]>
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> |
adf02123 | 22 | #include <linux/mm.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); | |
310712b2 OS |
31 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
32 | const struct btrfs_key *ins_key, struct btrfs_path *path, | |
33 | int data_size, int extend); | |
5f39d397 | 34 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
35 | struct btrfs_fs_info *fs_info, |
36 | struct extent_buffer *dst, | |
971a1f66 | 37 | struct extent_buffer *src, int empty); |
5f39d397 | 38 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 39 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
40 | struct extent_buffer *dst_buf, |
41 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
42 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
43 | int level, int slot); | |
5de865ee | 44 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 45 | struct extent_buffer *eb); |
d97e63b6 | 46 | |
df24a2b9 | 47 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 48 | { |
e2c89907 | 49 | return kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
2c90e5d6 CM |
50 | } |
51 | ||
b4ce94de CM |
52 | /* |
53 | * set all locked nodes in the path to blocking locks. This should | |
54 | * be done before scheduling | |
55 | */ | |
56 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
57 | { | |
58 | int i; | |
59 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
60 | if (!p->nodes[i] || !p->locks[i]) |
61 | continue; | |
62 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
63 | if (p->locks[i] == BTRFS_READ_LOCK) | |
64 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
65 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
66 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
67 | } |
68 | } | |
69 | ||
70 | /* | |
71 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
72 | * |
73 | * held is used to keep lockdep happy, when lockdep is enabled | |
74 | * we set held to a blocking lock before we go around and | |
75 | * retake all the spinlocks in the path. You can safely use NULL | |
76 | * for held | |
b4ce94de | 77 | */ |
4008c04a | 78 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 79 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
80 | { |
81 | int i; | |
4008c04a | 82 | |
bd681513 CM |
83 | if (held) { |
84 | btrfs_set_lock_blocking_rw(held, held_rw); | |
85 | if (held_rw == BTRFS_WRITE_LOCK) | |
86 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
87 | else if (held_rw == BTRFS_READ_LOCK) | |
88 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
89 | } | |
4008c04a | 90 | btrfs_set_path_blocking(p); |
4008c04a CM |
91 | |
92 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
93 | if (p->nodes[i] && p->locks[i]) { |
94 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
95 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
96 | p->locks[i] = BTRFS_WRITE_LOCK; | |
97 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
98 | p->locks[i] = BTRFS_READ_LOCK; | |
99 | } | |
b4ce94de | 100 | } |
4008c04a | 101 | |
4008c04a | 102 | if (held) |
bd681513 | 103 | btrfs_clear_lock_blocking_rw(held, held_rw); |
b4ce94de CM |
104 | } |
105 | ||
d352ac68 | 106 | /* this also releases the path */ |
df24a2b9 | 107 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 108 | { |
ff175d57 JJ |
109 | if (!p) |
110 | return; | |
b3b4aa74 | 111 | btrfs_release_path(p); |
df24a2b9 | 112 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
113 | } |
114 | ||
d352ac68 CM |
115 | /* |
116 | * path release drops references on the extent buffers in the path | |
117 | * and it drops any locks held by this path | |
118 | * | |
119 | * It is safe to call this on paths that no locks or extent buffers held. | |
120 | */ | |
b3b4aa74 | 121 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
122 | { |
123 | int i; | |
a2135011 | 124 | |
234b63a0 | 125 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 126 | p->slots[i] = 0; |
eb60ceac | 127 | if (!p->nodes[i]) |
925baedd CM |
128 | continue; |
129 | if (p->locks[i]) { | |
bd681513 | 130 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
131 | p->locks[i] = 0; |
132 | } | |
5f39d397 | 133 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 134 | p->nodes[i] = NULL; |
eb60ceac CM |
135 | } |
136 | } | |
137 | ||
d352ac68 CM |
138 | /* |
139 | * safely gets a reference on the root node of a tree. A lock | |
140 | * is not taken, so a concurrent writer may put a different node | |
141 | * at the root of the tree. See btrfs_lock_root_node for the | |
142 | * looping required. | |
143 | * | |
144 | * The extent buffer returned by this has a reference taken, so | |
145 | * it won't disappear. It may stop being the root of the tree | |
146 | * at any time because there are no locks held. | |
147 | */ | |
925baedd CM |
148 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
149 | { | |
150 | struct extent_buffer *eb; | |
240f62c8 | 151 | |
3083ee2e JB |
152 | while (1) { |
153 | rcu_read_lock(); | |
154 | eb = rcu_dereference(root->node); | |
155 | ||
156 | /* | |
157 | * RCU really hurts here, we could free up the root node because | |
01327610 | 158 | * it was COWed but we may not get the new root node yet so do |
3083ee2e JB |
159 | * the inc_not_zero dance and if it doesn't work then |
160 | * synchronize_rcu and try again. | |
161 | */ | |
162 | if (atomic_inc_not_zero(&eb->refs)) { | |
163 | rcu_read_unlock(); | |
164 | break; | |
165 | } | |
166 | rcu_read_unlock(); | |
167 | synchronize_rcu(); | |
168 | } | |
925baedd CM |
169 | return eb; |
170 | } | |
171 | ||
d352ac68 CM |
172 | /* loop around taking references on and locking the root node of the |
173 | * tree until you end up with a lock on the root. A locked buffer | |
174 | * is returned, with a reference held. | |
175 | */ | |
925baedd CM |
176 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
177 | { | |
178 | struct extent_buffer *eb; | |
179 | ||
d397712b | 180 | while (1) { |
925baedd CM |
181 | eb = btrfs_root_node(root); |
182 | btrfs_tree_lock(eb); | |
240f62c8 | 183 | if (eb == root->node) |
925baedd | 184 | break; |
925baedd CM |
185 | btrfs_tree_unlock(eb); |
186 | free_extent_buffer(eb); | |
187 | } | |
188 | return eb; | |
189 | } | |
190 | ||
bd681513 CM |
191 | /* loop around taking references on and locking the root node of the |
192 | * tree until you end up with a lock on the root. A locked buffer | |
193 | * is returned, with a reference held. | |
194 | */ | |
84f7d8e6 | 195 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
196 | { |
197 | struct extent_buffer *eb; | |
198 | ||
199 | while (1) { | |
200 | eb = btrfs_root_node(root); | |
201 | btrfs_tree_read_lock(eb); | |
202 | if (eb == root->node) | |
203 | break; | |
204 | btrfs_tree_read_unlock(eb); | |
205 | free_extent_buffer(eb); | |
206 | } | |
207 | return eb; | |
208 | } | |
209 | ||
d352ac68 CM |
210 | /* cowonly root (everything not a reference counted cow subvolume), just get |
211 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
212 | * get properly updated on disk. | |
213 | */ | |
0b86a832 CM |
214 | static void add_root_to_dirty_list(struct btrfs_root *root) |
215 | { | |
0b246afa JM |
216 | struct btrfs_fs_info *fs_info = root->fs_info; |
217 | ||
e7070be1 JB |
218 | if (test_bit(BTRFS_ROOT_DIRTY, &root->state) || |
219 | !test_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state)) | |
220 | return; | |
221 | ||
0b246afa | 222 | spin_lock(&fs_info->trans_lock); |
e7070be1 JB |
223 | if (!test_and_set_bit(BTRFS_ROOT_DIRTY, &root->state)) { |
224 | /* Want the extent tree to be the last on the list */ | |
225 | if (root->objectid == BTRFS_EXTENT_TREE_OBJECTID) | |
226 | list_move_tail(&root->dirty_list, | |
0b246afa | 227 | &fs_info->dirty_cowonly_roots); |
e7070be1 JB |
228 | else |
229 | list_move(&root->dirty_list, | |
0b246afa | 230 | &fs_info->dirty_cowonly_roots); |
0b86a832 | 231 | } |
0b246afa | 232 | spin_unlock(&fs_info->trans_lock); |
0b86a832 CM |
233 | } |
234 | ||
d352ac68 CM |
235 | /* |
236 | * used by snapshot creation to make a copy of a root for a tree with | |
237 | * a given objectid. The buffer with the new root node is returned in | |
238 | * cow_ret, and this func returns zero on success or a negative error code. | |
239 | */ | |
be20aa9d CM |
240 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
241 | struct btrfs_root *root, | |
242 | struct extent_buffer *buf, | |
243 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
244 | { | |
0b246afa | 245 | struct btrfs_fs_info *fs_info = root->fs_info; |
be20aa9d | 246 | struct extent_buffer *cow; |
be20aa9d CM |
247 | int ret = 0; |
248 | int level; | |
5d4f98a2 | 249 | struct btrfs_disk_key disk_key; |
be20aa9d | 250 | |
27cdeb70 | 251 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 252 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
253 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
254 | trans->transid != root->last_trans); | |
be20aa9d CM |
255 | |
256 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
257 | if (level == 0) |
258 | btrfs_item_key(buf, &disk_key, 0); | |
259 | else | |
260 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 261 | |
4d75f8a9 DS |
262 | cow = btrfs_alloc_tree_block(trans, root, 0, new_root_objectid, |
263 | &disk_key, level, buf->start, 0); | |
5d4f98a2 | 264 | if (IS_ERR(cow)) |
be20aa9d CM |
265 | return PTR_ERR(cow); |
266 | ||
58e8012c | 267 | copy_extent_buffer_full(cow, buf); |
be20aa9d CM |
268 | btrfs_set_header_bytenr(cow, cow->start); |
269 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
270 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
271 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
272 | BTRFS_HEADER_FLAG_RELOC); | |
273 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
274 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
275 | else | |
276 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 277 | |
0b246afa | 278 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 279 | |
be20aa9d | 280 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 281 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
e339a6b0 | 282 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 283 | else |
e339a6b0 | 284 | ret = btrfs_inc_ref(trans, root, cow, 0); |
4aec2b52 | 285 | |
be20aa9d CM |
286 | if (ret) |
287 | return ret; | |
288 | ||
289 | btrfs_mark_buffer_dirty(cow); | |
290 | *cow_ret = cow; | |
291 | return 0; | |
292 | } | |
293 | ||
bd989ba3 JS |
294 | enum mod_log_op { |
295 | MOD_LOG_KEY_REPLACE, | |
296 | MOD_LOG_KEY_ADD, | |
297 | MOD_LOG_KEY_REMOVE, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
299 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
300 | MOD_LOG_MOVE_KEYS, | |
301 | MOD_LOG_ROOT_REPLACE, | |
302 | }; | |
303 | ||
304 | struct tree_mod_move { | |
305 | int dst_slot; | |
306 | int nr_items; | |
307 | }; | |
308 | ||
309 | struct tree_mod_root { | |
310 | u64 logical; | |
311 | u8 level; | |
312 | }; | |
313 | ||
314 | struct tree_mod_elem { | |
315 | struct rb_node node; | |
298cfd36 | 316 | u64 logical; |
097b8a7c | 317 | u64 seq; |
bd989ba3 JS |
318 | enum mod_log_op op; |
319 | ||
320 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
321 | int slot; | |
322 | ||
323 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
324 | u64 generation; | |
325 | ||
326 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
327 | struct btrfs_disk_key key; | |
328 | u64 blockptr; | |
329 | ||
330 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
331 | struct tree_mod_move move; | |
332 | ||
333 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
334 | struct tree_mod_root old_root; | |
335 | }; | |
336 | ||
097b8a7c | 337 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 338 | { |
097b8a7c | 339 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
340 | } |
341 | ||
097b8a7c JS |
342 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
343 | { | |
344 | read_unlock(&fs_info->tree_mod_log_lock); | |
345 | } | |
346 | ||
347 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
348 | { | |
349 | write_lock(&fs_info->tree_mod_log_lock); | |
350 | } | |
351 | ||
352 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
353 | { | |
354 | write_unlock(&fs_info->tree_mod_log_lock); | |
355 | } | |
356 | ||
fc36ed7e | 357 | /* |
fcebe456 | 358 | * Pull a new tree mod seq number for our operation. |
fc36ed7e | 359 | */ |
fcebe456 | 360 | static inline u64 btrfs_inc_tree_mod_seq(struct btrfs_fs_info *fs_info) |
fc36ed7e JS |
361 | { |
362 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
363 | } | |
364 | ||
097b8a7c JS |
365 | /* |
366 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
367 | * passed does not already have a sequence number set. So when a caller expects | |
368 | * to record tree modifications, it should ensure to set elem->seq to zero | |
369 | * before calling btrfs_get_tree_mod_seq. | |
370 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
371 | * blocker was added. | |
372 | */ | |
373 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
374 | struct seq_list *elem) | |
bd989ba3 | 375 | { |
097b8a7c | 376 | tree_mod_log_write_lock(fs_info); |
bd989ba3 | 377 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 378 | if (!elem->seq) { |
fcebe456 | 379 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); |
097b8a7c JS |
380 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
381 | } | |
bd989ba3 | 382 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
383 | tree_mod_log_write_unlock(fs_info); |
384 | ||
fcebe456 | 385 | return elem->seq; |
bd989ba3 JS |
386 | } |
387 | ||
388 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
389 | struct seq_list *elem) | |
390 | { | |
391 | struct rb_root *tm_root; | |
392 | struct rb_node *node; | |
393 | struct rb_node *next; | |
394 | struct seq_list *cur_elem; | |
395 | struct tree_mod_elem *tm; | |
396 | u64 min_seq = (u64)-1; | |
397 | u64 seq_putting = elem->seq; | |
398 | ||
399 | if (!seq_putting) | |
400 | return; | |
401 | ||
bd989ba3 JS |
402 | spin_lock(&fs_info->tree_mod_seq_lock); |
403 | list_del(&elem->list); | |
097b8a7c | 404 | elem->seq = 0; |
bd989ba3 JS |
405 | |
406 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 407 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
408 | if (seq_putting > cur_elem->seq) { |
409 | /* | |
410 | * blocker with lower sequence number exists, we | |
411 | * cannot remove anything from the log | |
412 | */ | |
097b8a7c JS |
413 | spin_unlock(&fs_info->tree_mod_seq_lock); |
414 | return; | |
bd989ba3 JS |
415 | } |
416 | min_seq = cur_elem->seq; | |
417 | } | |
418 | } | |
097b8a7c JS |
419 | spin_unlock(&fs_info->tree_mod_seq_lock); |
420 | ||
bd989ba3 JS |
421 | /* |
422 | * anything that's lower than the lowest existing (read: blocked) | |
423 | * sequence number can be removed from the tree. | |
424 | */ | |
097b8a7c | 425 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
426 | tm_root = &fs_info->tree_mod_log; |
427 | for (node = rb_first(tm_root); node; node = next) { | |
428 | next = rb_next(node); | |
6b4df8b6 | 429 | tm = rb_entry(node, struct tree_mod_elem, node); |
097b8a7c | 430 | if (tm->seq > min_seq) |
bd989ba3 JS |
431 | continue; |
432 | rb_erase(node, tm_root); | |
bd989ba3 JS |
433 | kfree(tm); |
434 | } | |
097b8a7c | 435 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
436 | } |
437 | ||
438 | /* | |
439 | * key order of the log: | |
298cfd36 | 440 | * node/leaf start address -> sequence |
bd989ba3 | 441 | * |
298cfd36 CR |
442 | * The 'start address' is the logical address of the *new* root node |
443 | * for root replace operations, or the logical address of the affected | |
444 | * block for all other operations. | |
5de865ee FDBM |
445 | * |
446 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
447 | */ |
448 | static noinline int | |
449 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
450 | { | |
451 | struct rb_root *tm_root; | |
452 | struct rb_node **new; | |
453 | struct rb_node *parent = NULL; | |
454 | struct tree_mod_elem *cur; | |
c8cc6341 | 455 | |
fcebe456 | 456 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
bd989ba3 | 457 | |
bd989ba3 JS |
458 | tm_root = &fs_info->tree_mod_log; |
459 | new = &tm_root->rb_node; | |
460 | while (*new) { | |
6b4df8b6 | 461 | cur = rb_entry(*new, struct tree_mod_elem, node); |
bd989ba3 | 462 | parent = *new; |
298cfd36 | 463 | if (cur->logical < tm->logical) |
bd989ba3 | 464 | new = &((*new)->rb_left); |
298cfd36 | 465 | else if (cur->logical > tm->logical) |
bd989ba3 | 466 | new = &((*new)->rb_right); |
097b8a7c | 467 | else if (cur->seq < tm->seq) |
bd989ba3 | 468 | new = &((*new)->rb_left); |
097b8a7c | 469 | else if (cur->seq > tm->seq) |
bd989ba3 | 470 | new = &((*new)->rb_right); |
5de865ee FDBM |
471 | else |
472 | return -EEXIST; | |
bd989ba3 JS |
473 | } |
474 | ||
475 | rb_link_node(&tm->node, parent, new); | |
476 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 477 | return 0; |
bd989ba3 JS |
478 | } |
479 | ||
097b8a7c JS |
480 | /* |
481 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
482 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
483 | * this until all tree mod log insertions are recorded in the rb tree and then | |
484 | * call tree_mod_log_write_unlock() to release. | |
485 | */ | |
e9b7fd4d JS |
486 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
487 | struct extent_buffer *eb) { | |
488 | smp_mb(); | |
489 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
490 | return 1; | |
097b8a7c JS |
491 | if (eb && btrfs_header_level(eb) == 0) |
492 | return 1; | |
5de865ee FDBM |
493 | |
494 | tree_mod_log_write_lock(fs_info); | |
495 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
496 | tree_mod_log_write_unlock(fs_info); | |
497 | return 1; | |
498 | } | |
499 | ||
e9b7fd4d JS |
500 | return 0; |
501 | } | |
502 | ||
5de865ee FDBM |
503 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
504 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
505 | struct extent_buffer *eb) | |
506 | { | |
507 | smp_mb(); | |
508 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
509 | return 0; | |
510 | if (eb && btrfs_header_level(eb) == 0) | |
511 | return 0; | |
512 | ||
513 | return 1; | |
514 | } | |
515 | ||
516 | static struct tree_mod_elem * | |
517 | alloc_tree_mod_elem(struct extent_buffer *eb, int slot, | |
518 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 519 | { |
097b8a7c | 520 | struct tree_mod_elem *tm; |
bd989ba3 | 521 | |
c8cc6341 JB |
522 | tm = kzalloc(sizeof(*tm), flags); |
523 | if (!tm) | |
5de865ee | 524 | return NULL; |
bd989ba3 | 525 | |
298cfd36 | 526 | tm->logical = eb->start; |
bd989ba3 JS |
527 | if (op != MOD_LOG_KEY_ADD) { |
528 | btrfs_node_key(eb, &tm->key, slot); | |
529 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
530 | } | |
531 | tm->op = op; | |
532 | tm->slot = slot; | |
533 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
5de865ee | 534 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 535 | |
5de865ee | 536 | return tm; |
097b8a7c JS |
537 | } |
538 | ||
539 | static noinline int | |
c8cc6341 JB |
540 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
541 | struct extent_buffer *eb, int slot, | |
542 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 543 | { |
5de865ee FDBM |
544 | struct tree_mod_elem *tm; |
545 | int ret; | |
546 | ||
547 | if (!tree_mod_need_log(fs_info, eb)) | |
548 | return 0; | |
549 | ||
550 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
551 | if (!tm) | |
552 | return -ENOMEM; | |
553 | ||
554 | if (tree_mod_dont_log(fs_info, eb)) { | |
555 | kfree(tm); | |
097b8a7c | 556 | return 0; |
5de865ee FDBM |
557 | } |
558 | ||
559 | ret = __tree_mod_log_insert(fs_info, tm); | |
560 | tree_mod_log_write_unlock(fs_info); | |
561 | if (ret) | |
562 | kfree(tm); | |
097b8a7c | 563 | |
5de865ee | 564 | return ret; |
097b8a7c JS |
565 | } |
566 | ||
bd989ba3 JS |
567 | static noinline int |
568 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
569 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
176ef8f5 | 570 | int nr_items) |
bd989ba3 | 571 | { |
5de865ee FDBM |
572 | struct tree_mod_elem *tm = NULL; |
573 | struct tree_mod_elem **tm_list = NULL; | |
574 | int ret = 0; | |
bd989ba3 | 575 | int i; |
5de865ee | 576 | int locked = 0; |
bd989ba3 | 577 | |
5de865ee | 578 | if (!tree_mod_need_log(fs_info, eb)) |
f395694c | 579 | return 0; |
bd989ba3 | 580 | |
176ef8f5 | 581 | tm_list = kcalloc(nr_items, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
582 | if (!tm_list) |
583 | return -ENOMEM; | |
584 | ||
176ef8f5 | 585 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
586 | if (!tm) { |
587 | ret = -ENOMEM; | |
588 | goto free_tms; | |
589 | } | |
590 | ||
298cfd36 | 591 | tm->logical = eb->start; |
5de865ee FDBM |
592 | tm->slot = src_slot; |
593 | tm->move.dst_slot = dst_slot; | |
594 | tm->move.nr_items = nr_items; | |
595 | tm->op = MOD_LOG_MOVE_KEYS; | |
596 | ||
597 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
598 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
176ef8f5 | 599 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); |
5de865ee FDBM |
600 | if (!tm_list[i]) { |
601 | ret = -ENOMEM; | |
602 | goto free_tms; | |
603 | } | |
604 | } | |
605 | ||
606 | if (tree_mod_dont_log(fs_info, eb)) | |
607 | goto free_tms; | |
608 | locked = 1; | |
609 | ||
01763a2e JS |
610 | /* |
611 | * When we override something during the move, we log these removals. | |
612 | * This can only happen when we move towards the beginning of the | |
613 | * buffer, i.e. dst_slot < src_slot. | |
614 | */ | |
bd989ba3 | 615 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
5de865ee FDBM |
616 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
617 | if (ret) | |
618 | goto free_tms; | |
bd989ba3 JS |
619 | } |
620 | ||
5de865ee FDBM |
621 | ret = __tree_mod_log_insert(fs_info, tm); |
622 | if (ret) | |
623 | goto free_tms; | |
624 | tree_mod_log_write_unlock(fs_info); | |
625 | kfree(tm_list); | |
f395694c | 626 | |
5de865ee FDBM |
627 | return 0; |
628 | free_tms: | |
629 | for (i = 0; i < nr_items; i++) { | |
630 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
631 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
632 | kfree(tm_list[i]); | |
633 | } | |
634 | if (locked) | |
635 | tree_mod_log_write_unlock(fs_info); | |
636 | kfree(tm_list); | |
637 | kfree(tm); | |
bd989ba3 | 638 | |
5de865ee | 639 | return ret; |
bd989ba3 JS |
640 | } |
641 | ||
5de865ee FDBM |
642 | static inline int |
643 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
644 | struct tree_mod_elem **tm_list, | |
645 | int nritems) | |
097b8a7c | 646 | { |
5de865ee | 647 | int i, j; |
097b8a7c JS |
648 | int ret; |
649 | ||
097b8a7c | 650 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
651 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
652 | if (ret) { | |
653 | for (j = nritems - 1; j > i; j--) | |
654 | rb_erase(&tm_list[j]->node, | |
655 | &fs_info->tree_mod_log); | |
656 | return ret; | |
657 | } | |
097b8a7c | 658 | } |
5de865ee FDBM |
659 | |
660 | return 0; | |
097b8a7c JS |
661 | } |
662 | ||
bd989ba3 JS |
663 | static noinline int |
664 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
665 | struct extent_buffer *old_root, | |
bcc8e07f | 666 | struct extent_buffer *new_root, |
90f8d62e | 667 | int log_removal) |
bd989ba3 | 668 | { |
5de865ee FDBM |
669 | struct tree_mod_elem *tm = NULL; |
670 | struct tree_mod_elem **tm_list = NULL; | |
671 | int nritems = 0; | |
672 | int ret = 0; | |
673 | int i; | |
bd989ba3 | 674 | |
5de865ee | 675 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
676 | return 0; |
677 | ||
5de865ee FDBM |
678 | if (log_removal && btrfs_header_level(old_root) > 0) { |
679 | nritems = btrfs_header_nritems(old_root); | |
31e818fe | 680 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), |
bcc8e07f | 681 | GFP_NOFS); |
5de865ee FDBM |
682 | if (!tm_list) { |
683 | ret = -ENOMEM; | |
684 | goto free_tms; | |
685 | } | |
686 | for (i = 0; i < nritems; i++) { | |
687 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
bcc8e07f | 688 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); |
5de865ee FDBM |
689 | if (!tm_list[i]) { |
690 | ret = -ENOMEM; | |
691 | goto free_tms; | |
692 | } | |
693 | } | |
694 | } | |
d9abbf1c | 695 | |
bcc8e07f | 696 | tm = kzalloc(sizeof(*tm), GFP_NOFS); |
5de865ee FDBM |
697 | if (!tm) { |
698 | ret = -ENOMEM; | |
699 | goto free_tms; | |
700 | } | |
bd989ba3 | 701 | |
298cfd36 | 702 | tm->logical = new_root->start; |
bd989ba3 JS |
703 | tm->old_root.logical = old_root->start; |
704 | tm->old_root.level = btrfs_header_level(old_root); | |
705 | tm->generation = btrfs_header_generation(old_root); | |
706 | tm->op = MOD_LOG_ROOT_REPLACE; | |
707 | ||
5de865ee FDBM |
708 | if (tree_mod_dont_log(fs_info, NULL)) |
709 | goto free_tms; | |
710 | ||
711 | if (tm_list) | |
712 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
713 | if (!ret) | |
714 | ret = __tree_mod_log_insert(fs_info, tm); | |
715 | ||
716 | tree_mod_log_write_unlock(fs_info); | |
717 | if (ret) | |
718 | goto free_tms; | |
719 | kfree(tm_list); | |
720 | ||
721 | return ret; | |
722 | ||
723 | free_tms: | |
724 | if (tm_list) { | |
725 | for (i = 0; i < nritems; i++) | |
726 | kfree(tm_list[i]); | |
727 | kfree(tm_list); | |
728 | } | |
729 | kfree(tm); | |
730 | ||
731 | return ret; | |
bd989ba3 JS |
732 | } |
733 | ||
734 | static struct tree_mod_elem * | |
735 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
736 | int smallest) | |
737 | { | |
738 | struct rb_root *tm_root; | |
739 | struct rb_node *node; | |
740 | struct tree_mod_elem *cur = NULL; | |
741 | struct tree_mod_elem *found = NULL; | |
bd989ba3 | 742 | |
097b8a7c | 743 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
744 | tm_root = &fs_info->tree_mod_log; |
745 | node = tm_root->rb_node; | |
746 | while (node) { | |
6b4df8b6 | 747 | cur = rb_entry(node, struct tree_mod_elem, node); |
298cfd36 | 748 | if (cur->logical < start) { |
bd989ba3 | 749 | node = node->rb_left; |
298cfd36 | 750 | } else if (cur->logical > start) { |
bd989ba3 | 751 | node = node->rb_right; |
