Commit | Line | Data |
---|---|---|
6cbd5570 | 1 | /* |
d352ac68 | 2 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
a6b6e75e | 19 | #include <linux/sched.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
bd989ba3 | 21 | #include <linux/rbtree.h> |
8f282f71 | 22 | #include <linux/vmalloc.h> |
eb60ceac CM |
23 | #include "ctree.h" |
24 | #include "disk-io.h" | |
7f5c1516 | 25 | #include "transaction.h" |
5f39d397 | 26 | #include "print-tree.h" |
925baedd | 27 | #include "locking.h" |
9a8dd150 | 28 | |
e089f05c CM |
29 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
30 | *root, struct btrfs_path *path, int level); | |
31 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 32 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 33 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
34 | static int push_node_left(struct btrfs_trans_handle *trans, |
35 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 36 | struct extent_buffer *src, int empty); |
5f39d397 CM |
37 | static int balance_node_right(struct btrfs_trans_handle *trans, |
38 | struct btrfs_root *root, | |
39 | struct extent_buffer *dst_buf, | |
40 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
41 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
42 | int level, int slot); | |
5de865ee | 43 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 44 | struct extent_buffer *eb); |
d97e63b6 | 45 | |
df24a2b9 | 46 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 47 | { |
e2c89907 | 48 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
49 | } |
50 | ||
b4ce94de CM |
51 | /* |
52 | * set all locked nodes in the path to blocking locks. This should | |
53 | * be done before scheduling | |
54 | */ | |
55 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
56 | { | |
57 | int i; | |
58 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
59 | if (!p->nodes[i] || !p->locks[i]) |
60 | continue; | |
61 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
62 | if (p->locks[i] == BTRFS_READ_LOCK) | |
63 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
64 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
65 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
66 | } |
67 | } | |
68 | ||
69 | /* | |
70 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
71 | * |
72 | * held is used to keep lockdep happy, when lockdep is enabled | |
73 | * we set held to a blocking lock before we go around and | |
74 | * retake all the spinlocks in the path. You can safely use NULL | |
75 | * for held | |
b4ce94de | 76 | */ |
4008c04a | 77 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 78 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
79 | { |
80 | int i; | |
4008c04a | 81 | |
bd681513 CM |
82 | if (held) { |
83 | btrfs_set_lock_blocking_rw(held, held_rw); | |
84 | if (held_rw == BTRFS_WRITE_LOCK) | |
85 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
86 | else if (held_rw == BTRFS_READ_LOCK) | |
87 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
88 | } | |
4008c04a | 89 | btrfs_set_path_blocking(p); |
4008c04a CM |
90 | |
91 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
92 | if (p->nodes[i] && p->locks[i]) { |
93 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
94 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
95 | p->locks[i] = BTRFS_WRITE_LOCK; | |
96 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
97 | p->locks[i] = BTRFS_READ_LOCK; | |
98 | } | |
b4ce94de | 99 | } |
4008c04a | 100 | |
4008c04a | 101 | if (held) |
bd681513 | 102 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
103 | } |
104 | ||
d352ac68 | 105 | /* this also releases the path */ |
df24a2b9 | 106 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 107 | { |
ff175d57 JJ |
108 | if (!p) |
109 | return; | |
b3b4aa74 | 110 | btrfs_release_path(p); |
df24a2b9 | 111 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
112 | } |
113 | ||
d352ac68 CM |
114 | /* |
115 | * path release drops references on the extent buffers in the path | |
116 | * and it drops any locks held by this path | |
117 | * | |
118 | * It is safe to call this on paths that no locks or extent buffers held. | |
119 | */ | |
b3b4aa74 | 120 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
121 | { |
122 | int i; | |
a2135011 | 123 | |
234b63a0 | 124 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 125 | p->slots[i] = 0; |
eb60ceac | 126 | if (!p->nodes[i]) |
925baedd CM |
127 | continue; |
128 | if (p->locks[i]) { | |
bd681513 | 129 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
130 | p->locks[i] = 0; |
131 | } | |
5f39d397 | 132 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 133 | p->nodes[i] = NULL; |
eb60ceac CM |
134 | } |
135 | } | |
136 | ||
d352ac68 CM |
137 | /* |
138 | * safely gets a reference on the root node of a tree. A lock | |
139 | * is not taken, so a concurrent writer may put a different node | |
140 | * at the root of the tree. See btrfs_lock_root_node for the | |
141 | * looping required. | |
142 | * | |
143 | * The extent buffer returned by this has a reference taken, so | |
144 | * it won't disappear. It may stop being the root of the tree | |
145 | * at any time because there are no locks held. | |
146 | */ | |
925baedd CM |
147 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
148 | { | |
149 | struct extent_buffer *eb; | |
240f62c8 | 150 | |
3083ee2e JB |
151 | while (1) { |
152 | rcu_read_lock(); | |
153 | eb = rcu_dereference(root->node); | |
154 | ||
155 | /* | |
156 | * RCU really hurts here, we could free up the root node because | |
01327610 | 157 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
158 | * the inc_not_zero dance and if it doesn't work then |
159 | * synchronize_rcu and try again. | |
160 | */ | |
161 | if (atomic_inc_not_zero(&eb->refs)) { | |
162 | rcu_read_unlock(); | |
163 | break; | |
164 | } | |
165 | rcu_read_unlock(); | |
166 | synchronize_rcu(); | |
167 | } | |
925baedd CM |
168 | return eb; |
169 | } | |
170 | ||
d352ac68 CM |
171 | /* loop around taking references on and locking the root node of the |
172 | * tree until you end up with a lock on the root. A locked buffer | |
173 | * is returned, with a reference held. | |
174 | */ | |
925baedd CM |
175 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
176 | { | |
177 | struct extent_buffer *eb; | |
178 | ||
d397712b | 179 | while (1) { |
925baedd CM |
180 | eb = btrfs_root_node(root); |
181 | btrfs_tree_lock(eb); | |
240f62c8 | 182 | if (eb == root->node) |
925baedd | 183 | break; |
925baedd CM |
184 | btrfs_tree_unlock(eb); |
185 | free_extent_buffer(eb); | |
186 | } | |
187 | return eb; | |
188 | } | |
189 | ||
bd681513 CM |
190 | /* loop around taking references on and locking the root node of the |
191 | * tree until you end up with a lock on the root. A locked buffer | |
192 | * is returned, with a reference held. | |
193 | */ | |
48a3b636 | 194 | static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
195 | { |
196 | struct extent_buffer *eb; | |
197 | ||
198 | while (1) { | |
199 | eb = btrfs_root_node(root); | |
200 | btrfs_tree_read_lock(eb); | |
201 | if (eb == root->node) | |
202 | break; | |
203 | btrfs_tree_read_unlock(eb); | |
204 | free_extent_buffer(eb); | |
205 | } | |
206 | return eb; | |
207 | } | |
208 | ||
d352ac68 CM |
209 | /* cowonly root (everything not a reference counted cow subvolume), just get |
210 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
211 | * get properly updated on disk. | |
212 | */ | |
0b86a832 CM |
213 | static void add_root_to_dirty_list(struct btrfs_root *root) |
214 | { | |
0b246afa JM |
215 | struct btrfs_fs_info *fs_info = root->fs_info; |
216 | ||
e7070be1 JB |
217 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
218 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
219 | return; | |
220 | ||
0b246afa | 221 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
222 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
223 | /* Want the extent tree to be the last on the list */ | |
224 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
225 | list_move_tail(&root->dirty_list, | |
0b246afa | 226 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
227 | else |
228 | list_move(&root->dirty_list, | |
0b246afa | 229 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 230 | } |
0b246afa | 231 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
232 | } |
233 | ||
d352ac68 CM |
234 | /* |
235 | * used by snapshot creation to make a copy of a root for a tree with | |
236 | * a given objectid. The buffer with the new root node is returned in | |
237 | * cow_ret, and this func returns zero on success or a negative error code. | |
238 | */ | |
be20aa9d CM |
239 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
240 | struct btrfs_root *root, | |
241 | struct extent_buffer *buf, | |
242 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
243 | { | |
0b246afa | 244 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 245 | struct extent_buffer *cow; |
be20aa9d CM |
246 | int ret = 0; |
247 | int level; | |
5d4f98a2 | 248 | struct btrfs_disk_key disk_key; |
be20aa9d | 249 | |
27cdeb70 | 250 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 251 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
252 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
253 | trans->transid != root->last_trans); | |
be20aa9d CM |
254 | |
255 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
256 | if (level == 0) |
257 | btrfs_item_key(buf, &disk_key, 0); | |
258 | else | |
259 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 260 | |
4d75f8a9 DS |
261 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
262 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 263 | if (IS_ERR(cow)) |
be20aa9d CM |
264 | return PTR_ERR(cow); |
265 | ||
58e8012c | 266 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
267 | btrfs_set_header_bytenr(cow, cow->start); |
268 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
269 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
270 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
271 | BTRFS_HEADER_FLAG_RELOC); | |
272 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
273 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
274 | else | |
275 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 276 | |
0b246afa | 277 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 278 | |
be20aa9d | 279 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 280 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 281 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 282 | else |
e339a6b0 | 283 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 284 | |
be20aa9d CM |
285 | if (ret) |
286 | return ret; | |
287 | ||
288 | btrfs_mark_buffer_dirty(cow); | |
289 | *cow_ret = cow; | |
290 | return 0; | |
291 | } | |
292 | ||
bd989ba3 JS |
293 | enum mod_log_op { |
294 | MOD_LOG_KEY_REPLACE, | |
295 | MOD_LOG_KEY_ADD, | |
296 | MOD_LOG_KEY_REMOVE, | |
297 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
299 | MOD_LOG_MOVE_KEYS, | |
300 | MOD_LOG_ROOT_REPLACE, | |
301 | }; | |
302 | ||
303 | struct tree_mod_move { | |
304 | int dst_slot; | |
305 | int nr_items; | |
306 | }; | |
307 | ||
308 | struct tree_mod_root { | |
309 | u64 logical; | |
310 | u8 level; | |
311 | }; | |
312 | ||
313 | struct tree_mod_elem { | |
314 | struct rb_node node; | |
298cfd36 | 315 | u64 logical; |
097b8a7c | 316 | u64 seq; |
bd989ba3 JS |
317 | enum mod_log_op op; |
318 | ||
319 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
320 | int slot; | |
321 | ||
322 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
323 | u64 generation; | |
324 | ||
325 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
326 | struct btrfs_disk_key key; | |
327 | u64 blockptr; | |
328 | ||
329 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
330 | struct tree_mod_move move; | |
331 | ||
332 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
333 | struct tree_mod_root old_root; | |
334 | }; | |
335 | ||
097b8a7c | 336 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 337 | { |
097b8a7c | 338 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
339 | } |
340 | ||
097b8a7c JS |
341 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
342 | { | |
343 | read_unlock(&fs_info->tree_mod_log_lock); | |
344 | } | |
345 | ||
346 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
347 | { | |
348 | write_lock(&fs_info->tree_mod_log_lock); | |
349 | } | |
350 | ||
351 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
352 | { | |
353 | write_unlock(&fs_info->tree_mod_log_lock); | |
354 | } | |
355 | ||
fc36ed7e | 356 | /* |
fcebe456 | 357 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 358 | */ |
fcebe456 | 359 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
360 | { |
361 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
362 | } | |
363 | ||
097b8a7c JS |
364 | /* |
365 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
366 | * passed does not already have a sequence number set. So when a caller expects | |
367 | * to record tree modifications, it should ensure to set elem->seq to zero | |
368 | * before calling btrfs_get_tree_mod_seq. | |
369 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
370 | * blocker was added. | |
371 | */ | |
372 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
373 | struct seq_list *elem) | |
bd989ba3 | 374 | { |
097b8a7c | 375 | tree_mod_log_write_lock(fs_info); |
bd989ba3 | 376 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 377 | if (!elem->seq) { |
fcebe456 | 378 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
379 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
380 | } | |
bd989ba3 | 381 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
382 | tree_mod_log_write_unlock(fs_info); |
383 | ||
fcebe456 | 384 | return elem->seq; |
bd989ba3 JS |
385 | } |
386 | ||
387 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
388 | struct seq_list *elem) | |
389 | { | |
390 | struct rb_root *tm_root; | |
391 | struct rb_node *node; | |
392 | struct rb_node *next; | |
393 | struct seq_list *cur_elem; | |
394 | struct tree_mod_elem *tm; | |
395 | u64 min_seq = (u64)-1; | |
396 | u64 seq_putting = elem->seq; | |
397 | ||
398 | if (!seq_putting) | |
399 | return; | |
400 | ||
bd989ba3 JS |
401 | spin_lock(&fs_info->tree_mod_seq_lock); |
402 | list_del(&elem->list); | |
097b8a7c | 403 | elem->seq = 0; |
bd989ba3 JS |
404 | |
405 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 406 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
407 | if (seq_putting > cur_elem->seq) { |
408 | /* | |
409 | * blocker with lower sequence number exists, we | |
410 | * cannot remove anything from the log | |
411 | */ | |
097b8a7c JS |
412 | spin_unlock(&fs_info->tree_mod_seq_lock); |
413 | return; | |
bd989ba3 JS |
414 | } |
415 | min_seq = cur_elem->seq; | |
416 | } | |
417 | } | |
097b8a7c JS |
418 | spin_unlock(&fs_info->tree_mod_seq_lock); |
419 | ||
bd989ba3 JS |
420 | /* |
421 | * anything that's lower than the lowest existing (read: blocked) | |
422 | * sequence number can be removed from the tree. | |
423 | */ | |
097b8a7c | 424 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
425 | tm_root = &fs_info->tree_mod_log; |
426 | for (node = rb_first(tm_root); node; node = next) { | |
427 | next = rb_next(node); | |
428 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 429 | if (tm->seq > min_seq) |
bd989ba3 JS |
430 | continue; |
431 | rb_erase(node, tm_root); | |
bd989ba3 JS |
432 | kfree(tm); |
433 | } | |
097b8a7c | 434 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
435 | } |
436 | ||
437 | /* | |
438 | * key order of the log: | |
298cfd36 | 439 | * node/leaf start address -> sequence |
bd989ba3 | 440 | * |
298cfd36 CR |
441 | * The 'start address' is the logical address of the *new* root node |
442 | * for root replace operations, or the logical address of the affected | |
443 | * block for all other operations. | |
5de865ee FDBM |
444 | * |
445 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
446 | */ |
447 | static noinline int | |
448 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
449 | { | |
450 | struct rb_root *tm_root; | |
451 | struct rb_node **new; | |
452 | struct rb_node *parent = NULL; | |
453 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
454 | |
455 | BUG_ON(!tm); | |
456 | ||
fcebe456 | 457 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 458 | |
bd989ba3 JS |
459 | tm_root = &fs_info->tree_mod_log; |
460 | new = &tm_root->rb_node; | |
461 | while (*new) { | |
462 | cur = container_of(*new, struct tree_mod_elem, node); | |
463 | parent = *new; | |
298cfd36 | 464 | if (cur->logical < tm->logical) |
bd989ba3 | 465 | new = &((*new)->rb_left); |
298cfd36 | 466 | else if (cur->logical > tm->logical) |
bd989ba3 | 467 | new = &((*new)->rb_right); |
097b8a7c | 468 | else if (cur->seq < tm->seq) |
bd989ba3 | 469 | new = &((*new)->rb_left); |
097b8a7c | 470 | else if (cur->seq > tm->seq) |
bd989ba3 | 471 | new = &((*new)->rb_right); |
5de865ee FDBM |
472 | else |
473 | return -EEXIST; | |
bd989ba3 JS |
474 | } |
475 | ||
476 | rb_link_node(&tm->node, parent, new); | |
477 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 478 | return 0; |
bd989ba3 JS |
479 | } |
480 | ||
097b8a7c JS |
481 | /* |
482 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
483 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
484 | * this until all tree mod log insertions are recorded in the rb tree and then | |
485 | * call tree_mod_log_write_unlock() to release. | |
486 | */ | |
e9b7fd4d JS |
487 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
488 | struct extent_buffer *eb) { | |
489 | smp_mb(); | |
490 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
491 | return 1; | |
097b8a7c JS |
492 | if (eb && btrfs_header_level(eb) == 0) |
493 | return 1; | |
5de865ee FDBM |
494 | |
495 | tree_mod_log_write_lock(fs_info); | |
496 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
497 | tree_mod_log_write_unlock(fs_info); | |
498 | return 1; | |
499 | } | |
500 | ||
e9b7fd4d JS |
501 | return 0; |
502 | } | |
503 | ||
5de865ee FDBM |
504 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
505 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
506 | struct extent_buffer *eb) | |
507 | { | |
508 | smp_mb(); | |
509 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
510 | return 0; | |
511 | if (eb && btrfs_header_level(eb) == 0) | |
512 | return 0; | |
513 | ||
514 | return 1; | |
515 | } | |
516 | ||
517 | static struct tree_mod_elem * | |
518 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
519 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 520 | { |
097b8a7c | 521 | struct tree_mod_elem *tm; |
bd989ba3 | 522 | |
c8cc6341 JB |
523 | tm = kzalloc(sizeof(*tm), flags); |
524 | if (!tm) | |
5de865ee | 525 | return NULL; |
bd989ba3 | 526 | |
298cfd36 | 527 | tm->logical = eb->start; |
bd989ba3 JS |
528 | if (op != MOD_LOG_KEY_ADD) { |
529 | btrfs_node_key(eb, &tm->key, slot); | |
530 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
531 | } | |
532 | tm->op = op; | |
533 | tm->slot = slot; | |
534 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 535 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 536 | |
5de865ee | 537 | return tm; |
097b8a7c JS |
538 | } |
539 | ||
540 | static noinline int | |
c8cc6341 JB |
541 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
542 | struct extent_buffer *eb, int slot, | |
543 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 544 | { |
5de865ee FDBM |
545 | struct tree_mod_elem *tm; |
546 | int ret; | |
547 | ||
548 | if (!tree_mod_need_log(fs_info, eb)) | |
549 | return 0; | |
550 | ||
551 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
552 | if (!tm) | |
553 | return -ENOMEM; | |
554 | ||
555 | if (tree_mod_dont_log(fs_info, eb)) { | |
556 | kfree(tm); | |
097b8a7c | 557 | return 0; |
5de865ee FDBM |
558 | } |
559 | ||
560 | ret = __tree_mod_log_insert(fs_info, tm); | |
561 | tree_mod_log_write_unlock(fs_info); | |
562 | if (ret) | |
563 | kfree(tm); | |
097b8a7c | 564 | |
5de865ee | 565 | return ret; |
097b8a7c JS |
566 | } |
567 | ||
bd989ba3 JS |
568 | static noinline int |
569 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
570 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
571 | int nr_items, gfp_t flags) | |
572 | { | |
5de865ee FDBM |
573 | struct tree_mod_elem *tm = NULL; |
574 | struct tree_mod_elem **tm_list = NULL; | |
575 | int ret = 0; | |
bd989ba3 | 576 | int i; |
5de865ee | 577 | int locked = 0; |
bd989ba3 | 578 | |
5de865ee | 579 | if (!tree_mod_need_log(fs_info, eb)) |
f395694c | 580 | return 0; |
bd989ba3 | 581 | |
31e818fe | 582 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), flags); |
5de865ee FDBM |
583 | if (!tm_list) |
584 | return -ENOMEM; | |
585 | ||
586 | tm = kzalloc(sizeof(*tm), flags); | |
587 | if (!tm) { | |
588 | ret = -ENOMEM; | |
589 | goto free_tms; | |
590 | } | |
591 | ||
298cfd36 | 592 | tm->logical = eb->start; |
5de865ee FDBM |
593 | tm->slot = src_slot; |
594 | tm->move.dst_slot = dst_slot; | |
595 | tm->move.nr_items = nr_items; | |
596 | tm->op = MOD_LOG_MOVE_KEYS; | |
597 | ||
598 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
599 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
600 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags); | |
601 | if (!tm_list[i]) { | |
602 | ret = -ENOMEM; | |
603 | goto free_tms; | |
604 | } | |
605 | } | |
606 | ||
607 | if (tree_mod_dont_log(fs_info, eb)) | |
608 | goto free_tms; | |
609 | locked = 1; | |
610 | ||
01763a2e JS |
611 | /* |
612 | * When we override something during the move, we log these removals. | |
613 | * This can only happen when we move towards the beginning of the | |
614 | * buffer, i.e. dst_slot < src_slot. | |
615 | */ | |
bd989ba3 | 616 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
5de865ee FDBM |
617 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
618 | if (ret) | |
619 | goto free_tms; | |
bd989ba3 JS |
620 | } |
621 | ||
5de865ee FDBM |
622 | ret = __tree_mod_log_insert(fs_info, tm); |
623 | if (ret) | |
624 | goto free_tms; | |
625 | tree_mod_log_write_unlock(fs_info); | |
626 | kfree(tm_list); | |
f395694c | 627 | |
5de865ee FDBM |
628 | return 0; |
629 | free_tms: | |
630 | for (i = 0; i < nr_items; i++) { | |
631 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
632 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
633 | kfree(tm_list[i]); | |
634 | } | |
635 | if (locked) | |
636 | tree_mod_log_write_unlock(fs_info); | |
637 | kfree(tm_list); | |
638 | kfree(tm); | |
bd989ba3 | 639 | |
5de865ee | 640 | return ret; |
bd989ba3 JS |
641 | } |
642 | ||
5de865ee FDBM |
643 | static inline int |
644 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
645 | struct tree_mod_elem **tm_list, | |
646 | int nritems) | |
097b8a7c | 647 | { |
5de865ee | 648 | int i, j; |
097b8a7c JS |
649 | int ret; |
650 | ||
097b8a7c | 651 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
652 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
653 | if (ret) { | |
654 | for (j = nritems - 1; j > i; j--) | |
655 | rb_erase(&tm_list[j]->node, | |
656 | &fs_info->tree_mod_log); | |
657 | return ret; | |
658 | } | |
097b8a7c | 659 | } |
5de865ee FDBM |
660 | |
661 | return 0; | |
097b8a7c JS |
662 | } |
663 | ||
bd989ba3 JS |
664 | static noinline int |
665 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
666 | struct extent_buffer *old_root, | |
90f8d62e JS |
667 | struct extent_buffer *new_root, gfp_t flags, |
668 | int log_removal) | |
bd989ba3 | 669 | { |
5de865ee FDBM |
670 | struct tree_mod_elem *tm = NULL; |
671 | struct tree_mod_elem **tm_list = NULL; | |
672 | int nritems = 0; | |
673 | int ret = 0; | |
674 | int i; | |
bd989ba3 | 675 | |
5de865ee | 676 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
677 | return 0; |
678 | ||
5de865ee FDBM |
679 | if (log_removal && btrfs_header_level(old_root) > 0) { |
680 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 681 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
682 | flags); |
683 | if (!tm_list) { | |
684 | ret = -ENOMEM; | |
685 | goto free_tms; | |
686 | } | |
687 | for (i = 0; i < nritems; i++) { | |
688 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
689 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags); | |
690 | if (!tm_list[i]) { | |
691 | ret = -ENOMEM; | |
692 | goto free_tms; | |
693 | } | |
694 | } | |
695 | } | |
d9abbf1c | 696 | |
c8cc6341 | 697 | tm = kzalloc(sizeof(*tm), flags); |
5de865ee FDBM |
698 | if (!tm) { |
699 | ret = -ENOMEM; | |
700 | goto free_tms; | |
701 | } | |
bd989ba3 | 702 | |
298cfd36 | 703 | tm->logical = new_root->start; |
bd989ba3 JS |
704 | tm->old_root.logical = old_root->start; |
705 | tm->old_root.level = btrfs_header_level(old_root); | |
706 | tm->generation = btrfs_header_generation(old_root); | |
707 | tm->op = MOD_LOG_ROOT_REPLACE; | |
708 | ||
5de865ee FDBM |
709 | if (tree_mod_dont_log(fs_info, NULL)) |
710 | goto free_tms; | |
711 | ||
712 | if (tm_list) | |
713 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
714 | if (!ret) | |
715 | ret = __tree_mod_log_insert(fs_info, tm); | |
716 | ||
717 | tree_mod_log_write_unlock(fs_info); | |
718 | if (ret) | |
719 | goto free_tms; | |
720 | kfree(tm_list); | |
721 | ||
722 | return ret; | |
723 | ||
724 | free_tms: | |
725 | if (tm_list) { | |
726 | for (i = 0; i < nritems; i++) | |
727 | kfree(tm_list[i]); | |
728 | kfree(tm_list); | |
729 | } | |
730 | kfree(tm); | |
731 | ||
732 | return ret; | |
bd989ba3 JS |
733 | } |
734 | ||
735 | static struct tree_mod_elem * | |
736 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
737 | int smallest) | |
738 | { | |
739 | struct rb_root *tm_root; | |
740 | struct rb_node *node; | |
741 | struct tree_mod_elem *cur = NULL; | |
742 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 743 | |
097b8a7c | 744 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
745 | tm_root = &fs_info->tree_mod_log; |
746 | node = tm_root->rb_node; | |
747 | while (node) { | |
748 | cur = container_of(node, struct tree_mod_elem, node); | |
298cfd36 | 749 | if (cur->logical < start) { |
bd989ba3 | 750 | node = node->rb_left; |
298cfd36 | 751 | } else if (cur->logical > start) { |
bd989ba3 | 752 | node = node->rb_right; |
097b8a7c | 753 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
754 | node = node->rb_left; |
755 | } else if (!smallest) { | |
756 | /* we want the node with the highest seq */ | |
757 | if (found) | |
097b8a7c | 758 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
759 | found = cur; |
760 | node = node->rb_left; | |
097b8a7c | 761 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
762 | /* we want the node with the smallest seq */ |
763 | if (found) | |
097b8a7c | 764 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
765 | found = cur; |
766 | node = node->rb_right; | |
767 | } else { | |
768 | found = cur; | |
769 | break; | |
770 | } | |
771 | } | |
097b8a7c | 772 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
773 | |
774 | return found; | |
775 | } | |
776 | ||
777 | /* | |
778 | * this returns the element from the log with the smallest time sequence | |
779 | * value that's in the log (the oldest log item). any element with a time | |
780 | * sequence lower than min_seq will be ignored. | |
781 | */ | |
782 | static struct tree_mod_elem * | |
783 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
784 | u64 min_seq) | |
785 | { | |
786 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
787 | } | |
788 | ||
789 | /* | |
790 | * this returns the element from the log with the largest time sequence | |
791 | * value that's in the log (the most recent log item). any element with | |
792 | * a time sequence lower than min_seq will be ignored. | |
793 | */ | |
794 | static struct tree_mod_elem * | |
795 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
796 | { | |
