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> |
eb60ceac CM |
22 | #include "ctree.h" |
23 | #include "disk-io.h" | |
7f5c1516 | 24 | #include "transaction.h" |
5f39d397 | 25 | #include "print-tree.h" |
925baedd | 26 | #include "locking.h" |
9a8dd150 | 27 | |
e089f05c CM |
28 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
29 | *root, struct btrfs_path *path, int level); | |
30 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 31 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 32 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
33 | static int push_node_left(struct btrfs_trans_handle *trans, |
34 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 35 | struct extent_buffer *src, int empty); |
5f39d397 CM |
36 | static int balance_node_right(struct btrfs_trans_handle *trans, |
37 | struct btrfs_root *root, | |
38 | struct extent_buffer *dst_buf, | |
39 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
40 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
41 | int level, int slot); | |
f230475e JS |
42 | static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
43 | struct extent_buffer *eb); | |
48a3b636 | 44 | static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path); |
d97e63b6 | 45 | |
df24a2b9 | 46 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 47 | { |
df24a2b9 | 48 | struct btrfs_path *path; |
e00f7308 | 49 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 50 | return path; |
2c90e5d6 CM |
51 | } |
52 | ||
b4ce94de CM |
53 | /* |
54 | * set all locked nodes in the path to blocking locks. This should | |
55 | * be done before scheduling | |
56 | */ | |
57 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
58 | { | |
59 | int i; | |
60 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
61 | if (!p->nodes[i] || !p->locks[i]) |
62 | continue; | |
63 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
64 | if (p->locks[i] == BTRFS_READ_LOCK) | |
65 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
66 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
67 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
68 | } |
69 | } | |
70 | ||
71 | /* | |
72 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
73 | * |
74 | * held is used to keep lockdep happy, when lockdep is enabled | |
75 | * we set held to a blocking lock before we go around and | |
76 | * retake all the spinlocks in the path. You can safely use NULL | |
77 | * for held | |
b4ce94de | 78 | */ |
4008c04a | 79 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 80 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
81 | { |
82 | int i; | |
4008c04a CM |
83 | |
84 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
85 | /* lockdep really cares that we take all of these spinlocks | |
86 | * in the right order. If any of the locks in the path are not | |
87 | * currently blocking, it is going to complain. So, make really | |
88 | * really sure by forcing the path to blocking before we clear | |
89 | * the path blocking. | |
90 | */ | |
bd681513 CM |
91 | if (held) { |
92 | btrfs_set_lock_blocking_rw(held, held_rw); | |
93 | if (held_rw == BTRFS_WRITE_LOCK) | |
94 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
95 | else if (held_rw == BTRFS_READ_LOCK) | |
96 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
97 | } | |
4008c04a CM |
98 | btrfs_set_path_blocking(p); |
99 | #endif | |
100 | ||
101 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
102 | if (p->nodes[i] && p->locks[i]) { |
103 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
104 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
105 | p->locks[i] = BTRFS_WRITE_LOCK; | |
106 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
107 | p->locks[i] = BTRFS_READ_LOCK; | |
108 | } | |
b4ce94de | 109 | } |
4008c04a CM |
110 | |
111 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
112 | if (held) | |
bd681513 | 113 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 114 | #endif |
b4ce94de CM |
115 | } |
116 | ||
d352ac68 | 117 | /* this also releases the path */ |
df24a2b9 | 118 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 119 | { |
ff175d57 JJ |
120 | if (!p) |
121 | return; | |
b3b4aa74 | 122 | btrfs_release_path(p); |
df24a2b9 | 123 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
124 | } |
125 | ||
d352ac68 CM |
126 | /* |
127 | * path release drops references on the extent buffers in the path | |
128 | * and it drops any locks held by this path | |
129 | * | |
130 | * It is safe to call this on paths that no locks or extent buffers held. | |
131 | */ | |
b3b4aa74 | 132 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
133 | { |
134 | int i; | |
a2135011 | 135 | |
234b63a0 | 136 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 137 | p->slots[i] = 0; |
eb60ceac | 138 | if (!p->nodes[i]) |
925baedd CM |
139 | continue; |
140 | if (p->locks[i]) { | |
bd681513 | 141 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
142 | p->locks[i] = 0; |
143 | } | |
5f39d397 | 144 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 145 | p->nodes[i] = NULL; |
eb60ceac CM |
146 | } |
147 | } | |
148 | ||
d352ac68 CM |
149 | /* |
150 | * safely gets a reference on the root node of a tree. A lock | |
151 | * is not taken, so a concurrent writer may put a different node | |
152 | * at the root of the tree. See btrfs_lock_root_node for the | |
153 | * looping required. | |
154 | * | |
155 | * The extent buffer returned by this has a reference taken, so | |
156 | * it won't disappear. It may stop being the root of the tree | |
157 | * at any time because there are no locks held. | |
158 | */ | |
925baedd CM |
159 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
160 | { | |
161 | struct extent_buffer *eb; | |
240f62c8 | 162 | |
3083ee2e JB |
163 | while (1) { |
164 | rcu_read_lock(); | |
165 | eb = rcu_dereference(root->node); | |
166 | ||
167 | /* | |
168 | * RCU really hurts here, we could free up the root node because | |
169 | * it was cow'ed but we may not get the new root node yet so do | |
170 | * the inc_not_zero dance and if it doesn't work then | |
171 | * synchronize_rcu and try again. | |
172 | */ | |
173 | if (atomic_inc_not_zero(&eb->refs)) { | |
174 | rcu_read_unlock(); | |
175 | break; | |
176 | } | |
177 | rcu_read_unlock(); | |
178 | synchronize_rcu(); | |
179 | } | |
925baedd CM |
180 | return eb; |
181 | } | |
182 | ||
d352ac68 CM |
183 | /* loop around taking references on and locking the root node of the |
184 | * tree until you end up with a lock on the root. A locked buffer | |
185 | * is returned, with a reference held. | |
186 | */ | |
925baedd CM |
187 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
188 | { | |
189 | struct extent_buffer *eb; | |
190 | ||
d397712b | 191 | while (1) { |
925baedd CM |
192 | eb = btrfs_root_node(root); |
193 | btrfs_tree_lock(eb); | |
240f62c8 | 194 | if (eb == root->node) |
925baedd | 195 | break; |
925baedd CM |
196 | btrfs_tree_unlock(eb); |
197 | free_extent_buffer(eb); | |
198 | } | |
199 | return eb; | |
200 | } | |
201 | ||
bd681513 CM |
202 | /* loop around taking references on and locking the root node of the |
203 | * tree until you end up with a lock on the root. A locked buffer | |
204 | * is returned, with a reference held. | |
205 | */ | |
48a3b636 | 206 | static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
207 | { |
208 | struct extent_buffer *eb; | |
209 | ||
210 | while (1) { | |
211 | eb = btrfs_root_node(root); | |
212 | btrfs_tree_read_lock(eb); | |
213 | if (eb == root->node) | |
214 | break; | |
215 | btrfs_tree_read_unlock(eb); | |
216 | free_extent_buffer(eb); | |
217 | } | |
218 | return eb; | |
219 | } | |
220 | ||
d352ac68 CM |
221 | /* cowonly root (everything not a reference counted cow subvolume), just get |
222 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
223 | * get properly updated on disk. | |
224 | */ | |
0b86a832 CM |
225 | static void add_root_to_dirty_list(struct btrfs_root *root) |
226 | { | |
e5846fc6 | 227 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
228 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
229 | list_add(&root->dirty_list, | |
230 | &root->fs_info->dirty_cowonly_roots); | |
231 | } | |
e5846fc6 | 232 | spin_unlock(&root->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 | { | |
245 | struct extent_buffer *cow; | |
be20aa9d CM |
246 | int ret = 0; |
247 | int level; | |
5d4f98a2 | 248 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
249 | |
250 | WARN_ON(root->ref_cows && trans->transid != | |
251 | root->fs_info->running_transaction->transid); | |
252 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
253 | ||
254 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
255 | if (level == 0) |
256 | btrfs_item_key(buf, &disk_key, 0); | |
257 | else | |
258 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 259 | |
5d4f98a2 YZ |
260 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
261 | new_root_objectid, &disk_key, level, | |
5581a51a | 262 | buf->start, 0); |
5d4f98a2 | 263 | if (IS_ERR(cow)) |
be20aa9d CM |
264 | return PTR_ERR(cow); |
265 | ||
266 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
267 | btrfs_set_header_bytenr(cow, cow->start); | |
268 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
269 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
270 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
271 | BTRFS_HEADER_FLAG_RELOC); | |
272 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
273 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
274 | else | |
275 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 276 | |
2b82032c YZ |
277 | write_extent_buffer(cow, root->fs_info->fsid, |
278 | (unsigned long)btrfs_header_fsid(cow), | |
279 | BTRFS_FSID_SIZE); | |
280 | ||
be20aa9d | 281 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 282 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 283 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 284 | else |
66d7e7f0 | 285 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 286 | |
be20aa9d CM |
287 | if (ret) |
288 | return ret; | |
289 | ||
290 | btrfs_mark_buffer_dirty(cow); | |
291 | *cow_ret = cow; | |
292 | return 0; | |
293 | } | |
294 | ||
bd989ba3 JS |
295 | enum mod_log_op { |
296 | MOD_LOG_KEY_REPLACE, | |
297 | MOD_LOG_KEY_ADD, | |
298 | MOD_LOG_KEY_REMOVE, | |
299 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
300 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
301 | MOD_LOG_MOVE_KEYS, | |
302 | MOD_LOG_ROOT_REPLACE, | |
303 | }; | |
304 | ||
305 | struct tree_mod_move { | |
306 | int dst_slot; | |
307 | int nr_items; | |
308 | }; | |
309 | ||
310 | struct tree_mod_root { | |
311 | u64 logical; | |
312 | u8 level; | |
313 | }; | |
314 | ||
315 | struct tree_mod_elem { | |
316 | struct rb_node node; | |
317 | u64 index; /* shifted logical */ | |
097b8a7c | 318 | u64 seq; |
bd989ba3 JS |
319 | enum mod_log_op op; |
320 | ||
321 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
322 | int slot; | |
323 | ||
324 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
325 | u64 generation; | |
326 | ||
327 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
328 | struct btrfs_disk_key key; | |