097b8a7c | 752 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
753 | node = node->rb_left; |
754 | } else if (!smallest) { | |
755 | /* we want the node with the highest seq */ | |
756 | if (found) | |
097b8a7c | 757 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
758 | found = cur; |
759 | node = node->rb_left; | |
097b8a7c | 760 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
761 | /* we want the node with the smallest seq */ |
762 | if (found) | |
097b8a7c | 763 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
764 | found = cur; |
765 | node = node->rb_right; | |
766 | } else { | |
767 | found = cur; | |
768 | break; | |
769 | } | |
770 | } | |
097b8a7c | 771 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
772 | |
773 | return found; | |
774 | } | |
775 | ||
776 | /* | |
777 | * this returns the element from the log with the smallest time sequence | |
778 | * value that's in the log (the oldest log item). any element with a time | |
779 | * sequence lower than min_seq will be ignored. | |
780 | */ | |
781 | static struct tree_mod_elem * | |
782 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
783 | u64 min_seq) | |
784 | { | |
785 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
786 | } | |
787 | ||
788 | /* | |
789 | * this returns the element from the log with the largest time sequence | |
790 | * value that's in the log (the most recent log item). any element with | |
791 | * a time sequence lower than min_seq will be ignored. | |
792 | */ | |
793 | static struct tree_mod_elem * | |
794 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
795 | { | |
796 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
797 | } | |
798 | ||
5de865ee | 799 | static noinline int |
bd989ba3 JS |
800 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
801 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 802 | unsigned long src_offset, int nr_items) |
bd989ba3 | 803 | { |
5de865ee FDBM |
804 | int ret = 0; |
805 | struct tree_mod_elem **tm_list = NULL; | |
806 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 807 | int i; |
5de865ee | 808 | int locked = 0; |
bd989ba3 | 809 | |
5de865ee FDBM |
810 | if (!tree_mod_need_log(fs_info, NULL)) |
811 | return 0; | |
bd989ba3 | 812 | |
c8cc6341 | 813 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
814 | return 0; |
815 | ||
31e818fe | 816 | tm_list = kcalloc(nr_items * 2, sizeof(struct tree_mod_elem *), |
5de865ee FDBM |
817 | GFP_NOFS); |
818 | if (!tm_list) | |
819 | return -ENOMEM; | |
bd989ba3 | 820 | |
5de865ee FDBM |
821 | tm_list_add = tm_list; |
822 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 823 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
824 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
825 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
826 | if (!tm_list_rem[i]) { | |
827 | ret = -ENOMEM; | |
828 | goto free_tms; | |
829 | } | |
830 | ||
831 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
832 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
833 | if (!tm_list_add[i]) { | |
834 | ret = -ENOMEM; | |
835 | goto free_tms; | |
836 | } | |
837 | } | |
838 | ||
839 | if (tree_mod_dont_log(fs_info, NULL)) | |
840 | goto free_tms; | |
841 | locked = 1; | |
842 | ||
843 | for (i = 0; i < nr_items; i++) { | |
844 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
845 | if (ret) | |
846 | goto free_tms; | |
847 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
848 | if (ret) | |
849 | goto free_tms; | |
bd989ba3 | 850 | } |
5de865ee FDBM |
851 | |
852 | tree_mod_log_write_unlock(fs_info); | |
853 | kfree(tm_list); | |
854 | ||
855 | return 0; | |
856 | ||
857 | free_tms: | |
858 | for (i = 0; i < nr_items * 2; i++) { | |
859 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
860 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
861 | kfree(tm_list[i]); | |
862 | } | |
863 | if (locked) | |
864 | tree_mod_log_write_unlock(fs_info); | |
865 | kfree(tm_list); | |
866 | ||
867 | return ret; | |
bd989ba3 JS |
868 | } |
869 | ||
870 | static inline void | |
871 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
872 | int dst_offset, int src_offset, int nr_items) | |
873 | { | |
874 | int ret; | |
875 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
176ef8f5 | 876 | nr_items); |
bd989ba3 JS |
877 | BUG_ON(ret < 0); |
878 | } | |
879 | ||
097b8a7c | 880 | static noinline void |
bd989ba3 | 881 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 882 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
883 | { |
884 | int ret; | |
885 | ||
78357766 | 886 | ret = tree_mod_log_insert_key(fs_info, eb, slot, |
c8cc6341 JB |
887 | MOD_LOG_KEY_REPLACE, |
888 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
889 | BUG_ON(ret < 0); |
890 | } | |
891 | ||
5de865ee | 892 | static noinline int |
097b8a7c | 893 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 894 | { |
5de865ee FDBM |
895 | struct tree_mod_elem **tm_list = NULL; |
896 | int nritems = 0; | |
897 | int i; | |
898 | int ret = 0; | |
899 | ||
900 | if (btrfs_header_level(eb) == 0) | |
901 | return 0; | |
902 | ||
903 | if (!tree_mod_need_log(fs_info, NULL)) | |
904 | return 0; | |
905 | ||
906 | nritems = btrfs_header_nritems(eb); | |
31e818fe | 907 | tm_list = kcalloc(nritems, sizeof(struct tree_mod_elem *), GFP_NOFS); |
5de865ee FDBM |
908 | if (!tm_list) |
909 | return -ENOMEM; | |
910 | ||
911 | for (i = 0; i < nritems; i++) { | |
912 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
913 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
914 | if (!tm_list[i]) { | |
915 | ret = -ENOMEM; | |
916 | goto free_tms; | |
917 | } | |
918 | } | |
919 | ||
e9b7fd4d | 920 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
921 | goto free_tms; |
922 | ||
923 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
924 | tree_mod_log_write_unlock(fs_info); | |
925 | if (ret) | |
926 | goto free_tms; | |
927 | kfree(tm_list); | |
928 | ||
929 | return 0; | |
930 | ||
931 | free_tms: | |
932 | for (i = 0; i < nritems; i++) | |
933 | kfree(tm_list[i]); | |
934 | kfree(tm_list); | |
935 | ||
936 | return ret; | |
bd989ba3 JS |
937 | } |
938 | ||
097b8a7c | 939 | static noinline void |
bd989ba3 | 940 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
941 | struct extent_buffer *new_root_node, |
942 | int log_removal) | |
bd989ba3 JS |
943 | { |
944 | int ret; | |
bd989ba3 | 945 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
bcc8e07f | 946 | new_root_node, log_removal); |
bd989ba3 JS |
947 | BUG_ON(ret < 0); |
948 | } | |
949 | ||
5d4f98a2 YZ |
950 | /* |
951 | * check if the tree block can be shared by multiple trees | |
952 | */ | |
953 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
954 | struct extent_buffer *buf) | |
955 | { | |
956 | /* | |
01327610 | 957 | * Tree blocks not in reference counted trees and tree roots |
5d4f98a2 YZ |
958 | * are never shared. If a block was allocated after the last |
959 | * snapshot and the block was not allocated by tree relocation, | |
960 | * we know the block is not shared. | |
961 | */ | |
27cdeb70 | 962 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
963 | buf != root->node && buf != root->commit_root && |
964 | (btrfs_header_generation(buf) <= | |
965 | btrfs_root_last_snapshot(&root->root_item) || | |
966 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
967 | return 1; | |
968 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
27cdeb70 | 969 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
5d4f98a2 YZ |
970 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) |
971 | return 1; | |
972 | #endif | |
973 | return 0; | |
974 | } | |
975 | ||
976 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
977 | struct btrfs_root *root, | |
978 | struct extent_buffer *buf, | |
f0486c68 YZ |
979 | struct extent_buffer *cow, |
980 | int *last_ref) | |
5d4f98a2 | 981 | { |
0b246afa | 982 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
983 | u64 refs; |
984 | u64 owner; | |
985 | u64 flags; | |
986 | u64 new_flags = 0; | |
987 | int ret; | |
988 | ||
989 | /* | |
990 | * Backrefs update rules: | |
991 | * | |
992 | * Always use full backrefs for extent pointers in tree block | |
993 | * allocated by tree relocation. | |
994 | * | |
995 | * If a shared tree block is no longer referenced by its owner | |
996 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
997 | * use full backrefs for extent pointers in tree block. | |
998 | * | |
999 | * If a tree block is been relocating | |
1000 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
1001 | * use full backrefs for extent pointers in tree block. | |
1002 | * The reason for this is some operations (such as drop tree) | |
1003 | * are only allowed for blocks use full backrefs. | |
1004 | */ | |
1005 | ||
1006 | if (btrfs_block_can_be_shared(root, buf)) { | |
2ff7e61e | 1007 | ret = btrfs_lookup_extent_info(trans, fs_info, buf->start, |
3173a18f JB |
1008 | btrfs_header_level(buf), 1, |
1009 | &refs, &flags); | |
be1a5564 MF |
1010 | if (ret) |
1011 | return ret; | |
e5df9573 MF |
1012 | if (refs == 0) { |
1013 | ret = -EROFS; | |
0b246afa | 1014 | btrfs_handle_fs_error(fs_info, ret, NULL); |
e5df9573 MF |
1015 | return ret; |
1016 | } | |
5d4f98a2 YZ |
1017 | } else { |
1018 | refs = 1; | |
1019 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1020 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1021 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1022 | else | |
1023 | flags = 0; | |
1024 | } | |
1025 | ||
1026 | owner = btrfs_header_owner(buf); | |
1027 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1028 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1029 | ||
1030 | if (refs > 1) { | |
1031 | if ((owner == root->root_key.objectid || | |
1032 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1033 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
e339a6b0 | 1034 | ret = btrfs_inc_ref(trans, root, buf, 1); |
79787eaa | 1035 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1036 | |
1037 | if (root->root_key.objectid == | |
1038 | BTRFS_TREE_RELOC_OBJECTID) { | |
e339a6b0 | 1039 | ret = btrfs_dec_ref(trans, root, buf, 0); |
79787eaa | 1040 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1041 | ret = btrfs_inc_ref(trans, root, cow, 1); |
79787eaa | 1042 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1043 | } |
1044 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1045 | } else { | |
1046 | ||
1047 | if (root->root_key.objectid == | |
1048 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1049 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1050 | else |
e339a6b0 | 1051 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1052 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1053 | } |
1054 | if (new_flags != 0) { | |
b1c79e09 JB |
1055 | int level = btrfs_header_level(buf); |
1056 | ||
2ff7e61e | 1057 | ret = btrfs_set_disk_extent_flags(trans, fs_info, |
5d4f98a2 YZ |
1058 | buf->start, |
1059 | buf->len, | |
b1c79e09 | 1060 | new_flags, level, 0); |
be1a5564 MF |
1061 | if (ret) |
1062 | return ret; | |
5d4f98a2 YZ |
1063 | } |
1064 | } else { | |
1065 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1066 | if (root->root_key.objectid == | |
1067 | BTRFS_TREE_RELOC_OBJECTID) | |
e339a6b0 | 1068 | ret = btrfs_inc_ref(trans, root, cow, 1); |
5d4f98a2 | 1069 | else |
e339a6b0 | 1070 | ret = btrfs_inc_ref(trans, root, cow, 0); |
79787eaa | 1071 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 1072 | ret = btrfs_dec_ref(trans, root, buf, 1); |
79787eaa | 1073 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 | 1074 | } |
7c302b49 | 1075 | clean_tree_block(fs_info, buf); |
f0486c68 | 1076 | *last_ref = 1; |
5d4f98a2 YZ |
1077 | } |
1078 | return 0; | |
1079 | } | |
1080 | ||
d352ac68 | 1081 | /* |
d397712b CM |
1082 | * does the dirty work in cow of a single block. The parent block (if |
1083 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1084 | * dirty and returned locked. If you modify the block it needs to be marked | |
1085 | * dirty again. | |
d352ac68 CM |
1086 | * |
1087 | * search_start -- an allocation hint for the new block | |
1088 | * | |
d397712b CM |
1089 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1090 | * bytes the allocator should try to find free next to the block it returns. | |
1091 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1092 | */ |
d397712b | 1093 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1094 | struct btrfs_root *root, |
1095 | struct extent_buffer *buf, | |
1096 | struct extent_buffer *parent, int parent_slot, | |
1097 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1098 | u64 search_start, u64 empty_size) |
02217ed2 | 1099 | { |
0b246afa | 1100 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 | 1101 | struct btrfs_disk_key disk_key; |
5f39d397 | 1102 | struct extent_buffer *cow; |
be1a5564 | 1103 | int level, ret; |
f0486c68 | 1104 | int last_ref = 0; |
925baedd | 1105 | int unlock_orig = 0; |
0f5053eb | 1106 | u64 parent_start = 0; |
7bb86316 | 1107 | |
925baedd CM |
1108 | if (*cow_ret == buf) |
1109 | unlock_orig = 1; | |
1110 | ||
b9447ef8 | 1111 | btrfs_assert_tree_locked(buf); |
925baedd | 1112 | |
27cdeb70 | 1113 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
0b246afa | 1114 | trans->transid != fs_info->running_transaction->transid); |
27cdeb70 MX |
1115 | WARN_ON(test_bit(BTRFS_ROOT_REF_COWS, &root->state) && |
1116 | trans->transid != root->last_trans); | |
5f39d397 | 1117 | |
7bb86316 | 1118 | level = btrfs_header_level(buf); |
31840ae1 | 1119 | |
5d4f98a2 YZ |
1120 | if (level == 0) |
1121 | btrfs_item_key(buf, &disk_key, 0); | |
1122 | else | |
1123 | btrfs_node_key(buf, &disk_key, 0); | |
1124 | ||
0f5053eb GR |
1125 | if ((root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && parent) |
1126 | parent_start = parent->start; | |
5d4f98a2 | 1127 | |
4d75f8a9 DS |
1128 | cow = btrfs_alloc_tree_block(trans, root, parent_start, |
1129 | root->root_key.objectid, &disk_key, level, | |
1130 | search_start, empty_size); | |
54aa1f4d CM |
1131 | if (IS_ERR(cow)) |
1132 | return PTR_ERR(cow); | |
6702ed49 | 1133 | |
b4ce94de CM |
1134 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1135 | ||
58e8012c | 1136 | copy_extent_buffer_full(cow, buf); |
db94535d | 1137 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1138 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1139 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1140 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1141 | BTRFS_HEADER_FLAG_RELOC); | |
1142 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1143 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1144 | else | |
1145 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1146 | |
0b246afa | 1147 | write_extent_buffer_fsid(cow, fs_info->fsid); |
2b82032c | 1148 | |
be1a5564 | 1149 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1150 | if (ret) { |
66642832 | 1151 | btrfs_abort_transaction(trans, ret); |
b68dc2a9 MF |
1152 | return ret; |
1153 | } | |
1a40e23b | 1154 | |
27cdeb70 | 1155 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { |
83d4cfd4 | 1156 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); |
93314e3b | 1157 | if (ret) { |
66642832 | 1158 | btrfs_abort_transaction(trans, ret); |
83d4cfd4 | 1159 | return ret; |
93314e3b | 1160 | } |
83d4cfd4 | 1161 | } |
3fd0a558 | 1162 | |
02217ed2 | 1163 | if (buf == root->node) { |
925baedd | 1164 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1165 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1166 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1167 | parent_start = buf->start; | |
925baedd | 1168 | |
5f39d397 | 1169 | extent_buffer_get(cow); |
90f8d62e | 1170 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1171 | rcu_assign_pointer(root->node, cow); |
925baedd | 1172 | |
f0486c68 | 1173 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1174 | last_ref); |
5f39d397 | 1175 | free_extent_buffer(buf); |
0b86a832 | 1176 | add_root_to_dirty_list(root); |
02217ed2 | 1177 | } else { |
5d4f98a2 | 1178 | WARN_ON(trans->transid != btrfs_header_generation(parent)); |
0b246afa | 1179 | tree_mod_log_insert_key(fs_info, parent, parent_slot, |
c8cc6341 | 1180 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1181 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1182 | cow->start); |
74493f7a CM |
1183 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1184 | trans->transid); | |
d6025579 | 1185 | btrfs_mark_buffer_dirty(parent); |
5de865ee | 1186 | if (last_ref) { |
0b246afa | 1187 | ret = tree_mod_log_free_eb(fs_info, buf); |
5de865ee | 1188 | if (ret) { |
66642832 | 1189 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
1190 | return ret; |
1191 | } | |
1192 | } | |
f0486c68 | 1193 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1194 | last_ref); |
02217ed2 | 1195 | } |
925baedd CM |
1196 | if (unlock_orig) |
1197 | btrfs_tree_unlock(buf); | |
3083ee2e | 1198 | free_extent_buffer_stale(buf); |
ccd467d6 | 1199 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1200 | *cow_ret = cow; |
02217ed2 CM |
1201 | return 0; |
1202 | } | |
1203 | ||
5d9e75c4 JS |
1204 | /* |
1205 | * returns the logical address of the oldest predecessor of the given root. | |
1206 | * entries older than time_seq are ignored. | |
1207 | */ | |
1208 | static struct tree_mod_elem * | |
1209 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1210 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1211 | { |
1212 | struct tree_mod_elem *tm; | |
1213 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1214 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1215 | int looped = 0; |
1216 | ||
1217 | if (!time_seq) | |
35a3621b | 1218 | return NULL; |
5d9e75c4 JS |
1219 | |
1220 | /* | |
298cfd36 CR |
1221 | * the very last operation that's logged for a root is the |
1222 | * replacement operation (if it is replaced at all). this has | |
1223 | * the logical address of the *new* root, making it the very | |
1224 | * first operation that's logged for this root. | |
5d9e75c4 JS |
1225 | */ |
1226 | while (1) { | |
1227 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1228 | time_seq); | |
1229 | if (!looped && !tm) | |
35a3621b | 1230 | return NULL; |
5d9e75c4 | 1231 | /* |
28da9fb4 JS |
1232 | * if there are no tree operation for the oldest root, we simply |
1233 | * return it. this should only happen if that (old) root is at | |
1234 | * level 0. | |
5d9e75c4 | 1235 | */ |
28da9fb4 JS |
1236 | if (!tm) |
1237 | break; | |
5d9e75c4 | 1238 | |
28da9fb4 JS |
1239 | /* |
1240 | * if there's an operation that's not a root replacement, we | |
1241 | * found the oldest version of our root. normally, we'll find a | |
1242 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1243 | */ | |
5d9e75c4 JS |
1244 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1245 | break; | |
1246 | ||
1247 | found = tm; | |
1248 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1249 | looped = 1; |
1250 | } | |
1251 | ||
a95236d9 JS |
1252 | /* if there's no old root to return, return what we found instead */ |
1253 | if (!found) | |
1254 | found = tm; | |
1255 | ||
5d9e75c4 JS |
1256 | return found; |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * tm is a pointer to the first operation to rewind within eb. then, all | |
01327610 | 1261 | * previous operations will be rewound (until we reach something older than |
5d9e75c4 JS |
1262 | * time_seq). |
1263 | */ | |
1264 | static void | |
f1ca7e98 JB |
1265 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1266 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1267 | { |
1268 | u32 n; | |
1269 | struct rb_node *next; | |
1270 | struct tree_mod_elem *tm = first_tm; | |
1271 | unsigned long o_dst; | |
1272 | unsigned long o_src; | |
1273 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1274 | ||
1275 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1276 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1277 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1278 | /* |
1279 | * all the operations are recorded with the operator used for | |
1280 | * the modification. as we're going backwards, we do the | |
1281 | * opposite of each operation here. | |
1282 | */ | |
1283 | switch (tm->op) { | |
1284 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1285 | BUG_ON(tm->slot < n); | |
1c697d4a | 1286 | /* Fallthrough */ |
95c80bb1 | 1287 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1288 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1289 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1290 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1291 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1292 | tm->generation); | |
4c3e6969 | 1293 | n++; |
5d9e75c4 JS |
1294 | break; |
1295 | case MOD_LOG_KEY_REPLACE: | |
1296 | BUG_ON(tm->slot >= n); | |
1297 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1298 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1299 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1300 | tm->generation); | |
1301 | break; | |
1302 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1303 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1304 | n--; |
1305 | break; | |
1306 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1307 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1308 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1309 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1310 | tm->move.nr_items * p_size); |
1311 | break; | |
1312 | case MOD_LOG_ROOT_REPLACE: | |
1313 | /* | |
1314 | * this operation is special. for roots, this must be | |
1315 | * handled explicitly before rewinding. | |
1316 | * for non-roots, this operation may exist if the node | |
1317 | * was a root: root A -> child B; then A gets empty and | |
1318 | * B is promoted to the new root. in the mod log, we'll | |
1319 | * have a root-replace operation for B, a tree block | |
1320 | * that is no root. we simply ignore that operation. | |
1321 | */ | |
1322 | break; | |
1323 | } | |
1324 | next = rb_next(&tm->node); | |
1325 | if (!next) | |
1326 | break; | |
6b4df8b6 | 1327 | tm = rb_entry(next, struct tree_mod_elem, node); |
298cfd36 | 1328 | if (tm->logical != first_tm->logical) |
5d9e75c4 JS |
1329 | break; |
1330 | } | |
f1ca7e98 | 1331 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1332 | btrfs_set_header_nritems(eb, n); |
1333 | } | |
1334 | ||
47fb091f | 1335 | /* |
01327610 | 1336 | * Called with eb read locked. If the buffer cannot be rewound, the same buffer |
47fb091f JS |
1337 | * is returned. If rewind operations happen, a fresh buffer is returned. The |
1338 | * returned buffer is always read-locked. If the returned buffer is not the | |
1339 | * input buffer, the lock on the input buffer is released and the input buffer | |
1340 | * is freed (its refcount is decremented). | |
1341 | */ | |
5d9e75c4 | 1342 | static struct extent_buffer * |
9ec72677 JB |
1343 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1344 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1345 | { |
1346 | struct extent_buffer *eb_rewin; | |
1347 | struct tree_mod_elem *tm; | |
1348 | ||
1349 | if (!time_seq) | |
1350 | return eb; | |
1351 | ||
1352 | if (btrfs_header_level(eb) == 0) | |
1353 | return eb; | |
1354 | ||
1355 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1356 | if (!tm) | |
1357 | return eb; | |
1358 | ||
9ec72677 JB |
1359 | btrfs_set_path_blocking(path); |
1360 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1361 | ||
5d9e75c4 JS |
1362 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1363 | BUG_ON(tm->slot != 0); | |
da17066c | 1364 | eb_rewin = alloc_dummy_extent_buffer(fs_info, eb->start); |
db7f3436 | 1365 | if (!eb_rewin) { |
9ec72677 | 1366 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1367 | free_extent_buffer(eb); |
1368 | return NULL; | |
1369 | } | |
5d9e75c4 JS |
1370 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1371 | btrfs_set_header_backref_rev(eb_rewin, | |
1372 | btrfs_header_backref_rev(eb)); | |
1373 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1374 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1375 | } else { |
1376 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1377 | if (!eb_rewin) { |
9ec72677 | 1378 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1379 | free_extent_buffer(eb); |
1380 | return NULL; | |
1381 | } | |
5d9e75c4 JS |
1382 | } |
1383 | ||
9ec72677 JB |
1384 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1385 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1386 | free_extent_buffer(eb); |
1387 | ||
47fb091f JS |
1388 | extent_buffer_get(eb_rewin); |
1389 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1390 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1391 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
da17066c | 1392 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1393 | |
1394 | return eb_rewin; | |
1395 | } | |
1396 | ||
8ba97a15 JS |
1397 | /* |
1398 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1399 | * value. If there are no changes, the current root->root_node is returned. If | |
1400 | * anything changed in between, there's a fresh buffer allocated on which the | |
1401 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1402 | * Returns NULL on error (with no locks held). | |
1403 | */ | |
5d9e75c4 JS |
1404 | static inline struct extent_buffer * |
1405 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1406 | { | |
0b246afa | 1407 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 | 1408 | struct tree_mod_elem *tm; |
30b0463a JS |
1409 | struct extent_buffer *eb = NULL; |
1410 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1411 | struct extent_buffer *old; |
a95236d9 | 1412 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1413 | u64 old_generation = 0; |
a95236d9 | 1414 | u64 logical; |
5d9e75c4 | 1415 | |
30b0463a | 1416 | eb_root = btrfs_read_lock_root_node(root); |
0b246afa | 1417 | tm = __tree_mod_log_oldest_root(fs_info, eb_root, time_seq); |
5d9e75c4 | 1418 | if (!tm) |
30b0463a | 1419 | return eb_root; |
5d9e75c4 | 1420 | |
a95236d9 JS |