797 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
798 | } | |
799 | ||
5de865ee | 800 | static noinline int |
bd989ba3 JS |
801 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
802 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 803 | unsigned long src_offset, int nr_items) |
bd989ba3 | 804 | { |
5de865ee FDBM |
805 | int ret = 0; |
806 | struct tree_mod_elem **tm_list = NULL; | |
807 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 808 | int i; |
5de865ee | 809 | int locked = 0; |
bd989ba3 | 810 | |
5de865ee FDBM |
811 | if (!tree_mod_need_log(fs_info, NULL)) |
812 | return 0; | |
bd989ba3 | 813 | |
c8cc6341 | 814 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
815 | return 0; |
816 | ||
31e818fe | 817 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
818 | GFP_NOFS); |
819 | if (!tm_list) | |
820 | return -ENOMEM; | |
bd989ba3 | 821 | |
5de865ee FDBM |
822 | tm_list_add = tm_list; |
823 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 824 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
825 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
826 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
827 | if (!tm_list_rem[i]) { | |
828 | ret = -ENOMEM; | |
829 | goto free_tms; | |
830 | } | |
831 | ||
832 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
833 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
834 | if (!tm_list_add[i]) { | |
835 | ret = -ENOMEM; | |
836 | goto free_tms; | |
837 | } | |
838 | } | |
839 | ||
840 | if (tree_mod_dont_log(fs_info, NULL)) | |
841 | goto free_tms; | |
842 | locked = 1; | |
843 | ||
844 | for (i = 0; i < nr_items; i++) { | |
845 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
846 | if (ret) | |
847 | goto free_tms; | |
848 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
849 | if (ret) | |
850 | goto free_tms; | |
bd989ba3 | 851 | } |
5de865ee FDBM |
852 | |
853 | tree_mod_log_write_unlock(fs_info); | |
854 | kfree(tm_list); | |
855 | ||
856 | return 0; | |
857 | ||
858 | free_tms: | |
859 | for (i = 0; i < nr_items * 2; i++) { | |
860 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
861 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
862 | kfree(tm_list[i]); | |
863 | } | |
864 | if (locked) | |
865 | tree_mod_log_write_unlock(fs_info); | |
866 | kfree(tm_list); | |
867 | ||
868 | return ret; | |
bd989ba3 JS |
869 | } |
870 | ||
871 | static inline void | |
872 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
873 | int dst_offset, int src_offset, int nr_items) | |
874 | { | |
875 | int ret; | |
876 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
877 | nr_items, GFP_NOFS); | |
878 | BUG_ON(ret < 0); | |
879 | } | |
880 | ||
097b8a7c | 881 | static noinline void |
bd989ba3 | 882 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 883 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
884 | { |
885 | int ret; | |
886 | ||
78357766 | 887 | ret = tree_mod_log_insert_key(fs_info, eb, slot, |
c8cc6341 JB |
888 | MOD_LOG_KEY_REPLACE, |
889 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
890 | BUG_ON(ret < 0); |
891 | } | |
892 | ||
5de865ee | 893 | static noinline int |
097b8a7c | 894 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 895 | { |
5de865ee FDBM |
896 | struct tree_mod_elem **tm_list = NULL; |
897 | int nritems = 0; | |
898 | int i; | |
899 | int ret = 0; | |
900 | ||
901 | if (btrfs_header_level(eb) == 0) | |
902 | return 0; | |
903 | ||
904 | if (!tree_mod_need_log(fs_info, NULL)) | |
905 | return 0; | |
906 | ||
907 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 908 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
909 | if (!tm_list) |
910 | return -ENOMEM; | |
911 | ||
912 | for (i = 0; i < nritems; i++) { | |
913 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
914 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
915 | if (!tm_list[i]) { | |
916 | ret = -ENOMEM; | |
917 | goto free_tms; | |
918 | } | |
919 | } | |
920 | ||
e9b7fd4d | 921 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
922 | goto free_tms; |
923 | ||
924 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
925 | tree_mod_log_write_unlock(fs_info); | |
926 | if (ret) | |
927 | goto free_tms; | |
928 | kfree(tm_list); | |
929 | ||
930 | return 0; | |
931 | ||
932 | free_tms: | |
933 | for (i = 0; i < nritems; i++) | |
934 | kfree(tm_list[i]); | |
935 | kfree(tm_list); | |
936 | ||
937 | return ret; | |
bd989ba3 JS |
938 | } |
939 | ||
097b8a7c | 940 | static noinline void |
bd989ba3 | 941 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
942 | struct extent_buffer *new_root_node, |
943 | int log_removal) | |
bd989ba3 JS |
944 | { |
945 | int ret; | |
bd989ba3 | 946 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 947 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
948 | BUG_ON(ret < 0); |
949 | } | |
950 | ||
5d4f98a2 YZ |
951 | /* |
952 | * check if the tree block can be shared by multiple trees | |
953 | */ | |
954 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
955 | struct extent_buffer *buf) | |
956 | { | |
957 | /* | |
01327610 | 958 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
959 | * are never shared. If a block was allocated after the last |
960 | * snapshot and the block was not allocated by tree relocation, | |
961 | * we know the block is not shared. | |
962 | */ | |
27cdeb70 | 963 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
964 | buf != root->node && buf != root->commit_root && |
965 | (btrfs_header_generation(buf) <= | |
966 | btrfs_root_last_snapshot(&root->root_item) || | |
967 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
968 | return 1; | |
969 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 970 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
971 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
972 | return 1; | |
973 | #endif | |
974 | return 0; | |
975 | } | |
976 | ||
977 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
978 | struct btrfs_root *root, | |
979 | struct extent_buffer *buf, | |
f0486c68 YZ |
980 | struct extent_buffer *cow, |
981 | int *last_ref) | |
5d4f98a2 | 982 | { |
0b246afa | 983 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
984 | u64 refs; |
985 | u64 owner; | |
986 | u64 flags; | |
987 | u64 new_flags = 0; | |
988 | int ret; | |
989 | ||
990 | /* | |
991 | * Backrefs update rules: | |
992 | * | |
993 | * Always use full backrefs for extent pointers in tree block | |
994 | * allocated by tree relocation. | |
995 | * | |
996 | * If a shared tree block is no longer referenced by its owner | |
997 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
998 | * use full backrefs for extent pointers in tree block. | |
999 | * | |
1000 | * If a tree block is been relocating | |
1001 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
1002 | * use full backrefs for extent pointers in tree block. | |
1003 | * The reason for this is some operations (such as drop tree) | |
1004 | * are only allowed for blocks use full backrefs. | |
1005 | */ | |
1006 | ||
1007 | if (btrfs_block_can_be_shared(root, buf)) { | |
1008 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
1009 | btrfs_header_level(buf), 1, |
1010 | &refs, &flags); | |
be1a5564 MF |
1011 | if (ret) |
1012 | return ret; | |
e5df9573 MF |
1013 | if (refs == 0) { |
1014 | ret = -EROFS; | |
0b246afa | 1015 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
1016 | return ret; |
1017 | } | |
5d4f98a2 YZ |
1018 | } else { |
1019 | refs = 1; | |
1020 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1021 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1022 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1023 | else | |
1024 | flags = 0; | |
1025 | } | |
1026 | ||
1027 | owner = btrfs_header_owner(buf); | |
1028 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1029 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1030 | ||
1031 | if (refs > 1) { | |
1032 | if ((owner == root->root_key.objectid || | |
1033 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1034 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 1035 | ret = btrfs_inc_ref(trans, root, buf, 1); |
79787eaa | 1036 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1037 | |
1038 | if (root->root_key.objectid == | |
1039 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 1040 | ret = btrfs_dec_ref(trans, root, buf, 0); |
79787eaa | 1041 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1042 | ret = btrfs_inc_ref(trans, root, cow, 1); |
79787eaa | 1043 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1044 | } |
1045 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1046 | } else { | |
1047 | ||
1048 | if (root->root_key.objectid == | |
1049 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1050 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1051 | else |
e339a6b0 | 1052 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1053 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1054 | } |
1055 | if (new_flags != 0) { | |
b1c79e09 JB |
1056 | int level = btrfs_header_level(buf); |
1057 | ||
5d4f98a2 YZ |
1058 | ret = btrfs_set_disk_extent_flags(trans, root, |
1059 | buf->start, | |
1060 | buf->len, | |
b1c79e09 | 1061 | new_flags, level, 0); |
be1a5564 MF |
1062 | if (ret) |
1063 | return ret; | |
5d4f98a2 YZ |
1064 | } |
1065 | } else { | |
1066 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1067 | if (root->root_key.objectid == | |
1068 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1069 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1070 | else |
e339a6b0 | 1071 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1072 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1073 | ret = btrfs_dec_ref(trans, root, buf, 1); |
79787eaa | 1074 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 | 1075 | } |
0b246afa | 1076 | clean_tree_block(trans, fs_info, buf); |
f0486c68 | 1077 | *last_ref = 1; |
5d4f98a2 YZ |
1078 | } |
1079 | return 0; | |
1080 | } | |
1081 | ||
d352ac68 | 1082 | /* |
d397712b CM |
1083 | * does the dirty work in cow of a single block. The parent block (if |
1084 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1085 | * dirty and returned locked. If you modify the block it needs to be marked | |
1086 | * dirty again. | |
d352ac68 CM |
1087 | * |
1088 | * search_start -- an allocation hint for the new block | |
1089 | * | |
d397712b CM |
1090 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1091 | * bytes the allocator should try to find free next to the block it returns. | |
1092 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1093 | */ |
d397712b | 1094 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1095 | struct btrfs_root *root, |
1096 | struct extent_buffer *buf, | |
1097 | struct extent_buffer *parent, int parent_slot, | |
1098 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1099 | u64 search_start, u64 empty_size) |
02217ed2 | 1100 | { |
0b246afa | 1101 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1102 | struct btrfs_disk_key disk_key; |
5f39d397 | 1103 | struct extent_buffer *cow; |
be1a5564 | 1104 | int level, ret; |
f0486c68 | 1105 | int last_ref = 0; |
925baedd | 1106 | int unlock_orig = 0; |
0f5053eb | 1107 | u64 parent_start = 0; |
7bb86316 | 1108 | |
925baedd CM |
1109 | if (*cow_ret == buf) |
1110 | unlock_orig = 1; | |
1111 | ||
b9447ef8 | 1112 | btrfs_assert_tree_locked(buf); |
925baedd | 1113 | |
27cdeb70 | 1114 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1115 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1116 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1117 | trans->transid != root->last_trans); | |
5f39d397 | 1118 | |
7bb86316 | 1119 | level = btrfs_header_level(buf); |
31840ae1 | 1120 | |
5d4f98a2 YZ |
1121 | if (level == 0) |
1122 | btrfs_item_key(buf, &disk_key, 0); | |
1123 | else | |
1124 | btrfs_node_key(buf, &disk_key, 0); | |
1125 | ||
0f5053eb GR |
1126 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1127 | parent_start = parent->start; | |
5d4f98a2 | 1128 | |
4d75f8a9 DS |
1129 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1130 | root->root_key.objectid, &disk_key, level, | |
1131 | search_start, empty_size); | |
54aa1f4d CM |
1132 | if (IS_ERR(cow)) |
1133 | return PTR_ERR(cow); | |
6702ed49 | 1134 | |
b4ce94de CM |
1135 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1136 | ||
58e8012c | 1137 | copy_extent_buffer_full(cow, buf); |
db94535d | 1138 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1139 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1140 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1141 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1142 | BTRFS_HEADER_FLAG_RELOC); | |
1143 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1144 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1145 | else | |
1146 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1147 | |
0b246afa | 1148 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1149 | |
be1a5564 | 1150 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1151 | if (ret) { |
66642832 | 1152 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1153 | return ret; |
1154 | } | |
1a40e23b | 1155 | |
27cdeb70 | 1156 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1157 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1158 | if (ret) { |
66642832 | 1159 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1160 | return ret; |
93314e3b | 1161 | } |
83d4cfd4 | 1162 | } |
3fd0a558 | 1163 | |
02217ed2 | 1164 | if (buf == root->node) { |
925baedd | 1165 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1166 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1167 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1168 | parent_start = buf->start; | |
925baedd | 1169 | |
5f39d397 | 1170 | extent_buffer_get(cow); |
90f8d62e | 1171 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1172 | rcu_assign_pointer(root->node, cow); |
925baedd | 1173 | |
f0486c68 | 1174 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1175 | last_ref); |
5f39d397 | 1176 | free_extent_buffer(buf); |
0b86a832 | 1177 | add_root_to_dirty_list(root); |
02217ed2 | 1178 | } else { |
5d4f98a2 | 1179 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
0b246afa | 1180 | tree_mod_log_insert_key(fs_info, parent, parent_slot, |
c8cc6341 | 1181 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1182 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1183 | cow->start); |
74493f7a CM |
1184 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1185 | trans->transid); | |
d6025579 | 1186 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1187 | if (last_ref) { |
0b246afa | 1188 | ret = tree_mod_log_free_eb(fs_info, buf); |
5de865ee | 1189 | if (ret) { |
66642832 | 1190 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1191 | return ret; |
1192 | } | |
1193 | } | |
f0486c68 | 1194 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1195 | last_ref); |
02217ed2 | 1196 | } |
925baedd CM |
1197 | if (unlock_orig) |
1198 | btrfs_tree_unlock(buf); | |
3083ee2e | 1199 | free_extent_buffer_stale(buf); |
ccd467d6 | 1200 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1201 | *cow_ret = cow; |
02217ed2 CM |
1202 | return 0; |
1203 | } | |
1204 | ||
5d9e75c4 JS |
1205 | /* |
1206 | * returns the logical address of the oldest predecessor of the given root. | |
1207 | * entries older than time_seq are ignored. | |
1208 | */ | |
1209 | static struct tree_mod_elem * | |
1210 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1211 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1212 | { |
1213 | struct tree_mod_elem *tm; | |
1214 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1215 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1216 | int looped = 0; |
1217 | ||
1218 | if (!time_seq) | |
35a3621b | 1219 | return NULL; |
5d9e75c4 JS |
1220 | |
1221 | /* | |
298cfd36 CR |
1222 | * the very last operation that's logged for a root is the |
1223 | * replacement operation (if it is replaced at all). this has | |
1224 | * the logical address of the *new* root, making it the very | |
1225 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1226 | */ |
1227 | while (1) { | |
1228 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1229 | time_seq); | |
1230 | if (!looped && !tm) | |
35a3621b | 1231 | return NULL; |
5d9e75c4 | 1232 | /* |
28da9fb4 JS |
1233 | * if there are no tree operation for the oldest root, we simply |
1234 | * return it. this should only happen if that (old) root is at | |
1235 | * level 0. | |
5d9e75c4 | 1236 | */ |
28da9fb4 JS |
1237 | if (!tm) |
1238 | break; | |
5d9e75c4 | 1239 | |
28da9fb4 JS |
1240 | /* |
1241 | * if there's an operation that's not a root replacement, we | |
1242 | * found the oldest version of our root. normally, we'll find a | |
1243 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1244 | */ | |
5d9e75c4 JS |
1245 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1246 | break; | |
1247 | ||
1248 | found = tm; | |
1249 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1250 | looped = 1; |
1251 | } | |
1252 | ||
a95236d9 JS |
1253 | /* if there's no old root to return, return what we found instead */ |
1254 | if (!found) | |
1255 | found = tm; | |
1256 | ||
5d9e75c4 JS |
1257 | return found; |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1262 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1263 | * time_seq). |
1264 | */ | |
1265 | static void | |
f1ca7e98 JB |
1266 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1267 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1268 | { |
1269 | u32 n; | |
1270 | struct rb_node *next; | |
1271 | struct tree_mod_elem *tm = first_tm; | |
1272 | unsigned long o_dst; | |
1273 | unsigned long o_src; | |
1274 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1275 | ||
1276 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1277 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1278 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1279 | /* |
1280 | * all the operations are recorded with the operator used for | |
1281 | * the modification. as we're going backwards, we do the | |
1282 | * opposite of each operation here. | |
1283 | */ | |
1284 | switch (tm->op) { | |
1285 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1286 | BUG_ON(tm->slot < n); | |
1c697d4a | 1287 | /* Fallthrough */ |
95c80bb1 | 1288 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1289 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1290 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1291 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1292 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1293 | tm->generation); | |
4c3e6969 | 1294 | n++; |
5d9e75c4 JS |
1295 | break; |
1296 | case MOD_LOG_KEY_REPLACE: | |
1297 | BUG_ON(tm->slot >= n); | |
1298 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1299 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1300 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1301 | tm->generation); | |
1302 | break; | |
1303 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1304 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1305 | n--; |
1306 | break; | |
1307 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1308 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1309 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1310 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1311 | tm->move.nr_items * p_size); |
1312 | break; | |
1313 | case MOD_LOG_ROOT_REPLACE: | |
1314 | /* | |
1315 | * this operation is special. for roots, this must be | |
1316 | * handled explicitly before rewinding. | |
1317 | * for non-roots, this operation may exist if the node | |
1318 | * was a root: root A -> child B; then A gets empty and | |
1319 | * B is promoted to the new root. in the mod log, we'll | |
1320 | * have a root-replace operation for B, a tree block | |
1321 | * that is no root. we simply ignore that operation. | |
1322 | */ | |
1323 | break; | |
1324 | } | |
1325 | next = rb_next(&tm->node); | |
1326 | if (!next) | |
1327 | break; | |
1328 | tm = container_of(next, struct tree_mod_elem, node); | |
298cfd36 | 1329 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1330 | break; |
1331 | } | |
f1ca7e98 | 1332 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1333 | btrfs_set_header_nritems(eb, n); |
1334 | } | |
1335 | ||
47fb091f | 1336 | /* |
01327610 | 1337 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1338 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1339 | * returned buffer is always read-locked. If the returned buffer is not the | |
1340 | * input buffer, the lock on the input buffer is released and the input buffer | |
1341 | * is freed (its refcount is decremented). | |
1342 | */ | |
5d9e75c4 | 1343 | static struct extent_buffer * |
9ec72677 JB |
1344 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1345 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1346 | { |
1347 | struct extent_buffer *eb_rewin; | |
1348 | struct tree_mod_elem *tm; | |
1349 | ||
1350 | if (!time_seq) | |
1351 | return eb; | |
1352 | ||
1353 | if (btrfs_header_level(eb) == 0) | |
1354 | return eb; | |
1355 | ||
1356 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1357 | if (!tm) | |
1358 | return eb; | |
1359 | ||
9ec72677 JB |
1360 | btrfs_set_path_blocking(path); |
1361 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1362 | ||
5d9e75c4 JS |
1363 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1364 | BUG_ON(tm->slot != 0); | |
da17066c | 1365 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1366 | if (!eb_rewin) { |
9ec72677 | 1367 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1368 | free_extent_buffer(eb); |
1369 | return NULL; | |
1370 | } | |
5d9e75c4 JS |
1371 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1372 | btrfs_set_header_backref_rev(eb_rewin, | |
1373 | btrfs_header_backref_rev(eb)); | |
1374 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1375 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1376 | } else { |
1377 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1378 | if (!eb_rewin) { |
9ec72677 | 1379 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1380 | free_extent_buffer(eb); |
1381 | return NULL; | |
1382 | } | |
5d9e75c4 JS |
1383 | } |
1384 | ||
9ec72677 JB |
1385 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1386 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1387 | free_extent_buffer(eb); |
1388 | ||
47fb091f JS |
1389 | extent_buffer_get(eb_rewin); |
1390 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1391 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1392 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1393 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1394 | |
1395 | return eb_rewin; | |
1396 | } | |
1397 | ||
8ba97a15 JS |
1398 | /* |
1399 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1400 | * value. If there are no changes, the current root->root_node is returned. If | |
1401 | * anything changed in between, there's a fresh buffer allocated on which the | |
1402 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1403 | * Returns NULL on error (with no locks held). | |
1404 | */ | |
5d9e75c4 JS |
1405 | static inline struct extent_buffer * |
1406 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1407 | { | |
0b246afa | 1408 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1409 | struct tree_mod_elem *tm; |
30b0463a JS |
1410 | struct extent_buffer *eb = NULL; |
1411 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1412 | struct extent_buffer *old; |
a95236d9 | 1413 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1414 | u64 old_generation = 0; |
a95236d9 | 1415 | u64 logical; |
5d9e75c4 | 1416 | |
30b0463a | 1417 | eb_root = btrfs_read_lock_root_node(root); |
0b246afa | 1418 | tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq); |
5d9e75c4 | 1419 | if (!tm) |
30b0463a | 1420 | return eb_root; |
5d9e75c4 | 1421 | |
a95236d9 JS |
1422 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1423 | old_root = &tm->old_root; | |
1424 | old_generation = tm->generation; | |
1425 | logical = old_root->logical; | |
1426 | } else { | |
30b0463a | 1427 | logical = eb_root->start; |
a95236d9 | 1428 | } |
5d9e75c4 | 1429 | |
0b246afa | 1430 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1431 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1432 | btrfs_tree_read_unlock(eb_root); |
1433 | free_extent_buffer(eb_root); | |
ce86cd59 | 1434 | old = read_tree_block(root, logical, 0); |
64c043de LB |
1435 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1436 | if (!IS_ERR(old)) | |
1437 | free_extent_buffer(old); | |
0b246afa JM |
1438 | btrfs_warn(fs_info, |
1439 | "failed to read tree block %llu from get_old_root", | |
1440 | logical); | |
834328a8 | 1441 | } else { |
7bfdcf7f LB |
1442 | eb = btrfs_clone_extent_buffer(old); |
1443 | free_extent_buffer(old); | |
834328a8 JS |
1444 | } |
1445 | } else if (old_root) { | |
30b0463a JS |
1446 | btrfs_tree_read_unlock(eb_root); |
1447 | free_extent_buffer(eb_root); | |
0b246afa | 1448 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1449 | } else { |
9ec72677 | 1450 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1451 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1452 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1453 | free_extent_buffer(eb_root); |
834328a8 JS |
1454 | } |
1455 | ||
8ba97a15 JS |
1456 | if (!eb) |
1457 | return NULL; | |
d6381084 | 1458 | extent_buffer_get(eb); |
8ba97a15 | 1459 | btrfs_tree_read_lock(eb); |
a95236d9 | 1460 | if (old_root) { |
5d9e75c4 JS |
1461 | btrfs_set_header_bytenr(eb, eb->start); |
1462 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1463 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1464 | btrfs_set_header_level(eb, old_root->level); |
1465 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1466 | } |
28da9fb4 | 1467 | if (tm) |
0b246afa | 1468 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1469 | else |
1470 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1471 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1472 | |
1473 | return eb; | |
1474 | } | |
1475 | ||
5b6602e7 JS |
1476 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1477 | { | |
1478 | struct tree_mod_elem *tm; | |
1479 | int level; | |
30b0463a | 1480 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1481 | |
30b0463a | 1482 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1483 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1484 | level = tm->old_root.level; | |
1485 | } else { | |
30b0463a | 1486 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1487 | } |
30b0463a | 1488 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1489 | |
1490 | return level; | |
1491 | } | |
1492 | ||
5d4f98a2 YZ |
1493 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1494 | struct btrfs_root *root, | |
1495 | struct extent_buffer *buf) | |
1496 | { | |
f5ee5c9a | 1497 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1498 | return 0; |
fccb84c9 | 1499 | |
f1ebcc74 LB |
1500 | /* ensure we can see the force_cow */ |
1501 | smp_rmb(); | |
1502 | ||
1503 | /* | |
1504 | * We do not need to cow a block if | |
1505 | * 1) this block is not created or changed in this transaction; | |
1506 | * 2) this block does not belong to TREE_RELOC tree; | |
1507 | * 3) the root is not forced COW. | |
1508 | * | |
1509 | * What is forced COW: | |
01327610 | 1510 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1511 | * after we've finished coping src root, we must COW the shared |
1512 | * block to ensure the metadata consistency. | |
1513 | */ | |
5d4f98a2 YZ |
1514 | if (btrfs_header_generation(buf) == trans->transid && |
1515 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1516 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1517 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1518 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1519 | return 0; |