329 | u64 blockptr; | |
330 | ||
331 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
332 | struct tree_mod_move move; | |
333 | ||
334 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
335 | struct tree_mod_root old_root; | |
336 | }; | |
337 | ||
097b8a7c | 338 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 339 | { |
097b8a7c | 340 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
341 | } |
342 | ||
097b8a7c JS |
343 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
344 | { | |
345 | read_unlock(&fs_info->tree_mod_log_lock); | |
346 | } | |
347 | ||
348 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
349 | { | |
350 | write_lock(&fs_info->tree_mod_log_lock); | |
351 | } | |
352 | ||
353 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
354 | { | |
355 | write_unlock(&fs_info->tree_mod_log_lock); | |
356 | } | |
357 | ||
fc36ed7e JS |
358 | /* |
359 | * Increment the upper half of tree_mod_seq, set lower half zero. | |
360 | * | |
361 | * Must be called with fs_info->tree_mod_seq_lock held. | |
362 | */ | |
363 | static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info) | |
364 | { | |
365 | u64 seq = atomic64_read(&fs_info->tree_mod_seq); | |
366 | seq &= 0xffffffff00000000ull; | |
367 | seq += 1ull << 32; | |
368 | atomic64_set(&fs_info->tree_mod_seq, seq); | |
369 | return seq; | |
370 | } | |
371 | ||
372 | /* | |
373 | * Increment the lower half of tree_mod_seq. | |
374 | * | |
375 | * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers | |
376 | * are generated should not technically require a spin lock here. (Rationale: | |
377 | * incrementing the minor while incrementing the major seq number is between its | |
378 | * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it | |
379 | * just returns a unique sequence number as usual.) We have decided to leave | |
380 | * that requirement in here and rethink it once we notice it really imposes a | |
381 | * problem on some workload. | |
382 | */ | |
383 | static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info) | |
384 | { | |
385 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
386 | } | |
387 | ||
388 | /* | |
389 | * return the last minor in the previous major tree_mod_seq number | |
390 | */ | |
391 | u64 btrfs_tree_mod_seq_prev(u64 seq) | |
392 | { | |
393 | return (seq & 0xffffffff00000000ull) - 1ull; | |
394 | } | |
395 | ||
097b8a7c JS |
396 | /* |
397 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
398 | * passed does not already have a sequence number set. So when a caller expects | |
399 | * to record tree modifications, it should ensure to set elem->seq to zero | |
400 | * before calling btrfs_get_tree_mod_seq. | |
401 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
402 | * blocker was added. | |
403 | */ | |
404 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
405 | struct seq_list *elem) | |
bd989ba3 | 406 | { |
097b8a7c JS |
407 | u64 seq; |
408 | ||
409 | tree_mod_log_write_lock(fs_info); | |
bd989ba3 | 410 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 411 | if (!elem->seq) { |
fc36ed7e | 412 | elem->seq = btrfs_inc_tree_mod_seq_major(fs_info); |
097b8a7c JS |
413 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
414 | } | |
fc36ed7e | 415 | seq = btrfs_inc_tree_mod_seq_minor(fs_info); |
bd989ba3 | 416 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
417 | tree_mod_log_write_unlock(fs_info); |
418 | ||
419 | return seq; | |
bd989ba3 JS |
420 | } |
421 | ||
422 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
423 | struct seq_list *elem) | |
424 | { | |
425 | struct rb_root *tm_root; | |
426 | struct rb_node *node; | |
427 | struct rb_node *next; | |
428 | struct seq_list *cur_elem; | |
429 | struct tree_mod_elem *tm; | |
430 | u64 min_seq = (u64)-1; | |
431 | u64 seq_putting = elem->seq; | |
432 | ||
433 | if (!seq_putting) | |
434 | return; | |
435 | ||
bd989ba3 JS |
436 | spin_lock(&fs_info->tree_mod_seq_lock); |
437 | list_del(&elem->list); | |
097b8a7c | 438 | elem->seq = 0; |
bd989ba3 JS |
439 | |
440 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 441 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
442 | if (seq_putting > cur_elem->seq) { |
443 | /* | |
444 | * blocker with lower sequence number exists, we | |
445 | * cannot remove anything from the log | |
446 | */ | |
097b8a7c JS |
447 | spin_unlock(&fs_info->tree_mod_seq_lock); |
448 | return; | |
bd989ba3 JS |
449 | } |
450 | min_seq = cur_elem->seq; | |
451 | } | |
452 | } | |
097b8a7c JS |
453 | spin_unlock(&fs_info->tree_mod_seq_lock); |
454 | ||
bd989ba3 JS |
455 | /* |
456 | * anything that's lower than the lowest existing (read: blocked) | |
457 | * sequence number can be removed from the tree. | |
458 | */ | |
097b8a7c | 459 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
460 | tm_root = &fs_info->tree_mod_log; |
461 | for (node = rb_first(tm_root); node; node = next) { | |
462 | next = rb_next(node); | |
463 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 464 | if (tm->seq > min_seq) |
bd989ba3 JS |
465 | continue; |
466 | rb_erase(node, tm_root); | |
bd989ba3 JS |
467 | kfree(tm); |
468 | } | |
097b8a7c | 469 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
470 | } |
471 | ||
472 | /* | |
473 | * key order of the log: | |
474 | * index -> sequence | |
475 | * | |
476 | * the index is the shifted logical of the *new* root node for root replace | |
477 | * operations, or the shifted logical of the affected block for all other | |
478 | * operations. | |
479 | */ | |
480 | static noinline int | |
481 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
482 | { | |
483 | struct rb_root *tm_root; | |
484 | struct rb_node **new; | |
485 | struct rb_node *parent = NULL; | |
486 | struct tree_mod_elem *cur; | |
bd989ba3 | 487 | |
097b8a7c | 488 | BUG_ON(!tm || !tm->seq); |
bd989ba3 | 489 | |
bd989ba3 JS |
490 | tm_root = &fs_info->tree_mod_log; |
491 | new = &tm_root->rb_node; | |
492 | while (*new) { | |
493 | cur = container_of(*new, struct tree_mod_elem, node); | |
494 | parent = *new; | |
495 | if (cur->index < tm->index) | |
496 | new = &((*new)->rb_left); | |
497 | else if (cur->index > tm->index) | |
498 | new = &((*new)->rb_right); | |
097b8a7c | 499 | else if (cur->seq < tm->seq) |
bd989ba3 | 500 | new = &((*new)->rb_left); |
097b8a7c | 501 | else if (cur->seq > tm->seq) |
bd989ba3 JS |
502 | new = &((*new)->rb_right); |
503 | else { | |
504 | kfree(tm); | |
097b8a7c | 505 | return -EEXIST; |
bd989ba3 JS |
506 | } |
507 | } | |
508 | ||
509 | rb_link_node(&tm->node, parent, new); | |
510 | rb_insert_color(&tm->node, tm_root); | |
097b8a7c | 511 | return 0; |
bd989ba3 JS |
512 | } |
513 | ||
097b8a7c JS |
514 | /* |
515 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
516 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
517 | * this until all tree mod log insertions are recorded in the rb tree and then | |
518 | * call tree_mod_log_write_unlock() to release. | |
519 | */ | |
e9b7fd4d JS |
520 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
521 | struct extent_buffer *eb) { | |
522 | smp_mb(); | |
523 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
524 | return 1; | |
097b8a7c JS |
525 | if (eb && btrfs_header_level(eb) == 0) |
526 | return 1; | |
527 | ||
528 | tree_mod_log_write_lock(fs_info); | |
529 | if (list_empty(&fs_info->tree_mod_seq_list)) { | |
530 | /* | |
531 | * someone emptied the list while we were waiting for the lock. | |
532 | * we must not add to the list when no blocker exists. | |
533 | */ | |
534 | tree_mod_log_write_unlock(fs_info); | |
e9b7fd4d | 535 | return 1; |
097b8a7c JS |
536 | } |
537 | ||
e9b7fd4d JS |
538 | return 0; |
539 | } | |
540 | ||
3310c36e | 541 | /* |
097b8a7c | 542 | * This allocates memory and gets a tree modification sequence number. |
3310c36e | 543 | * |
097b8a7c JS |
544 | * Returns <0 on error. |
545 | * Returns >0 (the added sequence number) on success. | |
3310c36e | 546 | */ |
926dd8a6 JS |
547 | static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags, |
548 | struct tree_mod_elem **tm_ret) | |
bd989ba3 JS |
549 | { |
550 | struct tree_mod_elem *tm; | |
bd989ba3 | 551 | |
097b8a7c JS |
552 | /* |
553 | * once we switch from spin locks to something different, we should | |
554 | * honor the flags parameter here. | |
555 | */ | |
556 | tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC); | |
bd989ba3 JS |
557 | if (!tm) |
558 | return -ENOMEM; | |
559 | ||
fc36ed7e JS |
560 | spin_lock(&fs_info->tree_mod_seq_lock); |
561 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
562 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
563 | ||
097b8a7c | 564 | return tm->seq; |
bd989ba3 JS |
565 | } |
566 | ||
097b8a7c JS |
567 | static inline int |
568 | __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, | |
569 | struct extent_buffer *eb, int slot, | |
570 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 571 | { |
bd989ba3 | 572 | int ret; |
097b8a7c | 573 | struct tree_mod_elem *tm; |
bd989ba3 JS |
574 | |
575 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
097b8a7c | 576 | if (ret < 0) |
bd989ba3 JS |
577 | return ret; |
578 | ||
579 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
580 | if (op != MOD_LOG_KEY_ADD) { | |
581 | btrfs_node_key(eb, &tm->key, slot); | |
582 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
583 | } | |
584 | tm->op = op; | |
585 | tm->slot = slot; | |
586 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
587 | ||
097b8a7c JS |
588 | return __tree_mod_log_insert(fs_info, tm); |
589 | } | |
590 | ||
591 | static noinline int | |
592 | tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info, | |
593 | struct extent_buffer *eb, int slot, | |
594 | enum mod_log_op op, gfp_t flags) | |
595 | { | |
596 | int ret; | |
597 | ||
598 | if (tree_mod_dont_log(fs_info, eb)) | |
599 | return 0; | |
600 | ||
601 | ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags); | |
602 | ||
603 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 604 | return ret; |
bd989ba3 JS |
605 | } |
606 | ||
607 | static noinline int | |
608 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
609 | int slot, enum mod_log_op op) | |
610 | { | |
611 | return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS); | |
612 | } | |
613 | ||
097b8a7c JS |
614 | static noinline int |
615 | tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info, | |
616 | struct extent_buffer *eb, int slot, | |
617 | enum mod_log_op op) | |
618 | { | |
619 | return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS); | |
620 | } | |
621 | ||
bd989ba3 JS |
622 | static noinline int |
623 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
624 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
625 | int nr_items, gfp_t flags) | |
626 | { | |
627 | struct tree_mod_elem *tm; | |
628 | int ret; | |
629 | int i; | |
630 | ||
f395694c JS |
631 | if (tree_mod_dont_log(fs_info, eb)) |
632 | return 0; | |
bd989ba3 | 633 | |
01763a2e JS |
634 | /* |
635 | * When we override something during the move, we log these removals. | |
636 | * This can only happen when we move towards the beginning of the | |
637 | * buffer, i.e. dst_slot < src_slot. | |