1421 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1422 | old_root = &tm->old_root; | |
1423 | old_generation = tm->generation; | |
1424 | logical = old_root->logical; | |
1425 | } else { | |
30b0463a | 1426 | logical = eb_root->start; |
a95236d9 | 1427 | } |
5d9e75c4 | 1428 | |
0b246afa | 1429 | tm = tree_mod_log_search(fs_info, logical, time_seq); |
834328a8 | 1430 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1431 | btrfs_tree_read_unlock(eb_root); |
1432 | free_extent_buffer(eb_root); | |
2ff7e61e | 1433 | old = read_tree_block(fs_info, logical, 0); |
64c043de LB |
1434 | if (WARN_ON(IS_ERR(old) || !extent_buffer_uptodate(old))) { |
1435 | if (!IS_ERR(old)) | |
1436 | free_extent_buffer(old); | |
0b246afa JM |
1437 | btrfs_warn(fs_info, |
1438 | "failed to read tree block %llu from get_old_root", | |
1439 | logical); | |
834328a8 | 1440 | } else { |
7bfdcf7f LB |
1441 | eb = btrfs_clone_extent_buffer(old); |
1442 | free_extent_buffer(old); | |
834328a8 JS |
1443 | } |
1444 | } else if (old_root) { | |
30b0463a JS |
1445 | btrfs_tree_read_unlock(eb_root); |
1446 | free_extent_buffer(eb_root); | |
0b246afa | 1447 | eb = alloc_dummy_extent_buffer(fs_info, logical); |
834328a8 | 1448 | } else { |
9ec72677 | 1449 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1450 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1451 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1452 | free_extent_buffer(eb_root); |
834328a8 JS |
1453 | } |
1454 | ||
8ba97a15 JS |
1455 | if (!eb) |
1456 | return NULL; | |
d6381084 | 1457 | extent_buffer_get(eb); |
8ba97a15 | 1458 | btrfs_tree_read_lock(eb); |
a95236d9 | 1459 | if (old_root) { |
5d9e75c4 JS |
1460 | btrfs_set_header_bytenr(eb, eb->start); |
1461 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1462 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1463 | btrfs_set_header_level(eb, old_root->level); |
1464 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1465 | } |
28da9fb4 | 1466 | if (tm) |
0b246afa | 1467 | __tree_mod_log_rewind(fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1468 | else |
1469 | WARN_ON(btrfs_header_level(eb) != 0); | |
0b246afa | 1470 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
5d9e75c4 JS |
1471 | |
1472 | return eb; | |
1473 | } | |
1474 | ||
5b6602e7 JS |
1475 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1476 | { | |
1477 | struct tree_mod_elem *tm; | |
1478 | int level; | |
30b0463a | 1479 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1480 | |
30b0463a | 1481 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1482 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1483 | level = tm->old_root.level; | |
1484 | } else { | |
30b0463a | 1485 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1486 | } |
30b0463a | 1487 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1488 | |
1489 | return level; | |
1490 | } | |
1491 | ||
5d4f98a2 YZ |
1492 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1493 | struct btrfs_root *root, | |
1494 | struct extent_buffer *buf) | |
1495 | { | |
f5ee5c9a | 1496 | if (btrfs_is_testing(root->fs_info)) |
faa2dbf0 | 1497 | return 0; |
fccb84c9 | 1498 | |
f1ebcc74 LB |
1499 | /* ensure we can see the force_cow */ |
1500 | smp_rmb(); | |
1501 | ||
1502 | /* | |
1503 | * We do not need to cow a block if | |
1504 | * 1) this block is not created or changed in this transaction; | |
1505 | * 2) this block does not belong to TREE_RELOC tree; | |
1506 | * 3) the root is not forced COW. | |
1507 | * | |
1508 | * What is forced COW: | |
01327610 | 1509 | * when we create snapshot during committing the transaction, |
f1ebcc74 LB |
1510 | * after we've finished coping src root, we must COW the shared |
1511 | * block to ensure the metadata consistency. | |
1512 | */ | |
5d4f98a2 YZ |
1513 | if (btrfs_header_generation(buf) == trans->transid && |
1514 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1515 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 | 1516 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
27cdeb70 | 1517 | !test_bit(BTRFS_ROOT_FORCE_COW, &root->state)) |
5d4f98a2 YZ |
1518 | return 0; |
1519 | return 1; | |
1520 | } | |
1521 | ||
d352ac68 CM |
1522 | /* |
1523 | * cows a single block, see __btrfs_cow_block for the real work. | |
01327610 | 1524 | * This version of it has extra checks so that a block isn't COWed more than |
d352ac68 CM |
1525 | * once per transaction, as long as it hasn't been written yet |
1526 | */ | |
d397712b | 1527 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1528 | struct btrfs_root *root, struct extent_buffer *buf, |
1529 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1530 | struct extent_buffer **cow_ret) |
6702ed49 | 1531 | { |
0b246afa | 1532 | struct btrfs_fs_info *fs_info = root->fs_info; |
6702ed49 | 1533 | u64 search_start; |
f510cfec | 1534 | int ret; |
dc17ff8f | 1535 | |
0b246afa | 1536 | if (trans->transaction != fs_info->running_transaction) |
31b1a2bd | 1537 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
c1c9ff7c | 1538 | trans->transid, |
0b246afa | 1539 | fs_info->running_transaction->transid); |
31b1a2bd | 1540 | |
0b246afa | 1541 | if (trans->transid != fs_info->generation) |
31b1a2bd | 1542 | WARN(1, KERN_CRIT "trans %llu running %llu\n", |
0b246afa | 1543 | trans->transid, fs_info->generation); |
dc17ff8f | 1544 | |
5d4f98a2 | 1545 | if (!should_cow_block(trans, root, buf)) { |
64c12921 | 1546 | trans->dirty = true; |
6702ed49 CM |
1547 | *cow_ret = buf; |
1548 | return 0; | |
1549 | } | |
c487685d | 1550 | |
ee22184b | 1551 | search_start = buf->start & ~((u64)SZ_1G - 1); |
b4ce94de CM |
1552 | |
1553 | if (parent) | |
1554 | btrfs_set_lock_blocking(parent); | |
1555 | btrfs_set_lock_blocking(buf); | |
1556 | ||
f510cfec | 1557 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1558 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1559 | |
1560 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1561 | ||
f510cfec | 1562 | return ret; |
6702ed49 CM |
1563 | } |
1564 | ||
d352ac68 CM |
1565 | /* |
1566 | * helper function for defrag to decide if two blocks pointed to by a | |
1567 | * node are actually close by | |
1568 | */ | |
6b80053d | 1569 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1570 | { |
6b80053d | 1571 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1572 | return 1; |
6b80053d | 1573 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1574 | return 1; |
1575 | return 0; | |
1576 | } | |
1577 | ||
081e9573 CM |
1578 | /* |
1579 | * compare two keys in a memcmp fashion | |
1580 | */ | |
310712b2 OS |
1581 | static int comp_keys(const struct btrfs_disk_key *disk, |
1582 | const struct btrfs_key *k2) | |
081e9573 CM |
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 | */ | |
310712b2 | 1594 | int btrfs_comp_cpu_keys(const struct btrfs_key *k1, const 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) { |
2ff7e61e | 1685 | cur = read_tree_block(fs_info, 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 | 1733 | static noinline int generic_bin_search(struct extent_buffer *eb, |
310712b2 OS |
1734 | unsigned long p, int item_size, |
1735 | const struct btrfs_key *key, | |
e02119d5 | 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 | */ | |
310712b2 | 1804 | static int bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
5f39d397 | 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 | ||
310712b2 | 1821 | int btrfs_bin_search(struct extent_buffer *eb, const struct btrfs_key *key, |
5d4f98a2 YZ |
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 | */ |
2ff7e61e JM |
1846 | static noinline struct extent_buffer * |
1847 | read_node_slot(struct btrfs_fs_info *fs_info, struct extent_buffer *parent, | |
1848 | int slot) | |
bb803951 | 1849 | { |
ca7a79ad | 1850 | int level = btrfs_header_level(parent); |
416bc658 JB |
1851 | struct extent_buffer *eb; |
1852 | ||
fb770ae4 LB |
1853 | if (slot < 0 || slot >= btrfs_header_nritems(parent)) |
1854 | return ERR_PTR(-ENOENT); | |
ca7a79ad CM |
1855 | |
1856 | BUG_ON(level == 0); | |
1857 | ||
2ff7e61e | 1858 | eb = read_tree_block(fs_info, btrfs_node_blockptr(parent, slot), |
416bc658 | 1859 | btrfs_node_ptr_generation(parent, slot)); |
fb770ae4 LB |
1860 | if (!IS_ERR(eb) && !extent_buffer_uptodate(eb)) { |
1861 | free_extent_buffer(eb); | |
1862 | eb = ERR_PTR(-EIO); | |
416bc658 JB |
1863 | } |
1864 | ||
1865 | return eb; | |
bb803951 CM |
1866 | } |
1867 | ||
d352ac68 CM |
1868 | /* |
1869 | * node level balancing, used to make sure nodes are in proper order for | |
1870 | * item deletion. We balance from the top down, so we have to make sure | |
1871 | * that a deletion won't leave an node completely empty later on. | |
1872 | */ | |
e02119d5 | 1873 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1874 | struct btrfs_root *root, |
1875 | struct btrfs_path *path, int level) | |
bb803951 | 1876 | { |
0b246afa | 1877 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
1878 | struct extent_buffer *right = NULL; |
1879 | struct extent_buffer *mid; | |
1880 | struct extent_buffer *left = NULL; | |
1881 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1882 | int ret = 0; |
1883 | int wret; | |
1884 | int pslot; | |
bb803951 | 1885 | int orig_slot = path->slots[level]; |
79f95c82 | 1886 | u64 orig_ptr; |
bb803951 CM |
1887 | |
1888 | if (level == 0) | |
1889 | return 0; | |
1890 | ||
5f39d397 | 1891 | mid = path->nodes[level]; |
b4ce94de | 1892 | |
bd681513 CM |
1893 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1894 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1895 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1896 | ||
1d4f8a0c | 1897 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1898 | |
a05a9bb1 | 1899 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1900 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1901 | pslot = path->slots[level + 1]; |
1902 | } | |
bb803951 | 1903 | |
40689478 CM |
1904 | /* |
1905 | * deal with the case where there is only one pointer in the root | |
1906 | * by promoting the node below to a root | |
1907 | */ | |
5f39d397 CM |
1908 | if (!parent) { |
1909 | struct extent_buffer *child; | |
bb803951 | 1910 | |
5f39d397 | 1911 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1912 | return 0; |
1913 | ||
1914 | /* promote the child to a root */ | |
2ff7e61e | 1915 | child = read_node_slot(fs_info, mid, 0); |
fb770ae4 LB |
1916 | if (IS_ERR(child)) { |
1917 | ret = PTR_ERR(child); | |
0b246afa | 1918 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
1919 | goto enospc; |
1920 | } | |
1921 | ||
925baedd | 1922 | btrfs_tree_lock(child); |
b4ce94de | 1923 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1924 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1925 | if (ret) { |
1926 | btrfs_tree_unlock(child); | |
1927 | free_extent_buffer(child); | |
1928 | goto enospc; | |
1929 | } | |
2f375ab9 | 1930 | |
90f8d62e | 1931 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1932 | rcu_assign_pointer(root->node, child); |
925baedd | 1933 | |
0b86a832 | 1934 | add_root_to_dirty_list(root); |
925baedd | 1935 | btrfs_tree_unlock(child); |
b4ce94de | 1936 | |
925baedd | 1937 | path->locks[level] = 0; |
bb803951 | 1938 | path->nodes[level] = NULL; |
7c302b49 | 1939 | clean_tree_block(fs_info, mid); |
925baedd | 1940 | btrfs_tree_unlock(mid); |
bb803951 | 1941 | /* once for the path */ |
5f39d397 | 1942 | free_extent_buffer(mid); |
f0486c68 YZ |
1943 | |
1944 | root_sub_used(root, mid->len); | |
5581a51a | 1945 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1946 | /* once for the root ptr */ |
3083ee2e | 1947 | free_extent_buffer_stale(mid); |
f0486c68 | 1948 | return 0; |
bb803951 | 1949 | } |
5f39d397 | 1950 | if (btrfs_header_nritems(mid) > |
0b246afa | 1951 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 4) |
bb803951 CM |
1952 | return 0; |
1953 | ||
2ff7e61e | 1954 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
1955 | if (IS_ERR(left)) |
1956 | left = NULL; | |
1957 | ||
5f39d397 | 1958 | if (left) { |
925baedd | 1959 | btrfs_tree_lock(left); |
b4ce94de | 1960 | btrfs_set_lock_blocking(left); |
5f39d397 | 1961 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1962 | parent, pslot - 1, &left); |
54aa1f4d CM |
1963 | if (wret) { |
1964 | ret = wret; | |
1965 | goto enospc; | |
1966 | } | |
2cc58cf2 | 1967 | } |
fb770ae4 | 1968 | |
2ff7e61e | 1969 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
1970 | if (IS_ERR(right)) |
1971 | right = NULL; | |
1972 | ||
5f39d397 | 1973 | if (right) { |
925baedd | 1974 | btrfs_tree_lock(right); |
b4ce94de | 1975 | btrfs_set_lock_blocking(right); |
5f39d397 | 1976 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1977 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1978 | if (wret) { |
1979 | ret = wret; | |
1980 | goto enospc; | |
1981 | } | |
1982 | } | |
1983 | ||
1984 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1985 | if (left) { |
1986 | orig_slot += btrfs_header_nritems(left); | |
2ff7e61e | 1987 | wret = push_node_left(trans, fs_info, left, mid, 1); |
79f95c82 CM |
1988 | if (wret < 0) |
1989 | ret = wret; | |
bb803951 | 1990 | } |
79f95c82 CM |
1991 | |
1992 | /* | |
1993 | * then try to empty the right most buffer into the middle | |
1994 | */ | |
5f39d397 | 1995 | if (right) { |
2ff7e61e | 1996 | wret = push_node_left(trans, fs_info, mid, right, 1); |
54aa1f4d | 1997 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1998 | ret = wret; |
5f39d397 | 1999 | if (btrfs_header_nritems(right) == 0) { |
7c302b49 | 2000 | clean_tree_block(fs_info, right); |
925baedd | 2001 | btrfs_tree_unlock(right); |
afe5fea7 | 2002 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 2003 | root_sub_used(root, right->len); |
5581a51a | 2004 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2005 | free_extent_buffer_stale(right); |
f0486c68 | 2006 | right = NULL; |
bb803951 | 2007 | } else { |
5f39d397 CM |
2008 | struct btrfs_disk_key right_key; |
2009 | btrfs_node_key(right, &right_key, 0); | |
0b246afa | 2010 | tree_mod_log_set_node_key(fs_info, parent, |
32adf090 | 2011 | pslot + 1, 0); |
5f39d397 CM |
2012 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2013 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2014 | } |
2015 | } | |
5f39d397 | 2016 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2017 | /* |
2018 | * we're not allowed to leave a node with one item in the | |
2019 | * tree during a delete. A deletion from lower in the tree | |
2020 | * could try to delete the only pointer in this node. | |
2021 | * So, pull some keys from the left. | |
2022 | * There has to be a left pointer at this point because | |
2023 | * otherwise we would have pulled some pointers from the | |
2024 | * right | |
2025 | */ | |
305a26af MF |
2026 | if (!left) { |
2027 | ret = -EROFS; | |
0b246afa | 2028 | btrfs_handle_fs_error(fs_info, ret, NULL); |
305a26af MF |
2029 | goto enospc; |
2030 | } | |
2ff7e61e | 2031 | wret = balance_node_right(trans, fs_info, mid, left); |
54aa1f4d | 2032 | if (wret < 0) { |
79f95c82 | 2033 | ret = wret; |
54aa1f4d CM |
2034 | goto enospc; |
2035 | } | |
bce4eae9 | 2036 | if (wret == 1) { |
2ff7e61e | 2037 | wret = push_node_left(trans, fs_info, left, mid, 1); |
bce4eae9 CM |
2038 | if (wret < 0) |
2039 | ret = wret; | |
2040 | } | |
79f95c82 CM |
2041 | BUG_ON(wret == 1); |
2042 | } | |
5f39d397 | 2043 | if (btrfs_header_nritems(mid) == 0) { |
7c302b49 | 2044 | clean_tree_block(fs_info, mid); |
925baedd | 2045 | btrfs_tree_unlock(mid); |
afe5fea7 | 2046 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2047 | root_sub_used(root, mid->len); |
5581a51a | 2048 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2049 | free_extent_buffer_stale(mid); |
f0486c68 | 2050 | mid = NULL; |
79f95c82 CM |
2051 | } else { |
2052 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2053 | struct btrfs_disk_key mid_key; |
2054 | btrfs_node_key(mid, &mid_key, 0); | |
0b246afa | 2055 | tree_mod_log_set_node_key(fs_info, parent, pslot, 0); |
5f39d397 CM |
2056 | btrfs_set_node_key(parent, &mid_key, pslot); |
2057 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2058 | } |
bb803951 | 2059 | |
79f95c82 | 2060 | /* update the path */ |
5f39d397 CM |
2061 | if (left) { |
2062 | if (btrfs_header_nritems(left) > orig_slot) { | |
2063 | extent_buffer_get(left); | |
925baedd | 2064 | /* left was locked after cow */ |
5f39d397 | 2065 | path->nodes[level] = left; |
bb803951 CM |
2066 | path->slots[level + 1] -= 1; |
2067 | path->slots[level] = orig_slot; | |
925baedd CM |
2068 | if (mid) { |
2069 | btrfs_tree_unlock(mid); | |
5f39d397 | 2070 | free_extent_buffer(mid); |
925baedd | 2071 | } |
bb803951 | 2072 | } else { |
5f39d397 | 2073 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2074 | path->slots[level] = orig_slot; |
2075 | } | |
2076 | } | |
79f95c82 | 2077 | /* double check we haven't messed things up */ |
e20d96d6 | 2078 | if (orig_ptr != |
5f39d397 | 2079 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2080 | BUG(); |
54aa1f4d | 2081 | enospc: |
925baedd CM |
2082 | if (right) { |
2083 | btrfs_tree_unlock(right); | |
5f39d397 | 2084 | free_extent_buffer(right); |
925baedd CM |
2085 | } |
2086 | if (left) { | |
2087 | if (path->nodes[level] != left) | |
2088 | btrfs_tree_unlock(left); | |
5f39d397 | 2089 | free_extent_buffer(left); |
925baedd | 2090 | } |
bb803951 CM |
2091 | return ret; |
2092 | } | |
2093 | ||
d352ac68 CM |
2094 | /* Node balancing for insertion. Here we only split or push nodes around |
2095 | * when they are completely full. This is also done top down, so we | |
2096 | * have to be pessimistic. | |
2097 | */ | |
d397712b | 2098 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2099 | struct btrfs_root *root, |
2100 | struct btrfs_path *path, int level) | |
e66f709b | 2101 | { |
0b246afa | 2102 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
2103 | struct extent_buffer *right = NULL; |
2104 | struct extent_buffer *mid; | |
2105 | struct extent_buffer *left = NULL; | |
2106 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2107 | int ret = 0; |
2108 | int wret; | |
2109 | int pslot; | |
2110 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2111 | |
2112 | if (level == 0) | |
2113 | return 1; | |
2114 | ||
5f39d397 | 2115 | mid = path->nodes[level]; |
7bb86316 | 2116 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2117 | |
a05a9bb1 | 2118 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2119 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2120 | pslot = path->slots[level + 1]; |
2121 | } | |
e66f709b | 2122 | |
5f39d397 | 2123 | if (!parent) |
e66f709b | 2124 | return 1; |
e66f709b | 2125 | |
2ff7e61e | 2126 | left = read_node_slot(fs_info, parent, pslot - 1); |
fb770ae4 LB |
2127 | if (IS_ERR(left)) |
2128 | left = NULL; | |
e66f709b CM |
2129 | |
2130 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2131 | if (left) { |
e66f709b | 2132 | u32 left_nr; |
925baedd CM |
2133 | |
2134 | btrfs_tree_lock(left); | |
b4ce94de CM |
2135 | btrfs_set_lock_blocking(left); |
2136 | ||
5f39d397 | 2137 | left_nr = btrfs_header_nritems(left); |
0b246afa | 2138 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2139 | wret = 1; |
2140 | } else { | |
5f39d397 | 2141 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2142 | pslot - 1, &left); |
54aa1f4d CM |
2143 | if (ret) |
2144 | wret = 1; | |
2145 | else { | |
2ff7e61e | 2146 | wret = push_node_left(trans, fs_info, |
971a1f66 | 2147 | left, mid, 0); |
54aa1f4d | 2148 | } |
33ade1f8 | 2149 | } |
e66f709b CM |
2150 | if (wret < 0) |
2151 | ret = wret; | |
2152 | if (wret == 0) { | |
5f39d397 | 2153 | struct btrfs_disk_key disk_key; |
e66f709b | 2154 | orig_slot += left_nr; |
5f39d397 | 2155 | btrfs_node_key(mid, &disk_key, 0); |
0b246afa | 2156 | tree_mod_log_set_node_key(fs_info, parent, pslot, 0); |
5f39d397 CM |
2157 | btrfs_set_node_key(parent, &disk_key, pslot); |
2158 | btrfs_mark_buffer_dirty(parent); | |
2159 | if (btrfs_header_nritems(left) > orig_slot) { | |
2160 | path->nodes[level] = left; | |
e66f709b CM |
2161 | path->slots[level + 1] -= 1; |
2162 | path->slots[level] = orig_slot; | |
925baedd | 2163 | btrfs_tree_unlock(mid); |
5f39d397 | 2164 | free_extent_buffer(mid); |
e66f709b CM |
2165 | } else { |
2166 | orig_slot -= | |
5f39d397 | 2167 | btrfs_header_nritems(left); |
e66f709b | 2168 | path->slots[level] = orig_slot; |
925baedd | 2169 | btrfs_tree_unlock(left); |
5f39d397 | 2170 | free_extent_buffer(left); |
e66f709b | 2171 | } |
e66f709b CM |
2172 | return 0; |
2173 | } | |
925baedd | 2174 | btrfs_tree_unlock(left); |
5f39d397 | 2175 | free_extent_buffer(left); |
e66f709b | 2176 | } |
2ff7e61e | 2177 | right = read_node_slot(fs_info, parent, pslot + 1); |
fb770ae4 LB |
2178 | if (IS_ERR(right)) |
2179 | right = NULL; | |
e66f709b CM |
2180 | |
2181 | /* | |
2182 | * then try to empty the right most buffer into the middle | |
2183 | */ | |
5f39d397 | 2184 | if (right) { |
33ade1f8 | 2185 | u32 right_nr; |
b4ce94de | 2186 | |
925baedd | 2187 | btrfs_tree_lock(right); |
b4ce94de CM |
2188 | btrfs_set_lock_blocking(right); |
2189 | ||
5f39d397 | 2190 | right_nr = btrfs_header_nritems(right); |
0b246afa | 2191 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 1) { |
33ade1f8 CM |
2192 | wret = 1; |
2193 | } else { | |
5f39d397 CM |
2194 | ret = btrfs_cow_block(trans, root, right, |
2195 | parent, pslot + 1, | |
9fa8cfe7 | 2196 | &right); |
54aa1f4d CM |
2197 | if (ret) |
2198 | wret = 1; | |
2199 | else { | |
2ff7e61e | 2200 | wret = balance_node_right(trans, fs_info, |
5f39d397 | 2201 | right, mid); |
54aa1f4d | 2202 | } |
33ade1f8 | 2203 | } |
e66f709b CM |
2204 | if (wret < 0) |
2205 | ret = wret; | |
2206 | if (wret == 0) { | |
5f39d397 CM |
2207 | struct btrfs_disk_key disk_key; |
2208 | ||
2209 | btrfs_node_key(right, &disk_key, 0); | |
0b246afa | 2210 | tree_mod_log_set_node_key(fs_info, parent, |
32adf090 | 2211 | pslot + 1, 0); |
5f39d397 CM |
2212 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2213 | btrfs_mark_buffer_dirty(parent); | |
2214 | ||
2215 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2216 | path->nodes[level] = right; | |
e66f709b CM |
2217 | path->slots[level + 1] += 1; |
2218 | path->slots[level] = orig_slot - | |
5f39d397 | 2219 | btrfs_header_nritems(mid); |
925baedd | 2220 | btrfs_tree_unlock(mid); |
5f39d397 | 2221 | free_extent_buffer(mid); |
e66f709b | 2222 | } else { |
925baedd | 2223 | btrfs_tree_unlock(right); |
5f39d397 | 2224 | free_extent_buffer(right); |
e66f709b | 2225 | } |
e66f709b CM |
2226 | return 0; |
2227 | } | |
925baedd | 2228 | btrfs_tree_unlock(right); |
5f39d397 | 2229 | free_extent_buffer(right); |
e66f709b | 2230 | } |
e66f709b CM |
2231 | return 1; |
2232 | } | |
2233 | ||
3c69faec | 2234 | /* |
d352ac68 CM |
2235 | * readahead one full node of leaves, finding things that are close |
2236 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2237 | */ |
2ff7e61e | 2238 | static void reada_for_search(struct btrfs_fs_info *fs_info, |
c8c42864 CM |
2239 | struct btrfs_path *path, |
2240 | int level, int slot, u64 objectid) | |
3c69faec | 2241 | { |
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)) { | |
2ff7e61e | 2292 | readahead_tree_block(fs_info, 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 | |
2ff7e61e | 2301 | static noinline void reada_for_balance(struct btrfs_fs_info *fs_info, |
0b08851f | 2302 | struct btrfs_path *path, int level) |
b4ce94de CM |
2303 | { |
2304 | int slot; | |
2305 | int nritems; | |
2306 | struct extent_buffer *parent; | |
2307 | struct extent_buffer *eb; | |
2308 | u64 gen; | |
2309 | u64 block1 = 0; | |
2310 | u64 block2 = 0; | |
b4ce94de | 2311 | |
8c594ea8 | 2312 | parent = path->nodes[level + 1]; |
b4ce94de | 2313 | if (!parent) |
0b08851f | 2314 | return; |
b4ce94de CM |
2315 | |
2316 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2317 | slot = path->slots[level + 1]; |
b4ce94de CM |
2318 | |
2319 | if (slot > 0) { | |
2320 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2321 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
0b246afa | 2322 | eb = find_extent_buffer(fs_info, block1); |
b9fab919 CM |
2323 | /* |
2324 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2325 | * don't want to return eagain here. That will loop | |
2326 | * forever | |
2327 | */ | |
2328 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2329 | block1 = 0; |
2330 | free_extent_buffer(eb); | |
2331 | } | |
8c594ea8 | 2332 | if (slot + 1 < nritems) { |
b4ce94de CM |
2333 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2334 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
0b246afa | 2335 | eb = find_extent_buffer(fs_info, block2); |
b9fab919 | 2336 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2337 | block2 = 0; |
2338 | free_extent_buffer(eb); | |
2339 | } | |
8c594ea8 | 2340 | |
0b08851f | 2341 | if (block1) |
2ff7e61e | 2342 | readahead_tree_block(fs_info, block1); |
0b08851f | 2343 | if (block2) |
2ff7e61e | 2344 | readahead_tree_block(fs_info, block2); |
b4ce94de CM |
2345 | } |
2346 | ||
2347 | ||
d352ac68 | 2348 | /* |
d397712b CM |
2349 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2350 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2351 | * operations on the tree might require changing key pointers higher up in the | |
2352 | * tree. | |
d352ac68 | 2353 | * |
d397712b CM |
2354 | * callers might also have set path->keep_locks, which tells this code to keep |
2355 | * the lock if the path points to the last slot in the block. This is part of | |
2356 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2357 | * |
d397712b CM |
2358 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2359 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2360 | */ |
e02119d5 | 2361 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2362 | int lowest_unlock, int min_write_lock_level, |
2363 | int *write_lock_level) | |
925baedd CM |
2364 | { |
2365 | int i; | |
2366 | int skip_level = level; | |
051e1b9f | 2367 | int no_skips = 0; |
925baedd CM |
2368 | struct extent_buffer *t; |
2369 | ||
2370 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2371 | if (!path->nodes[i]) | |
2372 | break; | |
2373 | if (!path->locks[i]) | |