1520 | return 1; | |
1521 | } | |
1522 | ||
d352ac68 CM |
1523 | /* |
1524 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1525 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1526 | * once per transaction, as long as it hasn't been written yet |
1527 | */ | |
d397712b | 1528 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1529 | struct btrfs_root *root, struct extent_buffer *buf, |
1530 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1531 | struct extent_buffer **cow_ret) |
6702ed49 | 1532 | { |
0b246afa | 1533 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1534 | u64 search_start; |
f510cfec | 1535 | int ret; |
dc17ff8f | 1536 | |
0b246afa | 1537 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1538 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1539 | trans->transid, |
0b246afa | 1540 | fs_info->running_transaction->transid); |
31b1a2bd | 1541 | |
0b246afa | 1542 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1543 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1544 | trans->transid, fs_info->generation); |
dc17ff8f | 1545 | |
5d4f98a2 | 1546 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1547 | trans->dirty = true; |
6702ed49 CM |
1548 | *cow_ret = buf; |
1549 | return 0; | |
1550 | } | |
c487685d | 1551 | |
ee22184b | 1552 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1553 | |
1554 | if (parent) | |
1555 | btrfs_set_lock_blocking(parent); | |
1556 | btrfs_set_lock_blocking(buf); | |
1557 | ||
f510cfec | 1558 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1559 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1560 | |
1561 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1562 | ||
f510cfec | 1563 | return ret; |
6702ed49 CM |
1564 | } |
1565 | ||
d352ac68 CM |
1566 | /* |
1567 | * helper function for defrag to decide if two blocks pointed to by a | |
1568 | * node are actually close by | |
1569 | */ | |
6b80053d | 1570 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1571 | { |
6b80053d | 1572 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1573 | return 1; |
6b80053d | 1574 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1575 | return 1; |
1576 | return 0; | |
1577 | } | |
1578 | ||
081e9573 CM |
1579 | /* |
1580 | * compare two keys in a memcmp fashion | |
1581 | */ | |
1582 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1583 | { | |
1584 | struct btrfs_key k1; | |
1585 | ||
1586 | btrfs_disk_key_to_cpu(&k1, disk); | |
1587 | ||
20736aba | 1588 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1589 | } |
1590 | ||
f3465ca4 JB |
1591 | /* |
1592 | * same as comp_keys only with two btrfs_key's | |
1593 | */ | |
5d4f98a2 | 1594 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1595 | { |
1596 | if (k1->objectid > k2->objectid) | |
1597 | return 1; | |
1598 | if (k1->objectid < k2->objectid) | |
1599 | return -1; | |
1600 | if (k1->type > k2->type) | |
1601 | return 1; | |
1602 | if (k1->type < k2->type) | |
1603 | return -1; | |
1604 | if (k1->offset > k2->offset) | |
1605 | return 1; | |
1606 | if (k1->offset < k2->offset) | |
1607 | return -1; | |
1608 | return 0; | |
1609 | } | |
081e9573 | 1610 | |
d352ac68 CM |
1611 | /* |
1612 | * this is used by the defrag code to go through all the | |
1613 | * leaves pointed to by a node and reallocate them so that | |
1614 | * disk order is close to key order | |
1615 | */ | |
6702ed49 | 1616 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1617 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1618 | int start_slot, u64 *last_ret, |
a6b6e75e | 1619 | struct btrfs_key *progress) |
6702ed49 | 1620 | { |
0b246afa | 1621 | struct btrfs_fs_info *fs_info = root->fs_info; |
6b80053d | 1622 | struct extent_buffer *cur; |
6702ed49 | 1623 | u64 blocknr; |
ca7a79ad | 1624 | u64 gen; |
e9d0b13b CM |
1625 | u64 search_start = *last_ret; |
1626 | u64 last_block = 0; | |
6702ed49 CM |
1627 | u64 other; |
1628 | u32 parent_nritems; | |
6702ed49 CM |
1629 | int end_slot; |
1630 | int i; | |
1631 | int err = 0; | |
f2183bde | 1632 | int parent_level; |
6b80053d CM |
1633 | int uptodate; |
1634 | u32 blocksize; | |
081e9573 CM |
1635 | int progress_passed = 0; |
1636 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1637 | |
5708b959 | 1638 | parent_level = btrfs_header_level(parent); |
5708b959 | 1639 | |
0b246afa JM |
1640 | WARN_ON(trans->transaction != fs_info->running_transaction); |
1641 | WARN_ON(trans->transid != fs_info->generation); | |
86479a04 | 1642 | |
6b80053d | 1643 | parent_nritems = btrfs_header_nritems(parent); |
0b246afa | 1644 | blocksize = fs_info->nodesize; |
5dfe2be7 | 1645 | end_slot = parent_nritems - 1; |
6702ed49 | 1646 | |
5dfe2be7 | 1647 | if (parent_nritems <= 1) |
6702ed49 CM |
1648 | return 0; |
1649 | ||
b4ce94de CM |
1650 | btrfs_set_lock_blocking(parent); |
1651 | ||
5dfe2be7 | 1652 | for (i = start_slot; i <= end_slot; i++) { |
6702ed49 | 1653 | int close = 1; |
a6b6e75e | 1654 | |
081e9573 CM |
1655 | btrfs_node_key(parent, &disk_key, i); |
1656 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1657 | continue; | |
1658 | ||
1659 | progress_passed = 1; | |
6b80053d | 1660 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1661 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1662 | if (last_block == 0) |
1663 | last_block = blocknr; | |
5708b959 | 1664 | |
6702ed49 | 1665 | if (i > 0) { |
6b80053d CM |
1666 | other = btrfs_node_blockptr(parent, i - 1); |
1667 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1668 | } |
5dfe2be7 | 1669 | if (!close && i < end_slot) { |
6b80053d CM |
1670 | other = btrfs_node_blockptr(parent, i + 1); |
1671 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1672 | } |
e9d0b13b CM |
1673 | if (close) { |
1674 | last_block = blocknr; | |
6702ed49 | 1675 | continue; |
e9d0b13b | 1676 | } |
6702ed49 | 1677 | |
0b246afa | 1678 | cur = find_extent_buffer(fs_info, blocknr); |
6b80053d | 1679 | if (cur) |
b9fab919 | 1680 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1681 | else |
1682 | uptodate = 0; | |
5708b959 | 1683 | if (!cur || !uptodate) { |
6b80053d | 1684 | if (!cur) { |
ce86cd59 | 1685 | cur = read_tree_block(root, blocknr, gen); |
64c043de LB |
1686 | if (IS_ERR(cur)) { |
1687 | return PTR_ERR(cur); | |
1688 | } else if (!extent_buffer_uptodate(cur)) { | |
416bc658 | 1689 | free_extent_buffer(cur); |
97d9a8a4 | 1690 | return -EIO; |
416bc658 | 1691 | } |
6b80053d | 1692 | } else if (!uptodate) { |
018642a1 TI |
1693 | err = btrfs_read_buffer(cur, gen); |
1694 | if (err) { | |
1695 | free_extent_buffer(cur); | |
1696 | return err; | |
1697 | } | |
f2183bde | 1698 | } |
6702ed49 | 1699 | } |
e9d0b13b | 1700 | if (search_start == 0) |
6b80053d | 1701 | search_start = last_block; |
e9d0b13b | 1702 | |
e7a84565 | 1703 | btrfs_tree_lock(cur); |
b4ce94de | 1704 | btrfs_set_lock_blocking(cur); |
6b80053d | 1705 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1706 | &cur, search_start, |
6b80053d | 1707 | min(16 * blocksize, |
9fa8cfe7 | 1708 | (end_slot - i) * blocksize)); |
252c38f0 | 1709 | if (err) { |
e7a84565 | 1710 | btrfs_tree_unlock(cur); |
6b80053d | 1711 | free_extent_buffer(cur); |
6702ed49 | 1712 | break; |
252c38f0 | 1713 | } |
e7a84565 CM |
1714 | search_start = cur->start; |
1715 | last_block = cur->start; | |
f2183bde | 1716 | *last_ret = search_start; |
e7a84565 CM |
1717 | btrfs_tree_unlock(cur); |
1718 | free_extent_buffer(cur); | |
6702ed49 CM |
1719 | } |
1720 | return err; | |
1721 | } | |
1722 | ||
74123bd7 | 1723 | /* |
5f39d397 CM |
1724 | * search for key in the extent_buffer. The items start at offset p, |
1725 | * and they are item_size apart. There are 'max' items in p. | |
1726 | * | |
74123bd7 CM |
1727 | * the slot in the array is returned via slot, and it points to |
1728 | * the place where you would insert key if it is not found in | |
1729 | * the array. | |
1730 | * | |
1731 | * slot may point to max if the key is bigger than all of the keys | |
1732 | */ | |
e02119d5 CM |
1733 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1734 | unsigned long p, | |
1735 | int item_size, struct btrfs_key *key, | |
1736 | int max, int *slot) | |
be0e5c09 CM |
1737 | { |
1738 | int low = 0; | |
1739 | int high = max; | |
1740 | int mid; | |
1741 | int ret; | |
479965d6 | 1742 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1743 | struct btrfs_disk_key unaligned; |
1744 | unsigned long offset; | |
5f39d397 CM |
1745 | char *kaddr = NULL; |
1746 | unsigned long map_start = 0; | |
1747 | unsigned long map_len = 0; | |
479965d6 | 1748 | int err; |
be0e5c09 | 1749 | |
5e24e9af LB |
1750 | if (low > high) { |
1751 | btrfs_err(eb->fs_info, | |
1752 | "%s: low (%d) > high (%d) eb %llu owner %llu level %d", | |
1753 | __func__, low, high, eb->start, | |
1754 | btrfs_header_owner(eb), btrfs_header_level(eb)); | |
1755 | return -EINVAL; | |
1756 | } | |
1757 | ||
d397712b | 1758 | while (low < high) { |
be0e5c09 | 1759 | mid = (low + high) / 2; |
5f39d397 CM |
1760 | offset = p + mid * item_size; |
1761 | ||
a6591715 | 1762 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1763 | (offset + sizeof(struct btrfs_disk_key)) > |
1764 | map_start + map_len) { | |
934d375b CM |
1765 | |
1766 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1767 | sizeof(struct btrfs_disk_key), |
a6591715 | 1768 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1769 | |
1770 | if (!err) { | |
1771 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1772 | map_start); | |
415b35a5 | 1773 | } else if (err == 1) { |
479965d6 CM |
1774 | read_extent_buffer(eb, &unaligned, |
1775 | offset, sizeof(unaligned)); | |
1776 | tmp = &unaligned; | |
415b35a5 LB |
1777 | } else { |
1778 | return err; | |
479965d6 | 1779 | } |
5f39d397 | 1780 | |
5f39d397 CM |
1781 | } else { |
1782 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1783 | map_start); | |
1784 | } | |
be0e5c09 CM |
1785 | ret = comp_keys(tmp, key); |
1786 | ||
1787 | if (ret < 0) | |
1788 | low = mid + 1; | |
1789 | else if (ret > 0) | |
1790 | high = mid; | |
1791 | else { | |
1792 | *slot = mid; | |
1793 | return 0; | |
1794 | } | |
1795 | } | |
1796 | *slot = low; | |
1797 | return 1; | |
1798 | } | |
1799 | ||
97571fd0 CM |
1800 | /* |
1801 | * simple bin_search frontend that does the right thing for | |
1802 | * leaves vs nodes | |
1803 | */ | |
5f39d397 CM |
1804 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1805 | int level, int *slot) | |
be0e5c09 | 1806 | { |
f775738f | 1807 | if (level == 0) |
5f39d397 CM |
1808 | return generic_bin_search(eb, |
1809 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1810 | sizeof(struct btrfs_item), |
5f39d397 | 1811 | key, btrfs_header_nritems(eb), |
7518a238 | 1812 | slot); |
f775738f | 1813 | else |
5f39d397 CM |
1814 | return generic_bin_search(eb, |
1815 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1816 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1817 | key, btrfs_header_nritems(eb), |
7518a238 | 1818 | slot); |
be0e5c09 CM |
1819 | } |
1820 | ||
5d4f98a2 YZ |
1821 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1822 | int level, int *slot) | |
1823 | { | |
1824 | return bin_search(eb, key, level, slot); | |
1825 | } | |
1826 | ||
f0486c68 YZ |
1827 | static void root_add_used(struct btrfs_root *root, u32 size) |
1828 | { | |
1829 | spin_lock(&root->accounting_lock); | |
1830 | btrfs_set_root_used(&root->root_item, | |
1831 | btrfs_root_used(&root->root_item) + size); | |
1832 | spin_unlock(&root->accounting_lock); | |
1833 | } | |
1834 | ||
1835 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1836 | { | |
1837 | spin_lock(&root->accounting_lock); | |
1838 | btrfs_set_root_used(&root->root_item, | |
1839 | btrfs_root_used(&root->root_item) - size); | |
1840 | spin_unlock(&root->accounting_lock); | |
1841 | } | |
1842 | ||
d352ac68 CM |
1843 | /* given a node and slot number, this reads the blocks it points to. The |
1844 | * extent buffer is returned with a reference taken (but unlocked). | |
d352ac68 | 1845 | */ |
e02119d5 | 1846 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1847 | struct extent_buffer *parent, int slot) |
bb803951 | 1848 | { |
ca7a79ad | 1849 | int level = btrfs_header_level(parent); |
416bc658 JB |
1850 | struct extent_buffer *eb; |
1851 | ||
fb770ae4 LB |
1852 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1853 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1854 | |
1855 | BUG_ON(level == 0); | |
1856 | ||
416bc658 | 1857 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
416bc658 | 1858 | btrfs_node_ptr_generation(parent, slot)); |
fb770ae4 LB |
1859 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1860 | free_extent_buffer(eb); | |
1861 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1862 | } |
1863 | ||
1864 | return eb; | |
bb803951 CM |
1865 | } |
1866 | ||
d352ac68 CM |
1867 | /* |
1868 | * node level balancing, used to make sure nodes are in proper order for | |
1869 | * item deletion. We balance from the top down, so we have to make sure | |
1870 | * that a deletion won't leave an node completely empty later on. | |
1871 | */ | |
e02119d5 | 1872 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1873 | struct btrfs_root *root, |
1874 | struct btrfs_path *path, int level) | |
bb803951 | 1875 | { |
0b246afa | 1876 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1877 | struct extent_buffer *right = NULL; |
1878 | struct extent_buffer *mid; | |
1879 | struct extent_buffer *left = NULL; | |
1880 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1881 | int ret = 0; |
1882 | int wret; | |
1883 | int pslot; | |
bb803951 | 1884 | int orig_slot = path->slots[level]; |
79f95c82 | 1885 | u64 orig_ptr; |
bb803951 CM |
1886 | |
1887 | if (level == 0) | |
1888 | return 0; | |
1889 | ||
5f39d397 | 1890 | mid = path->nodes[level]; |
b4ce94de | 1891 | |
bd681513 CM |
1892 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1893 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1894 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1895 | ||
1d4f8a0c | 1896 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1897 | |
a05a9bb1 | 1898 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1899 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1900 | pslot = path->slots[level + 1]; |
1901 | } | |
bb803951 | 1902 | |
40689478 CM |
1903 | /* |
1904 | * deal with the case where there is only one pointer in the root | |
1905 | * by promoting the node below to a root | |
1906 | */ | |
5f39d397 CM |
1907 | if (!parent) { |
1908 | struct extent_buffer *child; | |
bb803951 | 1909 | |
5f39d397 | 1910 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1911 | return 0; |
1912 | ||
1913 | /* promote the child to a root */ | |
5f39d397 | 1914 | child = read_node_slot(root, mid, 0); |
fb770ae4 LB |
1915 | if (IS_ERR(child)) { |
1916 | ret = PTR_ERR(child); | |
0b246afa | 1917 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1918 | goto enospc; |
1919 | } | |
1920 | ||
925baedd | 1921 | btrfs_tree_lock(child); |
b4ce94de | 1922 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1923 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1924 | if (ret) { |
1925 | btrfs_tree_unlock(child); | |
1926 | free_extent_buffer(child); | |
1927 | goto enospc; | |
1928 | } | |
2f375ab9 | 1929 | |
90f8d62e | 1930 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1931 | rcu_assign_pointer(root->node, child); |
925baedd | 1932 | |
0b86a832 | 1933 | add_root_to_dirty_list(root); |
925baedd | 1934 | btrfs_tree_unlock(child); |
b4ce94de | 1935 | |
925baedd | 1936 | path->locks[level] = 0; |
bb803951 | 1937 | path->nodes[level] = NULL; |
0b246afa | 1938 | clean_tree_block(trans, fs_info, mid); |
925baedd | 1939 | btrfs_tree_unlock(mid); |
bb803951 | 1940 | /* once for the path */ |
5f39d397 | 1941 | free_extent_buffer(mid); |
f0486c68 YZ |
1942 | |
1943 | root_sub_used(root, mid->len); | |
5581a51a | 1944 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1945 | /* once for the root ptr */ |
3083ee2e | 1946 | free_extent_buffer_stale(mid); |
f0486c68 | 1947 | return 0; |
bb803951 | 1948 | } |
5f39d397 | 1949 | if (btrfs_header_nritems(mid) > |
0b246afa | 1950 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1951 | return 0; |
1952 | ||
5f39d397 | 1953 | left = read_node_slot(root, parent, pslot - 1); |
fb770ae4 LB |
1954 | if (IS_ERR(left)) |
1955 | left = NULL; | |
1956 | ||
5f39d397 | 1957 | if (left) { |
925baedd | 1958 | btrfs_tree_lock(left); |
b4ce94de | 1959 | btrfs_set_lock_blocking(left); |
5f39d397 | 1960 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1961 | parent, pslot - 1, &left); |
54aa1f4d CM |
1962 | if (wret) { |
1963 | ret = wret; | |
1964 | goto enospc; | |
1965 | } | |
2cc58cf2 | 1966 | } |
fb770ae4 | 1967 | |
5f39d397 | 1968 | right = read_node_slot(root, parent, pslot + 1); |
fb770ae4 LB |
1969 | if (IS_ERR(right)) |
1970 | right = NULL; | |
1971 | ||
5f39d397 | 1972 | if (right) { |
925baedd | 1973 | btrfs_tree_lock(right); |
b4ce94de | 1974 | btrfs_set_lock_blocking(right); |
5f39d397 | 1975 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1976 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1977 | if (wret) { |
1978 | ret = wret; | |
1979 | goto enospc; | |
1980 | } | |
1981 | } | |
1982 | ||
1983 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1984 | if (left) { |
1985 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1986 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1987 | if (wret < 0) |
1988 | ret = wret; | |
bb803951 | 1989 | } |
79f95c82 CM |
1990 | |
1991 | /* | |
1992 | * then try to empty the right most buffer into the middle | |
1993 | */ | |
5f39d397 | 1994 | if (right) { |
971a1f66 | 1995 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1996 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1997 | ret = wret; |
5f39d397 | 1998 | if (btrfs_header_nritems(right) == 0) { |
0b246afa | 1999 | clean_tree_block(trans, fs_info, right); |
925baedd | 2000 | btrfs_tree_unlock(right); |
afe5fea7 | 2001 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 2002 | root_sub_used(root, right->len); |
5581a51a | 2003 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2004 | free_extent_buffer_stale(right); |
f0486c68 | 2005 | right = NULL; |
bb803951 | 2006 | } else { |
5f39d397 CM |
2007 | struct btrfs_disk_key right_key; |
2008 | btrfs_node_key(right, &right_key, 0); | |
0b246afa | 2009 | tree_mod_log_set_node_key(fs_info, parent, |
32adf090 | 2010 | pslot + 1, 0); |
5f39d397 CM |
2011 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2012 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2013 | } |
2014 | } | |
5f39d397 | 2015 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2016 | /* |
2017 | * we're not allowed to leave a node with one item in the | |
2018 | * tree during a delete. A deletion from lower in the tree | |
2019 | * could try to delete the only pointer in this node. | |
2020 | * So, pull some keys from the left. | |
2021 | * There has to be a left pointer at this point because | |
2022 | * otherwise we would have pulled some pointers from the | |
2023 | * right | |
2024 | */ | |
305a26af MF |
2025 | if (!left) { |
2026 | ret = -EROFS; | |
0b246afa | 2027 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
2028 | goto enospc; |
2029 | } | |
5f39d397 | 2030 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 2031 | if (wret < 0) { |
79f95c82 | 2032 | ret = wret; |
54aa1f4d CM |
2033 | goto enospc; |
2034 | } | |
bce4eae9 CM |
2035 | if (wret == 1) { |
2036 | wret = push_node_left(trans, root, left, mid, 1); | |
2037 | if (wret < 0) | |
2038 | ret = wret; | |
2039 | } | |
79f95c82 CM |
2040 | BUG_ON(wret == 1); |
2041 | } | |
5f39d397 | 2042 | if (btrfs_header_nritems(mid) == 0) { |
0b246afa | 2043 | clean_tree_block(trans, fs_info, mid); |
925baedd | 2044 | btrfs_tree_unlock(mid); |
afe5fea7 | 2045 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2046 | root_sub_used(root, mid->len); |
5581a51a | 2047 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2048 | free_extent_buffer_stale(mid); |
f0486c68 | 2049 | mid = NULL; |
79f95c82 CM |
2050 | } else { |
2051 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2052 | struct btrfs_disk_key mid_key; |
2053 | btrfs_node_key(mid, &mid_key, 0); | |
0b246afa | 2054 | tree_mod_log_set_node_key(fs_info, parent, pslot, 0); |
5f39d397 CM |
2055 | btrfs_set_node_key(parent, &mid_key, pslot); |
2056 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2057 | } |
bb803951 | 2058 | |
79f95c82 | 2059 | /* update the path */ |
5f39d397 CM |
2060 | if (left) { |
2061 | if (btrfs_header_nritems(left) > orig_slot) { | |
2062 | extent_buffer_get(left); | |
925baedd | 2063 | /* left was locked after cow */ |
5f39d397 | 2064 | path->nodes[level] = left; |
bb803951 CM |
2065 | path->slots[level + 1] -= 1; |
2066 | path->slots[level] = orig_slot; | |
925baedd CM |
2067 | if (mid) { |
2068 | btrfs_tree_unlock(mid); | |
5f39d397 | 2069 | free_extent_buffer(mid); |
925baedd | 2070 | } |
bb803951 | 2071 | } else { |
5f39d397 | 2072 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2073 | path->slots[level] = orig_slot; |
2074 | } | |
2075 | } | |
79f95c82 | 2076 | /* double check we haven't messed things up */ |
e20d96d6 | 2077 | if (orig_ptr != |
5f39d397 | 2078 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2079 | BUG(); |
54aa1f4d | 2080 | enospc: |
925baedd CM |
2081 | if (right) { |
2082 | btrfs_tree_unlock(right); | |
5f39d397 | 2083 | free_extent_buffer(right); |
925baedd CM |
2084 | } |
2085 | if (left) { | |
2086 | if (path->nodes[level] != left) | |
2087 | btrfs_tree_unlock(left); | |
5f39d397 | 2088 | free_extent_buffer(left); |
925baedd | 2089 | } |
bb803951 CM |
2090 | return ret; |
2091 | } | |
2092 | ||
d352ac68 CM |
2093 | /* Node balancing for insertion. Here we only split or push nodes around |
2094 | * when they are completely full. This is also done top down, so we | |
2095 | * have to be pessimistic. | |
2096 | */ | |
d397712b | 2097 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2098 | struct btrfs_root *root, |
2099 | struct btrfs_path *path, int level) | |
e66f709b | 2100 | { |
0b246afa | 2101 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2102 | struct extent_buffer *right = NULL; |
2103 | struct extent_buffer *mid; | |
2104 | struct extent_buffer *left = NULL; | |
2105 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2106 | int ret = 0; |
2107 | int wret; | |
2108 | int pslot; | |
2109 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2110 | |
2111 | if (level == 0) | |
2112 | return 1; | |
2113 | ||
5f39d397 | 2114 | mid = path->nodes[level]; |
7bb86316 | 2115 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2116 | |
a05a9bb1 | 2117 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2118 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2119 | pslot = path->slots[level + 1]; |
2120 | } | |
e66f709b | 2121 | |
5f39d397 | 2122 | if (!parent) |
e66f709b | 2123 | return 1; |
e66f709b | 2124 | |
5f39d397 | 2125 | left = read_node_slot(root, parent, pslot - 1); |
fb770ae4 LB |
2126 | if (IS_ERR(left)) |
2127 | left = NULL; | |
e66f709b CM |
2128 | |
2129 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2130 | if (left) { |
e66f709b | 2131 | u32 left_nr; |
925baedd CM |
2132 | |
2133 | btrfs_tree_lock(left); | |
b4ce94de CM |
2134 | btrfs_set_lock_blocking(left); |
2135 | ||
5f39d397 | 2136 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2137 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2138 | wret = 1; |
2139 | } else { | |
5f39d397 | 2140 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2141 | pslot - 1, &left); |
54aa1f4d CM |
2142 | if (ret) |
2143 | wret = 1; | |
2144 | else { | |
54aa1f4d | 2145 | wret = push_node_left(trans, root, |
971a1f66 | 2146 | left, mid, 0); |
54aa1f4d | 2147 | } |
33ade1f8 | 2148 | } |
e66f709b CM |
2149 | if (wret < 0) |
2150 | ret = wret; | |
2151 | if (wret == 0) { | |
5f39d397 | 2152 | struct btrfs_disk_key disk_key; |
e66f709b | 2153 | orig_slot += left_nr; |
5f39d397 | 2154 | btrfs_node_key(mid, &disk_key, 0); |
0b246afa | 2155 | tree_mod_log_set_node_key(fs_info, parent, pslot, 0); |
5f39d397 CM |
2156 | btrfs_set_node_key(parent, &disk_key, pslot); |
2157 | btrfs_mark_buffer_dirty(parent); | |
2158 | if (btrfs_header_nritems(left) > orig_slot) { | |
2159 | path->nodes[level] = left; | |
e66f709b CM |
2160 | path->slots[level + 1] -= 1; |
2161 | path->slots[level] = orig_slot; | |
925baedd | 2162 | btrfs_tree_unlock(mid); |
5f39d397 | 2163 | free_extent_buffer(mid); |
e66f709b CM |
2164 | } else { |
2165 | orig_slot -= | |
5f39d397 | 2166 | btrfs_header_nritems(left); |
e66f709b | 2167 | path->slots[level] = orig_slot; |
925baedd | 2168 | btrfs_tree_unlock(left); |
5f39d397 | 2169 | free_extent_buffer(left); |
e66f709b | 2170 | } |
e66f709b CM |
2171 | return 0; |
2172 | } | |
925baedd | 2173 | btrfs_tree_unlock(left); |
5f39d397 | 2174 | free_extent_buffer(left); |
e66f709b | 2175 | } |
925baedd | 2176 | right = read_node_slot(root, parent, pslot + 1); |
fb770ae4 LB |
2177 | if (IS_ERR(right)) |
2178 | right = NULL; | |
e66f709b CM |
2179 | |
2180 | /* | |
2181 | * then try to empty the right most buffer into the middle | |
2182 | */ | |
5f39d397 | 2183 | if (right) { |
33ade1f8 | 2184 | u32 right_nr; |
b4ce94de | 2185 | |
925baedd | 2186 | btrfs_tree_lock(right); |
b4ce94de CM |
2187 | btrfs_set_lock_blocking(right); |
2188 | ||
5f39d397 | 2189 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2190 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2191 | wret = 1; |
2192 | } else { | |
5f39d397 CM |
2193 | ret = btrfs_cow_block(trans, root, right, |
2194 | parent, pslot + 1, | |
9fa8cfe7 | 2195 | &right); |
54aa1f4d CM |
2196 | if (ret) |
2197 | wret = 1; | |
2198 | else { | |
54aa1f4d | 2199 | wret = balance_node_right(trans, root, |
5f39d397 | 2200 | right, mid); |
54aa1f4d | 2201 | } |
33ade1f8 | 2202 | } |
e66f709b CM |
2203 | if (wret < 0) |
2204 | ret = wret; | |
2205 | if (wret == 0) { | |
5f39d397 CM |
2206 | struct btrfs_disk_key disk_key; |
2207 | ||
2208 | btrfs_node_key(right, &disk_key, 0); | |
0b246afa | 2209 | tree_mod_log_set_node_key(fs_info, parent, |
32adf090 | 2210 | pslot + 1, 0); |
5f39d397 CM |
2211 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2212 | btrfs_mark_buffer_dirty(parent); | |
2213 | ||
2214 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2215 | path->nodes[level] = right; | |
e66f709b CM |
2216 | path->slots[level + 1] += 1; |
2217 | path->slots[level] = orig_slot - | |
5f39d397 | 2218 | btrfs_header_nritems(mid); |
925baedd | 2219 | btrfs_tree_unlock(mid); |
5f39d397 | 2220 | free_extent_buffer(mid); |
e66f709b | 2221 | } else { |
925baedd | 2222 | btrfs_tree_unlock(right); |
5f39d397 | 2223 | free_extent_buffer(right); |
e66f709b | 2224 | } |
e66f709b CM |
2225 | return 0; |
2226 | } | |
925baedd | 2227 | btrfs_tree_unlock(right); |
5f39d397 | 2228 | free_extent_buffer(right); |
e66f709b | 2229 | } |
e66f709b CM |
2230 | return 1; |
2231 | } | |
2232 | ||
3c69faec | 2233 | /* |
d352ac68 CM |
2234 | * readahead one full node of leaves, finding things that are close |
2235 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2236 | */ |
c8c42864 CM |
2237 | static void reada_for_search(struct btrfs_root *root, |
2238 | struct btrfs_path *path, | |
2239 | int level, int slot, u64 objectid) | |
3c69faec | 2240 | { |
0b246afa | 2241 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2242 | struct extent_buffer *node; |
01f46658 | 2243 | struct btrfs_disk_key disk_key; |
3c69faec | 2244 | u32 nritems; |
3c69faec | 2245 | u64 search; |
a7175319 | 2246 | u64 target; |
6b80053d | 2247 | u64 nread = 0; |
5f39d397 | 2248 | struct extent_buffer *eb; |
6b80053d CM |
2249 | u32 nr; |
2250 | u32 blocksize; | |
2251 | u32 nscan = 0; | |
db94535d | 2252 | |