638 | */ | |
bd989ba3 | 639 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
097b8a7c | 640 | ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot, |
bd989ba3 JS |
641 | MOD_LOG_KEY_REMOVE_WHILE_MOVING); |
642 | BUG_ON(ret < 0); | |
643 | } | |
644 | ||
f395694c | 645 | ret = tree_mod_alloc(fs_info, flags, &tm); |
097b8a7c JS |
646 | if (ret < 0) |
647 | goto out; | |
f395694c | 648 | |
bd989ba3 JS |
649 | tm->index = eb->start >> PAGE_CACHE_SHIFT; |
650 | tm->slot = src_slot; | |
651 | tm->move.dst_slot = dst_slot; | |
652 | tm->move.nr_items = nr_items; | |
653 | tm->op = MOD_LOG_MOVE_KEYS; | |
654 | ||
3310c36e | 655 | ret = __tree_mod_log_insert(fs_info, tm); |
097b8a7c JS |
656 | out: |
657 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 658 | return ret; |
bd989ba3 JS |
659 | } |
660 | ||
097b8a7c JS |
661 | static inline void |
662 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
663 | { | |
664 | int i; | |
665 | u32 nritems; | |
666 | int ret; | |
667 | ||
b12a3b1e CM |
668 | if (btrfs_header_level(eb) == 0) |
669 | return; | |
670 | ||
097b8a7c JS |
671 | nritems = btrfs_header_nritems(eb); |
672 | for (i = nritems - 1; i >= 0; i--) { | |
673 | ret = tree_mod_log_insert_key_locked(fs_info, eb, i, | |
674 | MOD_LOG_KEY_REMOVE_WHILE_FREEING); | |
675 | BUG_ON(ret < 0); | |
676 | } | |
677 | } | |
678 | ||
bd989ba3 JS |
679 | static noinline int |
680 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
681 | struct extent_buffer *old_root, | |
90f8d62e JS |
682 | struct extent_buffer *new_root, gfp_t flags, |
683 | int log_removal) | |
bd989ba3 JS |
684 | { |
685 | struct tree_mod_elem *tm; | |
686 | int ret; | |
687 | ||
097b8a7c JS |
688 | if (tree_mod_dont_log(fs_info, NULL)) |
689 | return 0; | |
690 | ||
90f8d62e JS |
691 | if (log_removal) |
692 | __tree_mod_log_free_eb(fs_info, old_root); | |
d9abbf1c | 693 | |
bd989ba3 | 694 | ret = tree_mod_alloc(fs_info, flags, &tm); |
097b8a7c JS |
695 | if (ret < 0) |
696 | goto out; | |
bd989ba3 JS |
697 | |
698 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
699 | tm->old_root.logical = old_root->start; | |
700 | tm->old_root.level = btrfs_header_level(old_root); | |
701 | tm->generation = btrfs_header_generation(old_root); | |
702 | tm->op = MOD_LOG_ROOT_REPLACE; | |
703 | ||
3310c36e | 704 | ret = __tree_mod_log_insert(fs_info, tm); |
097b8a7c JS |
705 | out: |
706 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 707 | return ret; |
bd989ba3 JS |
708 | } |
709 | ||
710 | static struct tree_mod_elem * | |
711 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
712 | int smallest) | |
713 | { | |
714 | struct rb_root *tm_root; | |
715 | struct rb_node *node; | |
716 | struct tree_mod_elem *cur = NULL; | |
717 | struct tree_mod_elem *found = NULL; | |
718 | u64 index = start >> PAGE_CACHE_SHIFT; | |
719 | ||
097b8a7c | 720 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
721 | tm_root = &fs_info->tree_mod_log; |
722 | node = tm_root->rb_node; | |
723 | while (node) { | |
724 | cur = container_of(node, struct tree_mod_elem, node); | |
725 | if (cur->index < index) { | |
726 | node = node->rb_left; | |
727 | } else if (cur->index > index) { | |
728 | node = node->rb_right; | |
097b8a7c | 729 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
730 | node = node->rb_left; |
731 | } else if (!smallest) { | |
732 | /* we want the node with the highest seq */ | |
733 | if (found) | |
097b8a7c | 734 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
735 | found = cur; |
736 | node = node->rb_left; | |
097b8a7c | 737 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
738 | /* we want the node with the smallest seq */ |
739 | if (found) | |
097b8a7c | 740 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
741 | found = cur; |
742 | node = node->rb_right; | |
743 | } else { | |
744 | found = cur; | |
745 | break; | |
746 | } | |
747 | } | |
097b8a7c | 748 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
749 | |
750 | return found; | |
751 | } | |
752 | ||
753 | /* | |
754 | * this returns the element from the log with the smallest time sequence | |
755 | * value that's in the log (the oldest log item). any element with a time | |
756 | * sequence lower than min_seq will be ignored. | |
757 | */ | |
758 | static struct tree_mod_elem * | |
759 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
760 | u64 min_seq) | |
761 | { | |
762 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
763 | } | |
764 | ||
765 | /* | |
766 | * this returns the element from the log with the largest time sequence | |
767 | * value that's in the log (the most recent log item). any element with | |
768 | * a time sequence lower than min_seq will be ignored. | |
769 | */ | |
770 | static struct tree_mod_elem * | |
771 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
772 | { | |
773 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
774 | } | |
775 | ||
097b8a7c | 776 | static noinline void |
bd989ba3 JS |
777 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
778 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 779 | unsigned long src_offset, int nr_items) |
bd989ba3 JS |
780 | { |
781 | int ret; | |
782 | int i; | |
783 | ||
e9b7fd4d | 784 | if (tree_mod_dont_log(fs_info, NULL)) |
bd989ba3 JS |
785 | return; |
786 | ||
097b8a7c JS |
787 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) { |
788 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 | 789 | return; |
097b8a7c | 790 | } |
bd989ba3 | 791 | |
bd989ba3 | 792 | for (i = 0; i < nr_items; i++) { |
90f8d62e JS |
793 | ret = tree_mod_log_insert_key_locked(fs_info, src, |
794 | i + src_offset, | |
795 | MOD_LOG_KEY_REMOVE); | |
796 | BUG_ON(ret < 0); | |
097b8a7c JS |
797 | ret = tree_mod_log_insert_key_locked(fs_info, dst, |
798 | i + dst_offset, | |
799 | MOD_LOG_KEY_ADD); | |
bd989ba3 JS |
800 | BUG_ON(ret < 0); |
801 | } | |
097b8a7c JS |
802 | |
803 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 JS |
804 | } |
805 | ||
806 | static inline void | |
807 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
808 | int dst_offset, int src_offset, int nr_items) | |
809 | { | |
810 | int ret; | |
811 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
812 | nr_items, GFP_NOFS); | |
813 | BUG_ON(ret < 0); | |
814 | } | |
815 | ||
097b8a7c | 816 | static noinline void |
bd989ba3 | 817 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 818 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
819 | { |
820 | int ret; | |
821 | ||
822 | ret = tree_mod_log_insert_key_mask(fs_info, eb, slot, | |
823 | MOD_LOG_KEY_REPLACE, | |
824 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
825 | BUG_ON(ret < 0); | |
826 | } | |
827 | ||
097b8a7c JS |
828 | static noinline void |
829 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
bd989ba3 | 830 | { |
e9b7fd4d | 831 | if (tree_mod_dont_log(fs_info, eb)) |
bd989ba3 JS |
832 | return; |
833 | ||
097b8a7c JS |
834 | __tree_mod_log_free_eb(fs_info, eb); |
835 | ||
836 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 JS |
837 | } |
838 | ||
097b8a7c | 839 | static noinline void |
bd989ba3 | 840 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
841 | struct extent_buffer *new_root_node, |
842 | int log_removal) | |
bd989ba3 JS |
843 | { |
844 | int ret; | |
bd989ba3 | 845 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 846 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
847 | BUG_ON(ret < 0); |
848 | } | |
849 | ||
5d4f98a2 YZ |
850 | /* |
851 | * check if the tree block can be shared by multiple trees | |
852 | */ | |
853 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
854 | struct extent_buffer *buf) | |
855 | { | |
856 | /* | |
857 | * Tree blocks not in refernece counted trees and tree roots | |
858 | * are never shared. If a block was allocated after the last | |
859 | * snapshot and the block was not allocated by tree relocation, | |
860 | * we know the block is not shared. | |
861 | */ | |
862 | if (root->ref_cows && | |
863 | buf != root->node && buf != root->commit_root && | |
864 | (btrfs_header_generation(buf) <= | |
865 | btrfs_root_last_snapshot(&root->root_item) || | |
866 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
867 | return 1; | |
868 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
869 | if (root->ref_cows && | |
870 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
871 | return 1; | |
872 | #endif | |
873 | return 0; | |
874 | } | |
875 | ||
876 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
877 | struct btrfs_root *root, | |
878 | struct extent_buffer *buf, | |
f0486c68 YZ |
879 | struct extent_buffer *cow, |
880 | int *last_ref) | |
5d4f98a2 YZ |
881 | { |
882 | u64 refs; | |
883 | u64 owner; | |
884 | u64 flags; | |
885 | u64 new_flags = 0; | |
886 | int ret; | |
887 | ||
888 | /* | |
889 | * Backrefs update rules: | |
890 | * | |
891 | * Always use full backrefs for extent pointers in tree block | |
892 | * allocated by tree relocation. | |
893 | * | |
894 | * If a shared tree block is no longer referenced by its owner | |
895 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
896 | * use full backrefs for extent pointers in tree block. | |
897 | * | |
898 | * If a tree block is been relocating | |
899 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
900 | * use full backrefs for extent pointers in tree block. | |
901 | * The reason for this is some operations (such as drop tree) | |
902 | * are only allowed for blocks use full backrefs. | |
903 | */ | |
904 | ||
905 | if (btrfs_block_can_be_shared(root, buf)) { | |
906 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
907 | btrfs_header_level(buf), 1, |
908 | &refs, &flags); | |
be1a5564 MF |
909 | if (ret) |
910 | return ret; | |
e5df9573 MF |
911 | if (refs == 0) { |
912 | ret = -EROFS; | |
913 | btrfs_std_error(root->fs_info, ret); | |
914 | return ret; | |
915 | } | |
5d4f98a2 YZ |
916 | } else { |
917 | refs = 1; | |
918 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
919 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
920 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
921 | else | |
922 | flags = 0; | |
923 | } | |
924 | ||
925 | owner = btrfs_header_owner(buf); | |
926 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
927 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
928 | ||
929 | if (refs > 1) { | |
930 | if ((owner == root->root_key.objectid || | |
931 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
932 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 933 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 934 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
935 | |
936 | if (root->root_key.objectid == | |
937 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 938 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 939 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 940 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 941 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
942 | } |
943 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
944 | } else { | |
945 | ||
946 | if (root->root_key.objectid == | |
947 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 948 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 949 | else |