2374 | break; | |
051e1b9f | 2375 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2376 | skip_level = i + 1; |
2377 | continue; | |
2378 | } | |
051e1b9f | 2379 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2380 | u32 nritems; |
2381 | t = path->nodes[i]; | |
2382 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2383 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2384 | skip_level = i + 1; |
2385 | continue; | |
2386 | } | |
2387 | } | |
051e1b9f CM |
2388 | if (skip_level < i && i >= lowest_unlock) |
2389 | no_skips = 1; | |
2390 | ||
925baedd CM |
2391 | t = path->nodes[i]; |
2392 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2393 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2394 | path->locks[i] = 0; |
f7c79f30 CM |
2395 | if (write_lock_level && |
2396 | i > min_write_lock_level && | |
2397 | i <= *write_lock_level) { | |
2398 | *write_lock_level = i - 1; | |
2399 | } | |
925baedd CM |
2400 | } |
2401 | } | |
2402 | } | |
2403 | ||
b4ce94de CM |
2404 | /* |
2405 | * This releases any locks held in the path starting at level and | |
2406 | * going all the way up to the root. | |
2407 | * | |
2408 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2409 | * corner cases, such as COW of the block at slot zero in the node. This | |
2410 | * ignores those rules, and it should only be called when there are no | |
2411 | * more updates to be done higher up in the tree. | |
2412 | */ | |
2413 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2414 | { | |
2415 | int i; | |
2416 | ||
09a2a8f9 | 2417 | if (path->keep_locks) |
b4ce94de CM |
2418 | return; |
2419 | ||
2420 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2421 | if (!path->nodes[i]) | |
12f4dacc | 2422 | continue; |
b4ce94de | 2423 | if (!path->locks[i]) |
12f4dacc | 2424 | continue; |
bd681513 | 2425 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2426 | path->locks[i] = 0; |
2427 | } | |
2428 | } | |
2429 | ||
c8c42864 CM |
2430 | /* |
2431 | * helper function for btrfs_search_slot. The goal is to find a block | |
2432 | * in cache without setting the path to blocking. If we find the block | |
2433 | * we return zero and the path is unchanged. | |
2434 | * | |
2435 | * If we can't find the block, we set the path blocking and do some | |
2436 | * reada. -EAGAIN is returned and the search must be repeated. | |
2437 | */ | |
2438 | static int | |
d07b8528 LB |
2439 | read_block_for_search(struct btrfs_root *root, struct btrfs_path *p, |
2440 | struct extent_buffer **eb_ret, int level, int slot, | |
cda79c54 | 2441 | const struct btrfs_key *key) |
c8c42864 | 2442 | { |
0b246afa | 2443 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 CM |
2444 | u64 blocknr; |
2445 | u64 gen; | |
c8c42864 CM |
2446 | struct extent_buffer *b = *eb_ret; |
2447 | struct extent_buffer *tmp; | |
76a05b35 | 2448 | int ret; |
c8c42864 CM |
2449 | |
2450 | blocknr = btrfs_node_blockptr(b, slot); | |
2451 | gen = btrfs_node_ptr_generation(b, slot); | |
c8c42864 | 2452 | |
0b246afa | 2453 | tmp = find_extent_buffer(fs_info, blocknr); |
cb44921a | 2454 | if (tmp) { |
b9fab919 | 2455 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2456 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2457 | *eb_ret = tmp; | |
2458 | return 0; | |
2459 | } | |
2460 | ||
2461 | /* the pages were up to date, but we failed | |
2462 | * the generation number check. Do a full | |
2463 | * read for the generation number that is correct. | |
2464 | * We must do this without dropping locks so | |
2465 | * we can trust our generation number | |
2466 | */ | |
2467 | btrfs_set_path_blocking(p); | |
2468 | ||
2469 | /* now we're allowed to do a blocking uptodate check */ | |
2470 | ret = btrfs_read_buffer(tmp, gen); | |
2471 | if (!ret) { | |
2472 | *eb_ret = tmp; | |
2473 | return 0; | |
cb44921a | 2474 | } |
bdf7c00e JB |
2475 | free_extent_buffer(tmp); |
2476 | btrfs_release_path(p); | |
2477 | return -EIO; | |
c8c42864 CM |
2478 | } |
2479 | ||
2480 | /* | |
2481 | * reduce lock contention at high levels | |
2482 | * of the btree by dropping locks before | |
76a05b35 CM |
2483 | * we read. Don't release the lock on the current |
2484 | * level because we need to walk this node to figure | |
2485 | * out which blocks to read. | |
c8c42864 | 2486 | */ |
8c594ea8 CM |
2487 | btrfs_unlock_up_safe(p, level + 1); |
2488 | btrfs_set_path_blocking(p); | |
2489 | ||
cb44921a | 2490 | free_extent_buffer(tmp); |
e4058b54 | 2491 | if (p->reada != READA_NONE) |
2ff7e61e | 2492 | reada_for_search(fs_info, p, level, slot, key->objectid); |
c8c42864 | 2493 | |
b3b4aa74 | 2494 | btrfs_release_path(p); |
76a05b35 CM |
2495 | |
2496 | ret = -EAGAIN; | |
2ff7e61e | 2497 | tmp = read_tree_block(fs_info, blocknr, 0); |
64c043de | 2498 | if (!IS_ERR(tmp)) { |
76a05b35 CM |
2499 | /* |
2500 | * If the read above didn't mark this buffer up to date, | |
2501 | * it will never end up being up to date. Set ret to EIO now | |
2502 | * and give up so that our caller doesn't loop forever | |
2503 | * on our EAGAINs. | |
2504 | */ | |
b9fab919 | 2505 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2506 | ret = -EIO; |
c8c42864 | 2507 | free_extent_buffer(tmp); |
c871b0f2 LB |
2508 | } else { |
2509 | ret = PTR_ERR(tmp); | |
76a05b35 CM |
2510 | } |
2511 | return ret; | |
c8c42864 CM |
2512 | } |
2513 | ||
2514 | /* | |
2515 | * helper function for btrfs_search_slot. This does all of the checks | |
2516 | * for node-level blocks and does any balancing required based on | |
2517 | * the ins_len. | |
2518 | * | |
2519 | * If no extra work was required, zero is returned. If we had to | |
2520 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2521 | * start over | |
2522 | */ | |
2523 | static int | |
2524 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2525 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2526 | struct extent_buffer *b, int level, int ins_len, |
2527 | int *write_lock_level) | |
c8c42864 | 2528 | { |
0b246afa | 2529 | struct btrfs_fs_info *fs_info = root->fs_info; |
c8c42864 | 2530 | int ret; |
0b246afa | 2531 | |
c8c42864 | 2532 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= |
0b246afa | 2533 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) { |
c8c42864 CM |
2534 | int sret; |
2535 | ||
bd681513 CM |
2536 | if (*write_lock_level < level + 1) { |
2537 | *write_lock_level = level + 1; | |
2538 | btrfs_release_path(p); | |
2539 | goto again; | |
2540 | } | |
2541 | ||
c8c42864 | 2542 | btrfs_set_path_blocking(p); |
2ff7e61e | 2543 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2544 | sret = split_node(trans, root, p, level); |
bd681513 | 2545 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2546 | |
2547 | BUG_ON(sret > 0); | |
2548 | if (sret) { | |
2549 | ret = sret; | |
2550 | goto done; | |
2551 | } | |
2552 | b = p->nodes[level]; | |
2553 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
0b246afa | 2554 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) / 2) { |
c8c42864 CM |
2555 | int sret; |
2556 | ||
bd681513 CM |
2557 | if (*write_lock_level < level + 1) { |
2558 | *write_lock_level = level + 1; | |
2559 | btrfs_release_path(p); | |
2560 | goto again; | |
2561 | } | |
2562 | ||
c8c42864 | 2563 | btrfs_set_path_blocking(p); |
2ff7e61e | 2564 | reada_for_balance(fs_info, p, level); |
c8c42864 | 2565 | sret = balance_level(trans, root, p, level); |
bd681513 | 2566 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2567 | |
2568 | if (sret) { | |
2569 | ret = sret; | |
2570 | goto done; | |
2571 | } | |
2572 | b = p->nodes[level]; | |
2573 | if (!b) { | |
b3b4aa74 | 2574 | btrfs_release_path(p); |
c8c42864 CM |
2575 | goto again; |
2576 | } | |
2577 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2578 | } | |
2579 | return 0; | |
2580 | ||
2581 | again: | |
2582 | ret = -EAGAIN; | |
2583 | done: | |
2584 | return ret; | |
2585 | } | |
2586 | ||
d7396f07 | 2587 | static void key_search_validate(struct extent_buffer *b, |
310712b2 | 2588 | const struct btrfs_key *key, |
d7396f07 FDBM |
2589 | int level) |
2590 | { | |
2591 | #ifdef CONFIG_BTRFS_ASSERT | |
2592 | struct btrfs_disk_key disk_key; | |
2593 | ||
2594 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2595 | ||
2596 | if (level == 0) | |
2597 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2598 | offsetof(struct btrfs_leaf, items[0].key), | |
2599 | sizeof(disk_key))); | |
2600 | else | |
2601 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2602 | offsetof(struct btrfs_node, ptrs[0].key), | |
2603 | sizeof(disk_key))); | |
2604 | #endif | |
2605 | } | |
2606 | ||
310712b2 | 2607 | static int key_search(struct extent_buffer *b, const struct btrfs_key *key, |
d7396f07 FDBM |
2608 | int level, int *prev_cmp, int *slot) |
2609 | { | |
2610 | if (*prev_cmp != 0) { | |
2611 | *prev_cmp = bin_search(b, key, level, slot); | |
2612 | return *prev_cmp; | |
2613 | } | |
2614 | ||
2615 | key_search_validate(b, key, level); | |
2616 | *slot = 0; | |
2617 | ||
2618 | return 0; | |
2619 | } | |
2620 | ||
381cf658 | 2621 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *path, |
e33d5c3d KN |
2622 | u64 iobjectid, u64 ioff, u8 key_type, |
2623 | struct btrfs_key *found_key) | |
2624 | { | |
2625 | int ret; | |
2626 | struct btrfs_key key; | |
2627 | struct extent_buffer *eb; | |
381cf658 DS |
2628 | |
2629 | ASSERT(path); | |
1d4c08e0 | 2630 | ASSERT(found_key); |
e33d5c3d KN |
2631 | |
2632 | key.type = key_type; | |
2633 | key.objectid = iobjectid; | |
2634 | key.offset = ioff; | |
2635 | ||
2636 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); | |
1d4c08e0 | 2637 | if (ret < 0) |
e33d5c3d KN |
2638 | return ret; |
2639 | ||
2640 | eb = path->nodes[0]; | |
2641 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2642 | ret = btrfs_next_leaf(fs_root, path); | |
2643 | if (ret) | |
2644 | return ret; | |
2645 | eb = path->nodes[0]; | |
2646 | } | |
2647 | ||
2648 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2649 | if (found_key->type != key.type || | |
2650 | found_key->objectid != key.objectid) | |
2651 | return 1; | |
2652 | ||
2653 | return 0; | |
2654 | } | |
2655 | ||
74123bd7 CM |
2656 | /* |
2657 | * look for key in the tree. path is filled in with nodes along the way | |
2658 | * if key is found, we return zero and you can find the item in the leaf | |
2659 | * level of the path (level 0) | |
2660 | * | |
2661 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2662 | * be inserted, and 1 is returned. If there are other errors during the |
2663 | * search a negative error number is returned. | |
97571fd0 CM |
2664 | * |
2665 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2666 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2667 | * possible) | |
74123bd7 | 2668 | */ |
310712b2 OS |
2669 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
2670 | const struct btrfs_key *key, struct btrfs_path *p, | |
2671 | int ins_len, int cow) | |
be0e5c09 | 2672 | { |
0b246afa | 2673 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 2674 | struct extent_buffer *b; |
be0e5c09 CM |
2675 | int slot; |
2676 | int ret; | |
33c66f43 | 2677 | int err; |
be0e5c09 | 2678 | int level; |
925baedd | 2679 | int lowest_unlock = 1; |
bd681513 CM |
2680 | int root_lock; |
2681 | /* everything at write_lock_level or lower must be write locked */ | |
2682 | int write_lock_level = 0; | |
9f3a7427 | 2683 | u8 lowest_level = 0; |
f7c79f30 | 2684 | int min_write_lock_level; |
d7396f07 | 2685 | int prev_cmp; |
9f3a7427 | 2686 | |
6702ed49 | 2687 | lowest_level = p->lowest_level; |
323ac95b | 2688 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2689 | WARN_ON(p->nodes[0] != NULL); |
eb653de1 | 2690 | BUG_ON(!cow && ins_len); |
25179201 | 2691 | |
bd681513 | 2692 | if (ins_len < 0) { |
925baedd | 2693 | lowest_unlock = 2; |
65b51a00 | 2694 | |
bd681513 CM |
2695 | /* when we are removing items, we might have to go up to level |
2696 | * two as we update tree pointers Make sure we keep write | |
2697 | * for those levels as well | |
2698 | */ | |
2699 | write_lock_level = 2; | |
2700 | } else if (ins_len > 0) { | |
2701 | /* | |
2702 | * for inserting items, make sure we have a write lock on | |
2703 | * level 1 so we can update keys | |
2704 | */ | |
2705 | write_lock_level = 1; | |
2706 | } | |
2707 | ||
2708 | if (!cow) | |
2709 | write_lock_level = -1; | |
2710 | ||
09a2a8f9 | 2711 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2712 | write_lock_level = BTRFS_MAX_LEVEL; |
2713 | ||
f7c79f30 CM |
2714 | min_write_lock_level = write_lock_level; |
2715 | ||
bb803951 | 2716 | again: |
d7396f07 | 2717 | prev_cmp = -1; |
bd681513 CM |
2718 | /* |
2719 | * we try very hard to do read locks on the root | |
2720 | */ | |
2721 | root_lock = BTRFS_READ_LOCK; | |
2722 | level = 0; | |
5d4f98a2 | 2723 | if (p->search_commit_root) { |
bd681513 CM |
2724 | /* |
2725 | * the commit roots are read only | |
2726 | * so we always do read locks | |
2727 | */ | |
3f8a18cc | 2728 | if (p->need_commit_sem) |
0b246afa | 2729 | down_read(&fs_info->commit_root_sem); |
5d4f98a2 YZ |
2730 | b = root->commit_root; |
2731 | extent_buffer_get(b); | |
bd681513 | 2732 | level = btrfs_header_level(b); |
3f8a18cc | 2733 | if (p->need_commit_sem) |
0b246afa | 2734 | up_read(&fs_info->commit_root_sem); |
5d4f98a2 | 2735 | if (!p->skip_locking) |
bd681513 | 2736 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2737 | } else { |
bd681513 | 2738 | if (p->skip_locking) { |
5d4f98a2 | 2739 | b = btrfs_root_node(root); |
bd681513 CM |
2740 | level = btrfs_header_level(b); |
2741 | } else { | |
2742 | /* we don't know the level of the root node | |
2743 | * until we actually have it read locked | |
2744 | */ | |
2745 | b = btrfs_read_lock_root_node(root); | |
2746 | level = btrfs_header_level(b); | |
2747 | if (level <= write_lock_level) { | |
2748 | /* whoops, must trade for write lock */ | |
2749 | btrfs_tree_read_unlock(b); | |
2750 | free_extent_buffer(b); | |
2751 | b = btrfs_lock_root_node(root); | |
2752 | root_lock = BTRFS_WRITE_LOCK; | |
2753 | ||
2754 | /* the level might have changed, check again */ | |
2755 | level = btrfs_header_level(b); | |
2756 | } | |
2757 | } | |
5d4f98a2 | 2758 | } |
bd681513 CM |
2759 | p->nodes[level] = b; |
2760 | if (!p->skip_locking) | |
2761 | p->locks[level] = root_lock; | |
925baedd | 2762 | |
eb60ceac | 2763 | while (b) { |
5f39d397 | 2764 | level = btrfs_header_level(b); |
65b51a00 CM |
2765 | |
2766 | /* | |
2767 | * setup the path here so we can release it under lock | |
2768 | * contention with the cow code | |
2769 | */ | |
02217ed2 | 2770 | if (cow) { |
c8c42864 CM |
2771 | /* |
2772 | * if we don't really need to cow this block | |
2773 | * then we don't want to set the path blocking, | |
2774 | * so we test it here | |
2775 | */ | |
64c12921 JM |
2776 | if (!should_cow_block(trans, root, b)) { |
2777 | trans->dirty = true; | |
65b51a00 | 2778 | goto cow_done; |
64c12921 | 2779 | } |
5d4f98a2 | 2780 | |
bd681513 CM |
2781 | /* |
2782 | * must have write locks on this node and the | |
2783 | * parent | |
2784 | */ | |
5124e00e JB |
2785 | if (level > write_lock_level || |
2786 | (level + 1 > write_lock_level && | |
2787 | level + 1 < BTRFS_MAX_LEVEL && | |
2788 | p->nodes[level + 1])) { | |
bd681513 CM |
2789 | write_lock_level = level + 1; |
2790 | btrfs_release_path(p); | |
2791 | goto again; | |
2792 | } | |
2793 | ||
160f4089 | 2794 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2795 | err = btrfs_cow_block(trans, root, b, |
2796 | p->nodes[level + 1], | |
2797 | p->slots[level + 1], &b); | |
2798 | if (err) { | |
33c66f43 | 2799 | ret = err; |
65b51a00 | 2800 | goto done; |
54aa1f4d | 2801 | } |
02217ed2 | 2802 | } |
65b51a00 | 2803 | cow_done: |
eb60ceac | 2804 | p->nodes[level] = b; |
bd681513 | 2805 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2806 | |
2807 | /* | |
2808 | * we have a lock on b and as long as we aren't changing | |
2809 | * the tree, there is no way to for the items in b to change. | |
2810 | * It is safe to drop the lock on our parent before we | |
2811 | * go through the expensive btree search on b. | |
2812 | * | |
eb653de1 FDBM |
2813 | * If we're inserting or deleting (ins_len != 0), then we might |
2814 | * be changing slot zero, which may require changing the parent. | |
2815 | * So, we can't drop the lock until after we know which slot | |
2816 | * we're operating on. | |
b4ce94de | 2817 | */ |
eb653de1 FDBM |
2818 | if (!ins_len && !p->keep_locks) { |
2819 | int u = level + 1; | |
2820 | ||
2821 | if (u < BTRFS_MAX_LEVEL && p->locks[u]) { | |
2822 | btrfs_tree_unlock_rw(p->nodes[u], p->locks[u]); | |
2823 | p->locks[u] = 0; | |
2824 | } | |
2825 | } | |
b4ce94de | 2826 | |
d7396f07 | 2827 | ret = key_search(b, key, level, &prev_cmp, &slot); |
415b35a5 LB |
2828 | if (ret < 0) |
2829 | goto done; | |
b4ce94de | 2830 | |
5f39d397 | 2831 | if (level != 0) { |
33c66f43 YZ |
2832 | int dec = 0; |
2833 | if (ret && slot > 0) { | |
2834 | dec = 1; | |
be0e5c09 | 2835 | slot -= 1; |
33c66f43 | 2836 | } |
be0e5c09 | 2837 | p->slots[level] = slot; |
33c66f43 | 2838 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2839 | ins_len, &write_lock_level); |
33c66f43 | 2840 | if (err == -EAGAIN) |
c8c42864 | 2841 | goto again; |
33c66f43 YZ |
2842 | if (err) { |
2843 | ret = err; | |
c8c42864 | 2844 | goto done; |
33c66f43 | 2845 | } |
c8c42864 CM |
2846 | b = p->nodes[level]; |
2847 | slot = p->slots[level]; | |
b4ce94de | 2848 | |
bd681513 CM |
2849 | /* |
2850 | * slot 0 is special, if we change the key | |
2851 | * we have to update the parent pointer | |
2852 | * which means we must have a write lock | |
2853 | * on the parent | |
2854 | */ | |
eb653de1 | 2855 | if (slot == 0 && ins_len && |
bd681513 CM |
2856 | write_lock_level < level + 1) { |
2857 | write_lock_level = level + 1; | |
2858 | btrfs_release_path(p); | |
2859 | goto again; | |
2860 | } | |
2861 | ||
f7c79f30 CM |
2862 | unlock_up(p, level, lowest_unlock, |
2863 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2864 | |
925baedd | 2865 | if (level == lowest_level) { |
33c66f43 YZ |
2866 | if (dec) |
2867 | p->slots[level]++; | |
5b21f2ed | 2868 | goto done; |
925baedd | 2869 | } |
ca7a79ad | 2870 | |
d07b8528 | 2871 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 2872 | slot, key); |
33c66f43 | 2873 | if (err == -EAGAIN) |
c8c42864 | 2874 | goto again; |
33c66f43 YZ |
2875 | if (err) { |
2876 | ret = err; | |
76a05b35 | 2877 | goto done; |
33c66f43 | 2878 | } |
76a05b35 | 2879 | |
b4ce94de | 2880 | if (!p->skip_locking) { |
bd681513 CM |
2881 | level = btrfs_header_level(b); |
2882 | if (level <= write_lock_level) { | |
2883 | err = btrfs_try_tree_write_lock(b); | |
2884 | if (!err) { | |
2885 | btrfs_set_path_blocking(p); | |
2886 | btrfs_tree_lock(b); | |
2887 | btrfs_clear_path_blocking(p, b, | |
2888 | BTRFS_WRITE_LOCK); | |
2889 | } | |
2890 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2891 | } else { | |
f82c458a | 2892 | err = btrfs_tree_read_lock_atomic(b); |
bd681513 CM |
2893 | if (!err) { |
2894 | btrfs_set_path_blocking(p); | |
2895 | btrfs_tree_read_lock(b); | |
2896 | btrfs_clear_path_blocking(p, b, | |
2897 | BTRFS_READ_LOCK); | |
2898 | } | |
2899 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2900 | } |
bd681513 | 2901 | p->nodes[level] = b; |
b4ce94de | 2902 | } |
be0e5c09 CM |
2903 | } else { |
2904 | p->slots[level] = slot; | |
87b29b20 | 2905 | if (ins_len > 0 && |
2ff7e61e | 2906 | btrfs_leaf_free_space(fs_info, b) < ins_len) { |
bd681513 CM |
2907 | if (write_lock_level < 1) { |
2908 | write_lock_level = 1; | |
2909 | btrfs_release_path(p); | |
2910 | goto again; | |
2911 | } | |
2912 | ||
b4ce94de | 2913 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2914 | err = split_leaf(trans, root, key, |
2915 | p, ins_len, ret == 0); | |
bd681513 | 2916 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2917 | |
33c66f43 YZ |
2918 | BUG_ON(err > 0); |
2919 | if (err) { | |
2920 | ret = err; | |
65b51a00 CM |
2921 | goto done; |
2922 | } | |
5c680ed6 | 2923 | } |
459931ec | 2924 | if (!p->search_for_split) |
f7c79f30 CM |
2925 | unlock_up(p, level, lowest_unlock, |
2926 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2927 | goto done; |
be0e5c09 CM |
2928 | } |
2929 | } | |
65b51a00 CM |
2930 | ret = 1; |
2931 | done: | |
b4ce94de CM |
2932 | /* |
2933 | * we don't really know what they plan on doing with the path | |
2934 | * from here on, so for now just mark it as blocking | |
2935 | */ | |
b9473439 CM |
2936 | if (!p->leave_spinning) |
2937 | btrfs_set_path_blocking(p); | |
5f5bc6b1 | 2938 | if (ret < 0 && !p->skip_release_on_error) |
b3b4aa74 | 2939 | btrfs_release_path(p); |
65b51a00 | 2940 | return ret; |
be0e5c09 CM |
2941 | } |
2942 | ||
5d9e75c4 JS |
2943 | /* |
2944 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2945 | * current state of the tree together with the operations recorded in the tree | |
2946 | * modification log to search for the key in a previous version of this tree, as | |
2947 | * denoted by the time_seq parameter. | |
2948 | * | |
2949 | * Naturally, there is no support for insert, delete or cow operations. | |
2950 | * | |
2951 | * The resulting path and return value will be set up as if we called | |
2952 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2953 | */ | |
310712b2 | 2954 | int btrfs_search_old_slot(struct btrfs_root *root, const struct btrfs_key *key, |
5d9e75c4 JS |
2955 | struct btrfs_path *p, u64 time_seq) |
2956 | { | |
0b246afa | 2957 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d9e75c4 JS |
2958 | struct extent_buffer *b; |
2959 | int slot; | |
2960 | int ret; | |
2961 | int err; | |
2962 | int level; | |
2963 | int lowest_unlock = 1; | |
2964 | u8 lowest_level = 0; | |
d4b4087c | 2965 | int prev_cmp = -1; |
5d9e75c4 JS |
2966 | |
2967 | lowest_level = p->lowest_level; | |
2968 | WARN_ON(p->nodes[0] != NULL); | |
2969 | ||
2970 | if (p->search_commit_root) { | |
2971 | BUG_ON(time_seq); | |
2972 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2973 | } | |
2974 | ||
2975 | again: | |
5d9e75c4 | 2976 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2977 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2978 | p->locks[level] = BTRFS_READ_LOCK; |
2979 | ||
2980 | while (b) { | |
2981 | level = btrfs_header_level(b); | |
2982 | p->nodes[level] = b; | |
2983 | btrfs_clear_path_blocking(p, NULL, 0); | |
2984 | ||
2985 | /* | |
2986 | * we have a lock on b and as long as we aren't changing | |
2987 | * the tree, there is no way to for the items in b to change. | |
2988 | * It is safe to drop the lock on our parent before we | |
2989 | * go through the expensive btree search on b. | |
2990 | */ | |
2991 | btrfs_unlock_up_safe(p, level + 1); | |
2992 | ||
d4b4087c | 2993 | /* |
01327610 | 2994 | * Since we can unwind ebs we want to do a real search every |
d4b4087c JB |
2995 | * time. |
2996 | */ | |
2997 | prev_cmp = -1; | |
d7396f07 | 2998 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2999 | |
3000 | if (level != 0) { | |
3001 | int dec = 0; | |
3002 | if (ret && slot > 0) { | |
3003 | dec = 1; | |
3004 | slot -= 1; | |
3005 | } | |
3006 | p->slots[level] = slot; | |
3007 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3008 | ||
3009 | if (level == lowest_level) { | |
3010 | if (dec) | |
3011 | p->slots[level]++; | |
3012 | goto done; | |
3013 | } | |
3014 | ||
d07b8528 | 3015 | err = read_block_for_search(root, p, &b, level, |
cda79c54 | 3016 | slot, key); |
5d9e75c4 JS |
3017 | if (err == -EAGAIN) |
3018 | goto again; | |
3019 | if (err) { | |
3020 | ret = err; | |
3021 | goto done; | |
3022 | } | |
3023 | ||
3024 | level = btrfs_header_level(b); | |
f82c458a | 3025 | err = btrfs_tree_read_lock_atomic(b); |
5d9e75c4 JS |
3026 | if (!err) { |
3027 | btrfs_set_path_blocking(p); | |
3028 | btrfs_tree_read_lock(b); | |
3029 | btrfs_clear_path_blocking(p, b, | |
3030 | BTRFS_READ_LOCK); | |
3031 | } | |
0b246afa | 3032 | b = tree_mod_log_rewind(fs_info, p, b, time_seq); |
db7f3436 JB |
3033 | if (!b) { |
3034 | ret = -ENOMEM; | |
3035 | goto done; | |
3036 | } | |
5d9e75c4 JS |
3037 | p->locks[level] = BTRFS_READ_LOCK; |
3038 | p->nodes[level] = b; | |
5d9e75c4 JS |
3039 | } else { |
3040 | p->slots[level] = slot; | |
3041 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3042 | goto done; | |
3043 | } | |
3044 | } | |
3045 | ret = 1; | |
3046 | done: | |
3047 | if (!p->leave_spinning) | |
3048 | btrfs_set_path_blocking(p); | |
3049 | if (ret < 0) | |
3050 | btrfs_release_path(p); | |
3051 | ||
3052 | return ret; | |
3053 | } | |
3054 | ||
2f38b3e1 AJ |
3055 | /* |
3056 | * helper to use instead of search slot if no exact match is needed but | |
3057 | * instead the next or previous item should be returned. | |
3058 | * When find_higher is true, the next higher item is returned, the next lower | |
3059 | * otherwise. | |
3060 | * When return_any and find_higher are both true, and no higher item is found, | |
3061 | * return the next lower instead. | |
3062 | * When return_any is true and find_higher is false, and no lower item is found, | |
3063 | * return the next higher instead. | |
3064 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3065 | * < 0 on error | |
3066 | */ | |
3067 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
310712b2 OS |
3068 | const struct btrfs_key *key, |
3069 | struct btrfs_path *p, int find_higher, | |
3070 | int return_any) | |
2f38b3e1 AJ |
3071 | { |
3072 | int ret; | |
3073 | struct extent_buffer *leaf; | |
3074 | ||
3075 | again: | |
3076 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3077 | if (ret <= 0) | |
3078 | return ret; | |
3079 | /* | |
3080 | * a return value of 1 means the path is at the position where the | |
3081 | * item should be inserted. Normally this is the next bigger item, | |
3082 | * but in case the previous item is the last in a leaf, path points | |
3083 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3084 | * item. | |
3085 | */ | |
3086 | leaf = p->nodes[0]; | |
3087 | ||
3088 | if (find_higher) { | |
3089 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3090 | ret = btrfs_next_leaf(root, p); | |
3091 | if (ret <= 0) | |