a6b6e75e | 2253 | if (level != 1) |
6702ed49 CM |
2254 | return; |
2255 | ||
2256 | if (!path->nodes[level]) | |
3c69faec CM |
2257 | return; |
2258 | ||
5f39d397 | 2259 | node = path->nodes[level]; |
925baedd | 2260 | |
3c69faec | 2261 | search = btrfs_node_blockptr(node, slot); |
0b246afa JM |
2262 | blocksize = fs_info->nodesize; |
2263 | eb = find_extent_buffer(fs_info, search); | |
5f39d397 CM |
2264 | if (eb) { |
2265 | free_extent_buffer(eb); | |
3c69faec CM |
2266 | return; |
2267 | } | |
2268 | ||
a7175319 | 2269 | target = search; |
6b80053d | 2270 | |
5f39d397 | 2271 | nritems = btrfs_header_nritems(node); |
6b80053d | 2272 | nr = slot; |
25b8b936 | 2273 | |
d397712b | 2274 | while (1) { |
e4058b54 | 2275 | if (path->reada == READA_BACK) { |
6b80053d CM |
2276 | if (nr == 0) |
2277 | break; | |
2278 | nr--; | |
e4058b54 | 2279 | } else if (path->reada == READA_FORWARD) { |
6b80053d CM |
2280 | nr++; |
2281 | if (nr >= nritems) | |
2282 | break; | |
3c69faec | 2283 | } |
e4058b54 | 2284 | if (path->reada == READA_BACK && objectid) { |
01f46658 CM |
2285 | btrfs_node_key(node, &disk_key, nr); |
2286 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2287 | break; | |
2288 | } | |
6b80053d | 2289 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2290 | if ((search <= target && target - search <= 65536) || |
2291 | (search > target && search - target <= 65536)) { | |
d3e46fea | 2292 | readahead_tree_block(root, search); |
6b80053d CM |
2293 | nread += blocksize; |
2294 | } | |
2295 | nscan++; | |
a7175319 | 2296 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2297 | break; |
3c69faec CM |
2298 | } |
2299 | } | |
925baedd | 2300 | |
0b08851f JB |
2301 | static noinline void reada_for_balance(struct btrfs_root *root, |
2302 | struct btrfs_path *path, int level) | |
b4ce94de | 2303 | { |
0b246afa | 2304 | struct btrfs_fs_info *fs_info = root->fs_info; |
b4ce94de CM |
2305 | int slot; |
2306 | int nritems; | |
2307 | struct extent_buffer *parent; | |
2308 | struct extent_buffer *eb; | |
2309 | u64 gen; | |
2310 | u64 block1 = 0; | |
2311 | u64 block2 = 0; | |
b4ce94de | 2312 | |
8c594ea8 | 2313 | parent = path->nodes[level + 1]; |
b4ce94de | 2314 | if (!parent) |
0b08851f | 2315 | return; |
b4ce94de CM |
2316 | |
2317 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2318 | slot = path->slots[level + 1]; |
b4ce94de CM |
2319 | |
2320 | if (slot > 0) { | |
2321 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2322 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2323 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2324 | /* |
2325 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2326 | * don't want to return eagain here. That will loop | |
2327 | * forever | |
2328 | */ | |
2329 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2330 | block1 = 0; |
2331 | free_extent_buffer(eb); | |
2332 | } | |
8c594ea8 | 2333 | if (slot + 1 < nritems) { |
b4ce94de CM |
2334 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2335 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2336 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2337 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2338 | block2 = 0; |
2339 | free_extent_buffer(eb); | |
2340 | } | |
8c594ea8 | 2341 | |
0b08851f | 2342 | if (block1) |
d3e46fea | 2343 | readahead_tree_block(root, block1); |
0b08851f | 2344 | if (block2) |
d3e46fea | 2345 | readahead_tree_block(root, block2); |
b4ce94de CM |
2346 | } |
2347 | ||
2348 | ||
d352ac68 | 2349 | /* |
d397712b CM |
2350 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2351 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2352 | * operations on the tree might require changing key pointers higher up in the | |
2353 | * tree. | |
d352ac68 | 2354 | * |
d397712b CM |
2355 | * callers might also have set path->keep_locks, which tells this code to keep |
2356 | * the lock if the path points to the last slot in the block. This is part of | |
2357 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2358 | * |
d397712b CM |
2359 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2360 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2361 | */ |
e02119d5 | 2362 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2363 | int lowest_unlock, int min_write_lock_level, |
2364 | int *write_lock_level) | |
925baedd CM |
2365 | { |
2366 | int i; | |
2367 | int skip_level = level; | |
051e1b9f | 2368 | int no_skips = 0; |
925baedd CM |
2369 | struct extent_buffer *t; |
2370 | ||
2371 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2372 | if (!path->nodes[i]) | |
2373 | break; | |
2374 | if (!path->locks[i]) | |
2375 | break; | |
051e1b9f | 2376 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2377 | skip_level = i + 1; |
2378 | continue; | |
2379 | } | |
051e1b9f | 2380 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2381 | u32 nritems; |
2382 | t = path->nodes[i]; | |
2383 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2384 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2385 | skip_level = i + 1; |
2386 | continue; | |
2387 | } | |
2388 | } | |
051e1b9f CM |
2389 | if (skip_level < i && i >= lowest_unlock) |
2390 | no_skips = 1; | |
2391 | ||
925baedd CM |
2392 | t = path->nodes[i]; |
2393 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2394 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2395 | path->locks[i] = 0; |
f7c79f30 CM |
2396 | if (write_lock_level && |
2397 | i > min_write_lock_level && | |
2398 | i <= *write_lock_level) { | |
2399 | *write_lock_level = i - 1; | |
2400 | } | |
925baedd CM |
2401 | } |
2402 | } | |
2403 | } | |
2404 | ||
b4ce94de CM |
2405 | /* |
2406 | * This releases any locks held in the path starting at level and | |
2407 | * going all the way up to the root. | |
2408 | * | |
2409 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2410 | * corner cases, such as COW of the block at slot zero in the node. This | |
2411 | * ignores those rules, and it should only be called when there are no | |
2412 | * more updates to be done higher up in the tree. | |
2413 | */ | |
2414 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2415 | { | |
2416 | int i; | |
2417 | ||
09a2a8f9 | 2418 | if (path->keep_locks) |
b4ce94de CM |
2419 | return; |
2420 | ||
2421 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2422 | if (!path->nodes[i]) | |
12f4dacc | 2423 | continue; |
b4ce94de | 2424 | if (!path->locks[i]) |
12f4dacc | 2425 | continue; |
bd681513 | 2426 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2427 | path->locks[i] = 0; |
2428 | } | |
2429 | } | |
2430 | ||
c8c42864 CM |
2431 | /* |
2432 | * helper function for btrfs_search_slot. The goal is to find a block | |
2433 | * in cache without setting the path to blocking. If we find the block | |
2434 | * we return zero and the path is unchanged. | |
2435 | * | |
2436 | * If we can't find the block, we set the path blocking and do some | |
2437 | * reada. -EAGAIN is returned and the search must be repeated. | |
2438 | */ | |
2439 | static int | |
2440 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2441 | struct btrfs_root *root, struct btrfs_path *p, | |
2442 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2443 | struct btrfs_key *key, u64 time_seq) |
c8c42864 | 2444 | { |
0b246afa | 2445 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2446 | u64 blocknr; |
2447 | u64 gen; | |
c8c42864 CM |
2448 | struct extent_buffer *b = *eb_ret; |
2449 | struct extent_buffer *tmp; | |
76a05b35 | 2450 | int ret; |
c8c42864 CM |
2451 | |
2452 | blocknr = btrfs_node_blockptr(b, slot); | |
2453 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2454 | |
0b246afa | 2455 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2456 | if (tmp) { |
b9fab919 | 2457 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2458 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2459 | *eb_ret = tmp; | |
2460 | return 0; | |
2461 | } | |
2462 | ||
2463 | /* the pages were up to date, but we failed | |
2464 | * the generation number check. Do a full | |
2465 | * read for the generation number that is correct. | |
2466 | * We must do this without dropping locks so | |
2467 | * we can trust our generation number | |
2468 | */ | |
2469 | btrfs_set_path_blocking(p); | |
2470 | ||
2471 | /* now we're allowed to do a blocking uptodate check */ | |
2472 | ret = btrfs_read_buffer(tmp, gen); | |
2473 | if (!ret) { | |
2474 | *eb_ret = tmp; | |
2475 | return 0; | |
cb44921a | 2476 | } |
bdf7c00e JB |
2477 | free_extent_buffer(tmp); |
2478 | btrfs_release_path(p); | |
2479 | return -EIO; | |
c8c42864 CM |
2480 | } |
2481 | ||
2482 | /* | |
2483 | * reduce lock contention at high levels | |
2484 | * of the btree by dropping locks before | |
76a05b35 CM |
2485 | * we read. Don't release the lock on the current |
2486 | * level because we need to walk this node to figure | |
2487 | * out which blocks to read. | |
c8c42864 | 2488 | */ |
8c594ea8 CM |
2489 | btrfs_unlock_up_safe(p, level + 1); |
2490 | btrfs_set_path_blocking(p); | |
2491 | ||
cb44921a | 2492 | free_extent_buffer(tmp); |
e4058b54 | 2493 | if (p->reada != READA_NONE) |
c8c42864 CM |
2494 | reada_for_search(root, p, level, slot, key->objectid); |
2495 | ||
b3b4aa74 | 2496 | btrfs_release_path(p); |
76a05b35 CM |
2497 | |
2498 | ret = -EAGAIN; | |
ce86cd59 | 2499 | tmp = read_tree_block(root, blocknr, 0); |
64c043de | 2500 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2501 | /* |
2502 | * If the read above didn't mark this buffer up to date, | |
2503 | * it will never end up being up to date. Set ret to EIO now | |
2504 | * and give up so that our caller doesn't loop forever | |
2505 | * on our EAGAINs. | |
2506 | */ | |
b9fab919 | 2507 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2508 | ret = -EIO; |
c8c42864 | 2509 | free_extent_buffer(tmp); |
c871b0f2 LB |
2510 | } else { |
2511 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2512 | } |
2513 | return ret; | |
c8c42864 CM |
2514 | } |
2515 | ||
2516 | /* | |
2517 | * helper function for btrfs_search_slot. This does all of the checks | |
2518 | * for node-level blocks and does any balancing required based on | |
2519 | * the ins_len. | |
2520 | * | |
2521 | * If no extra work was required, zero is returned. If we had to | |
2522 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2523 | * start over | |
2524 | */ | |
2525 | static int | |
2526 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2527 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2528 | struct extent_buffer *b, int level, int ins_len, |
2529 | int *write_lock_level) | |
c8c42864 | 2530 | { |
0b246afa | 2531 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2532 | int ret; |
0b246afa | 2533 | |
c8c42864 | 2534 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2535 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2536 | int sret; |
2537 | ||
bd681513 CM |
2538 | if (*write_lock_level < level + 1) { |
2539 | *write_lock_level = level + 1; | |
2540 | btrfs_release_path(p); | |
2541 | goto again; | |
2542 | } | |
2543 | ||
c8c42864 | 2544 | btrfs_set_path_blocking(p); |
0b08851f | 2545 | reada_for_balance(root, p, level); |
c8c42864 | 2546 | sret = split_node(trans, root, p, level); |
bd681513 | 2547 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2548 | |
2549 | BUG_ON(sret > 0); | |
2550 | if (sret) { | |
2551 | ret = sret; | |
2552 | goto done; | |
2553 | } | |
2554 | b = p->nodes[level]; | |
2555 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2556 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2557 | int sret; |
2558 | ||
bd681513 CM |
2559 | if (*write_lock_level < level + 1) { |
2560 | *write_lock_level = level + 1; | |
2561 | btrfs_release_path(p); | |
2562 | goto again; | |
2563 | } | |
2564 | ||
c8c42864 | 2565 | btrfs_set_path_blocking(p); |
0b08851f | 2566 | reada_for_balance(root, p, level); |
c8c42864 | 2567 | sret = balance_level(trans, root, p, level); |
bd681513 | 2568 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2569 | |
2570 | if (sret) { | |
2571 | ret = sret; | |
2572 | goto done; | |
2573 | } | |
2574 | b = p->nodes[level]; | |
2575 | if (!b) { | |
b3b4aa74 | 2576 | btrfs_release_path(p); |
c8c42864 CM |
2577 | goto again; |
2578 | } | |
2579 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2580 | } | |
2581 | return 0; | |
2582 | ||
2583 | again: | |
2584 | ret = -EAGAIN; | |
2585 | done: | |
2586 | return ret; | |
2587 | } | |
2588 | ||
d7396f07 FDBM |
2589 | static void key_search_validate(struct extent_buffer *b, |
2590 | struct btrfs_key *key, | |
2591 | int level) | |
2592 | { | |
2593 | #ifdef CONFIG_BTRFS_ASSERT | |
2594 | struct btrfs_disk_key disk_key; | |
2595 | ||
2596 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2597 | ||
2598 | if (level == 0) | |
2599 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2600 | offsetof(struct btrfs_leaf, items[0].key), | |
2601 | sizeof(disk_key))); | |
2602 | else | |
2603 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2604 | offsetof(struct btrfs_node, ptrs[0].key), | |
2605 | sizeof(disk_key))); | |
2606 | #endif | |
2607 | } | |
2608 | ||
2609 | static int key_search(struct extent_buffer *b, struct btrfs_key *key, | |
2610 | int level, int *prev_cmp, int *slot) | |
2611 | { | |
2612 | if (*prev_cmp != 0) { | |
2613 | *prev_cmp = bin_search(b, key, level, slot); | |
2614 | return *prev_cmp; | |
2615 | } | |
2616 | ||
2617 | key_search_validate(b, key, level); | |
2618 | *slot = 0; | |
2619 | ||
2620 | return 0; | |
2621 | } | |
2622 | ||
381cf658 | 2623 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2624 | u64 iobjectid, u64 ioff, u8 key_type, |
2625 | struct btrfs_key *found_key) | |
2626 | { | |
2627 | int ret; | |
2628 | struct btrfs_key key; | |
2629 | struct extent_buffer *eb; | |
381cf658 DS |
2630 | |
2631 | ASSERT(path); | |
1d4c08e0 | 2632 | ASSERT(found_key); |
e33d5c3d KN |
2633 | |
2634 | key.type = key_type; | |
2635 | key.objectid = iobjectid; | |
2636 | key.offset = ioff; | |
2637 | ||
2638 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2639 | if (ret < 0) |
e33d5c3d KN |
2640 | return ret; |
2641 | ||
2642 | eb = path->nodes[0]; | |
2643 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2644 | ret = btrfs_next_leaf(fs_root, path); | |
2645 | if (ret) | |
2646 | return ret; | |
2647 | eb = path->nodes[0]; | |
2648 | } | |
2649 | ||
2650 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2651 | if (found_key->type != key.type || | |
2652 | found_key->objectid != key.objectid) | |
2653 | return 1; | |
2654 | ||
2655 | return 0; | |
2656 | } | |
2657 | ||
74123bd7 CM |
2658 | /* |
2659 | * look for key in the tree. path is filled in with nodes along the way | |
2660 | * if key is found, we return zero and you can find the item in the leaf | |
2661 | * level of the path (level 0) | |
2662 | * | |
2663 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2664 | * be inserted, and 1 is returned. If there are other errors during the |
2665 | * search a negative error number is returned. | |
97571fd0 CM |
2666 | * |
2667 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2668 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2669 | * possible) | |
74123bd7 | 2670 | */ |
e089f05c CM |
2671 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2672 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2673 | ins_len, int cow) | |
be0e5c09 | 2674 | { |
0b246afa | 2675 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2676 | struct extent_buffer *b; |
be0e5c09 CM |
2677 | int slot; |
2678 | int ret; | |
33c66f43 | 2679 | int err; |
be0e5c09 | 2680 | int level; |
925baedd | 2681 | int lowest_unlock = 1; |
bd681513 CM |
2682 | int root_lock; |
2683 | /* everything at write_lock_level or lower must be write locked */ | |
2684 | int write_lock_level = 0; | |
9f3a7427 | 2685 | u8 lowest_level = 0; |
f7c79f30 | 2686 | int min_write_lock_level; |
d7396f07 | 2687 | int prev_cmp; |
9f3a7427 | 2688 | |
6702ed49 | 2689 | lowest_level = p->lowest_level; |
323ac95b | 2690 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2691 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2692 | BUG_ON(!cow && ins_len); |
25179201 | 2693 | |
bd681513 | 2694 | if (ins_len < 0) { |
925baedd | 2695 | lowest_unlock = 2; |
65b51a00 | 2696 | |
bd681513 CM |
2697 | /* when we are removing items, we might have to go up to level |
2698 | * two as we update tree pointers Make sure we keep write | |
2699 | * for those levels as well | |
2700 | */ | |
2701 | write_lock_level = 2; | |
2702 | } else if (ins_len > 0) { | |
2703 | /* | |
2704 | * for inserting items, make sure we have a write lock on | |
2705 | * level 1 so we can update keys | |
2706 | */ | |
2707 | write_lock_level = 1; | |
2708 | } | |
2709 | ||
2710 | if (!cow) | |
2711 | write_lock_level = -1; | |
2712 | ||
09a2a8f9 | 2713 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2714 | write_lock_level = BTRFS_MAX_LEVEL; |
2715 | ||
f7c79f30 CM |
2716 | min_write_lock_level = write_lock_level; |
2717 | ||
bb803951 | 2718 | again: |
d7396f07 | 2719 | prev_cmp = -1; |
bd681513 CM |
2720 | /* |
2721 | * we try very hard to do read locks on the root | |
2722 | */ | |
2723 | root_lock = BTRFS_READ_LOCK; | |
2724 | level = 0; | |
5d4f98a2 | 2725 | if (p->search_commit_root) { |
bd681513 CM |
2726 | /* |
2727 | * the commit roots are read only | |
2728 | * so we always do read locks | |
2729 | */ | |
3f8a18cc | 2730 | if (p->need_commit_sem) |
0b246afa | 2731 | down_read(&fs_info->commit_root_sem); |
5d4f98a2 YZ |
2732 | b = root->commit_root; |
2733 | extent_buffer_get(b); | |
bd681513 | 2734 | level = btrfs_header_level(b); |
3f8a18cc | 2735 | if (p->need_commit_sem) |
0b246afa | 2736 | up_read(&fs_info->commit_root_sem); |
5d4f98a2 | 2737 | if (!p->skip_locking) |
bd681513 | 2738 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2739 | } else { |
bd681513 | 2740 | if (p->skip_locking) { |
5d4f98a2 | 2741 | b = btrfs_root_node(root); |
bd681513 CM |
2742 | level = btrfs_header_level(b); |
2743 | } else { | |
2744 | /* we don't know the level of the root node | |
2745 | * until we actually have it read locked | |
2746 | */ | |
2747 | b = btrfs_read_lock_root_node(root); | |
2748 | level = btrfs_header_level(b); | |
2749 | if (level <= write_lock_level) { | |
2750 | /* whoops, must trade for write lock */ | |
2751 | btrfs_tree_read_unlock(b); | |
2752 | free_extent_buffer(b); | |
2753 | b = btrfs_lock_root_node(root); | |
2754 | root_lock = BTRFS_WRITE_LOCK; | |
2755 | ||
2756 | /* the level might have changed, check again */ | |
2757 | level = btrfs_header_level(b); | |
2758 | } | |
2759 | } | |
5d4f98a2 | 2760 | } |
bd681513 CM |
2761 | p->nodes[level] = b; |
2762 | if (!p->skip_locking) | |
2763 | p->locks[level] = root_lock; | |
925baedd | 2764 | |
eb60ceac | 2765 | while (b) { |
5f39d397 | 2766 | level = btrfs_header_level(b); |
65b51a00 CM |
2767 | |
2768 | /* | |
2769 | * setup the path here so we can release it under lock | |
2770 | * contention with the cow code | |
2771 | */ | |
02217ed2 | 2772 | if (cow) { |
c8c42864 CM |
2773 | /* |
2774 | * if we don't really need to cow this block | |
2775 | * then we don't want to set the path blocking, | |
2776 | * so we test it here | |
2777 | */ | |
64c12921 JM |
2778 | if (!should_cow_block(trans, root, b)) { |
2779 | trans->dirty = true; | |
65b51a00 | 2780 | goto cow_done; |
64c12921 | 2781 | } |
5d4f98a2 | 2782 | |
bd681513 CM |
2783 | /* |
2784 | * must have write locks on this node and the | |
2785 | * parent | |
2786 | */ | |
5124e00e JB |
2787 | if (level > write_lock_level || |
2788 | (level + 1 > write_lock_level && | |
2789 | level + 1 < BTRFS_MAX_LEVEL && | |
2790 | p->nodes[level + 1])) { | |
bd681513 CM |
2791 | write_lock_level = level + 1; |
2792 | btrfs_release_path(p); | |
2793 | goto again; | |
2794 | } | |
2795 | ||
160f4089 | 2796 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2797 | err = btrfs_cow_block(trans, root, b, |
2798 | p->nodes[level + 1], | |
2799 | p->slots[level + 1], &b); | |
2800 | if (err) { | |
33c66f43 | 2801 | ret = err; |
65b51a00 | 2802 | goto done; |
54aa1f4d | 2803 | } |
02217ed2 | 2804 | } |
65b51a00 | 2805 | cow_done: |
eb60ceac | 2806 | p->nodes[level] = b; |
bd681513 | 2807 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2808 | |
2809 | /* | |
2810 | * we have a lock on b and as long as we aren't changing | |
2811 | * the tree, there is no way to for the items in b to change. | |
2812 | * It is safe to drop the lock on our parent before we | |
2813 | * go through the expensive btree search on b. | |
2814 | * | |
eb653de1 FDBM |
2815 | * If we're inserting or deleting (ins_len != 0), then we might |
2816 | * be changing slot zero, which may require changing the parent. | |
2817 | * So, we can't drop the lock until after we know which slot | |
2818 | * we're operating on. | |
b4ce94de | 2819 | */ |
eb653de1 FDBM |
2820 | if (!ins_len && !p->keep_locks) { |
2821 | int u = level + 1; | |
2822 | ||
2823 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2824 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2825 | p->locks[u] = 0; | |
2826 | } | |
2827 | } | |
b4ce94de | 2828 | |
d7396f07 | 2829 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2830 | if (ret < 0) |
2831 | goto done; | |
b4ce94de | 2832 | |
5f39d397 | 2833 | if (level != 0) { |
33c66f43 YZ |
2834 | int dec = 0; |
2835 | if (ret && slot > 0) { | |
2836 | dec = 1; | |
be0e5c09 | 2837 | slot -= 1; |
33c66f43 | 2838 | } |
be0e5c09 | 2839 | p->slots[level] = slot; |
33c66f43 | 2840 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2841 | ins_len, &write_lock_level); |
33c66f43 | 2842 | if (err == -EAGAIN) |
c8c42864 | 2843 | goto again; |
33c66f43 YZ |
2844 | if (err) { |
2845 | ret = err; | |
c8c42864 | 2846 | goto done; |
33c66f43 | 2847 | } |
c8c42864 CM |
2848 | b = p->nodes[level]; |
2849 | slot = p->slots[level]; | |
b4ce94de | 2850 | |
bd681513 CM |
2851 | /* |
2852 | * slot 0 is special, if we change the key | |
2853 | * we have to update the parent pointer | |
2854 | * which means we must have a write lock | |
2855 | * on the parent | |
2856 | */ | |
eb653de1 | 2857 | if (slot == 0 && ins_len && |
bd681513 CM |
2858 | write_lock_level < level + 1) { |
2859 | write_lock_level = level + 1; | |
2860 | btrfs_release_path(p); | |
2861 | goto again; | |
2862 | } | |
2863 | ||
f7c79f30 CM |
2864 | unlock_up(p, level, lowest_unlock, |
2865 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2866 | |
925baedd | 2867 | if (level == lowest_level) { |
33c66f43 YZ |
2868 | if (dec) |
2869 | p->slots[level]++; | |
5b21f2ed | 2870 | goto done; |
925baedd | 2871 | } |
ca7a79ad | 2872 | |
33c66f43 | 2873 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2874 | &b, level, slot, key, 0); |
33c66f43 | 2875 | if (err == -EAGAIN) |
c8c42864 | 2876 | goto again; |
33c66f43 YZ |
2877 | if (err) { |
2878 | ret = err; | |
76a05b35 | 2879 | goto done; |
33c66f43 | 2880 | } |
76a05b35 | 2881 | |
b4ce94de | 2882 | if (!p->skip_locking) { |
bd681513 CM |
2883 | level = btrfs_header_level(b); |
2884 | if (level <= write_lock_level) { | |
2885 | err = btrfs_try_tree_write_lock(b); | |
2886 | if (!err) { | |
2887 | btrfs_set_path_blocking(p); | |
2888 | btrfs_tree_lock(b); | |
2889 | btrfs_clear_path_blocking(p, b, | |
2890 | BTRFS_WRITE_LOCK); | |
2891 | } | |
2892 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2893 | } else { | |
f82c458a | 2894 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2895 | if (!err) { |
2896 | btrfs_set_path_blocking(p); | |
2897 | btrfs_tree_read_lock(b); | |
2898 | btrfs_clear_path_blocking(p, b, | |
2899 | BTRFS_READ_LOCK); | |
2900 | } | |
2901 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2902 | } |
bd681513 | 2903 | p->nodes[level] = b; |
b4ce94de | 2904 | } |
be0e5c09 CM |
2905 | } else { |
2906 | p->slots[level] = slot; | |
87b29b20 YZ |
2907 | if (ins_len > 0 && |
2908 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2909 | if (write_lock_level < 1) { |
2910 | write_lock_level = 1; | |
2911 | btrfs_release_path(p); | |
2912 | goto again; | |
2913 | } | |
2914 | ||
b4ce94de | 2915 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2916 | err = split_leaf(trans, root, key, |
2917 | p, ins_len, ret == 0); | |
bd681513 | 2918 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2919 | |
33c66f43 YZ |
2920 | BUG_ON(err > 0); |
2921 | if (err) { | |
2922 | ret = err; | |
65b51a00 CM |
2923 | goto done; |
2924 | } | |
5c680ed6 | 2925 | } |
459931ec | 2926 | if (!p->search_for_split) |
f7c79f30 CM |
2927 | unlock_up(p, level, lowest_unlock, |
2928 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2929 | goto done; |
be0e5c09 CM |
2930 | } |
2931 | } | |
65b51a00 CM |
2932 | ret = 1; |
2933 | done: | |
b4ce94de CM |
2934 | /* |
2935 | * we don't really know what they plan on doing with the path | |
2936 | * from here on, so for now just mark it as blocking | |
2937 | */ | |
b9473439 CM |
2938 | if (!p->leave_spinning) |
2939 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2940 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2941 | btrfs_release_path(p); |
65b51a00 | 2942 | return ret; |
be0e5c09 CM |
2943 | } |
2944 | ||
5d9e75c4 JS |
2945 | /* |
2946 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2947 | * current state of the tree together with the operations recorded in the tree | |
2948 | * modification log to search for the key in a previous version of this tree, as | |
2949 | * denoted by the time_seq parameter. | |
2950 | * | |
2951 | * Naturally, there is no support for insert, delete or cow operations. | |
2952 | * | |
2953 | * The resulting path and return value will be set up as if we called | |
2954 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2955 | */ | |
2956 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2957 | struct btrfs_path *p, u64 time_seq) | |
2958 | { | |
0b246afa | 2959 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2960 | struct extent_buffer *b; |
2961 | int slot; | |
2962 | int ret; | |
2963 | int err; | |
2964 | int level; | |
2965 | int lowest_unlock = 1; | |
2966 | u8 lowest_level = 0; | |
d4b4087c | 2967 | int prev_cmp = -1; |
5d9e75c4 JS |
2968 | |
2969 | lowest_level = p->lowest_level; | |
2970 | WARN_ON(p->nodes[0] != NULL); | |
2971 | ||
2972 | if (p->search_commit_root) { | |
2973 | BUG_ON(time_seq); | |
2974 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2975 | } | |
2976 | ||
2977 | again: | |
5d9e75c4 | 2978 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2979 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2980 | p->locks[level] = BTRFS_READ_LOCK; |
2981 | ||
2982 | while (b) { | |
2983 | level = btrfs_header_level(b); | |
2984 | p->nodes[level] = b; | |
2985 | btrfs_clear_path_blocking(p, NULL, 0); | |
2986 | ||
2987 | /* | |
2988 | * we have a lock on b and as long as we aren't changing | |
2989 | * the tree, there is no way to for the items in b to change. | |
2990 | * It is safe to drop the lock on our parent before we | |
2991 | * go through the expensive btree search on b. | |
2992 | */ | |
2993 | btrfs_unlock_up_safe(p, level + 1); | |
2994 | ||
d4b4087c | 2995 | /* |
01327610 | 2996 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2997 | * time. |
2998 | */ | |
2999 | prev_cmp = -1; | |
d7396f07 | 3000 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
3001 | |
3002 | if (level != 0) { | |
3003 | int dec = 0; | |
3004 | if (ret && slot > 0) { | |
3005 | dec = 1; | |
3006 | slot -= 1; | |
3007 | } | |
3008 | p->slots[level] = slot; | |
3009 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3010 | ||
3011 | if (level == lowest_level) { | |
3012 | if (dec) | |
3013 | p->slots[level]++; | |
3014 | goto done; | |
3015 | } | |
3016 | ||
3017 | err = read_block_for_search(NULL, root, p, &b, level, | |
3018 | slot, key, time_seq); | |
3019 | if (err == -EAGAIN) | |
3020 | goto again; | |
3021 | if (err) { | |
3022 | ret = err; | |