66d7e7f0 | 950 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 951 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
952 | } |
953 | if (new_flags != 0) { | |
b1c79e09 JB |
954 | int level = btrfs_header_level(buf); |
955 | ||
5d4f98a2 YZ |
956 | ret = btrfs_set_disk_extent_flags(trans, root, |
957 | buf->start, | |
958 | buf->len, | |
b1c79e09 | 959 | new_flags, level, 0); |
be1a5564 MF |
960 | if (ret) |
961 | return ret; | |
5d4f98a2 YZ |
962 | } |
963 | } else { | |
964 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
965 | if (root->root_key.objectid == | |
966 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 967 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 968 | else |
66d7e7f0 | 969 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 970 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 971 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 972 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
973 | } |
974 | clean_tree_block(trans, root, buf); | |
f0486c68 | 975 | *last_ref = 1; |
5d4f98a2 YZ |
976 | } |
977 | return 0; | |
978 | } | |
979 | ||
d352ac68 | 980 | /* |
d397712b CM |
981 | * does the dirty work in cow of a single block. The parent block (if |
982 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
983 | * dirty and returned locked. If you modify the block it needs to be marked | |
984 | * dirty again. | |
d352ac68 CM |
985 | * |
986 | * search_start -- an allocation hint for the new block | |
987 | * | |
d397712b CM |
988 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
989 | * bytes the allocator should try to find free next to the block it returns. | |
990 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 991 | */ |
d397712b | 992 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
993 | struct btrfs_root *root, |
994 | struct extent_buffer *buf, | |
995 | struct extent_buffer *parent, int parent_slot, | |
996 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 997 | u64 search_start, u64 empty_size) |
02217ed2 | 998 | { |
5d4f98a2 | 999 | struct btrfs_disk_key disk_key; |
5f39d397 | 1000 | struct extent_buffer *cow; |
be1a5564 | 1001 | int level, ret; |
f0486c68 | 1002 | int last_ref = 0; |
925baedd | 1003 | int unlock_orig = 0; |
5d4f98a2 | 1004 | u64 parent_start; |
7bb86316 | 1005 | |
925baedd CM |
1006 | if (*cow_ret == buf) |
1007 | unlock_orig = 1; | |
1008 | ||
b9447ef8 | 1009 | btrfs_assert_tree_locked(buf); |
925baedd | 1010 | |
7bb86316 CM |
1011 | WARN_ON(root->ref_cows && trans->transid != |
1012 | root->fs_info->running_transaction->transid); | |
6702ed49 | 1013 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 1014 | |
7bb86316 | 1015 | level = btrfs_header_level(buf); |
31840ae1 | 1016 | |
5d4f98a2 YZ |
1017 | if (level == 0) |
1018 | btrfs_item_key(buf, &disk_key, 0); | |
1019 | else | |
1020 | btrfs_node_key(buf, &disk_key, 0); | |
1021 | ||
1022 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
1023 | if (parent) | |
1024 | parent_start = parent->start; | |
1025 | else | |
1026 | parent_start = 0; | |
1027 | } else | |
1028 | parent_start = 0; | |
1029 | ||
1030 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
1031 | root->root_key.objectid, &disk_key, | |
5581a51a | 1032 | level, search_start, empty_size); |
54aa1f4d CM |
1033 | if (IS_ERR(cow)) |
1034 | return PTR_ERR(cow); | |
6702ed49 | 1035 | |
b4ce94de CM |
1036 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1037 | ||
5f39d397 | 1038 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 1039 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1040 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1041 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1042 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1043 | BTRFS_HEADER_FLAG_RELOC); | |
1044 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1045 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1046 | else | |
1047 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1048 | |
2b82032c YZ |
1049 | write_extent_buffer(cow, root->fs_info->fsid, |
1050 | (unsigned long)btrfs_header_fsid(cow), | |
1051 | BTRFS_FSID_SIZE); | |
1052 | ||
be1a5564 | 1053 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1054 | if (ret) { |
79787eaa | 1055 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1056 | return ret; |
1057 | } | |
1a40e23b | 1058 | |
3fd0a558 YZ |
1059 | if (root->ref_cows) |
1060 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
1061 | ||
02217ed2 | 1062 | if (buf == root->node) { |
925baedd | 1063 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1064 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1065 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1066 | parent_start = buf->start; | |
1067 | else | |
1068 | parent_start = 0; | |
925baedd | 1069 | |
5f39d397 | 1070 | extent_buffer_get(cow); |
90f8d62e | 1071 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1072 | rcu_assign_pointer(root->node, cow); |
925baedd | 1073 | |
f0486c68 | 1074 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1075 | last_ref); |
5f39d397 | 1076 | free_extent_buffer(buf); |
0b86a832 | 1077 | add_root_to_dirty_list(root); |
02217ed2 | 1078 | } else { |
5d4f98a2 YZ |
1079 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1080 | parent_start = parent->start; | |
1081 | else | |
1082 | parent_start = 0; | |
1083 | ||
1084 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e JS |
1085 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
1086 | MOD_LOG_KEY_REPLACE); | |
5f39d397 | 1087 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1088 | cow->start); |
74493f7a CM |
1089 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1090 | trans->transid); | |
d6025579 | 1091 | btrfs_mark_buffer_dirty(parent); |
d9abbf1c | 1092 | tree_mod_log_free_eb(root->fs_info, buf); |
f0486c68 | 1093 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1094 | last_ref); |
02217ed2 | 1095 | } |
925baedd CM |
1096 | if (unlock_orig) |
1097 | btrfs_tree_unlock(buf); | |
3083ee2e | 1098 | free_extent_buffer_stale(buf); |
ccd467d6 | 1099 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1100 | *cow_ret = cow; |
02217ed2 CM |
1101 | return 0; |
1102 | } | |
1103 | ||
5d9e75c4 JS |
1104 | /* |
1105 | * returns the logical address of the oldest predecessor of the given root. | |
1106 | * entries older than time_seq are ignored. | |
1107 | */ | |
1108 | static struct tree_mod_elem * | |
1109 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1110 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1111 | { |
1112 | struct tree_mod_elem *tm; | |
1113 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1114 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1115 | int looped = 0; |
1116 | ||
1117 | if (!time_seq) | |
1118 | return 0; | |
1119 | ||
1120 | /* | |
1121 | * the very last operation that's logged for a root is the replacement | |
1122 | * operation (if it is replaced at all). this has the index of the *new* | |
1123 | * root, making it the very first operation that's logged for this root. | |
1124 | */ | |
1125 | while (1) { | |
1126 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1127 | time_seq); | |
1128 | if (!looped && !tm) | |
1129 | return 0; | |
1130 | /* | |
28da9fb4 JS |
1131 | * if there are no tree operation for the oldest root, we simply |
1132 | * return it. this should only happen if that (old) root is at | |
1133 | * level 0. | |
5d9e75c4 | 1134 | */ |
28da9fb4 JS |
1135 | if (!tm) |
1136 | break; | |
5d9e75c4 | 1137 | |
28da9fb4 JS |
1138 | /* |
1139 | * if there's an operation that's not a root replacement, we | |
1140 | * found the oldest version of our root. normally, we'll find a | |
1141 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1142 | */ | |
5d9e75c4 JS |
1143 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1144 | break; | |
1145 | ||
1146 | found = tm; | |
1147 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1148 | looped = 1; |
1149 | } | |
1150 | ||
a95236d9 JS |
1151 | /* if there's no old root to return, return what we found instead */ |
1152 | if (!found) | |
1153 | found = tm; | |
1154 | ||
5d9e75c4 JS |
1155 | return found; |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1160 | * previous operations will be rewinded (until we reach something older than | |
1161 | * time_seq). | |
1162 | */ | |
1163 | static void | |
1164 | __tree_mod_log_rewind(struct extent_buffer *eb, u64 time_seq, | |
1165 | struct tree_mod_elem *first_tm) | |
1166 | { | |
1167 | u32 n; | |
1168 | struct rb_node *next; | |
1169 | struct tree_mod_elem *tm = first_tm; | |
1170 | unsigned long o_dst; | |
1171 | unsigned long o_src; | |
1172 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1173 | ||
1174 | n = btrfs_header_nritems(eb); | |
097b8a7c | 1175 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1176 | /* |
1177 | * all the operations are recorded with the operator used for | |
1178 | * the modification. as we're going backwards, we do the | |
1179 | * opposite of each operation here. | |
1180 | */ | |
1181 | switch (tm->op) { | |
1182 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1183 | BUG_ON(tm->slot < n); | |
1c697d4a | 1184 | /* Fallthrough */ |
95c80bb1 | 1185 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1186 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1187 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1188 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1189 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1190 | tm->generation); | |
4c3e6969 | 1191 | n++; |
5d9e75c4 JS |
1192 | break; |
1193 | case MOD_LOG_KEY_REPLACE: | |
1194 | BUG_ON(tm->slot >= n); | |
1195 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1196 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1197 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1198 | tm->generation); | |
1199 | break; | |
1200 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1201 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1202 | n--; |
1203 | break; | |
1204 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1205 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1206 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1207 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1208 | tm->move.nr_items * p_size); |
1209 | break; | |
1210 | case MOD_LOG_ROOT_REPLACE: | |
1211 | /* | |
1212 | * this operation is special. for roots, this must be | |
1213 | * handled explicitly before rewinding. | |
1214 | * for non-roots, this operation may exist if the node | |
1215 | * was a root: root A -> child B; then A gets empty and | |
1216 | * B is promoted to the new root. in the mod log, we'll | |
1217 | * have a root-replace operation for B, a tree block | |
1218 | * that is no root. we simply ignore that operation. | |
1219 | */ | |
1220 | break; | |
1221 | } | |
1222 | next = rb_next(&tm->node); | |
1223 | if (!next) | |
1224 | break; | |
1225 | tm = container_of(next, struct tree_mod_elem, node); | |
1226 | if (tm->index != first_tm->index) | |
1227 | break; | |
1228 | } | |
1229 | btrfs_set_header_nritems(eb, n); | |
1230 | } | |
1231 | ||