3092 | return ret; | |
3093 | if (!return_any) | |
3094 | return 1; | |
3095 | /* | |
3096 | * no higher item found, return the next | |
3097 | * lower instead | |
3098 | */ | |
3099 | return_any = 0; | |
3100 | find_higher = 0; | |
3101 | btrfs_release_path(p); | |
3102 | goto again; | |
3103 | } | |
3104 | } else { | |
e6793769 AJ |
3105 | if (p->slots[0] == 0) { |
3106 | ret = btrfs_prev_leaf(root, p); | |
3107 | if (ret < 0) | |
3108 | return ret; | |
3109 | if (!ret) { | |
23c6bf6a FDBM |
3110 | leaf = p->nodes[0]; |
3111 | if (p->slots[0] == btrfs_header_nritems(leaf)) | |
3112 | p->slots[0]--; | |
e6793769 | 3113 | return 0; |
2f38b3e1 | 3114 | } |
e6793769 AJ |
3115 | if (!return_any) |
3116 | return 1; | |
3117 | /* | |
3118 | * no lower item found, return the next | |
3119 | * higher instead | |
3120 | */ | |
3121 | return_any = 0; | |
3122 | find_higher = 1; | |
3123 | btrfs_release_path(p); | |
3124 | goto again; | |
3125 | } else { | |
2f38b3e1 AJ |
3126 | --p->slots[0]; |
3127 | } | |
3128 | } | |
3129 | return 0; | |
3130 | } | |
3131 | ||
74123bd7 CM |
3132 | /* |
3133 | * adjust the pointers going up the tree, starting at level | |
3134 | * making sure the right key of each node is points to 'key'. | |
3135 | * This is used after shifting pointers to the left, so it stops | |
3136 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3137 | * higher levels | |
aa5d6bed | 3138 | * |
74123bd7 | 3139 | */ |
b7a0365e DD |
3140 | static void fixup_low_keys(struct btrfs_fs_info *fs_info, |
3141 | struct btrfs_path *path, | |
143bede5 | 3142 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3143 | { |
3144 | int i; | |
5f39d397 CM |
3145 | struct extent_buffer *t; |
3146 | ||
234b63a0 | 3147 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3148 | int tslot = path->slots[i]; |
eb60ceac | 3149 | if (!path->nodes[i]) |
be0e5c09 | 3150 | break; |
5f39d397 | 3151 | t = path->nodes[i]; |
b7a0365e | 3152 | tree_mod_log_set_node_key(fs_info, t, tslot, 1); |
5f39d397 | 3153 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3154 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3155 | if (tslot != 0) |
3156 | break; | |
3157 | } | |
3158 | } | |
3159 | ||
31840ae1 ZY |
3160 | /* |
3161 | * update item key. | |
3162 | * | |
3163 | * This function isn't completely safe. It's the caller's responsibility | |
3164 | * that the new key won't break the order | |
3165 | */ | |
b7a0365e DD |
3166 | void btrfs_set_item_key_safe(struct btrfs_fs_info *fs_info, |
3167 | struct btrfs_path *path, | |
310712b2 | 3168 | const struct btrfs_key *new_key) |
31840ae1 ZY |
3169 | { |
3170 | struct btrfs_disk_key disk_key; | |
3171 | struct extent_buffer *eb; | |
3172 | int slot; | |
3173 | ||
3174 | eb = path->nodes[0]; | |
3175 | slot = path->slots[0]; | |
3176 | if (slot > 0) { | |
3177 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3178 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3179 | } |
3180 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3181 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3182 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3183 | } |
3184 | ||
3185 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3186 | btrfs_set_item_key(eb, &disk_key, slot); | |
3187 | btrfs_mark_buffer_dirty(eb); | |
3188 | if (slot == 0) | |
b7a0365e | 3189 | fixup_low_keys(fs_info, path, &disk_key, 1); |
31840ae1 ZY |
3190 | } |
3191 | ||
74123bd7 CM |
3192 | /* |
3193 | * try to push data from one node into the next node left in the | |
79f95c82 | 3194 | * tree. |
aa5d6bed CM |
3195 | * |
3196 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3197 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3198 | */ |
98ed5174 | 3199 | static int push_node_left(struct btrfs_trans_handle *trans, |
2ff7e61e JM |
3200 | struct btrfs_fs_info *fs_info, |
3201 | struct extent_buffer *dst, | |
971a1f66 | 3202 | struct extent_buffer *src, int empty) |
be0e5c09 | 3203 | { |
be0e5c09 | 3204 | int push_items = 0; |
bb803951 CM |
3205 | int src_nritems; |
3206 | int dst_nritems; | |
aa5d6bed | 3207 | int ret = 0; |
be0e5c09 | 3208 | |
5f39d397 CM |
3209 | src_nritems = btrfs_header_nritems(src); |
3210 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3211 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
7bb86316 CM |
3212 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3213 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3214 | |
bce4eae9 | 3215 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3216 | return 1; |
3217 | ||
d397712b | 3218 | if (push_items <= 0) |
be0e5c09 CM |
3219 | return 1; |
3220 | ||
bce4eae9 | 3221 | if (empty) { |
971a1f66 | 3222 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3223 | if (push_items < src_nritems) { |
3224 | /* leave at least 8 pointers in the node if | |
3225 | * we aren't going to empty it | |
3226 | */ | |
3227 | if (src_nritems - push_items < 8) { | |
3228 | if (push_items <= 8) | |
3229 | return 1; | |
3230 | push_items -= 8; | |
3231 | } | |
3232 | } | |
3233 | } else | |
3234 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3235 | |
0b246afa | 3236 | ret = tree_mod_log_eb_copy(fs_info, dst, src, dst_nritems, 0, |
5de865ee FDBM |
3237 | push_items); |
3238 | if (ret) { | |
66642832 | 3239 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3240 | return ret; |
3241 | } | |
5f39d397 CM |
3242 | copy_extent_buffer(dst, src, |
3243 | btrfs_node_key_ptr_offset(dst_nritems), | |
3244 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3245 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3246 | |
bb803951 | 3247 | if (push_items < src_nritems) { |
57911b8b JS |
3248 | /* |
3249 | * don't call tree_mod_log_eb_move here, key removal was already | |
3250 | * fully logged by tree_mod_log_eb_copy above. | |
3251 | */ | |
5f39d397 CM |
3252 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3253 | btrfs_node_key_ptr_offset(push_items), | |
3254 | (src_nritems - push_items) * | |
3255 | sizeof(struct btrfs_key_ptr)); | |
3256 | } | |
3257 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3258 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3259 | btrfs_mark_buffer_dirty(src); | |
3260 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3261 | |
79f95c82 CM |
3262 | return ret; |
3263 | } | |
3264 | ||
3265 | /* | |
3266 | * try to push data from one node into the next node right in the | |
3267 | * tree. | |
3268 | * | |
3269 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3270 | * error, and > 0 if there was no room in the right hand block. | |
3271 | * | |
3272 | * this will only push up to 1/2 the contents of the left node over | |
3273 | */ | |
5f39d397 | 3274 | static int balance_node_right(struct btrfs_trans_handle *trans, |
2ff7e61e | 3275 | struct btrfs_fs_info *fs_info, |
5f39d397 CM |
3276 | struct extent_buffer *dst, |
3277 | struct extent_buffer *src) | |
79f95c82 | 3278 | { |
79f95c82 CM |
3279 | int push_items = 0; |
3280 | int max_push; | |
3281 | int src_nritems; | |
3282 | int dst_nritems; | |
3283 | int ret = 0; | |
79f95c82 | 3284 | |
7bb86316 CM |
3285 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3286 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3287 | ||
5f39d397 CM |
3288 | src_nritems = btrfs_header_nritems(src); |
3289 | dst_nritems = btrfs_header_nritems(dst); | |
0b246afa | 3290 | push_items = BTRFS_NODEPTRS_PER_BLOCK(fs_info) - dst_nritems; |
d397712b | 3291 | if (push_items <= 0) |
79f95c82 | 3292 | return 1; |
bce4eae9 | 3293 | |
d397712b | 3294 | if (src_nritems < 4) |
bce4eae9 | 3295 | return 1; |
79f95c82 CM |
3296 | |
3297 | max_push = src_nritems / 2 + 1; | |
3298 | /* don't try to empty the node */ | |
d397712b | 3299 | if (max_push >= src_nritems) |
79f95c82 | 3300 | return 1; |
252c38f0 | 3301 | |
79f95c82 CM |
3302 | if (max_push < push_items) |
3303 | push_items = max_push; | |
3304 | ||
0b246afa | 3305 | tree_mod_log_eb_move(fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3306 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3307 | btrfs_node_key_ptr_offset(0), | |
3308 | (dst_nritems) * | |
3309 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3310 | |
0b246afa | 3311 | ret = tree_mod_log_eb_copy(fs_info, dst, src, 0, |
5de865ee FDBM |
3312 | src_nritems - push_items, push_items); |
3313 | if (ret) { | |
66642832 | 3314 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3315 | return ret; |
3316 | } | |
5f39d397 CM |
3317 | copy_extent_buffer(dst, src, |
3318 | btrfs_node_key_ptr_offset(0), | |
3319 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3320 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3321 | |
5f39d397 CM |
3322 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3323 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3324 | |
5f39d397 CM |
3325 | btrfs_mark_buffer_dirty(src); |
3326 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3327 | |
aa5d6bed | 3328 | return ret; |
be0e5c09 CM |
3329 | } |
3330 | ||
97571fd0 CM |
3331 | /* |
3332 | * helper function to insert a new root level in the tree. | |
3333 | * A new node is allocated, and a single item is inserted to | |
3334 | * point to the existing root | |
aa5d6bed CM |
3335 | * |
3336 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3337 | */ |
d397712b | 3338 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3339 | struct btrfs_root *root, |
fdd99c72 | 3340 | struct btrfs_path *path, int level) |
5c680ed6 | 3341 | { |
0b246afa | 3342 | struct btrfs_fs_info *fs_info = root->fs_info; |
7bb86316 | 3343 | u64 lower_gen; |
5f39d397 CM |
3344 | struct extent_buffer *lower; |
3345 | struct extent_buffer *c; | |
925baedd | 3346 | struct extent_buffer *old; |
5f39d397 | 3347 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3348 | |
3349 | BUG_ON(path->nodes[level]); | |
3350 | BUG_ON(path->nodes[level-1] != root->node); | |
3351 | ||
7bb86316 CM |
3352 | lower = path->nodes[level-1]; |
3353 | if (level == 1) | |
3354 | btrfs_item_key(lower, &lower_key, 0); | |
3355 | else | |
3356 | btrfs_node_key(lower, &lower_key, 0); | |
3357 | ||
4d75f8a9 DS |
3358 | c = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3359 | &lower_key, level, root->node->start, 0); | |
5f39d397 CM |
3360 | if (IS_ERR(c)) |
3361 | return PTR_ERR(c); | |
925baedd | 3362 | |
0b246afa | 3363 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3364 | |
b159fa28 | 3365 | memzero_extent_buffer(c, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3366 | btrfs_set_header_nritems(c, 1); |
3367 | btrfs_set_header_level(c, level); | |
db94535d | 3368 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3369 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3370 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3371 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3372 | |
0b246afa JM |
3373 | write_extent_buffer_fsid(c, fs_info->fsid); |
3374 | write_extent_buffer_chunk_tree_uuid(c, fs_info->chunk_tree_uuid); | |
e17cade2 | 3375 | |
5f39d397 | 3376 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3377 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3378 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3379 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3380 | |
3381 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3382 | |
5f39d397 | 3383 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3384 | |
925baedd | 3385 | old = root->node; |
fdd99c72 | 3386 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3387 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3388 | |
3389 | /* the super has an extra ref to root->node */ | |
3390 | free_extent_buffer(old); | |
3391 | ||
0b86a832 | 3392 | add_root_to_dirty_list(root); |
5f39d397 CM |
3393 | extent_buffer_get(c); |
3394 | path->nodes[level] = c; | |
95449a16 | 3395 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
5c680ed6 CM |
3396 | path->slots[level] = 0; |
3397 | return 0; | |
3398 | } | |
3399 | ||
74123bd7 CM |
3400 | /* |
3401 | * worker function to insert a single pointer in a node. | |
3402 | * the node should have enough room for the pointer already | |
97571fd0 | 3403 | * |
74123bd7 CM |
3404 | * slot and level indicate where you want the key to go, and |
3405 | * blocknr is the block the key points to. | |
3406 | */ | |
143bede5 | 3407 | static void insert_ptr(struct btrfs_trans_handle *trans, |
2ff7e61e | 3408 | struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 3409 | struct btrfs_disk_key *key, u64 bytenr, |
c3e06965 | 3410 | int slot, int level) |
74123bd7 | 3411 | { |
5f39d397 | 3412 | struct extent_buffer *lower; |
74123bd7 | 3413 | int nritems; |
f3ea38da | 3414 | int ret; |
5c680ed6 CM |
3415 | |
3416 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3417 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3418 | lower = path->nodes[level]; |
3419 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3420 | BUG_ON(slot > nritems); |
0b246afa | 3421 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
74123bd7 | 3422 | if (slot != nritems) { |
c3e06965 | 3423 | if (level) |
0b246afa | 3424 | tree_mod_log_eb_move(fs_info, lower, slot + 1, |
f3ea38da | 3425 | slot, nritems - slot); |
5f39d397 CM |
3426 | memmove_extent_buffer(lower, |
3427 | btrfs_node_key_ptr_offset(slot + 1), | |
3428 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3429 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3430 | } |
c3e06965 | 3431 | if (level) { |
0b246afa | 3432 | ret = tree_mod_log_insert_key(fs_info, lower, slot, |
c8cc6341 | 3433 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3434 | BUG_ON(ret < 0); |
3435 | } | |
5f39d397 | 3436 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3437 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3438 | WARN_ON(trans->transid == 0); |
3439 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3440 | btrfs_set_header_nritems(lower, nritems + 1); |
3441 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3442 | } |
3443 | ||
97571fd0 CM |
3444 | /* |
3445 | * split the node at the specified level in path in two. | |
3446 | * The path is corrected to point to the appropriate node after the split | |
3447 | * | |
3448 | * Before splitting this tries to make some room in the node by pushing | |
3449 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3450 | * |
3451 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3452 | */ |
e02119d5 CM |
3453 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3454 | struct btrfs_root *root, | |
3455 | struct btrfs_path *path, int level) | |
be0e5c09 | 3456 | { |
0b246afa | 3457 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
3458 | struct extent_buffer *c; |
3459 | struct extent_buffer *split; | |
3460 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3461 | int mid; |
5c680ed6 | 3462 | int ret; |
7518a238 | 3463 | u32 c_nritems; |
eb60ceac | 3464 | |
5f39d397 | 3465 | c = path->nodes[level]; |
7bb86316 | 3466 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3467 | if (c == root->node) { |
d9abbf1c | 3468 | /* |
90f8d62e JS |
3469 | * trying to split the root, lets make a new one |
3470 | * | |
fdd99c72 | 3471 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3472 | * insert_new_root, because that root buffer will be kept as a |
3473 | * normal node. We are going to log removal of half of the | |
3474 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3475 | * tree lock on the buffer, which is why we cannot race with | |
3476 | * other tree_mod_log users. | |
d9abbf1c | 3477 | */ |
fdd99c72 | 3478 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3479 | if (ret) |
3480 | return ret; | |
b3612421 | 3481 | } else { |
e66f709b | 3482 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3483 | c = path->nodes[level]; |
3484 | if (!ret && btrfs_header_nritems(c) < | |
0b246afa | 3485 | BTRFS_NODEPTRS_PER_BLOCK(fs_info) - 3) |
e66f709b | 3486 | return 0; |
54aa1f4d CM |
3487 | if (ret < 0) |
3488 | return ret; | |
be0e5c09 | 3489 | } |
e66f709b | 3490 | |
5f39d397 | 3491 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3492 | mid = (c_nritems + 1) / 2; |
3493 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3494 | |
4d75f8a9 DS |
3495 | split = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
3496 | &disk_key, level, c->start, 0); | |
5f39d397 CM |
3497 | if (IS_ERR(split)) |
3498 | return PTR_ERR(split); | |
3499 | ||
0b246afa | 3500 | root_add_used(root, fs_info->nodesize); |
f0486c68 | 3501 | |
b159fa28 | 3502 | memzero_extent_buffer(split, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3503 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3504 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3505 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3506 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3507 | btrfs_set_header_owner(split, root->root_key.objectid); |
0b246afa JM |
3508 | write_extent_buffer_fsid(split, fs_info->fsid); |
3509 | write_extent_buffer_chunk_tree_uuid(split, fs_info->chunk_tree_uuid); | |
54aa1f4d | 3510 | |
0b246afa | 3511 | ret = tree_mod_log_eb_copy(fs_info, split, c, 0, mid, c_nritems - mid); |
5de865ee | 3512 | if (ret) { |
66642832 | 3513 | btrfs_abort_transaction(trans, ret); |
5de865ee FDBM |
3514 | return ret; |
3515 | } | |
5f39d397 CM |
3516 | copy_extent_buffer(split, c, |
3517 | btrfs_node_key_ptr_offset(0), | |
3518 | btrfs_node_key_ptr_offset(mid), | |
3519 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3520 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3521 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3522 | ret = 0; |
3523 | ||
5f39d397 CM |
3524 | btrfs_mark_buffer_dirty(c); |
3525 | btrfs_mark_buffer_dirty(split); | |
3526 | ||
2ff7e61e | 3527 | insert_ptr(trans, fs_info, path, &disk_key, split->start, |
c3e06965 | 3528 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3529 | |
5de08d7d | 3530 | if (path->slots[level] >= mid) { |
5c680ed6 | 3531 | path->slots[level] -= mid; |
925baedd | 3532 | btrfs_tree_unlock(c); |
5f39d397 CM |
3533 | free_extent_buffer(c); |
3534 | path->nodes[level] = split; | |
5c680ed6 CM |
3535 | path->slots[level + 1] += 1; |
3536 | } else { | |
925baedd | 3537 | btrfs_tree_unlock(split); |
5f39d397 | 3538 | free_extent_buffer(split); |
be0e5c09 | 3539 | } |
aa5d6bed | 3540 | return ret; |
be0e5c09 CM |
3541 | } |
3542 | ||
74123bd7 CM |
3543 | /* |
3544 | * how many bytes are required to store the items in a leaf. start | |
3545 | * and nr indicate which items in the leaf to check. This totals up the | |
3546 | * space used both by the item structs and the item data | |
3547 | */ | |
5f39d397 | 3548 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3549 | { |
41be1f3b JB |
3550 | struct btrfs_item *start_item; |
3551 | struct btrfs_item *end_item; | |
3552 | struct btrfs_map_token token; | |
be0e5c09 | 3553 | int data_len; |
5f39d397 | 3554 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3555 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3556 | |
3557 | if (!nr) | |
3558 | return 0; | |
41be1f3b | 3559 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3560 | start_item = btrfs_item_nr(start); |
3561 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3562 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3563 | btrfs_token_item_size(l, start_item, &token); | |
3564 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3565 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3566 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3567 | return data_len; |
3568 | } | |
3569 | ||
d4dbff95 CM |
3570 | /* |
3571 | * The space between the end of the leaf items and | |
3572 | * the start of the leaf data. IOW, how much room | |
3573 | * the leaf has left for both items and data | |
3574 | */ | |
2ff7e61e | 3575 | noinline int btrfs_leaf_free_space(struct btrfs_fs_info *fs_info, |
e02119d5 | 3576 | struct extent_buffer *leaf) |
d4dbff95 | 3577 | { |
5f39d397 CM |
3578 | int nritems = btrfs_header_nritems(leaf); |
3579 | int ret; | |
0b246afa JM |
3580 | |
3581 | ret = BTRFS_LEAF_DATA_SIZE(fs_info) - leaf_space_used(leaf, 0, nritems); | |
5f39d397 | 3582 | if (ret < 0) { |
0b246afa JM |
3583 | btrfs_crit(fs_info, |
3584 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
3585 | ret, | |
3586 | (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info), | |
3587 | leaf_space_used(leaf, 0, nritems), nritems); | |
5f39d397 CM |
3588 | } |
3589 | return ret; | |
d4dbff95 CM |
3590 | } |
3591 | ||
99d8f83c CM |
3592 | /* |
3593 | * min slot controls the lowest index we're willing to push to the | |
3594 | * right. We'll push up to and including min_slot, but no lower | |
3595 | */ | |
1e47eef2 | 3596 | static noinline int __push_leaf_right(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3597 | struct btrfs_path *path, |
3598 | int data_size, int empty, | |
3599 | struct extent_buffer *right, | |
99d8f83c CM |
3600 | int free_space, u32 left_nritems, |
3601 | u32 min_slot) | |
00ec4c51 | 3602 | { |
5f39d397 | 3603 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3604 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3605 | struct btrfs_map_token token; |
5f39d397 | 3606 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3607 | int slot; |
34a38218 | 3608 | u32 i; |
00ec4c51 CM |
3609 | int push_space = 0; |
3610 | int push_items = 0; | |
0783fcfc | 3611 | struct btrfs_item *item; |
34a38218 | 3612 | u32 nr; |
7518a238 | 3613 | u32 right_nritems; |
5f39d397 | 3614 | u32 data_end; |
db94535d | 3615 | u32 this_item_size; |
00ec4c51 | 3616 | |
cfed81a0 CM |
3617 | btrfs_init_map_token(&token); |
3618 | ||
34a38218 CM |
3619 | if (empty) |
3620 | nr = 0; | |
3621 | else | |
99d8f83c | 3622 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3623 | |
31840ae1 | 3624 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3625 | push_space += data_size; |
31840ae1 | 3626 | |
44871b1b | 3627 | slot = path->slots[1]; |
34a38218 CM |
3628 | i = left_nritems - 1; |
3629 | while (i >= nr) { | |
dd3cc16b | 3630 | item = btrfs_item_nr(i); |
db94535d | 3631 | |
31840ae1 ZY |
3632 | if (!empty && push_items > 0) { |
3633 | if (path->slots[0] > i) | |
3634 | break; | |
3635 | if (path->slots[0] == i) { | |
2ff7e61e | 3636 | int space = btrfs_leaf_free_space(fs_info, left); |
31840ae1 ZY |
3637 | if (space + push_space * 2 > free_space) |
3638 | break; | |
3639 | } | |
3640 | } | |
3641 | ||
00ec4c51 | 3642 | if (path->slots[0] == i) |
87b29b20 | 3643 | push_space += data_size; |
db94535d | 3644 | |
db94535d CM |
3645 | this_item_size = btrfs_item_size(left, item); |
3646 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3647 | break; |
31840ae1 | 3648 | |
00ec4c51 | 3649 | push_items++; |
db94535d | 3650 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3651 | if (i == 0) |
3652 | break; | |
3653 | i--; | |
db94535d | 3654 | } |
5f39d397 | 3655 | |
925baedd CM |
3656 | if (push_items == 0) |
3657 | goto out_unlock; | |
5f39d397 | 3658 | |
6c1500f2 | 3659 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3660 | |
00ec4c51 | 3661 | /* push left to right */ |
5f39d397 | 3662 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3663 | |
5f39d397 | 3664 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
2ff7e61e | 3665 | push_space -= leaf_data_end(fs_info, left); |
5f39d397 | 3666 | |
00ec4c51 | 3667 | /* make room in the right data area */ |
2ff7e61e | 3668 | data_end = leaf_data_end(fs_info, right); |
5f39d397 | 3669 | memmove_extent_buffer(right, |
3d9ec8c4 NB |
3670 | BTRFS_LEAF_DATA_OFFSET + data_end - push_space, |
3671 | BTRFS_LEAF_DATA_OFFSET + data_end, | |
0b246afa | 3672 | BTRFS_LEAF_DATA_SIZE(fs_info) - data_end); |
5f39d397 | 3673 | |
00ec4c51 | 3674 | /* copy from the left data area */ |
3d9ec8c4 | 3675 | copy_extent_buffer(right, left, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3676 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3677 | BTRFS_LEAF_DATA_OFFSET + leaf_data_end(fs_info, left), |
d6025579 | 3678 | push_space); |
5f39d397 CM |
3679 | |
3680 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3681 | btrfs_item_nr_offset(0), | |
3682 | right_nritems * sizeof(struct btrfs_item)); | |
3683 | ||
00ec4c51 | 3684 | /* copy the items from left to right */ |
5f39d397 CM |
3685 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3686 | btrfs_item_nr_offset(left_nritems - push_items), | |
3687 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3688 | |
3689 | /* update the item pointers */ | |
7518a238 | 3690 | right_nritems += push_items; |
5f39d397 | 3691 | btrfs_set_header_nritems(right, right_nritems); |
0b246afa | 3692 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
7518a238 | 3693 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3694 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3695 | push_space -= btrfs_token_item_size(right, item, &token); |
3696 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3697 | } |
3698 | ||
7518a238 | 3699 | left_nritems -= push_items; |
5f39d397 | 3700 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3701 | |
34a38218 CM |
3702 | if (left_nritems) |
3703 | btrfs_mark_buffer_dirty(left); | |
f0486c68 | 3704 | else |
7c302b49 | 3705 | clean_tree_block(fs_info, left); |
f0486c68 | 3706 | |
5f39d397 | 3707 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3708 | |
5f39d397 CM |
3709 | btrfs_item_key(right, &disk_key, 0); |
3710 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3711 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3712 | |
00ec4c51 | 3713 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3714 | if (path->slots[0] >= left_nritems) { |
3715 | path->slots[0] -= left_nritems; | |
925baedd | 3716 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
7c302b49 | 3717 | clean_tree_block(fs_info, path->nodes[0]); |
925baedd | 3718 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3719 | free_extent_buffer(path->nodes[0]); |
3720 | path->nodes[0] = right; | |
00ec4c51 CM |
3721 | path->slots[1] += 1; |
3722 | } else { | |
925baedd | 3723 | btrfs_tree_unlock(right); |
5f39d397 | 3724 | free_extent_buffer(right); |
00ec4c51 CM |
3725 | } |
3726 | return 0; | |
925baedd CM |
3727 | |
3728 | out_unlock: | |
3729 | btrfs_tree_unlock(right); | |
3730 | free_extent_buffer(right); | |