3023 | goto done; | |
3024 | } | |
3025 | ||
3026 | level = btrfs_header_level(b); | |
f82c458a | 3027 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3028 | if (!err) { |
3029 | btrfs_set_path_blocking(p); | |
3030 | btrfs_tree_read_lock(b); | |
3031 | btrfs_clear_path_blocking(p, b, | |
3032 | BTRFS_READ_LOCK); | |
3033 | } | |
0b246afa | 3034 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3035 | if (!b) { |
3036 | ret = -ENOMEM; | |
3037 | goto done; | |
3038 | } | |
5d9e75c4 JS |
3039 | p->locks[level] = BTRFS_READ_LOCK; |
3040 | p->nodes[level] = b; | |
5d9e75c4 JS |
3041 | } else { |
3042 | p->slots[level] = slot; | |
3043 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3044 | goto done; | |
3045 | } | |
3046 | } | |
3047 | ret = 1; | |
3048 | done: | |
3049 | if (!p->leave_spinning) | |
3050 | btrfs_set_path_blocking(p); | |
3051 | if (ret < 0) | |
3052 | btrfs_release_path(p); | |
3053 | ||
3054 | return ret; | |
3055 | } | |
3056 | ||
2f38b3e1 AJ |
3057 | /* |
3058 | * helper to use instead of search slot if no exact match is needed but | |
3059 | * instead the next or previous item should be returned. | |
3060 | * When find_higher is true, the next higher item is returned, the next lower | |
3061 | * otherwise. | |
3062 | * When return_any and find_higher are both true, and no higher item is found, | |
3063 | * return the next lower instead. | |
3064 | * When return_any is true and find_higher is false, and no lower item is found, | |
3065 | * return the next higher instead. | |
3066 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3067 | * < 0 on error | |
3068 | */ | |
3069 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
3070 | struct btrfs_key *key, struct btrfs_path *p, | |
3071 | int find_higher, int return_any) | |
3072 | { | |
3073 | int ret; | |
3074 | struct extent_buffer *leaf; | |
3075 | ||
3076 | again: | |
3077 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3078 | if (ret <= 0) | |
3079 | return ret; | |
3080 | /* | |
3081 | * a return value of 1 means the path is at the position where the | |
3082 | * item should be inserted. Normally this is the next bigger item, | |
3083 | * but in case the previous item is the last in a leaf, path points | |
3084 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3085 | * item. | |
3086 | */ | |
3087 | leaf = p->nodes[0]; | |
3088 | ||
3089 | if (find_higher) { | |
3090 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3091 | ret = btrfs_next_leaf(root, p); | |
3092 | if (ret <= 0) | |
3093 | return ret; | |
3094 | if (!return_any) | |
3095 | return 1; | |
3096 | /* | |
3097 | * no higher item found, return the next | |
3098 | * lower instead | |
3099 | */ | |
3100 | return_any = 0; | |
3101 | find_higher = 0; | |
3102 | btrfs_release_path(p); | |
3103 | goto again; | |
3104 | } | |
3105 | } else { | |
e6793769 AJ |
3106 | if (p->slots[0] == 0) { |
3107 | ret = btrfs_prev_leaf(root, p); | |
3108 | if (ret < 0) | |
3109 | return ret; | |
3110 | if (!ret) { | |
23c6bf6a FDBM |
3111 | leaf = p->nodes[0]; |
3112 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3113 | p->slots[0]--; | |
e6793769 | 3114 | return 0; |
2f38b3e1 | 3115 | } |
e6793769 AJ |
3116 | if (!return_any) |
3117 | return 1; | |
3118 | /* | |
3119 | * no lower item found, return the next | |
3120 | * higher instead | |
3121 | */ | |
3122 | return_any = 0; | |
3123 | find_higher = 1; | |
3124 | btrfs_release_path(p); | |
3125 | goto again; | |
3126 | } else { | |
2f38b3e1 AJ |
3127 | --p->slots[0]; |
3128 | } | |
3129 | } | |
3130 | return 0; | |
3131 | } | |
3132 | ||
74123bd7 CM |
3133 | /* |
3134 | * adjust the pointers going up the tree, starting at level | |
3135 | * making sure the right key of each node is points to 'key'. | |
3136 | * This is used after shifting pointers to the left, so it stops | |
3137 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3138 | * higher levels | |
aa5d6bed | 3139 | * |
74123bd7 | 3140 | */ |
b7a0365e DD |
3141 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3142 | struct btrfs_path *path, | |
143bede5 | 3143 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3144 | { |
3145 | int i; | |
5f39d397 CM |
3146 | struct extent_buffer *t; |
3147 | ||
234b63a0 | 3148 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3149 | int tslot = path->slots[i]; |
eb60ceac | 3150 | if (!path->nodes[i]) |
be0e5c09 | 3151 | break; |
5f39d397 | 3152 | t = path->nodes[i]; |
b7a0365e | 3153 | tree_mod_log_set_node_key(fs_info, t, tslot, 1); |
5f39d397 | 3154 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3155 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3156 | if (tslot != 0) |
3157 | break; | |
3158 | } | |
3159 | } | |
3160 | ||
31840ae1 ZY |
3161 | /* |
3162 | * update item key. | |
3163 | * | |
3164 | * This function isn't completely safe. It's the caller's responsibility | |
3165 | * that the new key won't break the order | |
3166 | */ | |
b7a0365e DD |
3167 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3168 | struct btrfs_path *path, | |
143bede5 | 3169 | struct btrfs_key *new_key) |
31840ae1 ZY |
3170 | { |
3171 | struct btrfs_disk_key disk_key; | |
3172 | struct extent_buffer *eb; | |
3173 | int slot; | |
3174 | ||
3175 | eb = path->nodes[0]; | |
3176 | slot = path->slots[0]; | |
3177 | if (slot > 0) { | |
3178 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3179 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3180 | } |
3181 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3182 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3183 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3184 | } |
3185 | ||
3186 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3187 | btrfs_set_item_key(eb, &disk_key, slot); | |
3188 | btrfs_mark_buffer_dirty(eb); | |
3189 | if (slot == 0) | |
b7a0365e | 3190 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3191 | } |
3192 | ||
74123bd7 CM |
3193 | /* |
3194 | * try to push data from one node into the next node left in the | |
79f95c82 | 3195 | * tree. |
aa5d6bed CM |
3196 | * |
3197 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3198 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3199 | */ |
98ed5174 CM |
3200 | static int push_node_left(struct btrfs_trans_handle *trans, |
3201 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3202 | struct extent_buffer *src, int empty) |
be0e5c09 | 3203 | { |
0b246afa | 3204 | struct btrfs_fs_info *fs_info = root->fs_info; |
be0e5c09 | 3205 | int push_items = 0; |
bb803951 CM |
3206 | int src_nritems; |
3207 | int dst_nritems; | |
aa5d6bed | 3208 | int ret = 0; |
be0e5c09 | 3209 | |
5f39d397 CM |
3210 | src_nritems = btrfs_header_nritems(src); |
3211 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3212 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3213 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3214 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3215 | |
bce4eae9 | 3216 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3217 | return 1; |
3218 | ||
d397712b | 3219 | if (push_items <= 0) |
be0e5c09 CM |
3220 | return 1; |
3221 | ||
bce4eae9 | 3222 | if (empty) { |
971a1f66 | 3223 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3224 | if (push_items < src_nritems) { |
3225 | /* leave at least 8 pointers in the node if | |
3226 | * we aren't going to empty it | |
3227 | */ | |
3228 | if (src_nritems - push_items < 8) { | |
3229 | if (push_items <= 8) | |
3230 | return 1; | |
3231 | push_items -= 8; | |
3232 | } | |
3233 | } | |
3234 | } else | |
3235 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3236 | |
0b246afa | 3237 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3238 | push_items); |
3239 | if (ret) { | |
66642832 | 3240 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3241 | return ret; |
3242 | } | |
5f39d397 CM |
3243 | copy_extent_buffer(dst, src, |
3244 | btrfs_node_key_ptr_offset(dst_nritems), | |
3245 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3246 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3247 | |
bb803951 | 3248 | if (push_items < src_nritems) { |
57911b8b JS |
3249 | /* |
3250 | * don't call tree_mod_log_eb_move here, key removal was already | |
3251 | * fully logged by tree_mod_log_eb_copy above. | |
3252 | */ | |
5f39d397 CM |
3253 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3254 | btrfs_node_key_ptr_offset(push_items), | |
3255 | (src_nritems - push_items) * | |
3256 | sizeof(struct btrfs_key_ptr)); | |
3257 | } | |
3258 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3259 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3260 | btrfs_mark_buffer_dirty(src); | |
3261 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3262 | |
79f95c82 CM |
3263 | return ret; |
3264 | } | |
3265 | ||
3266 | /* | |
3267 | * try to push data from one node into the next node right in the | |
3268 | * tree. | |
3269 | * | |
3270 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3271 | * error, and > 0 if there was no room in the right hand block. | |
3272 | * | |
3273 | * this will only push up to 1/2 the contents of the left node over | |
3274 | */ | |
5f39d397 CM |
3275 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3276 | struct btrfs_root *root, | |
3277 | struct extent_buffer *dst, | |
3278 | struct extent_buffer *src) | |
79f95c82 | 3279 | { |
0b246afa | 3280 | struct btrfs_fs_info *fs_info = root->fs_info; |
79f95c82 CM |
3281 | int push_items = 0; |
3282 | int max_push; | |
3283 | int src_nritems; | |
3284 | int dst_nritems; | |
3285 | int ret = 0; | |
79f95c82 | 3286 | |
7bb86316 CM |
3287 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3288 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3289 | ||
5f39d397 CM |
3290 | src_nritems = btrfs_header_nritems(src); |
3291 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3292 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3293 | if (push_items <= 0) |
79f95c82 | 3294 | return 1; |
bce4eae9 | 3295 | |
d397712b | 3296 | if (src_nritems < 4) |
bce4eae9 | 3297 | return 1; |
79f95c82 CM |
3298 | |
3299 | max_push = src_nritems / 2 + 1; | |
3300 | /* don't try to empty the node */ | |
d397712b | 3301 | if (max_push >= src_nritems) |
79f95c82 | 3302 | return 1; |
252c38f0 | 3303 | |
79f95c82 CM |
3304 | if (max_push < push_items) |
3305 | push_items = max_push; | |
3306 | ||
0b246afa | 3307 | tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3308 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3309 | btrfs_node_key_ptr_offset(0), | |
3310 | (dst_nritems) * | |
3311 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3312 | |
0b246afa | 3313 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3314 | src_nritems - push_items, push_items); |
3315 | if (ret) { | |
66642832 | 3316 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3317 | return ret; |
3318 | } | |
5f39d397 CM |
3319 | copy_extent_buffer(dst, src, |
3320 | btrfs_node_key_ptr_offset(0), | |
3321 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3322 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3323 | |
5f39d397 CM |
3324 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3325 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3326 | |
5f39d397 CM |
3327 | btrfs_mark_buffer_dirty(src); |
3328 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3329 | |
aa5d6bed | 3330 | return ret; |
be0e5c09 CM |
3331 | } |
3332 | ||
97571fd0 CM |
3333 | /* |
3334 | * helper function to insert a new root level in the tree. | |
3335 | * A new node is allocated, and a single item is inserted to | |
3336 | * point to the existing root | |
aa5d6bed CM |
3337 | * |
3338 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3339 | */ |
d397712b | 3340 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3341 | struct btrfs_root *root, |
fdd99c72 | 3342 | struct btrfs_path *path, int level) |
5c680ed6 | 3343 | { |
0b246afa | 3344 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3345 | u64 lower_gen; |
5f39d397 CM |
3346 | struct extent_buffer *lower; |
3347 | struct extent_buffer *c; | |
925baedd | 3348 | struct extent_buffer *old; |
5f39d397 | 3349 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3350 | |
3351 | BUG_ON(path->nodes[level]); | |
3352 | BUG_ON(path->nodes[level-1] != root->node); | |
3353 | ||
7bb86316 CM |
3354 | lower = path->nodes[level-1]; |
3355 | if (level == 1) | |
3356 | btrfs_item_key(lower, &lower_key, 0); | |
3357 | else | |
3358 | btrfs_node_key(lower, &lower_key, 0); | |
3359 | ||
4d75f8a9 DS |
3360 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3361 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3362 | if (IS_ERR(c)) |
3363 | return PTR_ERR(c); | |
925baedd | 3364 | |
0b246afa | 3365 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3366 | |
b159fa28 | 3367 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3368 | btrfs_set_header_nritems(c, 1); |
3369 | btrfs_set_header_level(c, level); | |
db94535d | 3370 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3371 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3372 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3373 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3374 | |
0b246afa JM |
3375 | write_extent_buffer_fsid(c, fs_info->fsid); |
3376 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3377 | |
5f39d397 | 3378 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3379 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3380 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3381 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3382 | |
3383 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3384 | |
5f39d397 | 3385 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3386 | |
925baedd | 3387 | old = root->node; |
fdd99c72 | 3388 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3389 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3390 | |
3391 | /* the super has an extra ref to root->node */ | |
3392 | free_extent_buffer(old); | |
3393 | ||
0b86a832 | 3394 | add_root_to_dirty_list(root); |
5f39d397 CM |
3395 | extent_buffer_get(c); |
3396 | path->nodes[level] = c; | |
95449a16 | 3397 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3398 | path->slots[level] = 0; |
3399 | return 0; | |
3400 | } | |
3401 | ||
74123bd7 CM |
3402 | /* |
3403 | * worker function to insert a single pointer in a node. | |
3404 | * the node should have enough room for the pointer already | |
97571fd0 | 3405 | * |
74123bd7 CM |
3406 | * slot and level indicate where you want the key to go, and |
3407 | * blocknr is the block the key points to. | |
3408 | */ | |
143bede5 JM |
3409 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3410 | struct btrfs_root *root, struct btrfs_path *path, | |
3411 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3412 | int slot, int level) |
74123bd7 | 3413 | { |
0b246afa | 3414 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 3415 | struct extent_buffer *lower; |
74123bd7 | 3416 | int nritems; |
f3ea38da | 3417 | int ret; |
5c680ed6 CM |
3418 | |
3419 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3420 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3421 | lower = path->nodes[level]; |
3422 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3423 | BUG_ON(slot > nritems); |
0b246afa | 3424 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3425 | if (slot != nritems) { |
c3e06965 | 3426 | if (level) |
0b246afa | 3427 | tree_mod_log_eb_move(fs_info, lower, slot + 1, |
f3ea38da | 3428 | slot, nritems - slot); |
5f39d397 CM |
3429 | memmove_extent_buffer(lower, |
3430 | btrfs_node_key_ptr_offset(slot + 1), | |
3431 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3432 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3433 | } |
c3e06965 | 3434 | if (level) { |
0b246afa | 3435 | ret = tree_mod_log_insert_key(fs_info, lower, slot, |
c8cc6341 | 3436 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3437 | BUG_ON(ret < 0); |
3438 | } | |
5f39d397 | 3439 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3440 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3441 | WARN_ON(trans->transid == 0); |
3442 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3443 | btrfs_set_header_nritems(lower, nritems + 1); |
3444 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3445 | } |
3446 | ||
97571fd0 CM |
3447 | /* |
3448 | * split the node at the specified level in path in two. | |
3449 | * The path is corrected to point to the appropriate node after the split | |
3450 | * | |
3451 | * Before splitting this tries to make some room in the node by pushing | |
3452 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3453 | * |
3454 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3455 | */ |
e02119d5 CM |
3456 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3457 | struct btrfs_root *root, | |
3458 | struct btrfs_path *path, int level) | |
be0e5c09 | 3459 | { |
0b246afa | 3460 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3461 | struct extent_buffer *c; |
3462 | struct extent_buffer *split; | |
3463 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3464 | int mid; |
5c680ed6 | 3465 | int ret; |
7518a238 | 3466 | u32 c_nritems; |
eb60ceac | 3467 | |
5f39d397 | 3468 | c = path->nodes[level]; |
7bb86316 | 3469 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3470 | if (c == root->node) { |
d9abbf1c | 3471 | /* |
90f8d62e JS |
3472 | * trying to split the root, lets make a new one |
3473 | * | |
fdd99c72 | 3474 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3475 | * insert_new_root, because that root buffer will be kept as a |
3476 | * normal node. We are going to log removal of half of the | |
3477 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3478 | * tree lock on the buffer, which is why we cannot race with | |
3479 | * other tree_mod_log users. | |
d9abbf1c | 3480 | */ |
fdd99c72 | 3481 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3482 | if (ret) |
3483 | return ret; | |
b3612421 | 3484 | } else { |
e66f709b | 3485 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3486 | c = path->nodes[level]; |
3487 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3488 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3489 | return 0; |
54aa1f4d CM |
3490 | if (ret < 0) |
3491 | return ret; | |
be0e5c09 | 3492 | } |
e66f709b | 3493 | |
5f39d397 | 3494 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3495 | mid = (c_nritems + 1) / 2; |
3496 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3497 | |
4d75f8a9 DS |
3498 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3499 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3500 | if (IS_ERR(split)) |
3501 | return PTR_ERR(split); | |
3502 | ||
0b246afa | 3503 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3504 | |
b159fa28 | 3505 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3506 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3507 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3508 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3509 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3510 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3511 | write_extent_buffer_fsid(split, fs_info->fsid); |
3512 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3513 | |
0b246afa | 3514 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3515 | if (ret) { |
66642832 | 3516 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3517 | return ret; |
3518 | } | |
5f39d397 CM |
3519 | copy_extent_buffer(split, c, |
3520 | btrfs_node_key_ptr_offset(0), | |
3521 | btrfs_node_key_ptr_offset(mid), | |
3522 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3523 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3524 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3525 | ret = 0; |
3526 | ||
5f39d397 CM |
3527 | btrfs_mark_buffer_dirty(c); |
3528 | btrfs_mark_buffer_dirty(split); | |
3529 | ||
143bede5 | 3530 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3531 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3532 | |
5de08d7d | 3533 | if (path->slots[level] >= mid) { |
5c680ed6 | 3534 | path->slots[level] -= mid; |
925baedd | 3535 | btrfs_tree_unlock(c); |
5f39d397 CM |
3536 | free_extent_buffer(c); |
3537 | path->nodes[level] = split; | |
5c680ed6 CM |
3538 | path->slots[level + 1] += 1; |
3539 | } else { | |
925baedd | 3540 | btrfs_tree_unlock(split); |
5f39d397 | 3541 | free_extent_buffer(split); |
be0e5c09 | 3542 | } |
aa5d6bed | 3543 | return ret; |
be0e5c09 CM |
3544 | } |
3545 | ||
74123bd7 CM |
3546 | /* |
3547 | * how many bytes are required to store the items in a leaf. start | |
3548 | * and nr indicate which items in the leaf to check. This totals up the | |
3549 | * space used both by the item structs and the item data | |
3550 | */ | |
5f39d397 | 3551 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3552 | { |
41be1f3b JB |
3553 | struct btrfs_item *start_item; |
3554 | struct btrfs_item *end_item; | |
3555 | struct btrfs_map_token token; | |
be0e5c09 | 3556 | int data_len; |
5f39d397 | 3557 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3558 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3559 | |
3560 | if (!nr) | |
3561 | return 0; | |
41be1f3b | 3562 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3563 | start_item = btrfs_item_nr(start); |
3564 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3565 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3566 | btrfs_token_item_size(l, start_item, &token); | |
3567 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3568 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3569 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3570 | return data_len; |
3571 | } | |
3572 | ||
d4dbff95 CM |
3573 | /* |
3574 | * The space between the end of the leaf items and | |
3575 | * the start of the leaf data. IOW, how much room | |
3576 | * the leaf has left for both items and data | |
3577 | */ | |
d397712b | 3578 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3579 | struct extent_buffer *leaf) |
d4dbff95 | 3580 | { |
0b246afa | 3581 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3582 | int nritems = btrfs_header_nritems(leaf); |
3583 | int ret; | |
0b246afa JM |
3584 | |
3585 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3586 | if (ret < 0) { |
0b246afa JM |
3587 | btrfs_crit(fs_info, |
3588 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3589 | ret, | |
3590 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3591 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3592 | } |
3593 | return ret; | |
d4dbff95 CM |
3594 | } |
3595 | ||
99d8f83c CM |
3596 | /* |
3597 | * min slot controls the lowest index we're willing to push to the | |
3598 | * right. We'll push up to and including min_slot, but no lower | |
3599 | */ | |
44871b1b CM |
3600 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3601 | struct btrfs_root *root, | |
3602 | struct btrfs_path *path, | |
3603 | int data_size, int empty, | |
3604 | struct extent_buffer *right, | |
99d8f83c CM |
3605 | int free_space, u32 left_nritems, |
3606 | u32 min_slot) | |
00ec4c51 | 3607 | { |
0b246afa | 3608 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 3609 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3610 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3611 | struct btrfs_map_token token; |
5f39d397 | 3612 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3613 | int slot; |
34a38218 | 3614 | u32 i; |
00ec4c51 CM |
3615 | int push_space = 0; |
3616 | int push_items = 0; | |
0783fcfc | 3617 | struct btrfs_item *item; |
34a38218 | 3618 | u32 nr; |
7518a238 | 3619 | u32 right_nritems; |
5f39d397 | 3620 | u32 data_end; |
db94535d | 3621 | u32 this_item_size; |
00ec4c51 | 3622 | |
cfed81a0 CM |
3623 | btrfs_init_map_token(&token); |
3624 | ||
34a38218 CM |
3625 | if (empty) |
3626 | nr = 0; | |
3627 | else | |
99d8f83c | 3628 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3629 | |
31840ae1 | 3630 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3631 | push_space += data_size; |
31840ae1 | 3632 | |
44871b1b | 3633 | slot = path->slots[1]; |
34a38218 CM |
3634 | i = left_nritems - 1; |
3635 | while (i >= nr) { | |
dd3cc16b | 3636 | item = btrfs_item_nr(i); |
db94535d | 3637 | |
31840ae1 ZY |
3638 | if (!empty && push_items > 0) { |
3639 | if (path->slots[0] > i) | |
3640 | break; | |
3641 | if (path->slots[0] == i) { | |
3642 | int space = btrfs_leaf_free_space(root, left); | |
3643 | if (space + push_space * 2 > free_space) | |
3644 | break; | |
3645 | } | |
3646 | } | |
3647 | ||
00ec4c51 | 3648 | if (path->slots[0] == i) |
87b29b20 | 3649 | push_space += data_size; |
db94535d | 3650 | |
db94535d CM |
3651 | this_item_size = btrfs_item_size(left, item); |
3652 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3653 | break; |
31840ae1 | 3654 | |
00ec4c51 | 3655 | push_items++; |
db94535d | 3656 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3657 | if (i == 0) |
3658 | break; | |
3659 | i--; | |
db94535d | 3660 | } |
5f39d397 | 3661 | |
925baedd CM |
3662 | if (push_items == 0) |
3663 | goto out_unlock; | |
5f39d397 | 3664 | |
6c1500f2 | 3665 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3666 | |
00ec4c51 | 3667 | /* push left to right */ |
5f39d397 | 3668 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3669 | |
5f39d397 | 3670 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3671 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3672 | |
00ec4c51 | 3673 | /* make room in the right data area */ |
5f39d397 CM |
3674 | data_end = leaf_data_end(root, right); |
3675 | memmove_extent_buffer(right, | |
3676 | btrfs_leaf_data(right) + data_end - push_space, | |
3677 | btrfs_leaf_data(right) + data_end, | |
0b246afa | 3678 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3679 | |
00ec4c51 | 3680 | /* copy from the left data area */ |
5f39d397 | 3681 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
0b246afa | 3682 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
d6025579 CM |
3683 | btrfs_leaf_data(left) + leaf_data_end(root, left), |
3684 | push_space); | |
5f39d397 CM |
3685 | |
3686 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3687 | btrfs_item_nr_offset(0), | |
3688 | right_nritems * sizeof(struct btrfs_item)); | |
3689 | ||
00ec4c51 | 3690 | /* copy the items from left to right */ |
5f39d397 CM |
3691 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3692 | btrfs_item_nr_offset(left_nritems - push_items), | |
3693 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3694 | |
3695 | /* update the item pointers */ | |
7518a238 | 3696 | right_nritems += push_items; |
5f39d397 | 3697 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3698 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3699 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3700 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3701 | push_space -= btrfs_token_item_size(right, item, &token); |
3702 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3703 | } |
3704 | ||
7518a238 | 3705 | left_nritems -= push_items; |
5f39d397 | 3706 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3707 | |
34a38218 CM |
3708 | if (left_nritems) |
3709 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3710 | else |
0b246afa | 3711 | clean_tree_block(trans, fs_info, left); |
f0486c68 | 3712 | |
5f39d397 | 3713 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3714 | |
5f39d397 CM |
3715 | btrfs_item_key(right, &disk_key, 0); |
3716 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3717 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3718 | |
00ec4c51 | 3719 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3720 | if (path->slots[0] >= left_nritems) { |
3721 | path->slots[0] -= left_nritems; | |
925baedd | 3722 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
0b246afa | 3723 | clean_tree_block(trans, fs_info, path->nodes[0]); |
