycle. | |
4795 | */ | |
cf538830 | 4796 | free_extent_buffer(next); |
d42244a0 JS |
4797 | btrfs_release_path(path); |
4798 | cond_resched(); | |
4799 | goto again; | |
4800 | } | |
0e46318d JB |
4801 | if (!ret) |
4802 | btrfs_tree_read_lock(next); | |
5cd57b2c | 4803 | } |
d97e63b6 CM |
4804 | break; |
4805 | } | |
4806 | path->slots[level] = slot; | |
d397712b | 4807 | while (1) { |
d97e63b6 | 4808 | level--; |
d97e63b6 CM |
4809 | path->nodes[level] = next; |
4810 | path->slots[level] = 0; | |
a74a4b97 | 4811 | if (!path->skip_locking) |
ffeb03cf | 4812 | path->locks[level] = BTRFS_READ_LOCK; |
d97e63b6 CM |
4813 | if (!level) |
4814 | break; | |
b4ce94de | 4815 | |
d07b8528 | 4816 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 4817 | 0, &key); |
bdcdd86c | 4818 | if (ret == -EAGAIN && !path->nowait) |
8e73f275 CM |
4819 | goto again; |
4820 | ||
76a05b35 | 4821 | if (ret < 0) { |
b3b4aa74 | 4822 | btrfs_release_path(path); |
76a05b35 CM |
4823 | goto done; |
4824 | } | |
4825 | ||
bdcdd86c FM |
4826 | if (!path->skip_locking) { |
4827 | if (path->nowait) { | |
4828 | if (!btrfs_try_tree_read_lock(next)) { | |
4829 | ret = -EAGAIN; | |
4830 | goto done; | |
4831 | } | |
4832 | } else { | |
4833 | btrfs_tree_read_lock(next); | |
4834 | } | |
4835 | } | |
d97e63b6 | 4836 | } |
8e73f275 | 4837 | ret = 0; |
925baedd | 4838 | done: |
f7c79f30 | 4839 | unlock_up(path, 0, 1, 0, NULL); |
d96b3424 FM |
4840 | if (need_commit_sem) { |
4841 | int ret2; | |
4842 | ||
4843 | path->need_commit_sem = 1; | |
4844 | ret2 = finish_need_commit_sem_search(path); | |
4845 | up_read(&fs_info->commit_root_sem); | |
4846 | if (ret2) | |
4847 | ret = ret2; | |
4848 | } | |
8e73f275 CM |
4849 | |
4850 | return ret; | |
d97e63b6 | 4851 | } |
0b86a832 | 4852 | |
3f157a2f CM |
4853 | /* |
4854 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
4855 | * searching until it gets past min_objectid or finds an item of 'type' | |
4856 | * | |
4857 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4858 | */ | |
0b86a832 CM |
4859 | int btrfs_previous_item(struct btrfs_root *root, |
4860 | struct btrfs_path *path, u64 min_objectid, | |
4861 | int type) | |
4862 | { | |
4863 | struct btrfs_key found_key; | |
4864 | struct extent_buffer *leaf; | |
e02119d5 | 4865 | u32 nritems; |
0b86a832 CM |
4866 | int ret; |
4867 | ||
d397712b | 4868 | while (1) { |
0b86a832 CM |
4869 | if (path->slots[0] == 0) { |
4870 | ret = btrfs_prev_leaf(root, path); | |
4871 | if (ret != 0) | |
4872 | return ret; | |
4873 | } else { | |
4874 | path->slots[0]--; | |
4875 | } | |
4876 | leaf = path->nodes[0]; | |
e02119d5 CM |
4877 | nritems = btrfs_header_nritems(leaf); |
4878 | if (nritems == 0) | |
4879 | return 1; | |
4880 | if (path->slots[0] == nritems) | |
4881 | path->slots[0]--; | |
4882 | ||
0b86a832 | 4883 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
4884 | if (found_key.objectid < min_objectid) |
4885 | break; | |
0a4eefbb YZ |
4886 | if (found_key.type == type) |
4887 | return 0; | |
e02119d5 CM |
4888 | if (found_key.objectid == min_objectid && |
4889 | found_key.type < type) | |
4890 | break; | |
0b86a832 CM |
4891 | } |
4892 | return 1; | |
4893 | } | |
ade2e0b3 WS |
4894 | |
4895 | /* | |
4896 | * search in extent tree to find a previous Metadata/Data extent item with | |
4897 | * min objecitd. | |
4898 | * | |
4899 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
4900 | */ | |
4901 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
4902 | struct btrfs_path *path, u64 min_objectid) | |
4903 | { | |
4904 | struct btrfs_key found_key; | |
4905 | struct extent_buffer *leaf; | |
4906 | u32 nritems; | |
4907 | int ret; | |
4908 | ||
4909 | while (1) { | |
4910 | if (path->slots[0] == 0) { | |
ade2e0b3 WS |
4911 | ret = btrfs_prev_leaf(root, path); |
4912 | if (ret != 0) | |
4913 | return ret; | |
4914 | } else { | |
4915 | path->slots[0]--; | |
4916 | } | |
4917 | leaf = path->nodes[0]; | |
4918 | nritems = btrfs_header_nritems(leaf); | |
4919 | if (nritems == 0) | |
4920 | return 1; | |
4921 | if (path->slots[0] == nritems) | |
4922 | path->slots[0]--; | |
4923 | ||
4924 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
4925 | if (found_key.objectid < min_objectid) | |
4926 | break; | |
4927 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
4928 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
4929 | return 0; | |
4930 | if (found_key.objectid == min_objectid && | |
4931 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
4932 | break; | |
4933 | } | |
4934 | return 1; | |
4935 | } |