3731 | return 1; | |
00ec4c51 | 3732 | } |
925baedd | 3733 | |
44871b1b CM |
3734 | /* |
3735 | * push some data in the path leaf to the right, trying to free up at | |
3736 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3737 | * | |
3738 | * returns 1 if the push failed because the other node didn't have enough | |
3739 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3740 | * |
3741 | * this will push starting from min_slot to the end of the leaf. It won't | |
3742 | * push any slot lower than min_slot | |
44871b1b CM |
3743 | */ |
3744 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3745 | *root, struct btrfs_path *path, |
3746 | int min_data_size, int data_size, | |
3747 | int empty, u32 min_slot) | |
44871b1b | 3748 | { |
2ff7e61e | 3749 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3750 | struct extent_buffer *left = path->nodes[0]; |
3751 | struct extent_buffer *right; | |
3752 | struct extent_buffer *upper; | |
3753 | int slot; | |
3754 | int free_space; | |
3755 | u32 left_nritems; | |
3756 | int ret; | |
3757 | ||
3758 | if (!path->nodes[1]) | |
3759 | return 1; | |
3760 | ||
3761 | slot = path->slots[1]; | |
3762 | upper = path->nodes[1]; | |
3763 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3764 | return 1; | |
3765 | ||
3766 | btrfs_assert_tree_locked(path->nodes[1]); | |
3767 | ||
2ff7e61e | 3768 | right = read_node_slot(fs_info, upper, slot + 1); |
fb770ae4 LB |
3769 | /* |
3770 | * slot + 1 is not valid or we fail to read the right node, | |
3771 | * no big deal, just return. | |
3772 | */ | |
3773 | if (IS_ERR(right)) | |
91ca338d TI |
3774 | return 1; |
3775 | ||
44871b1b CM |
3776 | btrfs_tree_lock(right); |
3777 | btrfs_set_lock_blocking(right); | |
3778 | ||
2ff7e61e | 3779 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3780 | if (free_space < data_size) |
3781 | goto out_unlock; | |
3782 | ||
3783 | /* cow and double check */ | |
3784 | ret = btrfs_cow_block(trans, root, right, upper, | |
3785 | slot + 1, &right); | |
3786 | if (ret) | |
3787 | goto out_unlock; | |
3788 | ||
2ff7e61e | 3789 | free_space = btrfs_leaf_free_space(fs_info, right); |
44871b1b CM |
3790 | if (free_space < data_size) |
3791 | goto out_unlock; | |
3792 | ||
3793 | left_nritems = btrfs_header_nritems(left); | |
3794 | if (left_nritems == 0) | |
3795 | goto out_unlock; | |
3796 | ||
2ef1fed2 FDBM |
3797 | if (path->slots[0] == left_nritems && !empty) { |
3798 | /* Key greater than all keys in the leaf, right neighbor has | |
3799 | * enough room for it and we're not emptying our leaf to delete | |
3800 | * it, therefore use right neighbor to insert the new item and | |
3801 | * no need to touch/dirty our left leaft. */ | |
3802 | btrfs_tree_unlock(left); | |
3803 | free_extent_buffer(left); | |
3804 | path->nodes[0] = right; | |
3805 | path->slots[0] = 0; | |
3806 | path->slots[1]++; | |
3807 | return 0; | |
3808 | } | |
3809 | ||
1e47eef2 | 3810 | return __push_leaf_right(fs_info, path, min_data_size, empty, |
99d8f83c | 3811 | right, free_space, left_nritems, min_slot); |
44871b1b CM |
3812 | out_unlock: |
3813 | btrfs_tree_unlock(right); | |
3814 | free_extent_buffer(right); | |
3815 | return 1; | |
3816 | } | |
3817 | ||
74123bd7 CM |
3818 | /* |
3819 | * push some data in the path leaf to the left, trying to free up at | |
3820 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3821 | * |
3822 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3823 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3824 | * items | |
74123bd7 | 3825 | */ |
66cb7ddb | 3826 | static noinline int __push_leaf_left(struct btrfs_fs_info *fs_info, |
44871b1b CM |
3827 | struct btrfs_path *path, int data_size, |
3828 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3829 | int free_space, u32 right_nritems, |
3830 | u32 max_slot) | |
be0e5c09 | 3831 | { |
5f39d397 CM |
3832 | struct btrfs_disk_key disk_key; |
3833 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3834 | int i; |
be0e5c09 CM |
3835 | int push_space = 0; |
3836 | int push_items = 0; | |
0783fcfc | 3837 | struct btrfs_item *item; |
7518a238 | 3838 | u32 old_left_nritems; |
34a38218 | 3839 | u32 nr; |
aa5d6bed | 3840 | int ret = 0; |
db94535d CM |
3841 | u32 this_item_size; |
3842 | u32 old_left_item_size; | |
cfed81a0 CM |
3843 | struct btrfs_map_token token; |
3844 | ||
3845 | btrfs_init_map_token(&token); | |
be0e5c09 | 3846 | |
34a38218 | 3847 | if (empty) |
99d8f83c | 3848 | nr = min(right_nritems, max_slot); |
34a38218 | 3849 | else |
99d8f83c | 3850 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3851 | |
3852 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3853 | item = btrfs_item_nr(i); |
db94535d | 3854 | |
31840ae1 ZY |
3855 | if (!empty && push_items > 0) { |
3856 | if (path->slots[0] < i) | |
3857 | break; | |
3858 | if (path->slots[0] == i) { | |
2ff7e61e | 3859 | int space = btrfs_leaf_free_space(fs_info, right); |
31840ae1 ZY |
3860 | if (space + push_space * 2 > free_space) |
3861 | break; | |
3862 | } | |
3863 | } | |
3864 | ||
be0e5c09 | 3865 | if (path->slots[0] == i) |
87b29b20 | 3866 | push_space += data_size; |
db94535d CM |
3867 | |
3868 | this_item_size = btrfs_item_size(right, item); | |
3869 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3870 | break; |
db94535d | 3871 | |
be0e5c09 | 3872 | push_items++; |
db94535d CM |
3873 | push_space += this_item_size + sizeof(*item); |
3874 | } | |
3875 | ||
be0e5c09 | 3876 | if (push_items == 0) { |
925baedd CM |
3877 | ret = 1; |
3878 | goto out; | |
be0e5c09 | 3879 | } |
fae7f21c | 3880 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3881 | |
be0e5c09 | 3882 | /* push data from right to left */ |
5f39d397 CM |
3883 | copy_extent_buffer(left, right, |
3884 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3885 | btrfs_item_nr_offset(0), | |
3886 | push_items * sizeof(struct btrfs_item)); | |
3887 | ||
0b246afa | 3888 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info) - |
d397712b | 3889 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 | 3890 | |
3d9ec8c4 | 3891 | copy_extent_buffer(left, right, BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3892 | leaf_data_end(fs_info, left) - push_space, |
3d9ec8c4 | 3893 | BTRFS_LEAF_DATA_OFFSET + |
5f39d397 | 3894 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3895 | push_space); |
5f39d397 | 3896 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3897 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3898 | |
db94535d | 3899 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3900 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3901 | u32 ioff; |
db94535d | 3902 | |
dd3cc16b | 3903 | item = btrfs_item_nr(i); |
db94535d | 3904 | |
cfed81a0 CM |
3905 | ioff = btrfs_token_item_offset(left, item, &token); |
3906 | btrfs_set_token_item_offset(left, item, | |
0b246afa | 3907 | ioff - (BTRFS_LEAF_DATA_SIZE(fs_info) - old_left_item_size), |
cfed81a0 | 3908 | &token); |
be0e5c09 | 3909 | } |
5f39d397 | 3910 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3911 | |
3912 | /* fixup right node */ | |
31b1a2bd JL |
3913 | if (push_items > right_nritems) |
3914 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3915 | right_nritems); |
34a38218 CM |
3916 | |
3917 | if (push_items < right_nritems) { | |
3918 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
2ff7e61e | 3919 | leaf_data_end(fs_info, right); |
3d9ec8c4 | 3920 | memmove_extent_buffer(right, BTRFS_LEAF_DATA_OFFSET + |
0b246afa | 3921 | BTRFS_LEAF_DATA_SIZE(fs_info) - push_space, |
3d9ec8c4 | 3922 | BTRFS_LEAF_DATA_OFFSET + |
2ff7e61e | 3923 | leaf_data_end(fs_info, right), push_space); |
34a38218 CM |
3924 | |
3925 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3926 | btrfs_item_nr_offset(push_items), |
3927 | (btrfs_header_nritems(right) - push_items) * | |
3928 | sizeof(struct btrfs_item)); | |
34a38218 | 3929 | } |
eef1c494 Y |
3930 | right_nritems -= push_items; |
3931 | btrfs_set_header_nritems(right, right_nritems); | |
0b246afa | 3932 | push_space = BTRFS_LEAF_DATA_SIZE(fs_info); |
5f39d397 | 3933 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3934 | item = btrfs_item_nr(i); |
db94535d | 3935 | |
cfed81a0 CM |
3936 | push_space = push_space - btrfs_token_item_size(right, |
3937 | item, &token); | |
3938 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3939 | } |
eb60ceac | 3940 | |
5f39d397 | 3941 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3942 | if (right_nritems) |
3943 | btrfs_mark_buffer_dirty(right); | |
f0486c68 | 3944 | else |
7c302b49 | 3945 | clean_tree_block(fs_info, right); |
098f59c2 | 3946 | |
5f39d397 | 3947 | btrfs_item_key(right, &disk_key, 0); |
0b246afa | 3948 | fixup_low_keys(fs_info, path, &disk_key, 1); |
be0e5c09 CM |
3949 | |
3950 | /* then fixup the leaf pointer in the path */ | |
3951 | if (path->slots[0] < push_items) { | |
3952 | path->slots[0] += old_left_nritems; | |
925baedd | 3953 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3954 | free_extent_buffer(path->nodes[0]); |
3955 | path->nodes[0] = left; | |
be0e5c09 CM |
3956 | path->slots[1] -= 1; |
3957 | } else { | |
925baedd | 3958 | btrfs_tree_unlock(left); |
5f39d397 | 3959 | free_extent_buffer(left); |
be0e5c09 CM |
3960 | path->slots[0] -= push_items; |
3961 | } | |
eb60ceac | 3962 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3963 | return ret; |
925baedd CM |
3964 | out: |
3965 | btrfs_tree_unlock(left); | |
3966 | free_extent_buffer(left); | |
3967 | return ret; | |
be0e5c09 CM |
3968 | } |
3969 | ||
44871b1b CM |
3970 | /* |
3971 | * push some data in the path leaf to the left, trying to free up at | |
3972 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3973 | * |
3974 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3975 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3976 | * items | |
44871b1b CM |
3977 | */ |
3978 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3979 | *root, struct btrfs_path *path, int min_data_size, |
3980 | int data_size, int empty, u32 max_slot) | |
44871b1b | 3981 | { |
2ff7e61e | 3982 | struct btrfs_fs_info *fs_info = root->fs_info; |
44871b1b CM |
3983 | struct extent_buffer *right = path->nodes[0]; |
3984 | struct extent_buffer *left; | |
3985 | int slot; | |
3986 | int free_space; | |
3987 | u32 right_nritems; | |
3988 | int ret = 0; | |
3989 | ||
3990 | slot = path->slots[1]; | |
3991 | if (slot == 0) | |
3992 | return 1; | |
3993 | if (!path->nodes[1]) | |
3994 | return 1; | |
3995 | ||
3996 | right_nritems = btrfs_header_nritems(right); | |
3997 | if (right_nritems == 0) | |
3998 | return 1; | |
3999 | ||
4000 | btrfs_assert_tree_locked(path->nodes[1]); | |
4001 | ||
2ff7e61e | 4002 | left = read_node_slot(fs_info, path->nodes[1], slot - 1); |
fb770ae4 LB |
4003 | /* |
4004 | * slot - 1 is not valid or we fail to read the left node, | |
4005 | * no big deal, just return. | |
4006 | */ | |
4007 | if (IS_ERR(left)) | |
91ca338d TI |
4008 | return 1; |
4009 | ||
44871b1b CM |
4010 | btrfs_tree_lock(left); |
4011 | btrfs_set_lock_blocking(left); | |
4012 | ||
2ff7e61e | 4013 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4014 | if (free_space < data_size) { |
4015 | ret = 1; | |
4016 | goto out; | |
4017 | } | |
4018 | ||
4019 | /* cow and double check */ | |
4020 | ret = btrfs_cow_block(trans, root, left, | |
4021 | path->nodes[1], slot - 1, &left); | |
4022 | if (ret) { | |
4023 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4024 | if (ret == -ENOSPC) |
4025 | ret = 1; | |
44871b1b CM |
4026 | goto out; |
4027 | } | |
4028 | ||
2ff7e61e | 4029 | free_space = btrfs_leaf_free_space(fs_info, left); |
44871b1b CM |
4030 | if (free_space < data_size) { |
4031 | ret = 1; | |
4032 | goto out; | |
4033 | } | |
4034 | ||
66cb7ddb | 4035 | return __push_leaf_left(fs_info, path, min_data_size, |
99d8f83c CM |
4036 | empty, left, free_space, right_nritems, |
4037 | max_slot); | |
44871b1b CM |
4038 | out: |
4039 | btrfs_tree_unlock(left); | |
4040 | free_extent_buffer(left); | |
4041 | return ret; | |
4042 | } | |
4043 | ||
4044 | /* | |
4045 | * split the path's leaf in two, making sure there is at least data_size | |
4046 | * available for the resulting leaf level of the path. | |
44871b1b | 4047 | */ |
143bede5 | 4048 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
2ff7e61e | 4049 | struct btrfs_fs_info *fs_info, |
143bede5 JM |
4050 | struct btrfs_path *path, |
4051 | struct extent_buffer *l, | |
4052 | struct extent_buffer *right, | |
4053 | int slot, int mid, int nritems) | |
44871b1b CM |
4054 | { |
4055 | int data_copy_size; | |
4056 | int rt_data_off; | |
4057 | int i; | |
44871b1b | 4058 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4059 | struct btrfs_map_token token; |
4060 | ||
4061 | btrfs_init_map_token(&token); | |
44871b1b CM |
4062 | |
4063 | nritems = nritems - mid; | |
4064 | btrfs_set_header_nritems(right, nritems); | |
2ff7e61e | 4065 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(fs_info, l); |
44871b1b CM |
4066 | |
4067 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4068 | btrfs_item_nr_offset(mid), | |
4069 | nritems * sizeof(struct btrfs_item)); | |
4070 | ||
4071 | copy_extent_buffer(right, l, | |
3d9ec8c4 NB |
4072 | BTRFS_LEAF_DATA_OFFSET + BTRFS_LEAF_DATA_SIZE(fs_info) - |
4073 | data_copy_size, BTRFS_LEAF_DATA_OFFSET + | |
2ff7e61e | 4074 | leaf_data_end(fs_info, l), data_copy_size); |
44871b1b | 4075 | |
0b246afa | 4076 | rt_data_off = BTRFS_LEAF_DATA_SIZE(fs_info) - btrfs_item_end_nr(l, mid); |
44871b1b CM |
4077 | |
4078 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4079 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4080 | u32 ioff; |
4081 | ||
cfed81a0 CM |
4082 | ioff = btrfs_token_item_offset(right, item, &token); |
4083 | btrfs_set_token_item_offset(right, item, | |
4084 | ioff + rt_data_off, &token); | |
44871b1b CM |
4085 | } |
4086 | ||
44871b1b | 4087 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4088 | btrfs_item_key(right, &disk_key, 0); |
2ff7e61e | 4089 | insert_ptr(trans, fs_info, path, &disk_key, right->start, |
c3e06965 | 4090 | path->slots[1] + 1, 1); |
44871b1b CM |
4091 | |
4092 | btrfs_mark_buffer_dirty(right); | |
4093 | btrfs_mark_buffer_dirty(l); | |
4094 | BUG_ON(path->slots[0] != slot); | |
4095 | ||
44871b1b CM |
4096 | if (mid <= slot) { |
4097 | btrfs_tree_unlock(path->nodes[0]); | |
4098 | free_extent_buffer(path->nodes[0]); | |
4099 | path->nodes[0] = right; | |
4100 | path->slots[0] -= mid; | |
4101 | path->slots[1] += 1; | |
4102 | } else { | |
4103 | btrfs_tree_unlock(right); | |
4104 | free_extent_buffer(right); | |
4105 | } | |
4106 | ||
4107 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4108 | } |
4109 | ||
99d8f83c CM |
4110 | /* |
4111 | * double splits happen when we need to insert a big item in the middle | |
4112 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4113 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4114 | * A B C | |
4115 | * | |
4116 | * We avoid this by trying to push the items on either side of our target | |
4117 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4118 | * completely. | |
4119 | */ | |
4120 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4121 | struct btrfs_root *root, | |
4122 | struct btrfs_path *path, | |
4123 | int data_size) | |
4124 | { | |
2ff7e61e | 4125 | struct btrfs_fs_info *fs_info = root->fs_info; |
99d8f83c CM |
4126 | int ret; |
4127 | int progress = 0; | |
4128 | int slot; | |
4129 | u32 nritems; | |
5a4267ca | 4130 | int space_needed = data_size; |
99d8f83c CM |
4131 | |
4132 | slot = path->slots[0]; | |
5a4267ca | 4133 | if (slot < btrfs_header_nritems(path->nodes[0])) |
2ff7e61e | 4134 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); |
99d8f83c CM |
4135 | |
4136 | /* | |
4137 | * try to push all the items after our slot into the | |
4138 | * right leaf | |
4139 | */ | |
5a4267ca | 4140 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4141 | if (ret < 0) |
4142 | return ret; | |
4143 | ||
4144 | if (ret == 0) | |
4145 | progress++; | |
4146 | ||
4147 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4148 | /* | |
4149 | * our goal is to get our slot at the start or end of a leaf. If | |
4150 | * we've done so we're done | |
4151 | */ | |
4152 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4153 | return 0; | |
4154 | ||
2ff7e61e | 4155 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4156 | return 0; |
4157 | ||
4158 | /* try to push all the items before our slot into the next leaf */ | |
4159 | slot = path->slots[0]; | |
263d3995 FM |
4160 | space_needed = data_size; |
4161 | if (slot > 0) | |
4162 | space_needed -= btrfs_leaf_free_space(fs_info, path->nodes[0]); | |
5a4267ca | 4163 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4164 | if (ret < 0) |
4165 | return ret; | |
4166 | ||
4167 | if (ret == 0) | |
4168 | progress++; | |
4169 | ||
4170 | if (progress) | |
4171 | return 0; | |
4172 | return 1; | |
4173 | } | |
4174 | ||
74123bd7 CM |
4175 | /* |
4176 | * split the path's leaf in two, making sure there is at least data_size | |
4177 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4178 | * |
4179 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4180 | */ |
e02119d5 CM |
4181 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4182 | struct btrfs_root *root, | |
310712b2 | 4183 | const struct btrfs_key *ins_key, |
e02119d5 CM |
4184 | struct btrfs_path *path, int data_size, |
4185 | int extend) | |
be0e5c09 | 4186 | { |
5d4f98a2 | 4187 | struct btrfs_disk_key disk_key; |
5f39d397 | 4188 | struct extent_buffer *l; |
7518a238 | 4189 | u32 nritems; |
eb60ceac CM |
4190 | int mid; |
4191 | int slot; | |
5f39d397 | 4192 | struct extent_buffer *right; |
b7a0365e | 4193 | struct btrfs_fs_info *fs_info = root->fs_info; |
d4dbff95 | 4194 | int ret = 0; |
aa5d6bed | 4195 | int wret; |
5d4f98a2 | 4196 | int split; |
cc0c5538 | 4197 | int num_doubles = 0; |
99d8f83c | 4198 | int tried_avoid_double = 0; |
aa5d6bed | 4199 | |
a5719521 YZ |
4200 | l = path->nodes[0]; |
4201 | slot = path->slots[0]; | |
4202 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
0b246afa | 4203 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(fs_info)) |
a5719521 YZ |
4204 | return -EOVERFLOW; |
4205 | ||
40689478 | 4206 | /* first try to make some room by pushing left and right */ |
33157e05 | 4207 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4208 | int space_needed = data_size; |
4209 | ||
4210 | if (slot < btrfs_header_nritems(l)) | |
2ff7e61e | 4211 | space_needed -= btrfs_leaf_free_space(fs_info, l); |
5a4267ca FDBM |
4212 | |
4213 | wret = push_leaf_right(trans, root, path, space_needed, | |
4214 | space_needed, 0, 0); | |
d397712b | 4215 | if (wret < 0) |
eaee50e8 | 4216 | return wret; |
3685f791 | 4217 | if (wret) { |
263d3995 FM |
4218 | space_needed = data_size; |
4219 | if (slot > 0) | |
4220 | space_needed -= btrfs_leaf_free_space(fs_info, | |
4221 | l); | |
5a4267ca FDBM |
4222 | wret = push_leaf_left(trans, root, path, space_needed, |
4223 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4224 | if (wret < 0) |
4225 | return wret; | |
4226 | } | |
4227 | l = path->nodes[0]; | |
aa5d6bed | 4228 | |
3685f791 | 4229 | /* did the pushes work? */ |
2ff7e61e | 4230 | if (btrfs_leaf_free_space(fs_info, l) >= data_size) |
3685f791 | 4231 | return 0; |
3326d1b0 | 4232 | } |
aa5d6bed | 4233 | |
5c680ed6 | 4234 | if (!path->nodes[1]) { |
fdd99c72 | 4235 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4236 | if (ret) |
4237 | return ret; | |
4238 | } | |
cc0c5538 | 4239 | again: |
5d4f98a2 | 4240 | split = 1; |
cc0c5538 | 4241 | l = path->nodes[0]; |
eb60ceac | 4242 | slot = path->slots[0]; |
5f39d397 | 4243 | nritems = btrfs_header_nritems(l); |
d397712b | 4244 | mid = (nritems + 1) / 2; |
54aa1f4d | 4245 | |
5d4f98a2 YZ |
4246 | if (mid <= slot) { |
4247 | if (nritems == 1 || | |
4248 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
0b246afa | 4249 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4250 | if (slot >= nritems) { |
4251 | split = 0; | |
4252 | } else { | |
4253 | mid = slot; | |
4254 | if (mid != nritems && | |
4255 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4256 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4257 | if (data_size && !tried_avoid_double) |
4258 | goto push_for_double; | |
5d4f98a2 YZ |
4259 | split = 2; |
4260 | } | |
4261 | } | |
4262 | } | |
4263 | } else { | |
4264 | if (leaf_space_used(l, 0, mid) + data_size > | |
0b246afa | 4265 | BTRFS_LEAF_DATA_SIZE(fs_info)) { |
5d4f98a2 YZ |
4266 | if (!extend && data_size && slot == 0) { |
4267 | split = 0; | |
4268 | } else if ((extend || !data_size) && slot == 0) { | |
4269 | mid = 1; | |
4270 | } else { | |
4271 | mid = slot; | |
4272 | if (mid != nritems && | |
4273 | leaf_space_used(l, mid, nritems - mid) + | |
0b246afa | 4274 | data_size > BTRFS_LEAF_DATA_SIZE(fs_info)) { |
99d8f83c CM |
4275 | if (data_size && !tried_avoid_double) |
4276 | goto push_for_double; | |
67871254 | 4277 | split = 2; |
5d4f98a2 YZ |
4278 | } |
4279 | } | |
4280 | } | |
4281 | } | |
4282 | ||
4283 | if (split == 0) | |
4284 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4285 | else | |
4286 | btrfs_item_key(l, &disk_key, mid); | |
4287 | ||
4d75f8a9 DS |
4288 | right = btrfs_alloc_tree_block(trans, root, 0, root->root_key.objectid, |
4289 | &disk_key, 0, l->start, 0); | |
f0486c68 | 4290 | if (IS_ERR(right)) |
5f39d397 | 4291 | return PTR_ERR(right); |
f0486c68 | 4292 | |
0b246afa | 4293 | root_add_used(root, fs_info->nodesize); |
5f39d397 | 4294 | |
b159fa28 | 4295 | memzero_extent_buffer(right, 0, sizeof(struct btrfs_header)); |
db94535d | 4296 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4297 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4298 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4299 | btrfs_set_header_owner(right, root->root_key.objectid); |
4300 | btrfs_set_header_level(right, 0); | |
d24ee97b DS |
4301 | write_extent_buffer_fsid(right, fs_info->fsid); |
4302 | write_extent_buffer_chunk_tree_uuid(right, fs_info->chunk_tree_uuid); | |
44871b1b | 4303 | |
5d4f98a2 YZ |
4304 | if (split == 0) { |
4305 | if (mid <= slot) { | |
4306 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4307 | insert_ptr(trans, fs_info, path, &disk_key, |
4308 | right->start, path->slots[1] + 1, 1); | |
5d4f98a2 YZ |
4309 | btrfs_tree_unlock(path->nodes[0]); |
4310 | free_extent_buffer(path->nodes[0]); | |
4311 | path->nodes[0] = right; | |
4312 | path->slots[0] = 0; | |
4313 | path->slots[1] += 1; | |
4314 | } else { | |
4315 | btrfs_set_header_nritems(right, 0); | |
2ff7e61e JM |
4316 | insert_ptr(trans, fs_info, path, &disk_key, |
4317 | right->start, path->slots[1], 1); | |
5d4f98a2 YZ |
4318 | btrfs_tree_unlock(path->nodes[0]); |
4319 | free_extent_buffer(path->nodes[0]); | |
4320 | path->nodes[0] = right; | |
4321 | path->slots[0] = 0; | |
143bede5 | 4322 | if (path->slots[1] == 0) |
b7a0365e | 4323 | fixup_low_keys(fs_info, path, &disk_key, 1); |
d4dbff95 | 4324 | } |
196e0249 LB |
4325 | /* |
4326 | * We create a new leaf 'right' for the required ins_len and | |
4327 | * we'll do btrfs_mark_buffer_dirty() on this leaf after copying | |
4328 | * the content of ins_len to 'right'. | |
4329 | */ | |
5d4f98a2 | 4330 | return ret; |
d4dbff95 | 4331 | } |
74123bd7 | 4332 | |
2ff7e61e | 4333 | copy_for_split(trans, fs_info, path, l, right, slot, mid, nritems); |
31840ae1 | 4334 | |
5d4f98a2 | 4335 | if (split == 2) { |
cc0c5538 CM |
4336 | BUG_ON(num_doubles != 0); |
4337 | num_doubles++; | |
4338 | goto again; | |
a429e513 | 4339 | } |
44871b1b | 4340 | |
143bede5 | 4341 | return 0; |
99d8f83c CM |
4342 | |
4343 | push_for_double: | |
4344 | push_for_double_split(trans, root, path, data_size); | |
4345 | tried_avoid_double = 1; | |
2ff7e61e | 4346 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= data_size) |
99d8f83c CM |
4347 | return 0; |
4348 | goto again; | |
be0e5c09 CM |
4349 | } |
4350 | ||
ad48fd75 YZ |
4351 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4352 | struct btrfs_root *root, | |
4353 | struct btrfs_path *path, int ins_len) | |
459931ec | 4354 | { |
2ff7e61e | 4355 | struct btrfs_fs_info *fs_info = root->fs_info; |
ad48fd75 | 4356 | struct btrfs_key key; |
459931ec | 4357 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4358 | struct btrfs_file_extent_item *fi; |
4359 | u64 extent_len = 0; | |
4360 | u32 item_size; | |
4361 | int ret; | |
459931ec CM |
4362 | |
4363 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4364 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4365 | ||
4366 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4367 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4368 | ||
2ff7e61e | 4369 | if (btrfs_leaf_free_space(fs_info, leaf) >= ins_len) |
ad48fd75 | 4370 | return 0; |
459931ec CM |
4371 | |
4372 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4373 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4374 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4375 | struct btrfs_file_extent_item); | |
4376 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4377 | } | |
b3b4aa74 | 4378 | btrfs_release_path(path); |
459931ec | 4379 | |
459931ec | 4380 | path->keep_locks = 1; |
ad48fd75 YZ |
4381 | path->search_for_split = 1; |
4382 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4383 | path->search_for_split = 0; |
a8df6fe6 FM |
4384 | if (ret > 0) |
4385 | ret = -EAGAIN; | |
ad48fd75 YZ |
4386 | if (ret < 0) |
4387 | goto err; | |
459931ec | 4388 | |
ad48fd75 YZ |
4389 | ret = -EAGAIN; |
4390 | leaf = path->nodes[0]; | |
a8df6fe6 FM |
4391 | /* if our item isn't there, return now */ |
4392 | if (item_size != btrfs_item_size_nr(leaf, path->slots[0])) | |
ad48fd75 YZ |
4393 | goto err; |
4394 | ||
109f6aef | 4395 | /* the leaf has changed, it now has room. return now */ |
2ff7e61e | 4396 | if (btrfs_leaf_free_space(fs_info, path->nodes[0]) >= ins_len) |
109f6aef CM |
4397 | goto err; |
4398 | ||
ad48fd75 YZ |
4399 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4400 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4401 | struct btrfs_file_extent_item); | |
4402 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4403 | goto err; | |