925baedd | 3724 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3725 | free_extent_buffer(path->nodes[0]); |
3726 | path->nodes[0] = right; | |
00ec4c51 CM |
3727 | path->slots[1] += 1; |
3728 | } else { | |
925baedd | 3729 | btrfs_tree_unlock(right); |
5f39d397 | 3730 | free_extent_buffer(right); |
00ec4c51 CM |
3731 | } |
3732 | return 0; | |
925baedd CM |
3733 | |
3734 | out_unlock: | |
3735 | btrfs_tree_unlock(right); | |
3736 | free_extent_buffer(right); | |
3737 | return 1; | |
00ec4c51 | 3738 | } |
925baedd | 3739 | |
44871b1b CM |
3740 | /* |
3741 | * push some data in the path leaf to the right, trying to free up at | |
3742 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3743 | * | |
3744 | * returns 1 if the push failed because the other node didn't have enough | |
3745 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3746 | * |
3747 | * this will push starting from min_slot to the end of the leaf. It won't | |
3748 | * push any slot lower than min_slot | |
44871b1b CM |
3749 | */ |
3750 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3751 | *root, struct btrfs_path *path, |
3752 | int min_data_size, int data_size, | |
3753 | int empty, u32 min_slot) | |
44871b1b CM |
3754 | { |
3755 | struct extent_buffer *left = path->nodes[0]; | |
3756 | struct extent_buffer *right; | |
3757 | struct extent_buffer *upper; | |
3758 | int slot; | |
3759 | int free_space; | |
3760 | u32 left_nritems; | |
3761 | int ret; | |
3762 | ||
3763 | if (!path->nodes[1]) | |
3764 | return 1; | |
3765 | ||
3766 | slot = path->slots[1]; | |
3767 | upper = path->nodes[1]; | |
3768 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3769 | return 1; | |
3770 | ||
3771 | btrfs_assert_tree_locked(path->nodes[1]); | |
3772 | ||
3773 | right = read_node_slot(root, upper, slot + 1); | |
fb770ae4 LB |
3774 | /* |
3775 | * slot + 1 is not valid or we fail to read the right node, | |
3776 | * no big deal, just return. | |
3777 | */ | |
3778 | if (IS_ERR(right)) | |
91ca338d TI |
3779 | return 1; |
3780 | ||
44871b1b CM |
3781 | btrfs_tree_lock(right); |
3782 | btrfs_set_lock_blocking(right); | |
3783 | ||
3784 | free_space = btrfs_leaf_free_space(root, right); | |
3785 | if (free_space < data_size) | |
3786 | goto out_unlock; | |
3787 | ||
3788 | /* cow and double check */ | |
3789 | ret = btrfs_cow_block(trans, root, right, upper, | |
3790 | slot + 1, &right); | |
3791 | if (ret) | |
3792 | goto out_unlock; | |
3793 | ||
3794 | free_space = btrfs_leaf_free_space(root, right); | |
3795 | if (free_space < data_size) | |
3796 | goto out_unlock; | |
3797 | ||
3798 | left_nritems = btrfs_header_nritems(left); | |
3799 | if (left_nritems == 0) | |
3800 | goto out_unlock; | |
3801 | ||
2ef1fed2 FDBM |
3802 | if (path->slots[0] == left_nritems && !empty) { |
3803 | /* Key greater than all keys in the leaf, right neighbor has | |
3804 | * enough room for it and we're not emptying our leaf to delete | |
3805 | * it, therefore use right neighbor to insert the new item and | |
3806 | * no need to touch/dirty our left leaft. */ | |
3807 | btrfs_tree_unlock(left); | |
3808 | free_extent_buffer(left); | |
3809 | path->nodes[0] = right; | |
3810 | path->slots[0] = 0; | |
3811 | path->slots[1]++; | |
3812 | return 0; | |
3813 | } | |
3814 | ||
99d8f83c CM |
3815 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3816 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3817 | out_unlock: |
3818 | btrfs_tree_unlock(right); | |
3819 | free_extent_buffer(right); | |
3820 | return 1; | |
3821 | } | |
3822 | ||
74123bd7 CM |
3823 | /* |
3824 | * push some data in the path leaf to the left, trying to free up at | |
3825 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3826 | * |
3827 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3828 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3829 | * items | |
74123bd7 | 3830 | */ |
44871b1b CM |
3831 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3832 | struct btrfs_root *root, | |
3833 | struct btrfs_path *path, int data_size, | |
3834 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3835 | int free_space, u32 right_nritems, |
3836 | u32 max_slot) | |
be0e5c09 | 3837 | { |
0b246afa | 3838 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3839 | struct btrfs_disk_key disk_key; |
3840 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3841 | int i; |
be0e5c09 CM |
3842 | int push_space = 0; |
3843 | int push_items = 0; | |
0783fcfc | 3844 | struct btrfs_item *item; |
7518a238 | 3845 | u32 old_left_nritems; |
34a38218 | 3846 | u32 nr; |
aa5d6bed | 3847 | int ret = 0; |
db94535d CM |
3848 | u32 this_item_size; |
3849 | u32 old_left_item_size; | |
cfed81a0 CM |
3850 | struct btrfs_map_token token; |
3851 | ||
3852 | btrfs_init_map_token(&token); | |
be0e5c09 | 3853 | |
34a38218 | 3854 | if (empty) |
99d8f83c | 3855 | nr = min(right_nritems, max_slot); |
34a38218 | 3856 | else |
99d8f83c | 3857 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3858 | |
3859 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3860 | item = btrfs_item_nr(i); |
db94535d | 3861 | |
31840ae1 ZY |
3862 | if (!empty && push_items > 0) { |
3863 | if (path->slots[0] < i) | |
3864 | break; | |
3865 | if (path->slots[0] == i) { | |
3866 | int space = btrfs_leaf_free_space(root, right); | |
3867 | if (space + push_space * 2 > free_space) | |
3868 | break; | |
3869 | } | |
3870 | } | |
3871 | ||
be0e5c09 | 3872 | if (path->slots[0] == i) |
87b29b20 | 3873 | push_space += data_size; |
db94535d CM |
3874 | |
3875 | this_item_size = btrfs_item_size(right, item); | |
3876 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3877 | break; |
db94535d | 3878 | |
be0e5c09 | 3879 | push_items++; |
db94535d CM |
3880 | push_space += this_item_size + sizeof(*item); |
3881 | } | |
3882 | ||
be0e5c09 | 3883 | if (push_items == 0) { |
925baedd CM |
3884 | ret = 1; |
3885 | goto out; | |
be0e5c09 | 3886 | } |
fae7f21c | 3887 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3888 | |
be0e5c09 | 3889 | /* push data from right to left */ |
5f39d397 CM |
3890 | copy_extent_buffer(left, right, |
3891 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3892 | btrfs_item_nr_offset(0), | |
3893 | push_items * sizeof(struct btrfs_item)); | |
3894 | ||
0b246afa | 3895 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3896 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3897 | |
3898 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3899 | leaf_data_end(root, left) - push_space, |
3900 | btrfs_leaf_data(right) + | |
5f39d397 | 3901 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3902 | push_space); |
5f39d397 | 3903 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3904 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3905 | |
db94535d | 3906 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3907 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3908 | u32 ioff; |
db94535d | 3909 | |
dd3cc16b | 3910 | item = btrfs_item_nr(i); |
db94535d | 3911 | |
cfed81a0 CM |
3912 | ioff = btrfs_token_item_offset(left, item, &token); |
3913 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3914 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3915 | &token); |
be0e5c09 | 3916 | } |
5f39d397 | 3917 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3918 | |
3919 | /* fixup right node */ | |
31b1a2bd JL |
3920 | if (push_items > right_nritems) |
3921 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3922 | right_nritems); |
34a38218 CM |
3923 | |
3924 | if (push_items < right_nritems) { | |
3925 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3926 | leaf_data_end(root, right); | |
3927 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
0b246afa | 3928 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
34a38218 CM |
3929 | btrfs_leaf_data(right) + |
3930 | leaf_data_end(root, right), push_space); | |
3931 | ||
3932 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3933 | btrfs_item_nr_offset(push_items), |
3934 | (btrfs_header_nritems(right) - push_items) * | |
3935 | sizeof(struct btrfs_item)); | |
34a38218 | 3936 | } |
eef1c494 Y |
3937 | right_nritems -= push_items; |
3938 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3939 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3940 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3941 | item = btrfs_item_nr(i); |
db94535d | 3942 | |
cfed81a0 CM |
3943 | push_space = push_space - btrfs_token_item_size(right, |
3944 | item, &token); | |
3945 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3946 | } |
eb60ceac | 3947 | |
5f39d397 | 3948 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3949 | if (right_nritems) |
3950 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3951 | else |
0b246afa | 3952 | clean_tree_block(trans, fs_info, right); |
098f59c2 | 3953 | |
5f39d397 | 3954 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3955 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3956 | |
3957 | /* then fixup the leaf pointer in the path */ | |
3958 | if (path->slots[0] < push_items) { | |
3959 | path->slots[0] += old_left_nritems; | |
925baedd | 3960 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3961 | free_extent_buffer(path->nodes[0]); |
3962 | path->nodes[0] = left; | |
be0e5c09 CM |
3963 | path->slots[1] -= 1; |
3964 | } else { | |
925baedd | 3965 | btrfs_tree_unlock(left); |
5f39d397 | 3966 | free_extent_buffer(left); |
be0e5c09 CM |
3967 | path->slots[0] -= push_items; |
3968 | } | |
eb60ceac | 3969 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3970 | return ret; |
925baedd CM |
3971 | out: |
3972 | btrfs_tree_unlock(left); | |
3973 | free_extent_buffer(left); | |
3974 | return ret; | |
be0e5c09 CM |
3975 | } |
3976 | ||
44871b1b CM |
3977 | /* |
3978 | * push some data in the path leaf to the left, trying to free up at | |
3979 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3980 | * |
3981 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3982 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3983 | * items | |
44871b1b CM |
3984 | */ |
3985 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3986 | *root, struct btrfs_path *path, int min_data_size, |
3987 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3988 | { |
3989 | struct extent_buffer *right = path->nodes[0]; | |
3990 | struct extent_buffer *left; | |
3991 | int slot; | |
3992 | int free_space; | |
3993 | u32 right_nritems; | |
3994 | int ret = 0; | |
3995 | ||
3996 | slot = path->slots[1]; | |
3997 | if (slot == 0) | |
3998 | return 1; | |
3999 | if (!path->nodes[1]) | |
4000 | return 1; | |
4001 | ||
4002 | right_nritems = btrfs_header_nritems(right); | |
4003 | if (right_nritems == 0) | |
4004 | return 1; | |
4005 | ||
4006 | btrfs_assert_tree_locked(path->nodes[1]); | |
4007 | ||
4008 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
fb770ae4 LB |
4009 | /* |
4010 | * slot - 1 is not valid or we fail to read the left node, | |
4011 | * no big deal, just return. | |
4012 | */ | |
4013 | if (IS_ERR(left)) | |
91ca338d TI |
4014 | return 1; |
4015 | ||
44871b1b CM |
4016 | btrfs_tree_lock(left); |
4017 | btrfs_set_lock_blocking(left); | |
4018 | ||
4019 | free_space = btrfs_leaf_free_space(root, left); | |
4020 | if (free_space < data_size) { | |
4021 | ret = 1; | |
4022 | goto out; | |
4023 | } | |
4024 | ||
4025 | /* cow and double check */ | |
4026 | ret = btrfs_cow_block(trans, root, left, | |
4027 | path->nodes[1], slot - 1, &left); | |
4028 | if (ret) { | |
4029 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4030 | if (ret == -ENOSPC) |
4031 | ret = 1; | |
44871b1b CM |
4032 | goto out; |
4033 | } | |
4034 | ||
4035 | free_space = btrfs_leaf_free_space(root, left); | |
4036 | if (free_space < data_size) { | |
4037 | ret = 1; | |
4038 | goto out; | |
4039 | } | |
4040 | ||
99d8f83c CM |
4041 | return __push_leaf_left(trans, root, path, min_data_size, |
4042 | empty, left, free_space, right_nritems, | |
4043 | max_slot); | |
44871b1b CM |
4044 | out: |
4045 | btrfs_tree_unlock(left); | |
4046 | free_extent_buffer(left); | |
4047 | return ret; | |
4048 | } | |
4049 | ||
4050 | /* | |
4051 | * split the path's leaf in two, making sure there is at least data_size | |
4052 | * available for the resulting leaf level of the path. | |
44871b1b | 4053 | */ |
143bede5 JM |
4054 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
4055 | struct btrfs_root *root, | |
4056 | struct btrfs_path *path, | |
4057 | struct extent_buffer *l, | |
4058 | struct extent_buffer *right, | |
4059 | int slot, int mid, int nritems) | |
44871b1b | 4060 | { |
0b246afa | 4061 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
4062 | int data_copy_size; |
4063 | int rt_data_off; | |
4064 | int i; | |
44871b1b | 4065 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4066 | struct btrfs_map_token token; |
4067 | ||
4068 | btrfs_init_map_token(&token); | |
44871b1b CM |
4069 | |
4070 | nritems = nritems - mid; | |
4071 | btrfs_set_header_nritems(right, nritems); | |
4072 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
4073 | ||
4074 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4075 | btrfs_item_nr_offset(mid), | |
4076 | nritems * sizeof(struct btrfs_item)); | |
4077 | ||
4078 | copy_extent_buffer(right, l, | |
0b246afa | 4079 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(fs_info) - |
44871b1b CM |
4080 | data_copy_size, btrfs_leaf_data(l) + |
4081 | leaf_data_end(root, l), data_copy_size); | |
4082 | ||
0b246afa | 4083 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4084 | |
4085 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4086 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4087 | u32 ioff; |
4088 | ||
cfed81a0 CM |
4089 | ioff = btrfs_token_item_offset(right, item, &token); |
4090 | btrfs_set_token_item_offset(right, item, | |
4091 | ioff + rt_data_off, &token); | |
44871b1b CM |
4092 | } |
4093 | ||
44871b1b | 4094 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4095 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 4096 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4097 | path->slots[1] + 1, 1); |
44871b1b CM |
4098 | |
4099 | btrfs_mark_buffer_dirty(right); | |
4100 | btrfs_mark_buffer_dirty(l); | |
4101 | BUG_ON(path->slots[0] != slot); | |
4102 | ||
44871b1b CM |
4103 | if (mid <= slot) { |
4104 | btrfs_tree_unlock(path->nodes[0]); | |
4105 | free_extent_buffer(path->nodes[0]); | |
4106 | path->nodes[0] = right; | |
4107 | path->slots[0] -= mid; | |
4108 | path->slots[1] += 1; | |
4109 | } else { | |
4110 | btrfs_tree_unlock(right); | |
4111 | free_extent_buffer(right); | |
4112 | } | |
4113 | ||
4114 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4115 | } |
4116 | ||
99d8f83c CM |
4117 | /* |
4118 | * double splits happen when we need to insert a big item in the middle | |
4119 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4120 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4121 | * A B C | |
4122 | * | |
4123 | * We avoid this by trying to push the items on either side of our target | |
4124 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4125 | * completely. | |
4126 | */ | |
4127 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4128 | struct btrfs_root *root, | |
4129 | struct btrfs_path *path, | |
4130 | int data_size) | |
4131 | { | |
4132 | int ret; | |
4133 | int progress = 0; | |
4134 | int slot; | |
4135 | u32 nritems; | |
5a4267ca | 4136 | int space_needed = data_size; |
99d8f83c CM |
4137 | |
4138 | slot = path->slots[0]; | |
5a4267ca FDBM |
4139 | if (slot < btrfs_header_nritems(path->nodes[0])) |
4140 | space_needed -= btrfs_leaf_free_space(root, path->nodes[0]); | |
99d8f83c CM |
4141 | |
4142 | /* | |
4143 | * try to push all the items after our slot into the | |
4144 | * right leaf | |
4145 | */ | |
5a4267ca | 4146 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4147 | if (ret < 0) |
4148 | return ret; | |
4149 | ||
4150 | if (ret == 0) | |
4151 | progress++; | |
4152 | ||
4153 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4154 | /* | |
4155 | * our goal is to get our slot at the start or end of a leaf. If | |
4156 | * we've done so we're done | |
4157 | */ | |
4158 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4159 | return 0; | |
4160 | ||
4161 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4162 | return 0; | |
4163 | ||
4164 | /* try to push all the items before our slot into the next leaf */ | |
4165 | slot = path->slots[0]; | |
5a4267ca | 4166 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4167 | if (ret < 0) |
4168 | return ret; | |
4169 | ||
4170 | if (ret == 0) | |
4171 | progress++; | |
4172 | ||
4173 | if (progress) | |
4174 | return 0; | |
4175 | return 1; | |
4176 | } | |
4177 | ||
74123bd7 CM |
4178 | /* |
4179 | * split the path's leaf in two, making sure there is at least data_size | |
4180 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4181 | * |
4182 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4183 | */ |
e02119d5 CM |
4184 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4185 | struct btrfs_root *root, | |
4186 | struct btrfs_key *ins_key, | |
4187 | struct btrfs_path *path, int data_size, | |
4188 | int extend) | |
be0e5c09 | 4189 | { |
5d4f98a2 | 4190 | struct btrfs_disk_key disk_key; |
5f39d397 | 4191 | struct extent_buffer *l; |
7518a238 | 4192 | u32 nritems; |
eb60ceac CM |
4193 | int mid; |
4194 | int slot; | |
5f39d397 | 4195 | struct extent_buffer *right; |
b7a0365e | 4196 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4197 | int ret = 0; |
aa5d6bed | 4198 | int wret; |
5d4f98a2 | 4199 | int split; |
cc0c5538 | 4200 | int num_doubles = 0; |
99d8f83c | 4201 | int tried_avoid_double = 0; |
aa5d6bed | 4202 | |
a5719521 YZ |
4203 | l = path->nodes[0]; |
4204 | slot = path->slots[0]; | |
4205 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4206 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4207 | return -EOVERFLOW; |
4208 | ||
40689478 | 4209 | /* first try to make some room by pushing left and right */ |
33157e05 | 4210 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4211 | int space_needed = data_size; |
4212 | ||
4213 | if (slot < btrfs_header_nritems(l)) | |
4214 | space_needed -= btrfs_leaf_free_space(root, l); | |
4215 | ||
4216 | wret = push_leaf_right(trans, root, path, space_needed, | |
4217 | space_needed, 0, 0); | |
d397712b | 4218 | if (wret < 0) |
eaee50e8 | 4219 | return wret; |
3685f791 | 4220 | if (wret) { |
5a4267ca FDBM |
4221 | wret = push_leaf_left(trans, root, path, space_needed, |
4222 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4223 | if (wret < 0) |
4224 | return wret; | |
4225 | } | |
4226 | l = path->nodes[0]; | |
aa5d6bed | 4227 | |
3685f791 | 4228 | /* did the pushes work? */ |
87b29b20 | 4229 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 4230 | return 0; |
3326d1b0 | 4231 | } |
aa5d6bed | 4232 | |
5c680ed6 | 4233 | if (!path->nodes[1]) { |
fdd99c72 | 4234 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4235 | if (ret) |
4236 | return ret; | |
4237 | } | |
cc0c5538 | 4238 | again: |
5d4f98a2 | 4239 | split = 1; |
cc0c5538 | 4240 | l = path->nodes[0]; |
eb60ceac | 4241 | slot = path->slots[0]; |
5f39d397 | 4242 | nritems = btrfs_header_nritems(l); |
d397712b | 4243 | mid = (nritems + 1) / 2; |
54aa1f4d | 4244 | |
5d4f98a2 YZ |
4245 | if (mid <= slot) { |
4246 | if (nritems == 1 || | |
4247 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4248 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4249 | if (slot >= nritems) { |
4250 | split = 0; | |
4251 | } else { | |
4252 | mid = slot; | |
4253 | if (mid != nritems && | |
4254 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4255 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4256 | if (data_size && !tried_avoid_double) |
4257 | goto push_for_double; | |
5d4f98a2 YZ |
4258 | split = 2; |
4259 | } | |
4260 | } | |
4261 | } | |
4262 | } else { | |
4263 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4264 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4265 | if (!extend && data_size && slot == 0) { |
4266 | split = 0; | |
4267 | } else if ((extend || !data_size) && slot == 0) { | |
4268 | mid = 1; | |
4269 | } else { | |
4270 | mid = slot; | |
4271 | if (mid != nritems && | |
4272 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4273 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4274 | if (data_size && !tried_avoid_double) |
4275 | goto push_for_double; | |
67871254 | 4276 | split = 2; |
5d4f98a2 YZ |
4277 | } |
4278 | } | |
4279 | } | |
4280 | } | |
4281 | ||
4282 | if (split == 0) | |
4283 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4284 | else | |
4285 | btrfs_item_key(l, &disk_key, mid); | |
4286 | ||
4d75f8a9 DS |
4287 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4288 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4289 | if (IS_ERR(right)) |
5f39d397 | 4290 | return PTR_ERR(right); |
f0486c68 | 4291 | |
0b246afa | 4292 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4293 | |
b159fa28 | 4294 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4295 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4296 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4297 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4298 | btrfs_set_header_owner(right, root->root_key.objectid); |
4299 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4300 | write_extent_buffer_fsid(right, fs_info->fsid); |
4301 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4302 | |
5d4f98a2 YZ |
4303 | if (split == 0) { |
4304 | if (mid <= slot) { | |
4305 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4306 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4307 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4308 | btrfs_tree_unlock(path->nodes[0]); |
4309 | free_extent_buffer(path->nodes[0]); | |
4310 | path->nodes[0] = right; | |
4311 | path->slots[0] = 0; | |
4312 | path->slots[1] += 1; | |
4313 | } else { | |
4314 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4315 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4316 | path->slots[1], 1); |
5d4f98a2 YZ |
4317 | btrfs_tree_unlock(path->nodes[0]); |
4318 | free_extent_buffer(path->nodes[0]); | |
4319 | path->nodes[0] = right; | |
4320 | path->slots[0] = 0; | |
143bede5 | 4321 | if (path->slots[1] == 0) |
b7a0365e | 4322 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4323 | } |
196e0249 LB |
4324 | /* |
4325 | * We create a new leaf 'right' for the required ins_len and | |
4326 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4327 | * the content of ins_len to 'right'. | |
4328 | */ | |
5d4f98a2 | 4329 | return ret; |
d4dbff95 | 4330 | } |
74123bd7 | 4331 | |
143bede5 | 4332 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4333 | |
5d4f98a2 | 4334 | if (split == 2) { |
cc0c5538 CM |
4335 | BUG_ON(num_doubles != 0); |
4336 | num_doubles++; | |
4337 | goto again; | |
a429e513 | 4338 | } |
44871b1b | 4339 | |
143bede5 | 4340 | return 0; |
99d8f83c CM |
4341 | |
4342 | push_for_double: | |
4343 | push_for_double_split(trans, root, path, data_size); | |
4344 | tried_avoid_double = 1; | |
4345 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4346 | return 0; | |
4347 | goto again; | |
be0e5c09 CM |
4348 | } |
4349 | ||
ad48fd75 YZ |
4350 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4351 | struct btrfs_root *root, | |
4352 | struct btrfs_path *path, int ins_len) | |
459931ec | 4353 | { |
ad48fd75 | 4354 | struct btrfs_key key; |
459931ec | 4355 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4356 | struct btrfs_file_extent_item *fi; |
4357 | u64 extent_len = 0; | |
4358 | u32 item_size; | |
4359 | int ret; | |
459931ec CM |
4360 | |
4361 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4362 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4363 | ||
4364 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4365 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4366 | ||
4367 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4368 | return 0; | |
459931ec CM |
4369 | |
4370 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4371 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4372 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4373 | struct btrfs_file_extent_item); | |
4374 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4375 | } | |
b3b4aa74 | 4376 | btrfs_release_path(path); |
459931ec | 4377 | |
459931ec | 4378 | path->keep_locks = 1; |
ad48fd75 YZ |
4379 | path->search_for_split = 1; |
4380 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4381 | path->search_for_split = 0; |
a8df6fe6 FM |
4382 | if (ret > 0) |
4383 | ret = -EAGAIN; | |
ad48fd75 YZ |
4384 | if (ret < 0) |
4385 | goto err; | |
459931ec | 4386 | |
ad48fd75 YZ |
4387 | ret = -EAGAIN; |
4388 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4389 | /* if our item isn't there, return now */ |
4390 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4391 | goto err; |
4392 | ||
109f6aef CM |
4393 | /* the leaf has changed, it now has room. return now */ |
4394 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4395 | goto err; | |
4396 | ||
ad48fd75 YZ |
4397 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4398 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4399 | struct btrfs_file_extent_item); | |
4400 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4401 | goto err; | |
459931ec CM |
4402 | } |
4403 | ||
b9473439 | 4404 | btrfs_set_path_blocking(path); |
ad48fd75 | 4405 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4406 | if (ret) |
4407 | goto err; | |
459931ec | 4408 | |
ad48fd75 | 4409 | path->keep_locks = 0; |
b9473439 | 4410 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4411 | return 0; |
4412 | err: | |
4413 | path->keep_locks = 0; | |
4414 | return ret; | |
4415 | } | |
4416 | ||
4417 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4418 | struct btrfs_root *root, | |
4419 | struct btrfs_path *path, | |
4420 | struct btrfs_key *new_key, | |
4421 | unsigned long split_offset) | |
4422 | { | |
4423 | struct extent_buffer *leaf; | |
4424 | struct btrfs_item *item; | |
4425 | struct btrfs_item *new_item; | |
4426 | int slot; | |
4427 | char *buf; | |
4428 | u32 nritems; | |
4429 | u32 item_size; | |
4430 | u32 orig_offset; | |
4431 | struct btrfs_disk_key disk_key; | |
4432 | ||
b9473439 CM |
4433 | leaf = path->nodes[0]; |
4434 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4435 | ||
b4ce94de CM |
4436 | btrfs_set_path_blocking(path); |
4437 | ||
dd3cc16b | 4438 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4439 | orig_offset = btrfs_item_offset(leaf, item); |
4440 | item_size = btrfs_item_size(leaf, item); | |
4441 | ||
459931ec | 4442 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4443 | if (!buf) |
4444 | return -ENOMEM; | |
4445 | ||
459931ec CM |
4446 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4447 | path->slots[0]), item_size); | |
459931ec | 4448 | |
ad48fd75 | 4449 | slot = path->slots[0] + 1; |
459931ec | 4450 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4451 | if (slot != nritems) { |
4452 | /* shift the items */ | |
4453 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4454 | btrfs_item_nr_offset(slot), |
4455 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4456 | } |
4457 | ||
4458 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4459 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4460 | ||
dd3cc16b | 4461 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4462 | |
4463 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4464 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4465 | ||
4466 | btrfs_set_item_offset(leaf, item, | |
4467 | orig_offset + item_size - split_offset); | |
4468 | btrfs_set_item_size(leaf, item, split_offset); | |
4469 | ||
4470 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4471 | ||
4472 | /* write the data for the start of the original item */ | |
4473 | write_extent_buffer(leaf, buf, | |