47fb091f JS |
1232 | /* |
1233 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1234 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1235 | * returned buffer is always read-locked. If the returned buffer is not the | |
1236 | * input buffer, the lock on the input buffer is released and the input buffer | |
1237 | * is freed (its refcount is decremented). | |
1238 | */ | |
5d9e75c4 JS |
1239 | static struct extent_buffer * |
1240 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
1241 | u64 time_seq) | |
1242 | { | |
1243 | struct extent_buffer *eb_rewin; | |
1244 | struct tree_mod_elem *tm; | |
1245 | ||
1246 | if (!time_seq) | |
1247 | return eb; | |
1248 | ||
1249 | if (btrfs_header_level(eb) == 0) | |
1250 | return eb; | |
1251 | ||
1252 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1253 | if (!tm) | |
1254 | return eb; | |
1255 | ||
1256 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { | |
1257 | BUG_ON(tm->slot != 0); | |
1258 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1259 | fs_info->tree_root->nodesize); | |
1260 | BUG_ON(!eb_rewin); | |
1261 | btrfs_set_header_bytenr(eb_rewin, eb->start); | |
1262 | btrfs_set_header_backref_rev(eb_rewin, | |
1263 | btrfs_header_backref_rev(eb)); | |
1264 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1265 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1266 | } else { |
1267 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
1268 | BUG_ON(!eb_rewin); | |
1269 | } | |
1270 | ||
1271 | extent_buffer_get(eb_rewin); | |
47fb091f | 1272 | btrfs_tree_read_unlock(eb); |
5d9e75c4 JS |
1273 | free_extent_buffer(eb); |
1274 | ||
47fb091f JS |
1275 | extent_buffer_get(eb_rewin); |
1276 | btrfs_tree_read_lock(eb_rewin); | |
5d9e75c4 | 1277 | __tree_mod_log_rewind(eb_rewin, time_seq, tm); |
57911b8b | 1278 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1279 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1280 | |
1281 | return eb_rewin; | |
1282 | } | |
1283 | ||
8ba97a15 JS |
1284 | /* |
1285 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1286 | * value. If there are no changes, the current root->root_node is returned. If | |
1287 | * anything changed in between, there's a fresh buffer allocated on which the | |
1288 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1289 | * Returns NULL on error (with no locks held). | |
1290 | */ | |
5d9e75c4 JS |
1291 | static inline struct extent_buffer * |
1292 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1293 | { | |
1294 | struct tree_mod_elem *tm; | |
30b0463a JS |
1295 | struct extent_buffer *eb = NULL; |
1296 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1297 | struct extent_buffer *old; |
a95236d9 | 1298 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1299 | u64 old_generation = 0; |
a95236d9 | 1300 | u64 logical; |
834328a8 | 1301 | u32 blocksize; |
5d9e75c4 | 1302 | |
30b0463a JS |
1303 | eb_root = btrfs_read_lock_root_node(root); |
1304 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1305 | if (!tm) |
30b0463a | 1306 | return eb_root; |
5d9e75c4 | 1307 | |
a95236d9 JS |
1308 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1309 | old_root = &tm->old_root; | |
1310 | old_generation = tm->generation; | |
1311 | logical = old_root->logical; | |
1312 | } else { | |
30b0463a | 1313 | logical = eb_root->start; |
a95236d9 | 1314 | } |
5d9e75c4 | 1315 | |
a95236d9 | 1316 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1317 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1318 | btrfs_tree_read_unlock(eb_root); |
1319 | free_extent_buffer(eb_root); | |
834328a8 | 1320 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1321 | old = read_tree_block(root, logical, blocksize, 0); |
416bc658 JB |
1322 | if (!old || !extent_buffer_uptodate(old)) { |
1323 | free_extent_buffer(old); | |
834328a8 JS |
1324 | pr_warn("btrfs: failed to read tree block %llu from get_old_root\n", |
1325 | logical); | |
1326 | WARN_ON(1); | |
1327 | } else { | |
7bfdcf7f LB |
1328 | eb = btrfs_clone_extent_buffer(old); |
1329 | free_extent_buffer(old); | |
834328a8 JS |
1330 | } |
1331 | } else if (old_root) { | |
30b0463a JS |
1332 | btrfs_tree_read_unlock(eb_root); |
1333 | free_extent_buffer(eb_root); | |
28da9fb4 | 1334 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1335 | } else { |
30b0463a JS |
1336 | eb = btrfs_clone_extent_buffer(eb_root); |
1337 | btrfs_tree_read_unlock(eb_root); | |
1338 | free_extent_buffer(eb_root); | |
834328a8 JS |
1339 | } |
1340 | ||
8ba97a15 JS |
1341 | if (!eb) |
1342 | return NULL; | |
d6381084 | 1343 | extent_buffer_get(eb); |
8ba97a15 | 1344 | btrfs_tree_read_lock(eb); |
a95236d9 | 1345 | if (old_root) { |
5d9e75c4 JS |
1346 | btrfs_set_header_bytenr(eb, eb->start); |
1347 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1348 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1349 | btrfs_set_header_level(eb, old_root->level); |
1350 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1351 | } |
28da9fb4 JS |
1352 | if (tm) |
1353 | __tree_mod_log_rewind(eb, time_seq, tm); | |
1354 | else | |
1355 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1356 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1357 | |
1358 | return eb; | |
1359 | } | |
1360 | ||
5b6602e7 JS |
1361 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1362 | { | |
1363 | struct tree_mod_elem *tm; | |
1364 | int level; | |
30b0463a | 1365 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1366 | |
30b0463a | 1367 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1368 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1369 | level = tm->old_root.level; | |
1370 | } else { | |
30b0463a | 1371 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1372 | } |
30b0463a | 1373 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1374 | |
1375 | return level; | |
1376 | } | |
1377 | ||
5d4f98a2 YZ |
1378 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1379 | struct btrfs_root *root, | |
1380 | struct extent_buffer *buf) | |
1381 | { | |
f1ebcc74 LB |
1382 | /* ensure we can see the force_cow */ |
1383 | smp_rmb(); | |
1384 | ||
1385 | /* | |
1386 | * We do not need to cow a block if | |
1387 | * 1) this block is not created or changed in this transaction; | |
1388 | * 2) this block does not belong to TREE_RELOC tree; | |
1389 | * 3) the root is not forced COW. | |
1390 | * | |
1391 | * What is forced COW: | |
1392 | * when we create snapshot during commiting the transaction, | |
1393 | * after we've finished coping src root, we must COW the shared | |
1394 | * block to ensure the metadata consistency. | |
1395 | */ | |
5d4f98a2 YZ |
1396 | if (btrfs_header_generation(buf) == trans->transid && |
1397 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1398 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1399 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1400 | !root->force_cow) | |
5d4f98a2 YZ |
1401 | return 0; |
1402 | return 1; | |
1403 | } | |
1404 | ||
d352ac68 CM |
1405 | /* |
1406 | * cows a single block, see __btrfs_cow_block for the real work. | |
1407 | * This version of it has extra checks so that a block isn't cow'd more than | |
1408 | * once per transaction, as long as it hasn't been written yet | |
1409 | */ | |
d397712b | 1410 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1411 | struct btrfs_root *root, struct extent_buffer *buf, |
1412 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1413 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1414 | { |
1415 | u64 search_start; | |
f510cfec | 1416 | int ret; |
dc17ff8f | 1417 | |
31b1a2bd JL |
1418 | if (trans->transaction != root->fs_info->running_transaction) |
1419 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
d397712b CM |
1420 | (unsigned long long)trans->transid, |
1421 | (unsigned long long) | |
6702ed49 | 1422 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1423 | |
1424 | if (trans->transid != root->fs_info->generation) | |
1425 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
d397712b CM |
1426 | (unsigned long long)trans->transid, |
1427 | (unsigned long long)root->fs_info->generation); | |
dc17ff8f | 1428 | |
5d4f98a2 | 1429 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1430 | *cow_ret = buf; |
1431 | return 0; | |
1432 | } | |
c487685d | 1433 | |
0b86a832 | 1434 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1435 | |
1436 | if (parent) | |
1437 | btrfs_set_lock_blocking(parent); | |
1438 | btrfs_set_lock_blocking(buf); | |
1439 | ||
f510cfec | 1440 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1441 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1442 | |
1443 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1444 | ||
f510cfec | 1445 | return ret; |
6702ed49 CM |
1446 | } |
1447 | ||
d352ac68 CM |
1448 | /* |
1449 | * helper function for defrag to decide if two blocks pointed to by a | |
1450 | * node are actually close by | |
1451 | */ | |
6b80053d | 1452 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1453 | { |
6b80053d | 1454 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1455 | return 1; |
6b80053d | 1456 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1457 | return 1; |
1458 | return 0; | |
1459 | } | |
1460 | ||
081e9573 CM |
1461 | /* |
1462 | * compare two keys in a memcmp fashion | |
1463 | */ | |
1464 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1465 | { | |
1466 | struct btrfs_key k1; | |
1467 | ||
1468 | btrfs_disk_key_to_cpu(&k1, disk); | |
1469 | ||
20736aba | 1470 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1471 | } |
1472 | ||
f3465ca4 JB |
1473 | /* |
1474 | * same as comp_keys only with two btrfs_key's | |
1475 | */ | |
5d4f98a2 | 1476 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1477 | { |
1478 | if (k1->objectid > k2->objectid) | |
1479 | return 1; | |
1480 | if (k1->objectid < k2->objectid) | |
1481 | return -1; | |
1482 | if (k1->type > k2->type) | |
1483 | return 1; | |
1484 | if (k1->type < k2->type) | |
1485 | return -1; | |
1486 | if (k1->offset > k2->offset) | |
1487 | return 1; | |
1488 | if (k1->offset < k2->offset) | |
1489 | return -1; | |
1490 | return 0; | |
1491 | } | |
081e9573 | 1492 | |
d352ac68 CM |
1493 | /* |
1494 | * this is used by the defrag code to go through all the | |
1495 | * leaves pointed to by a node and reallocate them so that | |
1496 | * disk order is close to key order | |
1497 | */ | |
6702ed49 | 1498 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1499 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1500 | int start_slot, u64 *last_ret, |
a6b6e75e | 1501 | struct btrfs_key *progress) |
6702ed49 | 1502 | { |
6b80053d | 1503 | struct extent_buffer *cur; |
6702ed49 | 1504 | u64 blocknr; |
ca7a79ad | 1505 | u64 gen; |
e9d0b13b CM |
1506 | u64 search_start = *last_ret; |
1507 | u64 last_block = 0; | |
6702ed49 CM |
1508 | u64 other; |
1509 | u32 parent_nritems; | |
6702ed49 CM |
1510 | int end_slot; |
1511 | int i; | |
1512 | int err = 0; | |
f2183bde | 1513 | int parent_level; |
6b80053d CM |
1514 | int uptodate; |
1515 | u32 blocksize; | |
081e9573 CM |
1516 | int progress_passed = 0; |
1517 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1518 | |