459931ec CM |
4404 | } |
4405 | ||
b9473439 | 4406 | btrfs_set_path_blocking(path); |
ad48fd75 | 4407 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4408 | if (ret) |
4409 | goto err; | |
459931ec | 4410 | |
ad48fd75 | 4411 | path->keep_locks = 0; |
b9473439 | 4412 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4413 | return 0; |
4414 | err: | |
4415 | path->keep_locks = 0; | |
4416 | return ret; | |
4417 | } | |
4418 | ||
4961e293 | 4419 | static noinline int split_item(struct btrfs_fs_info *fs_info, |
ad48fd75 | 4420 | struct btrfs_path *path, |
310712b2 | 4421 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4422 | unsigned long split_offset) |
4423 | { | |
4424 | struct extent_buffer *leaf; | |
4425 | struct btrfs_item *item; | |
4426 | struct btrfs_item *new_item; | |
4427 | int slot; | |
4428 | char *buf; | |
4429 | u32 nritems; | |
4430 | u32 item_size; | |
4431 | u32 orig_offset; | |
4432 | struct btrfs_disk_key disk_key; | |
4433 | ||
b9473439 | 4434 | leaf = path->nodes[0]; |
2ff7e61e | 4435 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < sizeof(struct btrfs_item)); |
b9473439 | 4436 | |
b4ce94de CM |
4437 | btrfs_set_path_blocking(path); |
4438 | ||
dd3cc16b | 4439 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4440 | orig_offset = btrfs_item_offset(leaf, item); |
4441 | item_size = btrfs_item_size(leaf, item); | |
4442 | ||
459931ec | 4443 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4444 | if (!buf) |
4445 | return -ENOMEM; | |
4446 | ||
459931ec CM |
4447 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4448 | path->slots[0]), item_size); | |
459931ec | 4449 | |
ad48fd75 | 4450 | slot = path->slots[0] + 1; |
459931ec | 4451 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4452 | if (slot != nritems) { |
4453 | /* shift the items */ | |
4454 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4455 | btrfs_item_nr_offset(slot), |
4456 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4457 | } |
4458 | ||
4459 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4460 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4461 | ||
dd3cc16b | 4462 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4463 | |
4464 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4465 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4466 | ||
4467 | btrfs_set_item_offset(leaf, item, | |
4468 | orig_offset + item_size - split_offset); | |
4469 | btrfs_set_item_size(leaf, item, split_offset); | |
4470 | ||
4471 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4472 | ||
4473 | /* write the data for the start of the original item */ | |
4474 | write_extent_buffer(leaf, buf, | |
4475 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4476 | split_offset); | |
4477 | ||
4478 | /* write the data for the new item */ | |
4479 | write_extent_buffer(leaf, buf + split_offset, | |
4480 | btrfs_item_ptr_offset(leaf, slot), | |
4481 | item_size - split_offset); | |
4482 | btrfs_mark_buffer_dirty(leaf); | |
4483 | ||
2ff7e61e | 4484 | BUG_ON(btrfs_leaf_free_space(fs_info, leaf) < 0); |
459931ec | 4485 | kfree(buf); |
ad48fd75 YZ |
4486 | return 0; |
4487 | } | |
4488 | ||
4489 | /* | |
4490 | * This function splits a single item into two items, | |
4491 | * giving 'new_key' to the new item and splitting the | |
4492 | * old one at split_offset (from the start of the item). | |
4493 | * | |
4494 | * The path may be released by this operation. After | |
4495 | * the split, the path is pointing to the old item. The | |
4496 | * new item is going to be in the same node as the old one. | |
4497 | * | |
4498 | * Note, the item being split must be smaller enough to live alone on | |
4499 | * a tree block with room for one extra struct btrfs_item | |
4500 | * | |
4501 | * This allows us to split the item in place, keeping a lock on the | |
4502 | * leaf the entire time. | |
4503 | */ | |
4504 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4505 | struct btrfs_root *root, | |
4506 | struct btrfs_path *path, | |
310712b2 | 4507 | const struct btrfs_key *new_key, |
ad48fd75 YZ |
4508 | unsigned long split_offset) |
4509 | { | |
4510 | int ret; | |
4511 | ret = setup_leaf_for_split(trans, root, path, | |
4512 | sizeof(struct btrfs_item)); | |
4513 | if (ret) | |
4514 | return ret; | |
4515 | ||
4961e293 | 4516 | ret = split_item(root->fs_info, path, new_key, split_offset); |
459931ec CM |
4517 | return ret; |
4518 | } | |
4519 | ||
ad48fd75 YZ |
4520 | /* |
4521 | * This function duplicate a item, giving 'new_key' to the new item. | |
4522 | * It guarantees both items live in the same tree leaf and the new item | |
4523 | * is contiguous with the original item. | |
4524 | * | |
4525 | * This allows us to split file extent in place, keeping a lock on the | |
4526 | * leaf the entire time. | |
4527 | */ | |
4528 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4529 | struct btrfs_root *root, | |
4530 | struct btrfs_path *path, | |
310712b2 | 4531 | const struct btrfs_key *new_key) |
ad48fd75 YZ |
4532 | { |
4533 | struct extent_buffer *leaf; | |
4534 | int ret; | |
4535 | u32 item_size; | |
4536 | ||
4537 | leaf = path->nodes[0]; | |
4538 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4539 | ret = setup_leaf_for_split(trans, root, path, | |
4540 | item_size + sizeof(struct btrfs_item)); | |
4541 | if (ret) | |
4542 | return ret; | |
4543 | ||
4544 | path->slots[0]++; | |
afe5fea7 | 4545 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4546 | item_size, item_size + |
4547 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4548 | leaf = path->nodes[0]; |
4549 | memcpy_extent_buffer(leaf, | |
4550 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4551 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4552 | item_size); | |
4553 | return 0; | |
4554 | } | |
4555 | ||
d352ac68 CM |
4556 | /* |
4557 | * make the item pointed to by the path smaller. new_size indicates | |
4558 | * how small to make it, and from_end tells us if we just chop bytes | |
4559 | * off the end of the item or if we shift the item to chop bytes off | |
4560 | * the front. | |
4561 | */ | |
2ff7e61e JM |
4562 | void btrfs_truncate_item(struct btrfs_fs_info *fs_info, |
4563 | struct btrfs_path *path, u32 new_size, int from_end) | |
b18c6685 | 4564 | { |
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); |
2ff7e61e | 4586 | data_end = leaf_data_end(fs_info, leaf); |
b18c6685 | 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 | 4609 | if (from_end) { |
3d9ec8c4 NB |
4610 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4611 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
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 | ||
3d9ec8c4 NB |
4637 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4638 | data_end + size_diff, BTRFS_LEAF_DATA_OFFSET + | |
179e29e4 CM |
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 | |
2ff7e61e | 4652 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4653 | btrfs_print_leaf(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 | */ |
2ff7e61e | 4661 | void btrfs_extend_item(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
143bede5 | 4662 | u32 data_size) |
6567e837 | 4663 | { |
6567e837 | 4664 | int slot; |
5f39d397 CM |
4665 | struct extent_buffer *leaf; |
4666 | struct btrfs_item *item; | |
6567e837 CM |
4667 | u32 nritems; |
4668 | unsigned int data_end; | |
4669 | unsigned int old_data; | |
4670 | unsigned int old_size; | |
4671 | int i; | |
cfed81a0 CM |
4672 | struct btrfs_map_token token; |
4673 | ||
4674 | btrfs_init_map_token(&token); | |
6567e837 | 4675 | |
5f39d397 | 4676 | leaf = path->nodes[0]; |
6567e837 | 4677 | |
5f39d397 | 4678 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4679 | data_end = leaf_data_end(fs_info, leaf); |
6567e837 | 4680 | |
2ff7e61e | 4681 | if (btrfs_leaf_free_space(fs_info, leaf) < data_size) { |
a4f78750 | 4682 | btrfs_print_leaf(leaf); |
6567e837 | 4683 | BUG(); |
5f39d397 | 4684 | } |
6567e837 | 4685 | slot = path->slots[0]; |
5f39d397 | 4686 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4687 | |
4688 | BUG_ON(slot < 0); | |
3326d1b0 | 4689 | if (slot >= nritems) { |
a4f78750 | 4690 | btrfs_print_leaf(leaf); |
0b246afa JM |
4691 | btrfs_crit(fs_info, "slot %d too large, nritems %d", |
4692 | slot, nritems); | |
3326d1b0 CM |
4693 | BUG_ON(1); |
4694 | } | |
6567e837 CM |
4695 | |
4696 | /* | |
4697 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4698 | */ | |
4699 | /* first correct the data pointers */ | |
4700 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4701 | u32 ioff; |
dd3cc16b | 4702 | item = btrfs_item_nr(i); |
db94535d | 4703 | |
cfed81a0 CM |
4704 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4705 | btrfs_set_token_item_offset(leaf, item, | |
4706 | ioff - data_size, &token); | |
6567e837 | 4707 | } |
5f39d397 | 4708 | |
6567e837 | 4709 | /* shift the data */ |
3d9ec8c4 NB |
4710 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4711 | data_end - data_size, BTRFS_LEAF_DATA_OFFSET + | |
6567e837 | 4712 | data_end, old_data - data_end); |
5f39d397 | 4713 | |
6567e837 | 4714 | data_end = old_data; |
5f39d397 | 4715 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4716 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4717 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4718 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4719 | |
2ff7e61e | 4720 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4721 | btrfs_print_leaf(leaf); |
6567e837 | 4722 | BUG(); |
5f39d397 | 4723 | } |
6567e837 CM |
4724 | } |
4725 | ||
74123bd7 | 4726 | /* |
44871b1b CM |
4727 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4728 | * to save stack depth by doing the bulk of the work in a function | |
4729 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4730 | */ |
afe5fea7 | 4731 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
310712b2 | 4732 | const struct btrfs_key *cpu_key, u32 *data_size, |
143bede5 | 4733 | u32 total_data, u32 total_size, int nr) |
be0e5c09 | 4734 | { |
0b246afa | 4735 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4736 | struct btrfs_item *item; |
9c58309d | 4737 | int i; |
7518a238 | 4738 | u32 nritems; |
be0e5c09 | 4739 | unsigned int data_end; |
e2fa7227 | 4740 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4741 | struct extent_buffer *leaf; |
4742 | int slot; | |
cfed81a0 CM |
4743 | struct btrfs_map_token token; |
4744 | ||
24cdc847 FM |
4745 | if (path->slots[0] == 0) { |
4746 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
0b246afa | 4747 | fixup_low_keys(fs_info, path, &disk_key, 1); |
24cdc847 FM |
4748 | } |
4749 | btrfs_unlock_up_safe(path, 1); | |
4750 | ||
cfed81a0 | 4751 | btrfs_init_map_token(&token); |
e2fa7227 | 4752 | |
5f39d397 | 4753 | leaf = path->nodes[0]; |
44871b1b | 4754 | slot = path->slots[0]; |
74123bd7 | 4755 | |
5f39d397 | 4756 | nritems = btrfs_header_nritems(leaf); |
2ff7e61e | 4757 | data_end = leaf_data_end(fs_info, leaf); |
eb60ceac | 4758 | |
2ff7e61e | 4759 | if (btrfs_leaf_free_space(fs_info, leaf) < total_size) { |
a4f78750 | 4760 | btrfs_print_leaf(leaf); |
0b246afa | 4761 | btrfs_crit(fs_info, "not enough freespace need %u have %d", |
2ff7e61e | 4762 | total_size, btrfs_leaf_free_space(fs_info, leaf)); |
be0e5c09 | 4763 | BUG(); |
d4dbff95 | 4764 | } |
5f39d397 | 4765 | |
be0e5c09 | 4766 | if (slot != nritems) { |
5f39d397 | 4767 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4768 | |
5f39d397 | 4769 | if (old_data < data_end) { |
a4f78750 | 4770 | btrfs_print_leaf(leaf); |
0b246afa | 4771 | btrfs_crit(fs_info, "slot %d old_data %d data_end %d", |
5d163e0e | 4772 | slot, old_data, data_end); |
5f39d397 CM |
4773 | BUG_ON(1); |
4774 | } | |
be0e5c09 CM |
4775 | /* |
4776 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4777 | */ | |
4778 | /* first correct the data pointers */ | |
0783fcfc | 4779 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4780 | u32 ioff; |
db94535d | 4781 | |
62e85577 | 4782 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4783 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4784 | btrfs_set_token_item_offset(leaf, item, | |
4785 | ioff - total_data, &token); | |
0783fcfc | 4786 | } |
be0e5c09 | 4787 | /* shift the items */ |
9c58309d | 4788 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4789 | btrfs_item_nr_offset(slot), |
d6025579 | 4790 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4791 | |
4792 | /* shift the data */ | |
3d9ec8c4 NB |
4793 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
4794 | data_end - total_data, BTRFS_LEAF_DATA_OFFSET + | |
d6025579 | 4795 | data_end, old_data - data_end); |
be0e5c09 CM |
4796 | data_end = old_data; |
4797 | } | |
5f39d397 | 4798 | |
62e2749e | 4799 | /* setup the item for the new data */ |
9c58309d CM |
4800 | for (i = 0; i < nr; i++) { |
4801 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4802 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4803 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4804 | btrfs_set_token_item_offset(leaf, item, |
4805 | data_end - data_size[i], &token); | |
9c58309d | 4806 | data_end -= data_size[i]; |
cfed81a0 | 4807 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4808 | } |
44871b1b | 4809 | |
9c58309d | 4810 | btrfs_set_header_nritems(leaf, nritems + nr); |
b9473439 | 4811 | btrfs_mark_buffer_dirty(leaf); |
aa5d6bed | 4812 | |
2ff7e61e | 4813 | if (btrfs_leaf_free_space(fs_info, leaf) < 0) { |
a4f78750 | 4814 | btrfs_print_leaf(leaf); |
be0e5c09 | 4815 | BUG(); |
5f39d397 | 4816 | } |
44871b1b CM |
4817 | } |
4818 | ||
4819 | /* | |
4820 | * Given a key and some data, insert items into the tree. | |
4821 | * This does all the path init required, making room in the tree if needed. | |
4822 | */ | |
4823 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4824 | struct btrfs_root *root, | |
4825 | struct btrfs_path *path, | |
310712b2 | 4826 | const struct btrfs_key *cpu_key, u32 *data_size, |
44871b1b CM |
4827 | int nr) |
4828 | { | |
44871b1b CM |
4829 | int ret = 0; |
4830 | int slot; | |
4831 | int i; | |
4832 | u32 total_size = 0; | |
4833 | u32 total_data = 0; | |
4834 | ||
4835 | for (i = 0; i < nr; i++) | |
4836 | total_data += data_size[i]; | |
4837 | ||
4838 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4839 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4840 | if (ret == 0) | |
4841 | return -EEXIST; | |
4842 | if (ret < 0) | |
143bede5 | 4843 | return ret; |
44871b1b | 4844 | |
44871b1b CM |
4845 | slot = path->slots[0]; |
4846 | BUG_ON(slot < 0); | |
4847 | ||
afe5fea7 | 4848 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4849 | total_data, total_size, nr); |
143bede5 | 4850 | return 0; |
62e2749e CM |
4851 | } |
4852 | ||
4853 | /* | |
4854 | * Given a key and some data, insert an item into the tree. | |
4855 | * This does all the path init required, making room in the tree if needed. | |
4856 | */ | |
310712b2 OS |
4857 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4858 | const struct btrfs_key *cpu_key, void *data, | |
4859 | u32 data_size) | |
62e2749e CM |
4860 | { |
4861 | int ret = 0; | |
2c90e5d6 | 4862 | struct btrfs_path *path; |
5f39d397 CM |
4863 | struct extent_buffer *leaf; |
4864 | unsigned long ptr; | |
62e2749e | 4865 | |
2c90e5d6 | 4866 | path = btrfs_alloc_path(); |
db5b493a TI |
4867 | if (!path) |
4868 | return -ENOMEM; | |
2c90e5d6 | 4869 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4870 | if (!ret) { |
5f39d397 CM |
4871 | leaf = path->nodes[0]; |
4872 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4873 | write_extent_buffer(leaf, data, ptr, data_size); | |
4874 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4875 | } |
2c90e5d6 | 4876 | btrfs_free_path(path); |
aa5d6bed | 4877 | return ret; |
be0e5c09 CM |
4878 | } |
4879 | ||
74123bd7 | 4880 | /* |
5de08d7d | 4881 | * delete the pointer from a given node. |
74123bd7 | 4882 | * |
d352ac68 CM |
4883 | * the tree should have been previously balanced so the deletion does not |
4884 | * empty a node. | |
74123bd7 | 4885 | */ |
afe5fea7 TI |
4886 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4887 | int level, int slot) | |
be0e5c09 | 4888 | { |
0b246afa | 4889 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 | 4890 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4891 | u32 nritems; |
f3ea38da | 4892 | int ret; |
be0e5c09 | 4893 | |
5f39d397 | 4894 | nritems = btrfs_header_nritems(parent); |
d397712b | 4895 | if (slot != nritems - 1) { |
0e411ece | 4896 | if (level) |
0b246afa | 4897 | tree_mod_log_eb_move(fs_info, parent, slot, |
f3ea38da | 4898 | slot + 1, nritems - slot - 1); |
5f39d397 CM |
4899 | memmove_extent_buffer(parent, |
4900 | btrfs_node_key_ptr_offset(slot), | |
4901 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4902 | sizeof(struct btrfs_key_ptr) * |
4903 | (nritems - slot - 1)); | |
57ba86c0 | 4904 | } else if (level) { |
0b246afa | 4905 | ret = tree_mod_log_insert_key(fs_info, parent, slot, |
c8cc6341 | 4906 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4907 | BUG_ON(ret < 0); |
bb803951 | 4908 | } |
f3ea38da | 4909 | |
7518a238 | 4910 | nritems--; |
5f39d397 | 4911 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4912 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4913 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4914 | /* just turn the root into a leaf and break */ |
5f39d397 | 4915 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4916 | } else if (slot == 0) { |
5f39d397 CM |
4917 | struct btrfs_disk_key disk_key; |
4918 | ||
4919 | btrfs_node_key(parent, &disk_key, 0); | |
0b246afa | 4920 | fixup_low_keys(fs_info, path, &disk_key, level + 1); |
be0e5c09 | 4921 | } |
d6025579 | 4922 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4923 | } |
4924 | ||
323ac95b CM |
4925 | /* |
4926 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4927 | * path->nodes[1]. |
323ac95b CM |
4928 | * |
4929 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4930 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4931 | * | |
4932 | * The path must have already been setup for deleting the leaf, including | |
4933 | * all the proper balancing. path->nodes[1] must be locked. | |
4934 | */ | |
143bede5 JM |
4935 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4936 | struct btrfs_root *root, | |
4937 | struct btrfs_path *path, | |
4938 | struct extent_buffer *leaf) | |
323ac95b | 4939 | { |
5d4f98a2 | 4940 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4941 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4942 | |
4d081c41 CM |
4943 | /* |
4944 | * btrfs_free_extent is expensive, we want to make sure we | |
4945 | * aren't holding any locks when we call it | |
4946 | */ | |
4947 | btrfs_unlock_up_safe(path, 0); | |
4948 | ||
f0486c68 YZ |
4949 | root_sub_used(root, leaf->len); |
4950 | ||
3083ee2e | 4951 | extent_buffer_get(leaf); |
5581a51a | 4952 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4953 | free_extent_buffer_stale(leaf); |
323ac95b | 4954 | } |
74123bd7 CM |
4955 | /* |
4956 | * delete the item at the leaf level in path. If that empties | |
4957 | * the leaf, remove it from the tree | |
4958 | */ | |
85e21bac CM |
4959 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4960 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4961 | { |
0b246afa | 4962 | struct btrfs_fs_info *fs_info = root->fs_info; |
5f39d397 CM |
4963 | struct extent_buffer *leaf; |
4964 | struct btrfs_item *item; | |
ce0eac2a AM |
4965 | u32 last_off; |
4966 | u32 dsize = 0; | |
aa5d6bed CM |
4967 | int ret = 0; |
4968 | int wret; | |
85e21bac | 4969 | int i; |
7518a238 | 4970 | u32 nritems; |
cfed81a0 CM |
4971 | struct btrfs_map_token token; |
4972 | ||
4973 | btrfs_init_map_token(&token); | |
be0e5c09 | 4974 | |
5f39d397 | 4975 | leaf = path->nodes[0]; |
85e21bac CM |
4976 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4977 | ||
4978 | for (i = 0; i < nr; i++) | |
4979 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4980 | ||
5f39d397 | 4981 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4982 | |
85e21bac | 4983 | if (slot + nr != nritems) { |
2ff7e61e | 4984 | int data_end = leaf_data_end(fs_info, leaf); |
5f39d397 | 4985 | |
3d9ec8c4 | 4986 | memmove_extent_buffer(leaf, BTRFS_LEAF_DATA_OFFSET + |
d6025579 | 4987 | data_end + dsize, |
3d9ec8c4 | 4988 | BTRFS_LEAF_DATA_OFFSET + data_end, |
85e21bac | 4989 | last_off - data_end); |
5f39d397 | 4990 | |
85e21bac | 4991 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4992 | u32 ioff; |
db94535d | 4993 | |
dd3cc16b | 4994 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4995 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4996 | btrfs_set_token_item_offset(leaf, item, | |
4997 | ioff + dsize, &token); | |
0783fcfc | 4998 | } |
db94535d | 4999 | |
5f39d397 | 5000 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5001 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5002 | sizeof(struct btrfs_item) * |
85e21bac | 5003 | (nritems - slot - nr)); |
be0e5c09 | 5004 | } |
85e21bac CM |
5005 | btrfs_set_header_nritems(leaf, nritems - nr); |
5006 | nritems -= nr; | |
5f39d397 | 5007 | |
74123bd7 | 5008 | /* delete the leaf if we've emptied it */ |
7518a238 | 5009 | if (nritems == 0) { |
5f39d397 CM |
5010 | if (leaf == root->node) { |
5011 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5012 | } else { |
f0486c68 | 5013 | btrfs_set_path_blocking(path); |
7c302b49 | 5014 | clean_tree_block(fs_info, leaf); |
143bede5 | 5015 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5016 | } |
be0e5c09 | 5017 | } else { |
7518a238 | 5018 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5019 | if (slot == 0) { |
5f39d397 CM |
5020 | struct btrfs_disk_key disk_key; |
5021 | ||
5022 | btrfs_item_key(leaf, &disk_key, 0); | |
0b246afa | 5023 | fixup_low_keys(fs_info, path, &disk_key, 1); |
aa5d6bed | 5024 | } |
aa5d6bed | 5025 | |
74123bd7 | 5026 | /* delete the leaf if it is mostly empty */ |
0b246afa | 5027 | if (used < BTRFS_LEAF_DATA_SIZE(fs_info) / 3) { |
be0e5c09 CM |
5028 | /* push_leaf_left fixes the path. |
5029 | * make sure the path still points to our leaf | |
5030 | * for possible call to del_ptr below | |
5031 | */ | |
4920c9ac | 5032 | slot = path->slots[1]; |
5f39d397 CM |
5033 | extent_buffer_get(leaf); |
5034 | ||
b9473439 | 5035 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5036 | wret = push_leaf_left(trans, root, path, 1, 1, |
5037 | 1, (u32)-1); | |
54aa1f4d | 5038 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5039 | ret = wret; |
5f39d397 CM |
5040 | |
5041 | if (path->nodes[0] == leaf && | |
5042 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5043 | wret = push_leaf_right(trans, root, path, 1, |
5044 | 1, 1, 0); | |
54aa1f4d | 5045 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5046 | ret = wret; |
5047 | } | |
5f39d397 CM |
5048 | |
5049 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5050 | path->slots[1] = slot; |
143bede5 | 5051 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5052 | free_extent_buffer(leaf); |
143bede5 | 5053 | ret = 0; |
5de08d7d | 5054 | } else { |
925baedd CM |
5055 | /* if we're still in the path, make sure |
5056 | * we're dirty. Otherwise, one of the | |
5057 | * push_leaf functions must have already | |
5058 | * dirtied this buffer | |
5059 | */ | |
5060 | if (path->nodes[0] == leaf) | |
5061 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5062 | free_extent_buffer(leaf); |
be0e5c09 | 5063 | } |
d5719762 | 5064 | } else { |
5f39d397 | 5065 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5066 | } |
5067 | } | |
aa5d6bed | 5068 | return ret; |
be0e5c09 CM |
5069 | } |
5070 | ||
7bb86316 | 5071 | /* |
925baedd | 5072 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5073 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5074 | * returns < 0 on io errors. | |
d352ac68 CM |
5075 | * |
5076 | * This may release the path, and so you may lose any locks held at the | |
5077 | * time you call it. | |
7bb86316 | 5078 | */ |
16e7549f | 5079 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5080 | { |
925baedd CM |
5081 | struct btrfs_key key; |
5082 | struct btrfs_disk_key found_key; | |
5083 | int ret; | |
7bb86316 | 5084 | |
925baedd | 5085 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5086 | |
e8b0d724 | 5087 | if (key.offset > 0) { |
925baedd | 5088 | key.offset--; |
e8b0d724 | 5089 | } else if (key.type > 0) { |
925baedd | 5090 | key.type--; |
e8b0d724 FDBM |
5091 | key.offset = (u64)-1; |
5092 | } else if (key.objectid > 0) { | |
925baedd | 5093 | key.objectid--; |
e8b0d724 FDBM |
5094 | key.type = (u8)-1; |
5095 | key.offset = (u64)-1; | |
5096 | } else { | |
925baedd | 5097 | return 1; |
e8b0d724 | 5098 | } |
7bb86316 | 5099 | |
b3b4aa74 | 5100 | btrfs_release_path(path); |
925baedd CM |
5101 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5102 | if (ret < 0) | |
5103 | return ret; | |
5104 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5105 | ret = comp_keys(&found_key, &key); | |
337c6f68 FM |
5106 | /* |
5107 | * We might have had an item with the previous key in the tree right | |
5108 | * before we released our path. And after we released our path, that | |
5109 | * item might have been pushed to the first slot (0) of the leaf we | |
5110 | * were holding due to a tree balance. Alternatively, an item with the | |
5111 | * previous key can exist as the only element of a leaf (big fat item). | |
5112 | * Therefore account for these 2 cases, so that our callers (like | |
5113 | * btrfs_previous_item) don't miss an existing item with a key matching | |
5114 | * the previous key we computed above. | |
5115 | */ | |
5116 | if (ret <= 0) | |
925baedd CM |
5117 | return 0; |
5118 | return 1; | |
7bb86316 CM |
5119 | } |
5120 | ||
3f157a2f CM |
5121 | /* |
5122 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5123 | * for nodes or leaves that are have a minimum transaction id. |
5124 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5125 | * |
5126 | * This does not cow, but it does stuff the starting key it finds back | |
5127 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5128 | * key and get a writable path. | |