4474 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4475 | split_offset); | |
4476 | ||
4477 | /* write the data for the new item */ | |
4478 | write_extent_buffer(leaf, buf + split_offset, | |
4479 | btrfs_item_ptr_offset(leaf, slot), | |
4480 | item_size - split_offset); | |
4481 | btrfs_mark_buffer_dirty(leaf); | |
4482 | ||
ad48fd75 | 4483 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4484 | kfree(buf); |
ad48fd75 YZ |
4485 | return 0; |
4486 | } | |
4487 | ||
4488 | /* | |
4489 | * This function splits a single item into two items, | |
4490 | * giving 'new_key' to the new item and splitting the | |
4491 | * old one at split_offset (from the start of the item). | |
4492 | * | |
4493 | * The path may be released by this operation. After | |
4494 | * the split, the path is pointing to the old item. The | |
4495 | * new item is going to be in the same node as the old one. | |
4496 | * | |
4497 | * Note, the item being split must be smaller enough to live alone on | |
4498 | * a tree block with room for one extra struct btrfs_item | |
4499 | * | |
4500 | * This allows us to split the item in place, keeping a lock on the | |
4501 | * leaf the entire time. | |
4502 | */ | |
4503 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4504 | struct btrfs_root *root, | |
4505 | struct btrfs_path *path, | |
4506 | struct btrfs_key *new_key, | |
4507 | unsigned long split_offset) | |
4508 | { | |
4509 | int ret; | |
4510 | ret = setup_leaf_for_split(trans, root, path, | |
4511 | sizeof(struct btrfs_item)); | |
4512 | if (ret) | |
4513 | return ret; | |
4514 | ||
4515 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4516 | return ret; |
4517 | } | |
4518 | ||
ad48fd75 YZ |
4519 | /* |
4520 | * This function duplicate a item, giving 'new_key' to the new item. | |
4521 | * It guarantees both items live in the same tree leaf and the new item | |
4522 | * is contiguous with the original item. | |
4523 | * | |
4524 | * This allows us to split file extent in place, keeping a lock on the | |
4525 | * leaf the entire time. | |
4526 | */ | |
4527 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4528 | struct btrfs_root *root, | |
4529 | struct btrfs_path *path, | |
4530 | struct btrfs_key *new_key) | |
4531 | { | |
4532 | struct extent_buffer *leaf; | |
4533 | int ret; | |
4534 | u32 item_size; | |
4535 | ||
4536 | leaf = path->nodes[0]; | |
4537 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4538 | ret = setup_leaf_for_split(trans, root, path, | |
4539 | item_size + sizeof(struct btrfs_item)); | |
4540 | if (ret) | |
4541 | return ret; | |
4542 | ||
4543 | path->slots[0]++; | |
afe5fea7 | 4544 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4545 | item_size, item_size + |
4546 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4547 | leaf = path->nodes[0]; |
4548 | memcpy_extent_buffer(leaf, | |
4549 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4550 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4551 | item_size); | |
4552 | return 0; | |
4553 | } | |
4554 | ||
d352ac68 CM |
4555 | /* |
4556 | * make the item pointed to by the path smaller. new_size indicates | |
4557 | * how small to make it, and from_end tells us if we just chop bytes | |
4558 | * off the end of the item or if we shift the item to chop bytes off | |
4559 | * the front. | |
4560 | */ | |
afe5fea7 | 4561 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4562 | u32 new_size, int from_end) |
b18c6685 | 4563 | { |
0b246afa | 4564 | struct btrfs_fs_info *fs_info = root->fs_info; |
b18c6685 | 4565 | int slot; |
5f39d397 CM |
4566 | struct extent_buffer *leaf; |
4567 | struct btrfs_item *item; | |
b18c6685 CM |
4568 | u32 nritems; |
4569 | unsigned int data_end; | |
4570 | unsigned int old_data_start; | |
4571 | unsigned int old_size; | |
4572 | unsigned int size_diff; | |
4573 | int i; | |
cfed81a0 CM |
4574 | struct btrfs_map_token token; |
4575 | ||
4576 | btrfs_init_map_token(&token); | |
b18c6685 | 4577 | |
5f39d397 | 4578 | leaf = path->nodes[0]; |
179e29e4 CM |
4579 | slot = path->slots[0]; |
4580 | ||
4581 | old_size = btrfs_item_size_nr(leaf, slot); | |
4582 | if (old_size == new_size) | |
143bede5 | 4583 | return; |
b18c6685 | 4584 | |
5f39d397 | 4585 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4586 | data_end = leaf_data_end(root, leaf); |
4587 | ||
5f39d397 | 4588 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4589 | |
b18c6685 CM |
4590 | size_diff = old_size - new_size; |
4591 | ||
4592 | BUG_ON(slot < 0); | |
4593 | BUG_ON(slot >= nritems); | |
4594 | ||
4595 | /* | |
4596 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4597 | */ | |
4598 | /* first correct the data pointers */ | |
4599 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4600 | u32 ioff; |
dd3cc16b | 4601 | item = btrfs_item_nr(i); |
db94535d | 4602 | |
cfed81a0 CM |
4603 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4604 | btrfs_set_token_item_offset(leaf, item, | |
4605 | ioff + size_diff, &token); | |
b18c6685 | 4606 | } |
db94535d | 4607 | |
b18c6685 | 4608 | /* shift the data */ |
179e29e4 CM |
4609 | if (from_end) { |
4610 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4611 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4612 | data_end, old_data_start + new_size - data_end); | |
4613 | } else { | |
4614 | struct btrfs_disk_key disk_key; | |
4615 | u64 offset; | |
4616 | ||
4617 | btrfs_item_key(leaf, &disk_key, slot); | |
4618 | ||
4619 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4620 | unsigned long ptr; | |
4621 | struct btrfs_file_extent_item *fi; | |
4622 | ||
4623 | fi = btrfs_item_ptr(leaf, slot, | |
4624 | struct btrfs_file_extent_item); | |
4625 | fi = (struct btrfs_file_extent_item *)( | |
4626 | (unsigned long)fi - size_diff); | |
4627 | ||
4628 | if (btrfs_file_extent_type(leaf, fi) == | |
4629 | BTRFS_FILE_EXTENT_INLINE) { | |
4630 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4631 | memmove_extent_buffer(leaf, ptr, | |
d397712b | 4632 | (unsigned long)fi, |
7ec20afb | 4633 | BTRFS_FILE_EXTENT_INLINE_DATA_START); |
179e29e4 CM |
4634 | } |
4635 | } | |
4636 | ||
4637 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4638 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4639 | data_end, old_data_start - data_end); | |
4640 | ||
4641 | offset = btrfs_disk_key_offset(&disk_key); | |
4642 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4643 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4644 | if (slot == 0) | |
0b246afa | 4645 | fixup_low_keys(fs_info, path, &disk_key, 1); |
179e29e4 | 4646 | } |
5f39d397 | 4647 | |
dd3cc16b | 4648 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4649 | btrfs_set_item_size(leaf, item, new_size); |
4650 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4651 | |
5f39d397 CM |
4652 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4653 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4654 | BUG(); |
5f39d397 | 4655 | } |
b18c6685 CM |
4656 | } |
4657 | ||
d352ac68 | 4658 | /* |
8f69dbd2 | 4659 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4660 | */ |
4b90c680 | 4661 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4662 | u32 data_size) |
6567e837 | 4663 | { |
0b246afa | 4664 | struct btrfs_fs_info *fs_info = root->fs_info; |
6567e837 | 4665 | int slot; |
5f39d397 CM |
4666 | struct extent_buffer *leaf; |
4667 | struct btrfs_item *item; | |
6567e837 CM |
4668 | u32 nritems; |
4669 | unsigned int data_end; | |
4670 | unsigned int old_data; | |
4671 | unsigned int old_size; | |
4672 | int i; | |
cfed81a0 CM |
4673 | struct btrfs_map_token token; |
4674 | ||
4675 | btrfs_init_map_token(&token); | |
6567e837 | 4676 | |
5f39d397 | 4677 | leaf = path->nodes[0]; |
6567e837 | 4678 | |
5f39d397 | 4679 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4680 | data_end = leaf_data_end(root, leaf); |
4681 | ||
5f39d397 CM |
4682 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4683 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4684 | BUG(); |
5f39d397 | 4685 | } |
6567e837 | 4686 | slot = path->slots[0]; |
5f39d397 | 4687 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4688 | |
4689 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4690 | if (slot >= nritems) { |
4691 | btrfs_print_leaf(root, leaf); | |
0b246afa JM |
4692 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4693 | slot, nritems); | |
3326d1b0 CM |
4694 | BUG_ON(1); |
4695 | } | |
6567e837 CM |
4696 | |
4697 | /* | |
4698 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4699 | */ | |
4700 | /* first correct the data pointers */ | |
4701 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4702 | u32 ioff; |
dd3cc16b | 4703 | item = btrfs_item_nr(i); |
db94535d | 4704 | |
cfed81a0 CM |
4705 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4706 | btrfs_set_token_item_offset(leaf, item, | |
4707 | ioff - data_size, &token); | |
6567e837 | 4708 | } |
5f39d397 | 4709 | |
6567e837 | 4710 | /* shift the data */ |
5f39d397 | 4711 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4712 | data_end - data_size, btrfs_leaf_data(leaf) + |
4713 | data_end, old_data - data_end); | |
5f39d397 | 4714 | |
6567e837 | 4715 | data_end = old_data; |
5f39d397 | 4716 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4717 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4718 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4719 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4720 | |
5f39d397 CM |
4721 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4722 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4723 | BUG(); |
5f39d397 | 4724 | } |
6567e837 CM |
4725 | } |
4726 | ||
74123bd7 | 4727 | /* |
44871b1b CM |
4728 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4729 | * to save stack depth by doing the bulk of the work in a function | |
4730 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4731 | */ |
afe5fea7 | 4732 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4733 | struct btrfs_key *cpu_key, u32 *data_size, |
4734 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4735 | { |
0b246afa | 4736 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4737 | struct btrfs_item *item; |
9c58309d | 4738 | int i; |
7518a238 | 4739 | u32 nritems; |
be0e5c09 | 4740 | unsigned int data_end; |
e2fa7227 | 4741 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4742 | struct extent_buffer *leaf; |
4743 | int slot; | |
cfed81a0 CM |
4744 | struct btrfs_map_token token; |
4745 | ||
24cdc847 FM |
4746 | if (path->slots[0] == 0) { |
4747 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4748 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4749 | } |
4750 | btrfs_unlock_up_safe(path, 1); | |
4751 | ||
cfed81a0 | 4752 | btrfs_init_map_token(&token); |
e2fa7227 | 4753 | |
5f39d397 | 4754 | leaf = path->nodes[0]; |
44871b1b | 4755 | slot = path->slots[0]; |
74123bd7 | 4756 | |
5f39d397 | 4757 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4758 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4759 | |
f25956cc | 4760 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4761 | btrfs_print_leaf(root, leaf); |
0b246afa | 4762 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
5d163e0e | 4763 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4764 | BUG(); |
d4dbff95 | 4765 | } |
5f39d397 | 4766 | |
be0e5c09 | 4767 | if (slot != nritems) { |
5f39d397 | 4768 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4769 | |
5f39d397 CM |
4770 | if (old_data < data_end) { |
4771 | btrfs_print_leaf(root, leaf); | |
0b246afa | 4772 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4773 | slot, old_data, data_end); |
5f39d397 CM |
4774 | BUG_ON(1); |
4775 | } | |
be0e5c09 CM |
4776 | /* |
4777 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4778 | */ | |
4779 | /* first correct the data pointers */ | |
0783fcfc | 4780 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4781 | u32 ioff; |
db94535d | 4782 | |
62e85577 | 4783 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4784 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4785 | btrfs_set_token_item_offset(leaf, item, | |
4786 | ioff - total_data, &token); | |
0783fcfc | 4787 | } |
be0e5c09 | 4788 | /* shift the items */ |
9c58309d | 4789 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4790 | btrfs_item_nr_offset(slot), |
d6025579 | 4791 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4792 | |
4793 | /* shift the data */ | |
5f39d397 | 4794 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4795 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4796 | data_end, old_data - data_end); |
be0e5c09 CM |
4797 | data_end = old_data; |
4798 | } | |
5f39d397 | 4799 | |
62e2749e | 4800 | /* setup the item for the new data */ |
9c58309d CM |
4801 | for (i = 0; i < nr; i++) { |
4802 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4803 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4804 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4805 | btrfs_set_token_item_offset(leaf, item, |
4806 | data_end - data_size[i], &token); | |
9c58309d | 4807 | data_end -= data_size[i]; |
cfed81a0 | 4808 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4809 | } |
44871b1b | 4810 | |
9c58309d | 4811 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4812 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4813 | |
5f39d397 CM |
4814 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4815 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4816 | BUG(); |
5f39d397 | 4817 | } |
44871b1b CM |
4818 | } |
4819 | ||
4820 | /* | |
4821 | * Given a key and some data, insert items into the tree. | |
4822 | * This does all the path init required, making room in the tree if needed. | |
4823 | */ | |
4824 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4825 | struct btrfs_root *root, | |
4826 | struct btrfs_path *path, | |
4827 | struct btrfs_key *cpu_key, u32 *data_size, | |
4828 | int nr) | |
4829 | { | |
44871b1b CM |
4830 | int ret = 0; |
4831 | int slot; | |
4832 | int i; | |
4833 | u32 total_size = 0; | |
4834 | u32 total_data = 0; | |
4835 | ||
4836 | for (i = 0; i < nr; i++) | |
4837 | total_data += data_size[i]; | |
4838 | ||
4839 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4840 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4841 | if (ret == 0) | |
4842 | return -EEXIST; | |
4843 | if (ret < 0) | |
143bede5 | 4844 | return ret; |
44871b1b | 4845 | |
44871b1b CM |
4846 | slot = path->slots[0]; |
4847 | BUG_ON(slot < 0); | |
4848 | ||
afe5fea7 | 4849 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4850 | total_data, total_size, nr); |
143bede5 | 4851 | return 0; |
62e2749e CM |
4852 | } |
4853 | ||
4854 | /* | |
4855 | * Given a key and some data, insert an item into the tree. | |
4856 | * This does all the path init required, making room in the tree if needed. | |
4857 | */ | |
e089f05c CM |
4858 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4859 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4860 | data_size) | |
62e2749e CM |
4861 | { |
4862 | int ret = 0; | |
2c90e5d6 | 4863 | struct btrfs_path *path; |
5f39d397 CM |
4864 | struct extent_buffer *leaf; |
4865 | unsigned long ptr; | |
62e2749e | 4866 | |
2c90e5d6 | 4867 | path = btrfs_alloc_path(); |
db5b493a TI |
4868 | if (!path) |
4869 | return -ENOMEM; | |
2c90e5d6 | 4870 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4871 | if (!ret) { |
5f39d397 CM |
4872 | leaf = path->nodes[0]; |
4873 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4874 | write_extent_buffer(leaf, data, ptr, data_size); | |
4875 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4876 | } |
2c90e5d6 | 4877 | btrfs_free_path(path); |
aa5d6bed | 4878 | return ret; |
be0e5c09 CM |
4879 | } |
4880 | ||
74123bd7 | 4881 | /* |
5de08d7d | 4882 | * delete the pointer from a given node. |
74123bd7 | 4883 | * |
d352ac68 CM |
4884 | * the tree should have been previously balanced so the deletion does not |
4885 | * empty a node. | |
74123bd7 | 4886 | */ |
afe5fea7 TI |
4887 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4888 | int level, int slot) | |
be0e5c09 | 4889 | { |
0b246afa | 4890 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4891 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4892 | u32 nritems; |
f3ea38da | 4893 | int ret; |
be0e5c09 | 4894 | |
5f39d397 | 4895 | nritems = btrfs_header_nritems(parent); |
d397712b | 4896 | if (slot != nritems - 1) { |
0e411ece | 4897 | if (level) |
0b246afa | 4898 | tree_mod_log_eb_move(fs_info, parent, slot, |
f3ea38da | 4899 | slot + 1, nritems - slot - 1); |
5f39d397 CM |
4900 | memmove_extent_buffer(parent, |
4901 | btrfs_node_key_ptr_offset(slot), | |
4902 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4903 | sizeof(struct btrfs_key_ptr) * |
4904 | (nritems - slot - 1)); | |
57ba86c0 | 4905 | } else if (level) { |
0b246afa | 4906 | ret = tree_mod_log_insert_key(fs_info, parent, slot, |
c8cc6341 | 4907 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4908 | BUG_ON(ret < 0); |
bb803951 | 4909 | } |
f3ea38da | 4910 | |
7518a238 | 4911 | nritems--; |
5f39d397 | 4912 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4913 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4914 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4915 | /* just turn the root into a leaf and break */ |
5f39d397 | 4916 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4917 | } else if (slot == 0) { |
5f39d397 CM |
4918 | struct btrfs_disk_key disk_key; |
4919 | ||
4920 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4921 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4922 | } |
d6025579 | 4923 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4924 | } |
4925 | ||
323ac95b CM |
4926 | /* |
4927 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4928 | * path->nodes[1]. |
323ac95b CM |
4929 | * |
4930 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4931 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4932 | * | |
4933 | * The path must have already been setup for deleting the leaf, including | |
4934 | * all the proper balancing. path->nodes[1] must be locked. | |
4935 | */ | |
143bede5 JM |
4936 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4937 | struct btrfs_root *root, | |
4938 | struct btrfs_path *path, | |
4939 | struct extent_buffer *leaf) | |
323ac95b | 4940 | { |
5d4f98a2 | 4941 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4942 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4943 | |
4d081c41 CM |
4944 | /* |
4945 | * btrfs_free_extent is expensive, we want to make sure we | |
4946 | * aren't holding any locks when we call it | |
4947 | */ | |
4948 | btrfs_unlock_up_safe(path, 0); | |
4949 | ||
f0486c68 YZ |
4950 | root_sub_used(root, leaf->len); |
4951 | ||
3083ee2e | 4952 | extent_buffer_get(leaf); |
5581a51a | 4953 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4954 | free_extent_buffer_stale(leaf); |
323ac95b | 4955 | } |
74123bd7 CM |
4956 | /* |
4957 | * delete the item at the leaf level in path. If that empties | |
4958 | * the leaf, remove it from the tree | |
4959 | */ | |
85e21bac CM |
4960 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4961 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4962 | { |
0b246afa | 4963 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4964 | struct extent_buffer *leaf; |
4965 | struct btrfs_item *item; | |
ce0eac2a AM |
4966 | u32 last_off; |
4967 | u32 dsize = 0; | |
aa5d6bed CM |
4968 | int ret = 0; |
4969 | int wret; | |
85e21bac | 4970 | int i; |
7518a238 | 4971 | u32 nritems; |
cfed81a0 CM |
4972 | struct btrfs_map_token token; |
4973 | ||
4974 | btrfs_init_map_token(&token); | |
be0e5c09 | 4975 | |
5f39d397 | 4976 | leaf = path->nodes[0]; |
85e21bac CM |
4977 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4978 | ||
4979 | for (i = 0; i < nr; i++) | |
4980 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4981 | ||
5f39d397 | 4982 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4983 | |
85e21bac | 4984 | if (slot + nr != nritems) { |
123abc88 | 4985 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4986 | |
4987 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4988 | data_end + dsize, |
4989 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4990 | last_off - data_end); |
5f39d397 | 4991 | |
85e21bac | 4992 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4993 | u32 ioff; |
db94535d | 4994 | |
dd3cc16b | 4995 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4996 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4997 | btrfs_set_token_item_offset(leaf, item, | |
4998 | ioff + dsize, &token); | |
0783fcfc | 4999 | } |
db94535d | 5000 | |
5f39d397 | 5001 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5002 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5003 | sizeof(struct btrfs_item) * |
85e21bac | 5004 | (nritems - slot - nr)); |
be0e5c09 | 5005 | } |
85e21bac CM |
5006 | btrfs_set_header_nritems(leaf, nritems - nr); |
5007 | nritems -= nr; | |
5f39d397 | 5008 | |
74123bd7 | 5009 | /* delete the leaf if we've emptied it */ |
7518a238 | 5010 | if (nritems == 0) { |
5f39d397 CM |
5011 | if (leaf == root->node) { |
5012 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5013 | } else { |
f0486c68 | 5014 | btrfs_set_path_blocking(path); |
0b246afa | 5015 | clean_tree_block(trans, fs_info, leaf); |
143bede5 | 5016 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5017 | } |
be0e5c09 | 5018 | } else { |
7518a238 | 5019 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5020 | if (slot == 0) { |
5f39d397 CM |
5021 | struct btrfs_disk_key disk_key; |
5022 | ||
5023 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 5024 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 5025 | } |
aa5d6bed | 5026 | |
74123bd7 | 5027 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5028 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5029 | /* push_leaf_left fixes the path. |
5030 | * make sure the path still points to our leaf | |
5031 | * for possible call to del_ptr below | |
5032 | */ | |
4920c9ac | 5033 | slot = path->slots[1]; |
5f39d397 CM |
5034 | extent_buffer_get(leaf); |
5035 | ||
b9473439 | 5036 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5037 | wret = push_leaf_left(trans, root, path, 1, 1, |
5038 | 1, (u32)-1); | |
54aa1f4d | 5039 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5040 | ret = wret; |
5f39d397 CM |
5041 | |
5042 | if (path->nodes[0] == leaf && | |
5043 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5044 | wret = push_leaf_right(trans, root, path, 1, |
5045 | 1, 1, 0); | |
54aa1f4d | 5046 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5047 | ret = wret; |
5048 | } | |
5f39d397 CM |
5049 | |
5050 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5051 | path->slots[1] = slot; |
143bede5 | 5052 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5053 | free_extent_buffer(leaf); |
143bede5 | 5054 | ret = 0; |
5de08d7d | 5055 | } else { |
925baedd CM |
5056 | /* if we're still in the path, make sure |
5057 | * we're dirty. Otherwise, one of the | |
5058 | * push_leaf functions must have already | |
5059 | * dirtied this buffer | |
5060 | */ | |
5061 | if (path->nodes[0] == leaf) | |
5062 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5063 | free_extent_buffer(leaf); |
be0e5c09 | 5064 | } |
d5719762 | 5065 | } else { |
5f39d397 | 5066 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5067 | } |
5068 | } | |
aa5d6bed | 5069 | return ret; |
be0e5c09 CM |
5070 | } |
5071 | ||
7bb86316 | 5072 | /* |
925baedd | 5073 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5074 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5075 | * returns < 0 on io errors. | |
d352ac68 CM |
5076 | * |
5077 | * This may release the path, and so you may lose any locks held at the | |
5078 | * time you call it. | |
7bb86316 | 5079 | */ |
16e7549f | 5080 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5081 | { |
925baedd CM |
5082 | struct btrfs_key key; |
5083 | struct btrfs_disk_key found_key; | |
5084 | int ret; | |
7bb86316 | 5085 | |
925baedd | 5086 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5087 | |
e8b0d724 | 5088 | if (key.offset > 0) { |
925baedd | 5089 | key.offset--; |
e8b0d724 | 5090 | } else if (key.type > 0) { |
925baedd | 5091 | key.type--; |
e8b0d724 FDBM |
5092 | key.offset = (u64)-1; |
5093 | } else if (key.objectid > 0) { | |
925baedd | 5094 | key.objectid--; |
e8b0d724 FDBM |
5095 | key.type = (u8)-1; |
5096 | key.offset = (u64)-1; | |
5097 | } else { | |
925baedd | 5098 | return 1; |
e8b0d724 | 5099 | } |
7bb86316 | 5100 | |
b3b4aa74 | 5101 | btrfs_release_path(path); |
925baedd CM |
5102 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5103 | if (ret < 0) | |
5104 | return ret; | |
5105 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5106 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5107 | /* |
5108 | * We might have had an item with the previous key in the tree right | |
5109 | * before we released our path. And after we released our path, that | |
5110 | * item might have been pushed to the first slot (0) of the leaf we | |
5111 | * were holding due to a tree balance. Alternatively, an item with the | |
5112 | * previous key can exist as the only element of a leaf (big fat item). | |
5113 | * Therefore account for these 2 cases, so that our callers (like | |
5114 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5115 | * the previous key we computed above. | |
5116 | */ | |
5117 | if (ret <= 0) | |
925baedd CM |
5118 | return 0; |
5119 | return 1; | |
7bb86316 CM |
5120 | } |
5121 | ||
3f157a2f CM |
5122 | /* |
5123 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5124 | * for nodes or leaves that are have a minimum transaction id. |
5125 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5126 | * |
5127 | * This does not cow, but it does stuff the starting key it finds back | |
5128 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5129 | * key and get a writable path. | |
5130 | * | |
5131 | * This does lock as it descends, and path->keep_locks should be set | |
5132 | * to 1 by the caller. | |
5133 | * | |
5134 | * This honors path->lowest_level to prevent descent past a given level | |
5135 | * of the tree. | |
5136 | * | |
d352ac68 CM |
5137 | * min_trans indicates the oldest transaction that you are interested |
5138 | * in walking through. Any nodes or leaves older than min_trans are | |
5139 | * skipped over (without reading them). | |
5140 | * | |
3f157a2f CM |
5141 | * returns zero if something useful was found, < 0 on error and 1 if there |
5142 | * was nothing in the tree that matched the search criteria. | |
5143 | */ | |
5144 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5145 | struct btrfs_path *path, |
3f157a2f CM |
5146 | u64 min_trans) |
5147 | { | |
5148 | struct extent_buffer *cur; | |
5149 | struct btrfs_key found_key; | |
5150 | int slot; | |
9652480b | 5151 | int sret; |
3f157a2f CM |
5152 | u32 nritems; |
5153 | int level; | |
5154 | int ret = 1; | |
f98de9b9 | 5155 | int keep_locks = path->keep_locks; |
3f157a2f | 5156 | |
f98de9b9 | 5157 | path->keep_locks = 1; |
3f157a2f | 5158 | again: |
bd681513 | 5159 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5160 | level = btrfs_header_level(cur); |
e02119d5 | 5161 | WARN_ON(path->nodes[level]); |
3f157a2f | 5162 | path->nodes[level] = cur; |
bd681513 | 5163 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5164 | |
5165 | if (btrfs_header_generation(cur) < min_trans) { | |
5166 | ret = 1; | |
5167 | goto out; | |
5168 | } | |
d397712b | 5169 | while (1) { |
3f157a2f CM |
5170 | nritems = btrfs_header_nritems(cur); |
5171 | level = btrfs_header_level(cur); | |
9652480b | 5172 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 5173 | |
323ac95b CM |
5174 | /* at the lowest level, we're done, setup the path and exit */ |
5175 | if (level == path->lowest_level) { | |
e02119d5 CM |