5708b959 | 1519 | parent_level = btrfs_header_level(parent); |
5708b959 | 1520 | |
6c1500f2 JL |
1521 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1522 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1523 | |
6b80053d | 1524 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1525 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1526 | end_slot = parent_nritems; |
1527 | ||
1528 | if (parent_nritems == 1) | |
1529 | return 0; | |
1530 | ||
b4ce94de CM |
1531 | btrfs_set_lock_blocking(parent); |
1532 | ||
6702ed49 CM |
1533 | for (i = start_slot; i < end_slot; i++) { |
1534 | int close = 1; | |
a6b6e75e | 1535 | |
081e9573 CM |
1536 | btrfs_node_key(parent, &disk_key, i); |
1537 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1538 | continue; | |
1539 | ||
1540 | progress_passed = 1; | |
6b80053d | 1541 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1542 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1543 | if (last_block == 0) |
1544 | last_block = blocknr; | |
5708b959 | 1545 | |
6702ed49 | 1546 | if (i > 0) { |
6b80053d CM |
1547 | other = btrfs_node_blockptr(parent, i - 1); |
1548 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1549 | } |
0ef3e66b | 1550 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1551 | other = btrfs_node_blockptr(parent, i + 1); |
1552 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1553 | } |
e9d0b13b CM |
1554 | if (close) { |
1555 | last_block = blocknr; | |
6702ed49 | 1556 | continue; |
e9d0b13b | 1557 | } |
6702ed49 | 1558 | |
6b80053d CM |
1559 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1560 | if (cur) | |
b9fab919 | 1561 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1562 | else |
1563 | uptodate = 0; | |
5708b959 | 1564 | if (!cur || !uptodate) { |
6b80053d CM |
1565 | if (!cur) { |
1566 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1567 | blocksize, gen); |
416bc658 JB |
1568 | if (!cur || !extent_buffer_uptodate(cur)) { |
1569 | free_extent_buffer(cur); | |
97d9a8a4 | 1570 | return -EIO; |
416bc658 | 1571 | } |
6b80053d | 1572 | } else if (!uptodate) { |
018642a1 TI |
1573 | err = btrfs_read_buffer(cur, gen); |
1574 | if (err) { | |
1575 | free_extent_buffer(cur); | |
1576 | return err; | |
1577 | } | |
f2183bde | 1578 | } |
6702ed49 | 1579 | } |
e9d0b13b | 1580 | if (search_start == 0) |
6b80053d | 1581 | search_start = last_block; |
e9d0b13b | 1582 | |
e7a84565 | 1583 | btrfs_tree_lock(cur); |
b4ce94de | 1584 | btrfs_set_lock_blocking(cur); |
6b80053d | 1585 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1586 | &cur, search_start, |
6b80053d | 1587 | min(16 * blocksize, |
9fa8cfe7 | 1588 | (end_slot - i) * blocksize)); |
252c38f0 | 1589 | if (err) { |
e7a84565 | 1590 | btrfs_tree_unlock(cur); |
6b80053d | 1591 | free_extent_buffer(cur); |
6702ed49 | 1592 | break; |
252c38f0 | 1593 | } |
e7a84565 CM |
1594 | search_start = cur->start; |
1595 | last_block = cur->start; | |
f2183bde | 1596 | *last_ret = search_start; |
e7a84565 CM |
1597 | btrfs_tree_unlock(cur); |
1598 | free_extent_buffer(cur); | |
6702ed49 CM |
1599 | } |
1600 | return err; | |
1601 | } | |
1602 | ||
74123bd7 CM |
1603 | /* |
1604 | * The leaf data grows from end-to-front in the node. | |
1605 | * this returns the address of the start of the last item, | |
1606 | * which is the stop of the leaf data stack | |
1607 | */ | |
123abc88 | 1608 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1609 | struct extent_buffer *leaf) |
be0e5c09 | 1610 | { |
5f39d397 | 1611 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1612 | if (nr == 0) |
123abc88 | 1613 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1614 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1615 | } |
1616 | ||
aa5d6bed | 1617 | |
74123bd7 | 1618 | /* |
5f39d397 CM |
1619 | * search for key in the extent_buffer. The items start at offset p, |
1620 | * and they are item_size apart. There are 'max' items in p. | |
1621 | * | |
74123bd7 CM |
1622 | * the slot in the array is returned via slot, and it points to |
1623 | * the place where you would insert key if it is not found in | |
1624 | * the array. | |
1625 | * | |
1626 | * slot may point to max if the key is bigger than all of the keys | |
1627 | */ | |
e02119d5 CM |
1628 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1629 | unsigned long p, | |
1630 | int item_size, struct btrfs_key *key, | |
1631 | int max, int *slot) | |
be0e5c09 CM |
1632 | { |
1633 | int low = 0; | |
1634 | int high = max; | |
1635 | int mid; | |
1636 | int ret; | |
479965d6 | 1637 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1638 | struct btrfs_disk_key unaligned; |
1639 | unsigned long offset; | |
5f39d397 CM |
1640 | char *kaddr = NULL; |
1641 | unsigned long map_start = 0; | |
1642 | unsigned long map_len = 0; | |
479965d6 | 1643 | int err; |
be0e5c09 | 1644 | |
d397712b | 1645 | while (low < high) { |
be0e5c09 | 1646 | mid = (low + high) / 2; |
5f39d397 CM |
1647 | offset = p + mid * item_size; |
1648 | ||
a6591715 | 1649 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1650 | (offset + sizeof(struct btrfs_disk_key)) > |
1651 | map_start + map_len) { | |
934d375b CM |
1652 | |
1653 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1654 | sizeof(struct btrfs_disk_key), |
a6591715 | 1655 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1656 | |
1657 | if (!err) { | |
1658 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1659 | map_start); | |
1660 | } else { | |
1661 | read_extent_buffer(eb, &unaligned, | |
1662 | offset, sizeof(unaligned)); | |
1663 | tmp = &unaligned; | |
1664 | } | |
5f39d397 | 1665 | |
5f39d397 CM |
1666 | } else { |
1667 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1668 | map_start); | |
1669 | } | |
be0e5c09 CM |
1670 | ret = comp_keys(tmp, key); |
1671 | ||
1672 | if (ret < 0) | |
1673 | low = mid + 1; | |
1674 | else if (ret > 0) | |
1675 | high = mid; | |
1676 | else { | |
1677 | *slot = mid; | |
1678 | return 0; | |
1679 | } | |
1680 | } | |
1681 | *slot = low; | |
1682 | return 1; | |
1683 | } | |
1684 | ||
97571fd0 CM |
1685 | /* |
1686 | * simple bin_search frontend that does the right thing for | |
1687 | * leaves vs nodes | |
1688 | */ | |
5f39d397 CM |
1689 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1690 | int level, int *slot) | |
be0e5c09 | 1691 | { |
f775738f | 1692 | if (level == 0) |
5f39d397 CM |
1693 | return generic_bin_search(eb, |
1694 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1695 | sizeof(struct btrfs_item), |
5f39d397 | 1696 | key, btrfs_header_nritems(eb), |
7518a238 | 1697 | slot); |
f775738f | 1698 | else |
5f39d397 CM |
1699 | return generic_bin_search(eb, |
1700 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1701 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1702 | key, btrfs_header_nritems(eb), |
7518a238 | 1703 | slot); |
be0e5c09 CM |
1704 | } |
1705 | ||
5d4f98a2 YZ |
1706 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1707 | int level, int *slot) | |
1708 | { | |
1709 | return bin_search(eb, key, level, slot); | |
1710 | } | |
1711 | ||
f0486c68 YZ |
1712 | static void root_add_used(struct btrfs_root *root, u32 size) |
1713 | { | |
1714 | spin_lock(&root->accounting_lock); | |
1715 | btrfs_set_root_used(&root->root_item, | |
1716 | btrfs_root_used(&root->root_item) + size); | |
1717 | spin_unlock(&root->accounting_lock); | |
1718 | } | |
1719 | ||
1720 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1721 | { | |
1722 | spin_lock(&root->accounting_lock); | |
1723 | btrfs_set_root_used(&root->root_item, | |
1724 | btrfs_root_used(&root->root_item) - size); | |
1725 | spin_unlock(&root->accounting_lock); | |
1726 | } | |
1727 | ||
d352ac68 CM |
1728 | /* given a node and slot number, this reads the blocks it points to. The |
1729 | * extent buffer is returned with a reference taken (but unlocked). | |
1730 | * NULL is returned on error. | |
1731 | */ | |
e02119d5 | 1732 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1733 | struct extent_buffer *parent, int slot) |
bb803951 | 1734 | { |
ca7a79ad | 1735 | int level = btrfs_header_level(parent); |
416bc658 JB |
1736 | struct extent_buffer *eb; |
1737 | ||
bb803951 CM |
1738 | if (slot < 0) |
1739 | return NULL; | |
5f39d397 | 1740 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1741 | return NULL; |
ca7a79ad CM |
1742 | |
1743 | BUG_ON(level == 0); | |
1744 | ||
416bc658 JB |
1745 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1746 | btrfs_level_size(root, level - 1), | |
1747 | btrfs_node_ptr_generation(parent, slot)); | |
1748 | if (eb && !extent_buffer_uptodate(eb)) { | |
1749 | free_extent_buffer(eb); | |
1750 | eb = NULL; | |
1751 | } | |
1752 | ||
1753 | return eb; | |
bb803951 CM |
1754 | } |
1755 | ||
d352ac68 CM |
1756 | /* |
1757 | * node level balancing, used to make sure nodes are in proper order for | |
1758 | * item deletion. We balance from the top down, so we have to make sure | |
1759 | * that a deletion won't leave an node completely empty later on. | |
1760 | */ | |
e02119d5 | 1761 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1762 | struct btrfs_root *root, |
1763 | struct btrfs_path *path, int level) | |
bb803951 | 1764 | { |
5f39d397 CM |
1765 | struct extent_buffer *right = NULL; |
1766 | struct extent_buffer *mid; | |
1767 | struct extent_buffer *left = NULL; | |
1768 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1769 | int ret = 0; |
1770 | int wret; | |
1771 | int pslot; | |
bb803951 | 1772 | int orig_slot = path->slots[level]; |
79f95c82 | 1773 | u64 orig_ptr; |
bb803951 CM |
1774 | |
1775 | if (level == 0) | |
1776 | return 0; | |
1777 | ||
5f39d397 | 1778 | mid = path->nodes[level]; |
b4ce94de | 1779 | |
bd681513 CM |
1780 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1781 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1782 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1783 | ||
1d4f8a0c | 1784 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1785 | |
a05a9bb1 | 1786 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1787 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1788 | pslot = path->slots[level + 1]; |
1789 | } | |
bb803951 | 1790 | |
40689478 CM |
1791 | /* |
1792 | * deal with the case where there is only one pointer in the root | |
1793 | * by promoting the node below to a root | |
1794 | */ | |
5f39d397 CM |
1795 | if (!parent) { |
1796 | struct extent_buffer *child; | |
bb803951 | 1797 | |
5f39d397 | 1798 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1799 | return 0; |
1800 | ||
1801 | /* promote the child to a root */ | |
5f39d397 | 1802 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1803 | if (!child) { |
1804 | ret = -EROFS; | |
1805 | btrfs_std_error(root->fs_info, ret); | |
1806 | goto enospc; | |
1807 | } | |
1808 | ||
925baedd | 1809 | btrfs_tree_lock(child); |
b4ce94de | 1810 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1811 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1812 | if (ret) { |
1813 | btrfs_tree_unlock(child); | |
1814 | free_extent_buffer(child); | |
1815 | goto enospc; | |
1816 | } | |
2f375ab9 | 1817 | |
90f8d62e | 1818 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1819 | rcu_assign_pointer(root->node, child); |