5129 | * | |
5130 | * This does lock as it descends, and path->keep_locks should be set | |
5131 | * to 1 by the caller. | |
5132 | * | |
5133 | * This honors path->lowest_level to prevent descent past a given level | |
5134 | * of the tree. | |
5135 | * | |
d352ac68 CM |
5136 | * min_trans indicates the oldest transaction that you are interested |
5137 | * in walking through. Any nodes or leaves older than min_trans are | |
5138 | * skipped over (without reading them). | |
5139 | * | |
3f157a2f CM |
5140 | * returns zero if something useful was found, < 0 on error and 1 if there |
5141 | * was nothing in the tree that matched the search criteria. | |
5142 | */ | |
5143 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5144 | struct btrfs_path *path, |
3f157a2f CM |
5145 | u64 min_trans) |
5146 | { | |
2ff7e61e | 5147 | struct btrfs_fs_info *fs_info = root->fs_info; |
3f157a2f CM |
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); |
2ff7e61e | 5224 | cur = read_node_slot(fs_info, 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 | ||
2ff7e61e | 5247 | static int tree_move_down(struct btrfs_fs_info *fs_info, |
7069830a | 5248 | struct btrfs_path *path, |
ab6a43e1 | 5249 | int *level) |
7069830a | 5250 | { |
fb770ae4 LB |
5251 | struct extent_buffer *eb; |
5252 | ||
74dd17fb | 5253 | BUG_ON(*level == 0); |
2ff7e61e | 5254 | eb = read_node_slot(fs_info, path->nodes[*level], path->slots[*level]); |
fb770ae4 LB |
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 | ||
f1e30261 | 5264 | static int tree_move_next_or_upnext(struct btrfs_path *path, |
7069830a AB |
5265 | int *level, int root_level) |
5266 | { | |
5267 | int ret = 0; | |
5268 | int nritems; | |
5269 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5270 | ||
5271 | path->slots[*level]++; | |
5272 | ||
74dd17fb | 5273 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5274 | if (*level == root_level) |
5275 | return -1; | |
5276 | ||
5277 | /* move upnext */ | |
5278 | path->slots[*level] = 0; | |
5279 | free_extent_buffer(path->nodes[*level]); | |
5280 | path->nodes[*level] = NULL; | |
5281 | (*level)++; | |
5282 | path->slots[*level]++; | |
5283 | ||
5284 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5285 | ret = 1; | |
5286 | } | |
5287 | return ret; | |
5288 | } | |
5289 | ||
5290 | /* | |
5291 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5292 | * or down. | |
5293 | */ | |
2ff7e61e | 5294 | static int tree_advance(struct btrfs_fs_info *fs_info, |
7069830a AB |
5295 | struct btrfs_path *path, |
5296 | int *level, int root_level, | |
5297 | int allow_down, | |
5298 | struct btrfs_key *key) | |
5299 | { | |
5300 | int ret; | |
5301 | ||
5302 | if (*level == 0 || !allow_down) { | |
f1e30261 | 5303 | ret = tree_move_next_or_upnext(path, level, root_level); |
7069830a | 5304 | } else { |
ab6a43e1 | 5305 | ret = tree_move_down(fs_info, path, level); |
7069830a AB |
5306 | } |
5307 | if (ret >= 0) { | |
5308 | if (*level == 0) | |
5309 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5310 | path->slots[*level]); | |
5311 | else | |
5312 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5313 | path->slots[*level]); | |
5314 | } | |
5315 | return ret; | |
5316 | } | |
5317 | ||
2ff7e61e | 5318 | static int tree_compare_item(struct btrfs_path *left_path, |
7069830a AB |
5319 | struct btrfs_path *right_path, |
5320 | char *tmp_buf) | |
5321 | { | |
5322 | int cmp; | |
5323 | int len1, len2; | |
5324 | unsigned long off1, off2; | |
5325 | ||
5326 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5327 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5328 | if (len1 != len2) | |
5329 | return 1; | |
5330 | ||
5331 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5332 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5333 | right_path->slots[0]); | |
5334 | ||
5335 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5336 | ||
5337 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5338 | if (cmp) | |
5339 | return 1; | |
5340 | return 0; | |
5341 | } | |
5342 | ||
5343 | #define ADVANCE 1 | |
5344 | #define ADVANCE_ONLY_NEXT -1 | |
5345 | ||
5346 | /* | |
5347 | * This function compares two trees and calls the provided callback for | |
5348 | * every changed/new/deleted item it finds. | |
5349 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5350 | * the compare pretty fast on snapshotted subvolumes. | |
5351 | * | |
5352 | * This currently works on commit roots only. As commit roots are read only, | |
5353 | * we don't do any locking. The commit roots are protected with transactions. | |
5354 | * Transactions are ended and rejoined when a commit is tried in between. | |
5355 | * | |
5356 | * This function checks for modifications done to the trees while comparing. | |
5357 | * If it detects a change, it aborts immediately. | |
5358 | */ | |
5359 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5360 | struct btrfs_root *right_root, | |
5361 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5362 | { | |
0b246afa | 5363 | struct btrfs_fs_info *fs_info = left_root->fs_info; |
7069830a AB |
5364 | int ret; |
5365 | int cmp; | |
7069830a AB |
5366 | struct btrfs_path *left_path = NULL; |
5367 | struct btrfs_path *right_path = NULL; | |
5368 | struct btrfs_key left_key; | |
5369 | struct btrfs_key right_key; | |
5370 | char *tmp_buf = NULL; | |
5371 | int left_root_level; | |
5372 | int right_root_level; | |
5373 | int left_level; | |
5374 | int right_level; | |
5375 | int left_end_reached; | |
5376 | int right_end_reached; | |
5377 | int advance_left; | |
5378 | int advance_right; | |
5379 | u64 left_blockptr; | |
5380 | u64 right_blockptr; | |
6baa4293 FM |
5381 | u64 left_gen; |
5382 | u64 right_gen; | |
7069830a AB |
5383 | |
5384 | left_path = btrfs_alloc_path(); | |
5385 | if (!left_path) { | |
5386 | ret = -ENOMEM; | |
5387 | goto out; | |
5388 | } | |
5389 | right_path = btrfs_alloc_path(); | |
5390 | if (!right_path) { | |
5391 | ret = -ENOMEM; | |
5392 | goto out; | |
5393 | } | |
5394 | ||
752ade68 | 5395 | tmp_buf = kvmalloc(fs_info->nodesize, GFP_KERNEL); |
7069830a | 5396 | if (!tmp_buf) { |
752ade68 MH |
5397 | ret = -ENOMEM; |
5398 | goto out; | |
7069830a AB |
5399 | } |
5400 | ||
5401 | left_path->search_commit_root = 1; | |
5402 | left_path->skip_locking = 1; | |
5403 | right_path->search_commit_root = 1; | |
5404 | right_path->skip_locking = 1; | |
5405 | ||
7069830a AB |
5406 | /* |
5407 | * Strategy: Go to the first items of both trees. Then do | |
5408 | * | |
5409 | * If both trees are at level 0 | |
5410 | * Compare keys of current items | |
5411 | * If left < right treat left item as new, advance left tree | |
5412 | * and repeat | |
5413 | * If left > right treat right item as deleted, advance right tree | |
5414 | * and repeat | |
5415 | * If left == right do deep compare of items, treat as changed if | |
5416 | * needed, advance both trees and repeat | |
5417 | * If both trees are at the same level but not at level 0 | |
5418 | * Compare keys of current nodes/leafs | |
5419 | * If left < right advance left tree and repeat | |
5420 | * If left > right advance right tree and repeat | |
5421 | * If left == right compare blockptrs of the next nodes/leafs | |
5422 | * If they match advance both trees but stay at the same level | |
5423 | * and repeat | |
5424 | * If they don't match advance both trees while allowing to go | |
5425 | * deeper and repeat | |
5426 | * If tree levels are different | |
5427 | * Advance the tree that needs it and repeat | |
5428 | * | |
5429 | * Advancing a tree means: | |
5430 | * If we are at level 0, try to go to the next slot. If that's not | |
5431 | * possible, go one level up and repeat. Stop when we found a level | |
5432 | * where we could go to the next slot. We may at this point be on a | |
5433 | * node or a leaf. | |
5434 | * | |
5435 | * If we are not at level 0 and not on shared tree blocks, go one | |
5436 | * level deeper. | |
5437 | * | |
5438 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5439 | * the right if possible or go up and right. | |
5440 | */ | |
5441 | ||
0b246afa | 5442 | down_read(&fs_info->commit_root_sem); |
7069830a AB |
5443 | left_level = btrfs_header_level(left_root->commit_root); |
5444 | left_root_level = left_level; | |
5445 | left_path->nodes[left_level] = left_root->commit_root; | |
5446 | extent_buffer_get(left_path->nodes[left_level]); | |
5447 | ||
5448 | right_level = btrfs_header_level(right_root->commit_root); | |
5449 | right_root_level = right_level; | |
5450 | right_path->nodes[right_level] = right_root->commit_root; | |
5451 | extent_buffer_get(right_path->nodes[right_level]); | |
0b246afa | 5452 | up_read(&fs_info->commit_root_sem); |
7069830a AB |
5453 | |
5454 | if (left_level == 0) | |
5455 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5456 | &left_key, left_path->slots[left_level]); | |
5457 | else | |
5458 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5459 | &left_key, left_path->slots[left_level]); | |
5460 | if (right_level == 0) | |
5461 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5462 | &right_key, right_path->slots[right_level]); | |
5463 | else | |
5464 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5465 | &right_key, right_path->slots[right_level]); | |
5466 | ||
5467 | left_end_reached = right_end_reached = 0; | |
5468 | advance_left = advance_right = 0; | |
5469 | ||
5470 | while (1) { | |
7069830a | 5471 | if (advance_left && !left_end_reached) { |
2ff7e61e | 5472 | ret = tree_advance(fs_info, left_path, &left_level, |
7069830a AB |
5473 | left_root_level, |
5474 | advance_left != ADVANCE_ONLY_NEXT, | |
5475 | &left_key); | |
fb770ae4 | 5476 | if (ret == -1) |
7069830a | 5477 | left_end_reached = ADVANCE; |
fb770ae4 LB |
5478 | else if (ret < 0) |
5479 | goto out; | |
7069830a AB |
5480 | advance_left = 0; |
5481 | } | |
5482 | if (advance_right && !right_end_reached) { | |
2ff7e61e | 5483 | ret = tree_advance(fs_info, right_path, &right_level, |
7069830a AB |
5484 | right_root_level, |
5485 | advance_right != ADVANCE_ONLY_NEXT, | |
5486 | &right_key); | |
fb770ae4 | 5487 | if (ret == -1) |
7069830a | 5488 | right_end_reached = ADVANCE; |
fb770ae4 LB |
5489 | else if (ret < 0) |
5490 | goto out; | |
7069830a AB |
5491 | advance_right = 0; |
5492 | } | |
5493 | ||
5494 | if (left_end_reached && right_end_reached) { | |
5495 | ret = 0; | |
5496 | goto out; | |
5497 | } else if (left_end_reached) { | |
5498 | if (right_level == 0) { | |
ee8c494f | 5499 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5500 | &right_key, |
5501 | BTRFS_COMPARE_TREE_DELETED, | |
5502 | ctx); | |
5503 | if (ret < 0) | |
5504 | goto out; | |
5505 | } | |
5506 | advance_right = ADVANCE; | |
5507 | continue; | |
5508 | } else if (right_end_reached) { | |
5509 | if (left_level == 0) { | |
ee8c494f | 5510 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5511 | &left_key, |
5512 | BTRFS_COMPARE_TREE_NEW, | |
5513 | ctx); | |
5514 | if (ret < 0) | |
5515 | goto out; | |
5516 | } | |
5517 | advance_left = ADVANCE; | |
5518 | continue; | |
5519 | } | |
5520 | ||
5521 | if (left_level == 0 && right_level == 0) { | |
5522 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5523 | if (cmp < 0) { | |
ee8c494f | 5524 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5525 | &left_key, |
5526 | BTRFS_COMPARE_TREE_NEW, | |
5527 | ctx); | |
5528 | if (ret < 0) | |
5529 | goto out; | |
5530 | advance_left = ADVANCE; | |
5531 | } else if (cmp > 0) { | |
ee8c494f | 5532 | ret = changed_cb(left_path, right_path, |
7069830a AB |
5533 | &right_key, |
5534 | BTRFS_COMPARE_TREE_DELETED, | |
5535 | ctx); | |
5536 | if (ret < 0) | |
5537 | goto out; | |
5538 | advance_right = ADVANCE; | |
5539 | } else { | |
b99d9a6a | 5540 | enum btrfs_compare_tree_result result; |
ba5e8f2e | 5541 | |
74dd17fb | 5542 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
2ff7e61e JM |
5543 | ret = tree_compare_item(left_path, right_path, |
5544 | tmp_buf); | |
ba5e8f2e | 5545 | if (ret) |
b99d9a6a | 5546 | result = BTRFS_COMPARE_TREE_CHANGED; |
ba5e8f2e | 5547 | else |
b99d9a6a | 5548 | result = BTRFS_COMPARE_TREE_SAME; |
ee8c494f | 5549 | ret = changed_cb(left_path, right_path, |
b99d9a6a | 5550 | &left_key, result, ctx); |
ba5e8f2e JB |
5551 | if (ret < 0) |
5552 | goto out; | |
7069830a AB |
5553 | advance_left = ADVANCE; |
5554 | advance_right = ADVANCE; | |
5555 | } | |
5556 | } else if (left_level == right_level) { | |
5557 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5558 | if (cmp < 0) { | |
5559 | advance_left = ADVANCE; | |
5560 | } else if (cmp > 0) { | |
5561 | advance_right = ADVANCE; | |
5562 | } else { | |
5563 | left_blockptr = btrfs_node_blockptr( | |
5564 | left_path->nodes[left_level], | |
5565 | left_path->slots[left_level]); | |
5566 | right_blockptr = btrfs_node_blockptr( | |
5567 | right_path->nodes[right_level], | |
5568 | right_path->slots[right_level]); | |
6baa4293 FM |
5569 | left_gen = btrfs_node_ptr_generation( |
5570 | left_path->nodes[left_level], | |
5571 | left_path->slots[left_level]); | |
5572 | right_gen = btrfs_node_ptr_generation( | |
5573 | right_path->nodes[right_level], | |
5574 | right_path->slots[right_level]); | |
5575 | if (left_blockptr == right_blockptr && | |
5576 | left_gen == right_gen) { | |
7069830a AB |
5577 | /* |
5578 | * As we're on a shared block, don't | |
5579 | * allow to go deeper. | |
5580 | */ | |
5581 | advance_left = ADVANCE_ONLY_NEXT; | |
5582 | advance_right = ADVANCE_ONLY_NEXT; | |
5583 | } else { | |
5584 | advance_left = ADVANCE; | |
5585 | advance_right = ADVANCE; | |
5586 | } | |
5587 | } | |
5588 | } else if (left_level < right_level) { | |
5589 | advance_right = ADVANCE; | |
5590 | } else { | |
5591 | advance_left = ADVANCE; | |
5592 | } | |
5593 | } | |
5594 | ||
5595 | out: | |
5596 | btrfs_free_path(left_path); | |
5597 | btrfs_free_path(right_path); | |
8f282f71 | 5598 | kvfree(tmp_buf); |
7069830a AB |
5599 | return ret; |
5600 | } | |
5601 | ||
3f157a2f CM |
5602 | /* |
5603 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5604 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5605 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5606 | * |
5607 | * 0 is returned if another key is found, < 0 if there are any errors | |
5608 | * and 1 is returned if there are no higher keys in the tree | |
5609 | * | |
5610 | * path->keep_locks should be set to 1 on the search made before | |
5611 | * calling this function. | |
5612 | */ | |
e7a84565 | 5613 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5614 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5615 | { |
e7a84565 CM |
5616 | int slot; |
5617 | struct extent_buffer *c; | |
5618 | ||
934d375b | 5619 | WARN_ON(!path->keep_locks); |
d397712b | 5620 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5621 | if (!path->nodes[level]) |
5622 | return 1; | |
5623 | ||
5624 | slot = path->slots[level] + 1; | |
5625 | c = path->nodes[level]; | |
3f157a2f | 5626 | next: |
e7a84565 | 5627 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5628 | int ret; |
5629 | int orig_lowest; | |
5630 | struct btrfs_key cur_key; | |
5631 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5632 | !path->nodes[level + 1]) | |
e7a84565 | 5633 | return 1; |
33c66f43 YZ |
5634 | |
5635 | if (path->locks[level + 1]) { | |
5636 | level++; | |
5637 | continue; | |
5638 | } | |
5639 | ||
5640 | slot = btrfs_header_nritems(c) - 1; | |
5641 | if (level == 0) | |
5642 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5643 | else | |
5644 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5645 | ||
5646 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5647 | btrfs_release_path(path); |
33c66f43 YZ |
5648 | path->lowest_level = level; |
5649 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5650 | 0, 0); | |
5651 | path->lowest_level = orig_lowest; | |
5652 | if (ret < 0) | |
5653 | return ret; | |
5654 | ||
5655 | c = path->nodes[level]; | |
5656 | slot = path->slots[level]; | |
5657 | if (ret == 0) | |
5658 | slot++; | |
5659 | goto next; | |
e7a84565 | 5660 | } |
33c66f43 | 5661 | |
e7a84565 CM |
5662 | if (level == 0) |
5663 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5664 | else { |
3f157a2f CM |
5665 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5666 | ||
3f157a2f CM |
5667 | if (gen < min_trans) { |
5668 | slot++; | |
5669 | goto next; | |
5670 | } | |
e7a84565 | 5671 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5672 | } |
e7a84565 CM |
5673 | return 0; |
5674 | } | |
5675 | return 1; | |
5676 | } | |
5677 | ||
97571fd0 | 5678 | /* |
925baedd | 5679 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5680 | * returns 0 if it found something or 1 if there are no greater leaves. |
5681 | * returns < 0 on io errors. | |
97571fd0 | 5682 | */ |
234b63a0 | 5683 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5684 | { |
5685 | return btrfs_next_old_leaf(root, path, 0); | |
5686 | } | |
5687 | ||
5688 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5689 | u64 time_seq) | |
d97e63b6 CM |
5690 | { |
5691 | int slot; | |
8e73f275 | 5692 | int level; |
5f39d397 | 5693 | struct extent_buffer *c; |
8e73f275 | 5694 | struct extent_buffer *next; |
925baedd CM |
5695 | struct btrfs_key key; |
5696 | u32 nritems; | |
5697 | int ret; | |
8e73f275 | 5698 | int old_spinning = path->leave_spinning; |
bd681513 | 5699 | int next_rw_lock = 0; |
925baedd CM |
5700 | |
5701 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5702 | if (nritems == 0) |
925baedd | 5703 | return 1; |
925baedd | 5704 | |
8e73f275 CM |
5705 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5706 | again: | |
5707 | level = 1; | |
5708 | next = NULL; | |
bd681513 | 5709 | next_rw_lock = 0; |
b3b4aa74 | 5710 | btrfs_release_path(path); |
8e73f275 | 5711 | |
a2135011 | 5712 | path->keep_locks = 1; |
31533fb2 | 5713 | path->leave_spinning = 1; |
8e73f275 | 5714 | |
3d7806ec JS |
5715 | if (time_seq) |
5716 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5717 | else | |
5718 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5719 | path->keep_locks = 0; |
5720 | ||
5721 | if (ret < 0) | |
5722 | return ret; | |
5723 | ||
a2135011 | 5724 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5725 | /* |
5726 | * by releasing the path above we dropped all our locks. A balance | |
5727 | * could have added more items next to the key that used to be | |
5728 | * at the very end of the block. So, check again here and | |
5729 | * advance the path if there are now more items available. | |
5730 | */ | |
a2135011 | 5731 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5732 | if (ret == 0) |
5733 | path->slots[0]++; | |
8e73f275 | 5734 | ret = 0; |
925baedd CM |
5735 | goto done; |
5736 | } | |
0b43e04f LB |
5737 | /* |
5738 | * So the above check misses one case: | |
5739 | * - after releasing the path above, someone has removed the item that | |
5740 | * used to be at the very end of the block, and balance between leafs | |
5741 | * gets another one with bigger key.offset to replace it. | |
5742 | * | |
5743 | * This one should be returned as well, or we can get leaf corruption | |
5744 | * later(esp. in __btrfs_drop_extents()). | |
5745 | * | |
5746 | * And a bit more explanation about this check, | |
5747 | * with ret > 0, the key isn't found, the path points to the slot | |
5748 | * where it should be inserted, so the path->slots[0] item must be the | |
5749 | * bigger one. | |
5750 | */ | |
5751 | if (nritems > 0 && ret > 0 && path->slots[0] == nritems - 1) { | |
5752 | ret = 0; | |
5753 | goto done; | |
5754 | } | |
d97e63b6 | 5755 | |
d397712b | 5756 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5757 | if (!path->nodes[level]) { |
5758 | ret = 1; | |
5759 | goto done; | |
5760 | } | |
5f39d397 | 5761 | |
d97e63b6 CM |
5762 | slot = path->slots[level] + 1; |
5763 | c = path->nodes[level]; | |
5f39d397 | 5764 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5765 | level++; |
8e73f275 CM |
5766 | if (level == BTRFS_MAX_LEVEL) { |
5767 | ret = 1; | |
5768 | goto done; | |
5769 | } | |
d97e63b6 CM |
5770 | continue; |
5771 | } | |
5f39d397 | 5772 | |
925baedd | 5773 | if (next) { |
bd681513 | 5774 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5775 | free_extent_buffer(next); |
925baedd | 5776 | } |
5f39d397 | 5777 | |
8e73f275 | 5778 | next = c; |
bd681513 | 5779 | next_rw_lock = path->locks[level]; |
d07b8528 | 5780 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5781 | slot, &key); |
8e73f275 CM |
5782 | if (ret == -EAGAIN) |
5783 | goto again; | |
5f39d397 | 5784 | |
76a05b35 | 5785 | if (ret < 0) { |
b3b4aa74 | 5786 | btrfs_release_path(path); |
76a05b35 CM |
5787 | goto done; |
5788 | } | |
5789 | ||
5cd57b2c | 5790 | if (!path->skip_locking) { |
bd681513 | 5791 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5792 | if (!ret && time_seq) { |
5793 | /* | |
5794 | * If we don't get the lock, we may be racing | |
5795 | * with push_leaf_left, holding that lock while | |
5796 | * itself waiting for the leaf we've currently | |
5797 | * locked. To solve this situation, we give up | |
5798 | * on our lock and cycle. | |
5799 | */ | |
cf538830 | 5800 | free_extent_buffer(next); |
d42244a0 JS |
5801 | btrfs_release_path(path); |
5802 | cond_resched(); | |
5803 | goto again; | |
5804 | } | |
8e73f275 CM |
5805 | if (!ret) { |
5806 | btrfs_set_path_blocking(path); | |
bd681513 | 5807 | btrfs_tree_read_lock(next); |
31533fb2 | 5808 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5809 | BTRFS_READ_LOCK); |
8e73f275 | 5810 | } |
31533fb2 | 5811 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5812 | } |
d97e63b6 CM |
5813 | break; |
5814 | } | |
5815 | path->slots[level] = slot; | |
d397712b | 5816 | while (1) { |
d97e63b6 CM |
5817 | level--; |
5818 | c = path->nodes[level]; | |
925baedd | 5819 | if (path->locks[level]) |
bd681513 | 5820 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5821 | |
5f39d397 | 5822 | free_extent_buffer(c); |
d97e63b6 CM |
5823 | path->nodes[level] = next; |
5824 | path->slots[level] = 0; | |
a74a4b97 | 5825 | if (!path->skip_locking) |
bd681513 | 5826 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5827 | if (!level) |
5828 | break; | |
b4ce94de | 5829 | |
d07b8528 | 5830 | ret = read_block_for_search(root, path, &next, level, |
cda79c54 | 5831 | 0, &key); |
8e73f275 CM |
5832 | if (ret == -EAGAIN) |
5833 | goto again; | |
5834 | ||
76a05b35 | 5835 | if (ret < 0) { |
b3b4aa74 | 5836 | btrfs_release_path(path); |
76a05b35 CM |
5837 | goto done; |
5838 | } | |
5839 | ||
5cd57b2c | 5840 | if (!path->skip_locking) { |
bd681513 | 5841 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5842 | if (!ret) { |
5843 | btrfs_set_path_blocking(path); | |
bd681513 | 5844 | btrfs_tree_read_lock(next); |
31533fb2 | 5845 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5846 | BTRFS_READ_LOCK); |
5847 | } | |
31533fb2 | 5848 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5849 | } |
d97e63b6 | 5850 | } |
8e73f275 | 5851 | ret = 0; |
925baedd | 5852 | done: |
f7c79f30 | 5853 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5854 | path->leave_spinning = old_spinning; |
5855 | if (!old_spinning) | |
5856 | btrfs_set_path_blocking(path); | |
5857 | ||
5858 | return ret; | |
d97e63b6 | 5859 | } |
0b86a832 | 5860 | |
3f157a2f CM |
5861 | /* |
5862 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5863 | * searching until it gets past min_objectid or finds an item of 'type' | |
5864 | * | |
5865 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5866 | */ | |
0b86a832 CM |
5867 | int btrfs_previous_item(struct btrfs_root *root, |
5868 | struct btrfs_path *path, u64 min_objectid, | |
5869 | int type) | |
5870 | { | |
5871 | struct btrfs_key found_key; | |
5872 | struct extent_buffer *leaf; | |
e02119d5 | 5873 | u32 nritems; |
0b86a832 CM |
5874 | int ret; |
5875 | ||
d397712b | 5876 | while (1) { |
0b86a832 | 5877 | if (path->slots[0] == 0) { |
b4ce94de | 5878 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5879 | ret = btrfs_prev_leaf(root, path); |
5880 | if (ret != 0) | |
5881 | return ret; | |
5882 | } else { | |
5883 | path->slots[0]--; | |
5884 | } | |
5885 | leaf = path->nodes[0]; | |
e02119d5 CM |
5886 | nritems = btrfs_header_nritems(leaf); |
5887 | if (nritems == 0) | |
5888 | return 1; | |
5889 | if (path->slots[0] == nritems) | |
5890 | path->slots[0]--; | |
5891 | ||
0b86a832 | 5892 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5893 | if (found_key.objectid < min_objectid) |
5894 | break; | |
0a4eefbb YZ |
5895 | if (found_key.type == type) |
5896 | return 0; | |
e02119d5 CM |
5897 | if (found_key.objectid == min_objectid && |
5898 | found_key.type < type) | |
5899 | break; | |
0b86a832 CM |
5900 | } |
5901 | return 1; | |
5902 | } | |
ade2e0b3 WS |
5903 | |
5904 | /* | |
5905 | * search in extent tree to find a previous Metadata/Data extent item with | |
5906 | * min objecitd. | |
5907 | * | |
5908 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5909 | */ | |
5910 | int btrfs_previous_extent_item(struct btrfs_root *root, | |
5911 | struct btrfs_path *path, u64 min_objectid) | |
5912 | { | |
5913 | struct btrfs_key found_key; | |
5914 | struct extent_buffer *leaf; | |
5915 | u32 nritems; | |
5916 | int ret; | |
5917 | ||
5918 | while (1) { | |
5919 | if (path->slots[0] == 0) { | |
5920 | btrfs_set_path_blocking(path); | |
5921 | ret = btrfs_prev_leaf(root, path); | |
5922 | if (ret != 0) | |
5923 | return ret; | |
5924 | } else { | |
5925 | path->slots[0]--; | |
5926 | } | |
5927 | leaf = path->nodes[0]; | |
5928 | nritems = btrfs_header_nritems(leaf); | |
5929 | if (nritems == 0) | |
5930 | return 1; | |
5931 | if (path->slots[0] == nritems) | |
5932 | path->slots[0]--; | |
5933 | ||
5934 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
5935 | if (found_key.objectid < min_objectid) | |
5936 | break; | |
5937 | if (found_key.type == BTRFS_EXTENT_ITEM_KEY || | |
5938 | found_key.type == BTRFS_METADATA_ITEM_KEY) | |
5939 | return 0; | |
5940 | if (found_key.objectid == min_objectid && | |
5941 | found_key.type < BTRFS_EXTENT_ITEM_KEY) | |
5942 | break; | |
5943 | } | |
5944 | return 1; | |
5945 | } |