5176 | if (slot >= nritems) |
5177 | goto find_next_key; | |
3f157a2f CM |
5178 | ret = 0; |
5179 | path->slots[level] = slot; | |
5180 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5181 | goto out; | |
5182 | } | |
9652480b Y |
5183 | if (sret && slot > 0) |
5184 | slot--; | |
3f157a2f | 5185 | /* |
de78b51a ES |
5186 | * check this node pointer against the min_trans parameters. |
5187 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5188 | */ |
d397712b | 5189 | while (slot < nritems) { |
3f157a2f | 5190 | u64 gen; |
e02119d5 | 5191 | |
3f157a2f CM |
5192 | gen = btrfs_node_ptr_generation(cur, slot); |
5193 | if (gen < min_trans) { | |
5194 | slot++; | |
5195 | continue; | |
5196 | } | |
de78b51a | 5197 | break; |
3f157a2f | 5198 | } |
e02119d5 | 5199 | find_next_key: |
3f157a2f CM |
5200 | /* |
5201 | * we didn't find a candidate key in this node, walk forward | |
5202 | * and find another one | |
5203 | */ | |
5204 | if (slot >= nritems) { | |
e02119d5 | 5205 | path->slots[level] = slot; |
b4ce94de | 5206 | btrfs_set_path_blocking(path); |
e02119d5 | 5207 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5208 | min_trans); |
e02119d5 | 5209 | if (sret == 0) { |
b3b4aa74 | 5210 | btrfs_release_path(path); |
3f157a2f CM |
5211 | goto again; |
5212 | } else { | |
5213 | goto out; | |
5214 | } | |
5215 | } | |
5216 | /* save our key for returning back */ | |
5217 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5218 | path->slots[level] = slot; | |
5219 | if (level == path->lowest_level) { | |
5220 | ret = 0; | |
3f157a2f CM |
5221 | goto out; |
5222 | } | |
b4ce94de | 5223 | btrfs_set_path_blocking(path); |
3f157a2f | 5224 | cur = read_node_slot(root, cur, slot); |
fb770ae4 LB |
5225 | if (IS_ERR(cur)) { |
5226 | ret = PTR_ERR(cur); | |
5227 | goto out; | |
5228 | } | |
3f157a2f | 5229 | |
bd681513 | 5230 | btrfs_tree_read_lock(cur); |
b4ce94de | 5231 | |
bd681513 | 5232 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5233 | path->nodes[level - 1] = cur; |
f7c79f30 | 5234 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5235 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5236 | } |
5237 | out: | |
f98de9b9 FM |
5238 | path->keep_locks = keep_locks; |
5239 | if (ret == 0) { | |
5240 | btrfs_unlock_up_safe(path, path->lowest_level + 1); | |
5241 | btrfs_set_path_blocking(path); | |
3f157a2f | 5242 | memcpy(min_key, &found_key, sizeof(found_key)); |
f98de9b9 | 5243 | } |
3f157a2f CM |
5244 | return ret; |
5245 | } | |
5246 | ||
fb770ae4 | 5247 | static int tree_move_down(struct btrfs_root *root, |
7069830a AB |
5248 | struct btrfs_path *path, |
5249 | int *level, int root_level) | |
5250 | { | |
fb770ae4 LB |
5251 | struct extent_buffer *eb; |
5252 | ||
74dd17fb | 5253 | BUG_ON(*level == 0); |
fb770ae4 LB |
5254 | eb = read_node_slot(root, path->nodes[*level], path->slots[*level]); |
5255 | if (IS_ERR(eb)) | |
5256 | return PTR_ERR(eb); | |
5257 | ||
5258 | path->nodes[*level - 1] = eb; | |
7069830a AB |
5259 | path->slots[*level - 1] = 0; |
5260 | (*level)--; | |
fb770ae4 | 5261 | return 0; |
7069830a AB |
5262 | } |
5263 | ||
5264 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5265 | struct btrfs_path *path, | |
5266 | int *level, int root_level) | |
5267 | { | |
5268 | int ret = 0; | |
5269 | int nritems; | |
5270 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5271 | ||
5272 | path->slots[*level]++; | |
5273 | ||
74dd17fb | 5274 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5275 | if (*level == root_level) |
5276 | return -1; | |
5277 | ||
5278 | /* move upnext */ | |
5279 | path->slots[*level] = 0; | |
5280 | free_extent_buffer(path->nodes[*level]); | |
5281 | path->nodes[*level] = NULL; | |
5282 | (*level)++; | |
5283 | path->slots[*level]++; | |
5284 | ||
5285 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5286 | ret = 1; | |
5287 | } | |
5288 | return ret; | |
5289 | } | |
5290 | ||
5291 | /* | |
5292 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5293 | * or down. | |
5294 | */ | |
5295 | static int tree_advance(struct btrfs_root *root, | |
5296 | struct btrfs_path *path, | |
5297 | int *level, int root_level, | |
5298 | int allow_down, | |
5299 | struct btrfs_key *key) | |
5300 | { | |
5301 | int ret; | |
5302 | ||
5303 | if (*level == 0 || !allow_down) { | |
5304 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5305 | } else { | |
fb770ae4 | 5306 | ret = tree_move_down(root, path, level, root_level); |
7069830a AB |
5307 | } |
5308 | if (ret >= 0) { | |
5309 | if (*level == 0) | |
5310 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5311 | path->slots[*level]); | |
5312 | else | |
5313 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5314 | path->slots[*level]); | |
5315 | } | |
5316 | return ret; | |
5317 | } | |
5318 | ||
5319 | static int tree_compare_item(struct btrfs_root *left_root, | |
5320 | struct btrfs_path *left_path, | |
5321 | struct btrfs_path *right_path, | |
5322 | char *tmp_buf) | |
5323 | { | |
5324 | int cmp; | |
5325 | int len1, len2; | |
5326 | unsigned long off1, off2; | |
5327 | ||
5328 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5329 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5330 | if (len1 != len2) | |
5331 | return 1; | |
5332 | ||
5333 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5334 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5335 | right_path->slots[0]); | |
5336 | ||
5337 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5338 | ||
5339 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5340 | if (cmp) | |
5341 | return 1; | |
5342 | return 0; | |
5343 | } | |
5344 | ||
5345 | #define ADVANCE 1 | |
5346 | #define ADVANCE_ONLY_NEXT -1 | |
5347 | ||
5348 | /* | |
5349 | * This function compares two trees and calls the provided callback for | |
5350 | * every changed/new/deleted item it finds. | |
5351 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5352 | * the compare pretty fast on snapshotted subvolumes. | |
5353 | * | |
5354 | * This currently works on commit roots only. As commit roots are read only, | |
5355 | * we don't do any locking. The commit roots are protected with transactions. | |
5356 | * Transactions are ended and rejoined when a commit is tried in between. | |
5357 | * | |
5358 | * This function checks for modifications done to the trees while comparing. | |
5359 | * If it detects a change, it aborts immediately. | |
5360 | */ | |
5361 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5362 | struct btrfs_root *right_root, | |
5363 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5364 | { | |
0b246afa | 5365 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5366 | int ret; |
5367 | int cmp; | |
7069830a AB |
5368 | struct btrfs_path *left_path = NULL; |
5369 | struct btrfs_path *right_path = NULL; | |
5370 | struct btrfs_key left_key; | |
5371 | struct btrfs_key right_key; | |
5372 | char *tmp_buf = NULL; | |
5373 | int left_root_level; | |
5374 | int right_root_level; | |
5375 | int left_level; | |
5376 | int right_level; | |
5377 | int left_end_reached; | |
5378 | int right_end_reached; | |
5379 | int advance_left; | |
5380 | int advance_right; | |
5381 | u64 left_blockptr; | |
5382 | u64 right_blockptr; | |
6baa4293 FM |
5383 | u64 left_gen; |
5384 | u64 right_gen; | |
7069830a AB |
5385 | |
5386 | left_path = btrfs_alloc_path(); | |
5387 | if (!left_path) { | |
5388 | ret = -ENOMEM; | |
5389 | goto out; | |
5390 | } | |
5391 | right_path = btrfs_alloc_path(); | |
5392 | if (!right_path) { | |
5393 | ret = -ENOMEM; | |
5394 | goto out; | |
5395 | } | |
5396 | ||
0b246afa | 5397 | tmp_buf = kmalloc(fs_info->nodesize, GFP_KERNEL | __GFP_NOWARN); |
7069830a | 5398 | if (!tmp_buf) { |
0b246afa | 5399 | tmp_buf = vmalloc(fs_info->nodesize); |
8f282f71 DS |
5400 | if (!tmp_buf) { |
5401 | ret = -ENOMEM; | |
5402 | goto out; | |
5403 | } | |
7069830a AB |
5404 | } |
5405 | ||
5406 | left_path->search_commit_root = 1; | |
5407 | left_path->skip_locking = 1; | |
5408 | right_path->search_commit_root = 1; | |
5409 | right_path->skip_locking = 1; | |
5410 | ||
7069830a AB |
5411 | /* |
5412 | * Strategy: Go to the first items of both trees. Then do | |
5413 | * | |
5414 | * If both trees are at level 0 | |
5415 | * Compare keys of current items | |
5416 | * If left < right treat left item as new, advance left tree | |
5417 | * and repeat | |
5418 | * If left > right treat right item as deleted, advance right tree | |
5419 | * and repeat | |
5420 | * If left == right do deep compare of items, treat as changed if | |
5421 | * needed, advance both trees and repeat | |
5422 | * If both trees are at the same level but not at level 0 | |
5423 | * Compare keys of current nodes/leafs | |
5424 | * If left < right advance left tree and repeat | |
5425 | * If left > right advance right tree and repeat | |
5426 | * If left == right compare blockptrs of the next nodes/leafs | |
5427 | * If they match advance both trees but stay at the same level | |
5428 | * and repeat | |
5429 | * If they don't match advance both trees while allowing to go | |
5430 | * deeper and repeat | |
5431 | * If tree levels are different | |
5432 | * Advance the tree that needs it and repeat | |
5433 | * | |
5434 | * Advancing a tree means: | |
5435 | * If we are at level 0, try to go to the next slot. If that's not | |
5436 | * possible, go one level up and repeat. Stop when we found a level | |
5437 | * where we could go to the next slot. We may at this point be on a | |
5438 | * node or a leaf. | |
5439 | * | |
5440 | * If we are not at level 0 and not on shared tree blocks, go one | |
5441 | * level deeper. | |
5442 | * | |
5443 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5444 | * the right if possible or go up and right. | |
5445 | */ | |
5446 | ||
0b246afa | 5447 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5448 | left_level = btrfs_header_level(left_root->commit_root); |
5449 | left_root_level = left_level; | |
5450 | left_path->nodes[left_level] = left_root->commit_root; | |
5451 | extent_buffer_get(left_path->nodes[left_level]); | |
5452 | ||
5453 | right_level = btrfs_header_level(right_root->commit_root); | |
5454 | right_root_level = right_level; | |
5455 | right_path->nodes[right_level] = right_root->commit_root; | |
5456 | extent_buffer_get(right_path->nodes[right_level]); | |
0b246afa | 5457 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5458 | |
5459 | if (left_level == 0) | |
5460 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5461 | &left_key, left_path->slots[left_level]); | |
5462 | else | |
5463 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5464 | &left_key, left_path->slots[left_level]); | |
5465 | if (right_level == 0) | |
5466 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5467 | &right_key, right_path->slots[right_level]); | |
5468 | else | |
5469 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5470 | &right_key, right_path->slots[right_level]); | |
5471 | ||
5472 | left_end_reached = right_end_reached = 0; | |
5473 | advance_left = advance_right = 0; | |
5474 | ||
5475 | while (1) { | |
7069830a AB |
5476 | if (advance_left && !left_end_reached) { |
5477 | ret = tree_advance(left_root, left_path, &left_level, | |
5478 | left_root_level, | |
5479 | advance_left != ADVANCE_ONLY_NEXT, | |
5480 | &left_key); | |
fb770ae4 | 5481 | if (ret == -1) |
7069830a | 5482 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5483 | else if (ret < 0) |
5484 | goto out; | |
7069830a AB |
5485 | advance_left = 0; |
5486 | } | |
5487 | if (advance_right && !right_end_reached) { | |
5488 | ret = tree_advance(right_root, right_path, &right_level, | |
5489 | right_root_level, | |
5490 | advance_right != ADVANCE_ONLY_NEXT, | |
5491 | &right_key); | |
fb770ae4 | 5492 | if (ret == -1) |
7069830a | 5493 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5494 | else if (ret < 0) |
5495 | goto out; | |
7069830a AB |
5496 | advance_right = 0; |
5497 | } | |
5498 | ||
5499 | if (left_end_reached && right_end_reached) { | |
5500 | ret = 0; | |
5501 | goto out; | |
5502 | } else if (left_end_reached) { | |
5503 | if (right_level == 0) { | |
5504 | ret = changed_cb(left_root, right_root, | |
5505 | left_path, right_path, | |
5506 | &right_key, | |
5507 | BTRFS_COMPARE_TREE_DELETED, | |
5508 | ctx); | |
5509 | if (ret < 0) | |
5510 | goto out; | |
5511 | } | |
5512 | advance_right = ADVANCE; | |
5513 | continue; | |
5514 | } else if (right_end_reached) { | |
5515 | if (left_level == 0) { | |
5516 | ret = changed_cb(left_root, right_root, | |
5517 | left_path, right_path, | |
5518 | &left_key, | |
5519 | BTRFS_COMPARE_TREE_NEW, | |
5520 | ctx); | |
5521 | if (ret < 0) | |
5522 | goto out; | |
5523 | } | |
5524 | advance_left = ADVANCE; | |
5525 | continue; | |
5526 | } | |
5527 | ||
5528 | if (left_level == 0 && right_level == 0) { | |
5529 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5530 | if (cmp < 0) { | |
5531 | ret = changed_cb(left_root, right_root, | |
5532 | left_path, right_path, | |
5533 | &left_key, | |
5534 | BTRFS_COMPARE_TREE_NEW, | |
5535 | ctx); | |
5536 | if (ret < 0) | |
5537 | goto out; | |
5538 | advance_left = ADVANCE; | |
5539 | } else if (cmp > 0) { | |
5540 | ret = changed_cb(left_root, right_root, | |
5541 | left_path, right_path, | |
5542 | &right_key, | |
5543 | BTRFS_COMPARE_TREE_DELETED, | |
5544 | ctx); | |
5545 | if (ret < 0) | |
5546 | goto out; | |
5547 | advance_right = ADVANCE; | |
5548 | } else { | |
b99d9a6a | 5549 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5550 | |
74dd17fb | 5551 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5552 | ret = tree_compare_item(left_root, left_path, |
5553 | right_path, tmp_buf); | |
ba5e8f2e | 5554 | if (ret) |
b99d9a6a | 5555 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5556 | else |
b99d9a6a | 5557 | result = BTRFS_COMPARE_TREE_SAME; |
ba5e8f2e JB |
5558 | ret = changed_cb(left_root, right_root, |
5559 | left_path, right_path, | |
b99d9a6a | 5560 | &left_key, result, ctx); |
ba5e8f2e JB |
5561 | if (ret < 0) |
5562 | goto out; | |
7069830a AB |
5563 | advance_left = ADVANCE; |
5564 | advance_right = ADVANCE; | |
5565 | } | |
5566 | } else if (left_level == right_level) { | |
5567 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5568 | if (cmp < 0) { | |
5569 | advance_left = ADVANCE; | |
5570 | } else if (cmp > 0) { | |
5571 | advance_right = ADVANCE; | |
5572 | } else { | |
5573 | left_blockptr = btrfs_node_blockptr( | |
5574 | left_path->nodes[left_level], | |
5575 | left_path->slots[left_level]); | |
5576 | right_blockptr = btrfs_node_blockptr( | |
5577 | right_path->nodes[right_level], | |
5578 | right_path->slots[right_level]); | |
6baa4293 FM |
5579 | left_gen = btrfs_node_ptr_generation( |
5580 | left_path->nodes[left_level], | |
5581 | left_path->slots[left_level]); | |
5582 | right_gen = btrfs_node_ptr_generation( | |
5583 | right_path->nodes[right_level], | |
5584 | right_path->slots[right_level]); | |
5585 | if (left_blockptr == right_blockptr && | |
5586 | left_gen == right_gen) { | |
7069830a AB |
5587 | /* |
5588 | * As we're on a shared block, don't | |
5589 | * allow to go deeper. | |
5590 | */ | |
5591 | advance_left = ADVANCE_ONLY_NEXT; | |
5592 | advance_right = ADVANCE_ONLY_NEXT; | |
5593 | } else { | |
5594 | advance_left = ADVANCE; | |
5595 | advance_right = ADVANCE; | |
5596 | } | |
5597 | } | |
5598 | } else if (left_level < right_level) { | |
5599 | advance_right = ADVANCE; | |
5600 | } else { | |
5601 | advance_left = ADVANCE; | |
5602 | } | |
5603 | } | |
5604 | ||
5605 | out: | |
5606 | btrfs_free_path(left_path); | |
5607 | btrfs_free_path(right_path); | |
8f282f71 | 5608 | kvfree(tmp_buf); |
7069830a AB |
5609 | return ret; |
5610 | } | |
5611 | ||
3f157a2f CM |
5612 | /* |
5613 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5614 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5615 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5616 | * |
5617 | * 0 is returned if another key is found, < 0 if there are any errors | |
5618 | * and 1 is returned if there are no higher keys in the tree | |
5619 | * | |
5620 | * path->keep_locks should be set to 1 on the search made before | |
5621 | * calling this function. | |
5622 | */ | |
e7a84565 | 5623 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5624 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5625 | { |
e7a84565 CM |
5626 | int slot; |
5627 | struct extent_buffer *c; | |
5628 | ||
934d375b | 5629 | WARN_ON(!path->keep_locks); |
d397712b | 5630 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5631 | if (!path->nodes[level]) |
5632 | return 1; | |
5633 | ||
5634 | slot = path->slots[level] + 1; | |
5635 | c = path->nodes[level]; | |
3f157a2f | 5636 | next: |
e7a84565 | 5637 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5638 | int ret; |
5639 | int orig_lowest; | |
5640 | struct btrfs_key cur_key; | |
5641 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5642 | !path->nodes[level + 1]) | |
e7a84565 | 5643 | return 1; |
33c66f43 YZ |
5644 | |
5645 | if (path->locks[level + 1]) { | |
5646 | level++; | |
5647 | continue; | |
5648 | } | |
5649 | ||
5650 | slot = btrfs_header_nritems(c) - 1; | |
5651 | if (level == 0) | |
5652 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5653 | else | |
5654 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5655 | ||
5656 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5657 | btrfs_release_path(path); |
33c66f43 YZ |
5658 | path->lowest_level = level; |
5659 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5660 | 0, 0); | |
5661 | path->lowest_level = orig_lowest; | |
5662 | if (ret < 0) | |
5663 | return ret; | |
5664 | ||
5665 | c = path->nodes[level]; | |
5666 | slot = path->slots[level]; | |
5667 | if (ret == 0) | |
5668 | slot++; | |
5669 | goto next; | |
e7a84565 | 5670 | } |
33c66f43 | 5671 | |
e7a84565 CM |
5672 | if (level == 0) |
5673 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5674 | else { |
3f157a2f CM |
5675 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5676 | ||
3f157a2f CM |
5677 | if (gen < min_trans) { |
5678 | slot++; | |
5679 | goto next; | |
5680 | } | |
e7a84565 | 5681 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5682 | } |
e7a84565 CM |
5683 | return 0; |
5684 | } | |
5685 | return 1; | |
5686 | } | |
5687 | ||
97571fd0 | 5688 | /* |
925baedd | 5689 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5690 | * returns 0 if it found something or 1 if there are no greater leaves. |
5691 | * returns < 0 on io errors. | |
97571fd0 | 5692 | */ |
234b63a0 | 5693 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5694 | { |
5695 | return btrfs_next_old_leaf(root, path, 0); | |
5696 | } | |
5697 | ||
5698 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5699 | u64 time_seq) | |
d97e63b6 CM |
5700 | { |
5701 | int slot; | |
8e73f275 | 5702 | int level; |
5f39d397 | 5703 | struct extent_buffer *c; |
8e73f275 | 5704 | struct extent_buffer *next; |
925baedd CM |
5705 | struct btrfs_key key; |
5706 | u32 nritems; | |
5707 | int ret; | |
8e73f275 | 5708 | int old_spinning = path->leave_spinning; |
bd681513 | 5709 | int next_rw_lock = 0; |
925baedd CM |
5710 | |
5711 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5712 | if (nritems == 0) |
925baedd | 5713 | return 1; |
925baedd | 5714 | |
8e73f275 CM |
5715 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5716 | again: | |
5717 | level = 1; | |
5718 | next = NULL; | |
bd681513 | 5719 | next_rw_lock = 0; |
b3b4aa74 | 5720 | btrfs_release_path(path); |
8e73f275 | 5721 | |
a2135011 | 5722 | path->keep_locks = 1; |
31533fb2 | 5723 | path->leave_spinning = 1; |
8e73f275 | 5724 | |
3d7806ec JS |
5725 | if (time_seq) |
5726 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5727 | else | |
5728 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5729 | path->keep_locks = 0; |
5730 | ||
5731 | if (ret < 0) | |
5732 | return ret; | |
5733 | ||
a2135011 | 5734 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5735 | /* |
5736 | * by releasing the path above we dropped all our locks. A balance | |
5737 | * could have added more items next to the key that used to be | |
5738 | * at the very end of the block. So, check again here and | |
5739 | * advance the path if there are now more items available. | |
5740 | */ | |
a2135011 | 5741 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5742 | if (ret == 0) |
5743 | path->slots[0]++; | |
8e73f275 | 5744 | ret = 0; |
925baedd CM |
5745 | goto done; |
5746 | } | |
0b43e04f LB |
5747 | /* |
5748 | * So the above check misses one case: | |
5749 | * - after releasing the path above, someone has removed the item that | |
5750 | * used to be at the very end of the block, and balance between leafs | |
5751 | * gets another one with bigger key.offset to replace it. | |
5752 | * | |
5753 | * This one should be returned as well, or we can get leaf corruption | |
5754 | * later(esp. in __btrfs_drop_extents()). | |
5755 | * | |
5756 | * And a bit more explanation about this check, | |
5757 | * with ret > 0, the key isn't found, the path points to the slot | |
5758 | * where it should be inserted, so the path->slots[0] item must be the | |
5759 | * bigger one. | |
5760 | */ | |
5761 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5762 | ret = 0; | |
5763 | goto done; | |
5764 | } | |
d97e63b6 | 5765 | |
d397712b | 5766 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5767 | if (!path->nodes[level]) { |
5768 | ret = 1; | |
5769 | goto done; | |
5770 | } | |
5f39d397 | 5771 | |
d97e63b6 CM |
5772 | slot = path->slots[level] + 1; |
5773 | c = path->nodes[level]; | |
5f39d397 | 5774 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5775 | level++; |
8e73f275 CM |
5776 | if (level == BTRFS_MAX_LEVEL) { |
5777 | ret = 1; | |
5778 | goto done; | |
5779 | } | |
d97e63b6 CM |
5780 | continue; |
5781 | } | |
5f39d397 | 5782 | |
925baedd | 5783 | if (next) { |
bd681513 | 5784 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5785 | free_extent_buffer(next); |
925baedd | 5786 | } |
5f39d397 | 5787 | |
8e73f275 | 5788 | next = c; |
bd681513 | 5789 | next_rw_lock = path->locks[level]; |
8e73f275 | 5790 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5791 | slot, &key, 0); |
8e73f275 CM |
5792 | if (ret == -EAGAIN) |
5793 | goto again; | |
5f39d397 | 5794 | |
76a05b35 | 5795 | if (ret < 0) { |
b3b4aa74 | 5796 | btrfs_release_path(path); |
76a05b35 CM |
5797 | goto done; |
5798 | } | |
5799 | ||
5cd57b2c | 5800 | if (!path->skip_locking) { |
bd681513 | 5801 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5802 | if (!ret && time_seq) { |
5803 | /* | |
5804 | * If we don't get the lock, we may be racing | |
5805 | * with push_leaf_left, holding that lock while | |
5806 | * itself waiting for the leaf we've currently | |
5807 | * locked. To solve this situation, we give up | |
5808 | * on our lock and cycle. | |
5809 | */ | |
cf538830 | 5810 | free_extent_buffer(next); |
d42244a0 JS |
5811 | btrfs_release_path(path); |
5812 | cond_resched(); | |
5813 | goto again; | |
5814 | } | |
8e73f275 CM |
5815 | if (!ret) { |
5816 | btrfs_set_path_blocking(path); | |
bd681513 | 5817 | btrfs_tree_read_lock(next); |
31533fb2 | 5818 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5819 | BTRFS_READ_LOCK); |
8e73f275 | 5820 | } |
31533fb2 | 5821 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5822 | } |
d97e63b6 CM |
5823 | break; |
5824 | } | |
5825 | path->slots[level] = slot; | |
d397712b | 5826 | while (1) { |
d97e63b6 CM |
5827 | level--; |
5828 | c = path->nodes[level]; | |
925baedd | 5829 | if (path->locks[level]) |
bd681513 | 5830 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5831 | |
5f39d397 | 5832 | free_extent_buffer(c); |
d97e63b6 CM |
5833 | path->nodes[level] = next; |
5834 | path->slots[level] = 0; | |
a74a4b97 | 5835 | if (!path->skip_locking) |
bd681513 | 5836 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5837 | if (!level) |
5838 | break; | |
b4ce94de | 5839 | |
8e73f275 | 5840 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5841 | 0, &key, 0); |
8e73f275 CM |
5842 | if (ret == -EAGAIN) |
5843 | goto again; | |
5844 | ||
76a05b35 | 5845 | if (ret < 0) { |
b3b4aa74 | 5846 | btrfs_release_path(path); |
76a05b35 CM |
5847 | goto done; |
5848 | } | |
5849 | ||
5cd57b2c | 5850 | if (!path->skip_locking) { |
bd681513 | 5851 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5852 | if (!ret) { |
5853 | btrfs_set_path_blocking(path); | |
bd681513 | 5854 | btrfs_tree_read_lock(next); |
31533fb2 | 5855 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5856 | BTRFS_READ_LOCK); |
5857 | } | |
31533fb2 | 5858 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5859 | } |
d97e63b6 | 5860 | } |
8e73f275 | 5861 | ret = 0; |
925baedd | 5862 | done: |
f7c79f30 | 5863 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5864 | path->leave_spinning = old_spinning; |
5865 | if (!old_spinning) | |
5866 | btrfs_set_path_blocking(path); | |
5867 | ||
5868 | return ret; | |
d97e63b6 | 5869 | } |
0b86a832 | 5870 | |
3f157a2f CM |
5871 | /* |
5872 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5873 | * searching until it gets past min_objectid or finds an item of 'type' | |
5874 | * | |
5875 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5876 | */ | |
0b86a832 CM |
5877 | int btrfs_previous_item(struct btrfs_root *root, |
5878 | struct btrfs_path *path, u64 min_objectid, | |
5879 | int type) | |
5880 | { | |
5881 | struct btrfs_key found_key; | |
5882 | struct extent_buffer *leaf; | |
e02119d5 | 5883 | u32 nritems; |
0b86a832 CM |
5884 | int ret; |
5885 | ||
d397712b | 5886 | while (1) { |
0b86a832 | 5887 | if (path->slots[0] == 0) { |
b4ce94de | 5888 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5889 | ret = btrfs_prev_leaf(root, path); |
5890 | if (ret != 0) | |
5891 | return ret; | |
5892 | } else { | |
5893 | path->slots[0]--; | |
5894 | } | |
5895 | leaf = path->nodes[0]; | |
e02119d5 CM |
5896 | nritems = btrfs_header_nritems(leaf); |
5897 | if (nritems == 0) | |
5898 | return 1; | |
5899 | if (path->slots[0] == nritems) | |
5900 | path->slots[0]--; | |
5901 | ||
0b86a832 | 5902 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5903 | if (found_key.objectid < min_objectid) |
5904 | break; | |
0a4eefbb YZ |
5905 | if (found_key.type == type) |
5906 | return 0; | |
e02119d5 CM |
5907 | if (found_key.objectid == min_objectid && |
5908 | found_key.type < type) | |
5909 | break; | |
0b86a832 CM |
5910 | } |
5911 | return 1; | |
5912 | } | |
ade2e0b3 WS |
5913 | |
5914 | /* | |
5915 | * search in extent tree to find a previous Metadata/Data extent item with | |
5916 | * min objecitd. | |
5917 | * | |
5918 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5919 | */ | |
5920 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5921 | struct btrfs_path *path, u64 min_objectid) | |
5922 | { | |
5923 | struct btrfs_key found_key; | |
5924 | struct extent_buffer *leaf; | |
5925 | u32 nritems; | |
5926 | int ret; | |
5927 | ||
5928 | while (1) { | |
5929 | if (path->slots[0] == 0) { | |
5930 | btrfs_set_path_blocking(path); | |
5931 | ret = btrfs_prev_leaf(root, path); | |
5932 | if (ret != 0) | |
5933 | return ret; | |
5934 | } else { | |
5935 | path->slots[0]--; | |
5936 | } | |
5937 | leaf = path->nodes[0]; | |
5938 | nritems = btrfs_header_nritems(leaf); | |
5939 | if (nritems == 0) | |
5940 | return 1; | |
5941 | if (path->slots[0] == nritems) | |
5942 | path->slots[0]--; | |
5943 | ||
5944 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5945 | if (found_key.objectid < min_objectid) | |
5946 | break; | |
5947 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5948 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5949 | return 0; | |
5950 | if (found_key.objectid == min_objectid && | |
5951 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5952 | break; | |
5953 | } | |
5954 | return 1; | |
5955 | } |