925baedd | 1820 | |
0b86a832 | 1821 | add_root_to_dirty_list(root); |
925baedd | 1822 | btrfs_tree_unlock(child); |
b4ce94de | 1823 | |
925baedd | 1824 | path->locks[level] = 0; |
bb803951 | 1825 | path->nodes[level] = NULL; |
5f39d397 | 1826 | clean_tree_block(trans, root, mid); |
925baedd | 1827 | btrfs_tree_unlock(mid); |
bb803951 | 1828 | /* once for the path */ |
5f39d397 | 1829 | free_extent_buffer(mid); |
f0486c68 YZ |
1830 | |
1831 | root_sub_used(root, mid->len); | |
5581a51a | 1832 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1833 | /* once for the root ptr */ |
3083ee2e | 1834 | free_extent_buffer_stale(mid); |
f0486c68 | 1835 | return 0; |
bb803951 | 1836 | } |
5f39d397 | 1837 | if (btrfs_header_nritems(mid) > |
123abc88 | 1838 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1839 | return 0; |
1840 | ||
5f39d397 CM |
1841 | left = read_node_slot(root, parent, pslot - 1); |
1842 | if (left) { | |
925baedd | 1843 | btrfs_tree_lock(left); |
b4ce94de | 1844 | btrfs_set_lock_blocking(left); |
5f39d397 | 1845 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1846 | parent, pslot - 1, &left); |
54aa1f4d CM |
1847 | if (wret) { |
1848 | ret = wret; | |
1849 | goto enospc; | |
1850 | } | |
2cc58cf2 | 1851 | } |
5f39d397 CM |
1852 | right = read_node_slot(root, parent, pslot + 1); |
1853 | if (right) { | |
925baedd | 1854 | btrfs_tree_lock(right); |
b4ce94de | 1855 | btrfs_set_lock_blocking(right); |
5f39d397 | 1856 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1857 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1858 | if (wret) { |
1859 | ret = wret; | |
1860 | goto enospc; | |
1861 | } | |
1862 | } | |
1863 | ||
1864 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1865 | if (left) { |
1866 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1867 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1868 | if (wret < 0) |
1869 | ret = wret; | |
bb803951 | 1870 | } |
79f95c82 CM |
1871 | |
1872 | /* | |
1873 | * then try to empty the right most buffer into the middle | |
1874 | */ | |
5f39d397 | 1875 | if (right) { |
971a1f66 | 1876 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1877 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1878 | ret = wret; |
5f39d397 | 1879 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1880 | clean_tree_block(trans, root, right); |
925baedd | 1881 | btrfs_tree_unlock(right); |
afe5fea7 | 1882 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1883 | root_sub_used(root, right->len); |
5581a51a | 1884 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1885 | free_extent_buffer_stale(right); |
f0486c68 | 1886 | right = NULL; |
bb803951 | 1887 | } else { |
5f39d397 CM |
1888 | struct btrfs_disk_key right_key; |
1889 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 1890 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1891 | pslot + 1, 0); |
5f39d397 CM |
1892 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1893 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1894 | } |
1895 | } | |
5f39d397 | 1896 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1897 | /* |
1898 | * we're not allowed to leave a node with one item in the | |
1899 | * tree during a delete. A deletion from lower in the tree | |
1900 | * could try to delete the only pointer in this node. | |
1901 | * So, pull some keys from the left. | |
1902 | * There has to be a left pointer at this point because | |
1903 | * otherwise we would have pulled some pointers from the | |
1904 | * right | |
1905 | */ | |
305a26af MF |
1906 | if (!left) { |
1907 | ret = -EROFS; | |
1908 | btrfs_std_error(root->fs_info, ret); | |
1909 | goto enospc; | |
1910 | } | |
5f39d397 | 1911 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1912 | if (wret < 0) { |
79f95c82 | 1913 | ret = wret; |
54aa1f4d CM |
1914 | goto enospc; |
1915 | } | |
bce4eae9 CM |
1916 | if (wret == 1) { |
1917 | wret = push_node_left(trans, root, left, mid, 1); | |
1918 | if (wret < 0) | |
1919 | ret = wret; | |
1920 | } | |
79f95c82 CM |
1921 | BUG_ON(wret == 1); |
1922 | } | |
5f39d397 | 1923 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1924 | clean_tree_block(trans, root, mid); |
925baedd | 1925 | btrfs_tree_unlock(mid); |
afe5fea7 | 1926 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1927 | root_sub_used(root, mid->len); |
5581a51a | 1928 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1929 | free_extent_buffer_stale(mid); |
f0486c68 | 1930 | mid = NULL; |
79f95c82 CM |
1931 | } else { |
1932 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1933 | struct btrfs_disk_key mid_key; |
1934 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 1935 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 1936 | pslot, 0); |
5f39d397 CM |
1937 | btrfs_set_node_key(parent, &mid_key, pslot); |
1938 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1939 | } |
bb803951 | 1940 | |
79f95c82 | 1941 | /* update the path */ |
5f39d397 CM |
1942 | if (left) { |
1943 | if (btrfs_header_nritems(left) > orig_slot) { | |
1944 | extent_buffer_get(left); | |
925baedd | 1945 | /* left was locked after cow */ |
5f39d397 | 1946 | path->nodes[level] = left; |
bb803951 CM |
1947 | path->slots[level + 1] -= 1; |
1948 | path->slots[level] = orig_slot; | |
925baedd CM |
1949 | if (mid) { |
1950 | btrfs_tree_unlock(mid); | |
5f39d397 | 1951 | free_extent_buffer(mid); |
925baedd | 1952 | } |
bb803951 | 1953 | } else { |
5f39d397 | 1954 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1955 | path->slots[level] = orig_slot; |
1956 | } | |
1957 | } | |
79f95c82 | 1958 | /* double check we haven't messed things up */ |
e20d96d6 | 1959 | if (orig_ptr != |
5f39d397 | 1960 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1961 | BUG(); |
54aa1f4d | 1962 | enospc: |
925baedd CM |
1963 | if (right) { |
1964 | btrfs_tree_unlock(right); | |
5f39d397 | 1965 | free_extent_buffer(right); |
925baedd CM |
1966 | } |
1967 | if (left) { | |
1968 | if (path->nodes[level] != left) | |
1969 | btrfs_tree_unlock(left); | |
5f39d397 | 1970 | free_extent_buffer(left); |
925baedd | 1971 | } |
bb803951 CM |
1972 | return ret; |
1973 | } | |
1974 | ||
d352ac68 CM |
1975 | /* Node balancing for insertion. Here we only split or push nodes around |
1976 | * when they are completely full. This is also done top down, so we | |
1977 | * have to be pessimistic. | |
1978 | */ | |
d397712b | 1979 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1980 | struct btrfs_root *root, |
1981 | struct btrfs_path *path, int level) | |
e66f709b | 1982 | { |
5f39d397 CM |
1983 | struct extent_buffer *right = NULL; |
1984 | struct extent_buffer *mid; | |
1985 | struct extent_buffer *left = NULL; | |
1986 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1987 | int ret = 0; |
1988 | int wret; | |
1989 | int pslot; | |
1990 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1991 | |
1992 | if (level == 0) | |
1993 | return 1; | |
1994 | ||
5f39d397 | 1995 | mid = path->nodes[level]; |
7bb86316 | 1996 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1997 | |
a05a9bb1 | 1998 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1999 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2000 | pslot = path->slots[level + 1]; |
2001 | } | |
e66f709b | 2002 | |
5f39d397 | 2003 | if (!parent) |
e66f709b | 2004 | return 1; |
e66f709b | 2005 | |
5f39d397 | 2006 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
2007 | |
2008 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2009 | if (left) { |
e66f709b | 2010 | u32 left_nr; |
925baedd CM |
2011 | |
2012 | btrfs_tree_lock(left); | |
b4ce94de CM |
2013 | btrfs_set_lock_blocking(left); |
2014 | ||
5f39d397 | 2015 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
2016 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2017 | wret = 1; | |
2018 | } else { | |
5f39d397 | 2019 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2020 | pslot - 1, &left); |
54aa1f4d CM |
2021 | if (ret) |
2022 | wret = 1; | |
2023 | else { | |
54aa1f4d | 2024 | wret = push_node_left(trans, root, |
971a1f66 | 2025 | left, mid, 0); |
54aa1f4d | 2026 | } |
33ade1f8 | 2027 | } |
e66f709b CM |
2028 | if (wret < 0) |
2029 | ret = wret; | |
2030 | if (wret == 0) { | |
5f39d397 | 2031 | struct btrfs_disk_key disk_key; |
e66f709b | 2032 | orig_slot += left_nr; |
5f39d397 | 2033 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 2034 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2035 | pslot, 0); |
5f39d397 CM |
2036 | btrfs_set_node_key(parent, &disk_key, pslot); |
2037 | btrfs_mark_buffer_dirty(parent); | |
2038 | if (btrfs_header_nritems(left) > orig_slot) { | |
2039 | path->nodes[level] = left; | |
e66f709b CM |
2040 | path->slots[level + 1] -= 1; |
2041 | path->slots[level] = orig_slot; | |
925baedd | 2042 | btrfs_tree_unlock(mid); |
5f39d397 | 2043 | free_extent_buffer(mid); |
e66f709b CM |
2044 | } else { |
2045 | orig_slot -= | |
5f39d397 | 2046 | btrfs_header_nritems(left); |
e66f709b | 2047 | path->slots[level] = orig_slot; |
925baedd | 2048 | btrfs_tree_unlock(left); |
5f39d397 | 2049 | free_extent_buffer(left); |
e66f709b | 2050 | } |
e66f709b CM |
2051 | return 0; |
2052 | } | |
925baedd | 2053 | btrfs_tree_unlock(left); |
5f39d397 | 2054 | free_extent_buffer(left); |
e66f709b | 2055 | } |
925baedd | 2056 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2057 | |
2058 | /* | |
2059 | * then try to empty the right most buffer into the middle | |
2060 | */ | |
5f39d397 | 2061 | if (right) { |
33ade1f8 | 2062 | u32 right_nr; |
b4ce94de | 2063 | |
925baedd | 2064 | btrfs_tree_lock(right); |
b4ce94de CM |
2065 | btrfs_set_lock_blocking(right); |
2066 | ||
5f39d397 | 2067 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2068 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2069 | wret = 1; | |
2070 | } else { | |
5f39d397 CM |
2071 | ret = btrfs_cow_block(trans, root, right, |
2072 | parent, pslot + 1, | |
9fa8cfe7 | 2073 | &right); |
54aa1f4d CM |
2074 | if (ret) |
2075 | wret = 1; | |
2076 | else { | |
54aa1f4d | 2077 | wret = balance_node_right(trans, root, |
5f39d397 | 2078 | right, mid); |
54aa1f4d | 2079 | } |
33ade1f8 | 2080 | } |
e66f709b CM |
2081 | if (wret < 0) |
2082 | ret = wret; | |
2083 | if (wret == 0) { | |
5f39d397 CM |
2084 | struct btrfs_disk_key disk_key; |
2085 | ||
2086 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2087 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2088 | pslot + 1, 0); |
5f39d397 CM |
2089 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2090 | btrfs_mark_buffer_dirty(parent); | |
2091 | ||
2092 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2093 | path->nodes[level] = right; | |
e66f709b CM |
2094 | path->slots[level + 1] += 1; |
2095 | path->slots[level] = orig_slot - | |
5f39d397 | 2096 | btrfs_header_nritems(mid); |
925baedd | 2097 | btrfs_tree_unlock(mid); |
5f39d397 | 2098 | free_extent_buffer(mid); |
e66f709b | 2099 | } else { |
925baedd | 2100 | btrfs_tree_unlock(right); |
5f39d397 | 2101 | free_extent_buffer(right); |
e66f709b | 2102 | } |
e66f709b CM |
2103 | return 0; |
2104 | } | |
925baedd | 2105 | btrfs_tree_unlock(right); |
5f39d397 | 2106 | free_extent_buffer(right); |
e66f709b | 2107 | } |
e66f709b CM |
2108 | return 1; |
2109 | } | |
2110 | ||
3c69faec | 2111 | /* |
d352ac68 CM |
2112 | * readahead one full node of leaves, finding things that are close |
2113 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2114 | */ |
c8c42864 CM |
2115 | static void reada_for_search(struct btrfs_root *root, |
2116 | struct btrfs_path *path, | |
2117 | int level, int slot, u64 objectid) | |
3c69faec | 2118 | { |
5f39d397 | 2119 | struct extent_buffer *node; |
01f46658 | 2120 | struct btrfs_disk_key disk_key; |
3c69faec | 2121 | u32 nritems; |
3c69faec | 2122 | u64 search; |
a7175319 | 2123 | u64 target; |
6b80053d | 2124 | u64 nread = 0; |
cb25c2ea | 2125 | u64 gen; |
3c69faec | 2126 | int direction = path->reada; |
5f39d397 | 2127 | struct extent_buffer *eb; |
6b80053d CM |
2128 | u32 nr; |
2129 | u32 blocksize; | |
2130 | u32 nscan = 0; | |
db94535d | 2131 | |
a6b6e75e | 2132 | if (level != 1) |
6702ed49 CM |
2133 | return; |
2134 | ||
2135 | if (!path->nodes[level]) | |
3c69faec CM |
2136 | return; |
2137 | ||
5f39d397 | 2138 | node = path->nodes[level]; |
925baedd | 2139 | |
3c69faec | 2140 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2141 | blocksize = btrfs_level_size(root, level - 1); |
2142 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2143 | if (eb) { |
2144 | free_extent_buffer(eb); | |
3c69faec CM |
2145 | return; |
2146 | } | |
2147 | ||
a7175319 | 2148 | target = search; |
6b80053d | 2149 | |
5f39d397 | 2150 | nritems = btrfs_header_nritems(node); |
6b80053d | 2151 | nr = slot; |
25b8b936 | 2152 | |
d397712b | 2153 | while (1) { |
6b80053d CM |
2154 | if (direction < 0) { |
2155 | if (nr == 0) | |
2156 | break; | |
2157 | nr--; | |
2158 | } else if (direction > 0) { | |
2159 | nr++; | |
2160 | if (nr >= nritems) | |
2161 | break; | |
3c69faec | 2162 | } |
01f46658 CM |
2163 | if (path->reada < 0 && objectid) { |
2164 | btrfs_node_key(node, &disk_key, nr); | |
2165 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2166 | break; | |
2167 | } | |
6b80053d | 2168 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2169 | if ((search <= target && target - search <= 65536) || |
2170 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2171 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2172 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2173 | nread += blocksize; |
2174 | } | |
2175 | nscan++; | |
a7175319 | 2176 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2177 | break; |
3c69faec CM |
2178 | } |
2179 | } | |
925baedd | 2180 | |
b4ce94de CM |
2181 | /* |
2182 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
2183 | * cache | |
2184 | */ | |
2185 | static noinline int reada_for_balance(struct btrfs_root *root, | |
2186 | struct btrfs_path *path, int level) | |
2187 | { | |
2188 | int slot; | |
2189 | int nritems; | |
2190 | struct extent_buffer *parent; | |
2191 | struct extent_buffer *eb; | |
2192 | u64 gen; | |
2193 | u64 block1 = 0; | |
2194 | u64 block2 = 0; | |
2195 | int ret = 0; | |
2196 | int blocksize; | |
2197 | ||
8c594ea8 | 2198 | parent = path->nodes[level + 1]; |
b4ce94de CM |
2199 | if (!parent) |
2200 | return 0; | |
2201 | ||
2202 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2203 | slot = path->slots[level + 1]; |
b4ce94de CM |
2204 | blocksize = btrfs_level_size(root, level); |
2205 | ||
2206 | if (slot > 0) { | |
2207 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2208 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2209 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2210 | /* |
2211 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2212 | * don't want to return eagain here. That will loop | |
2213 | * forever | |
2214 | */ | |
2215 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2216 | block1 = 0; |
2217 | free_extent_buffer(eb); | |
2218 | } | |
8c594ea8 | 2219 | if (slot + 1 < nritems) { |
b4ce94de CM |
2220 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2221 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2222 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2223 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2224 | block2 = 0; |
2225 | free_extent_buffer(eb); | |
2226 | } | |
2227 | if (block1 || block2) { | |
2228 | ret = -EAGAIN; | |
8c594ea8 CM |
2229 | |
2230 | /* release the whole path */ | |
b3b4aa74 | 2231 | btrfs_release_path(path); |
8c594ea8 CM |
2232 | |
2233 | /* read the blocks */ | |
b4ce94de CM |
2234 | if (block1) |
2235 | readahead_tree_block(root, block1, blocksize, 0); | |
2236 | if (block2) | |
2237 | readahead_tree_block(root, block2, blocksize, 0); | |
2238 | ||
2239 | if (block1) { | |
2240 | eb = read_tree_block(root, block1, blocksize, 0); | |
2241 | free_extent_buffer(eb); | |
2242 | } | |
8c594ea8 | 2243 | if (block2) { |
b4ce94de CM |
2244 | eb = read_tree_block(root, block2, blocksize, 0); |
2245 | free_extent_buffer(eb); | |
2246 | } | |
2247 | } | |
2248 | return ret; | |
2249 | } | |
2250 | ||
2251 | ||
d352ac68 | 2252 | /* |
d397712b CM |
2253 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2254 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2255 | * operations on the tree might require changing key pointers higher up in the | |
2256 | * tree. | |
d352ac68 | 2257 | * |
d397712b CM |
2258 | * callers might also have set path->keep_locks, which tells this code to keep |
2259 | * the lock if the path points to the last slot in the block. This is part of | |
2260 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2261 | * |
d397712b CM |
2262 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2263 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2264 | */ |
e02119d5 | 2265 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2266 | int lowest_unlock, int min_write_lock_level, |
2267 | int *write_lock_level) | |
925baedd CM |
2268 | { |
2269 | int i; | |
2270 | int skip_level = level; | |
051e1b9f | 2271 | int no_skips = 0; |
925baedd CM |
2272 | struct extent_buffer *t; |
2273 | ||
2274 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2275 | if (!path->nodes[i]) | |
2276 | break; | |
2277 | if (!path->locks[i]) | |
2278 | break; | |
051e1b9f | 2279 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2280 | skip_level = i + 1; |
2281 | continue; | |
2282 | } | |
051e1b9f | 2283 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2284 | u32 nritems; |
2285 | t = path->nodes[i]; | |
2286 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2287 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2288 | skip_level = i + 1; |
2289 | continue; | |
2290 | } | |
2291 | } | |
051e1b9f CM |
2292 | if (skip_level < i && i >= lowest_unlock) |
2293 | no_skips = 1; | |
2294 | ||
925baedd CM |
2295 | t = path->nodes[i]; |
2296 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2297 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2298 | path->locks[i] = 0; |
f7c79f30 CM |
2299 | if (write_lock_level && |
2300 | i > min_write_lock_level && | |
2301 | i <= *write_lock_level) { | |
2302 | *write_lock_level = i - 1; | |
2303 | } | |
925baedd CM |
2304 | } |
2305 | } | |
2306 | } | |
2307 | ||
b4ce94de CM |
2308 | /* |
2309 | * This releases any locks held in the path starting at level and | |
2310 | * going all the way up to the root. | |
2311 | * | |
2312 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2313 | * corner cases, such as COW of the block at slot zero in the node. This | |
2314 | * ignores those rules, and it should only be called when there are no | |
2315 | * more updates to be done higher up in the tree. | |
2316 | */ | |
2317 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2318 | { | |
2319 | int i; | |
2320 | ||
09a2a8f9 | 2321 | if (path->keep_locks) |
b4ce94de CM |
2322 | return; |
2323 | ||
2324 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2325 | if (!path->nodes[i]) | |
12f4dacc | 2326 | continue; |
b4ce94de | 2327 | if (!path->locks[i]) |
12f4dacc | 2328 | continue; |
bd681513 | 2329 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2330 | path->locks[i] = 0; |
2331 | } | |
2332 | } | |
2333 | ||
c8c42864 CM |
2334 | /* |
2335 | * helper function for btrfs_search_slot. The goal is to find a block | |
2336 | * in cache without setting the path to blocking. If we find the block | |
2337 | * we return zero and the path is unchanged. | |
2338 | * | |
2339 | * If we can't find the block, we set the path blocking and do some | |
2340 | * reada. -EAGAIN is returned and the search must be repeated. | |
2341 | */ | |
2342 | static int | |
2343 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2344 | struct btrfs_root *root, struct btrfs_path *p, | |
2345 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2346 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2347 | { |
2348 | u64 blocknr; | |
2349 | u64 gen; | |
2350 | u32 blocksize; | |
2351 | struct extent_buffer *b = *eb_ret; | |
2352 | struct extent_buffer *tmp; | |
76a05b35 | 2353 | int ret; |
c8c42864 CM |
2354 | |
2355 | blocknr = btrfs_node_blockptr(b, slot); | |
2356 | gen = btrfs_node_ptr_generation(b, slot); | |
2357 | blocksize = btrfs_level_size(root, level - 1); | |
2358 | ||
2359 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2360 | if (tmp) { |
b9fab919 | 2361 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2362 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2363 | *eb_ret = tmp; | |
2364 | return 0; | |
2365 | } | |
2366 | ||
2367 | /* the pages were up to date, but we failed | |
2368 | * the generation number check. Do a full | |
2369 | * read for the generation number that is correct. | |
2370 | * We must do this without dropping locks so | |
2371 | * we can trust our generation number | |
2372 | */ | |
2373 | btrfs_set_path_blocking(p); | |
2374 | ||
2375 | /* now we're allowed to do a blocking uptodate check */ | |
2376 | ret = btrfs_read_buffer(tmp, gen); | |
2377 | if (!ret) { | |
2378 | *eb_ret = tmp; | |
2379 | return 0; | |
cb44921a | 2380 | } |
bdf7c00e JB |
2381 | free_extent_buffer(tmp); |
2382 | btrfs_release_path(p); | |
2383 | return -EIO; | |
c8c42864 CM |
2384 | } |
2385 | ||
2386 | /* | |
2387 | * reduce lock contention at high levels | |
2388 | * of the btree by dropping locks before | |
76a05b35 CM |
2389 | * we read. Don't release the lock on the current |
2390 | * level because we need to walk this node to figure | |
2391 | * out which blocks to read. | |
c8c42864 | 2392 | */ |
8c594ea8 CM |
2393 | btrfs_unlock_up_safe(p, level + 1); |
2394 | btrfs_set_path_blocking(p); | |
2395 | ||
cb44921a | 2396 | free_extent_buffer(tmp); |
c8c42864 CM |
2397 | if (p->reada) |
2398 | reada_for_search(root, p, level, slot, key->objectid); | |
2399 | ||
b3b4aa74 | 2400 | btrfs_release_path(p); |
76a05b35 CM |
2401 | |
2402 | ret = -EAGAIN; | |
5bdd3536 | 2403 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2404 | if (tmp) { |
2405 | /* | |
2406 | * If the read above didn't mark this buffer up to date, | |
2407 | * it will never end up being up to date. Set ret to EIO now | |
2408 | * and give up so that our caller doesn't loop forever | |
2409 | * on our EAGAINs. | |
2410 | */ | |
b9fab919 | 2411 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2412 | ret = -EIO; |
c8c42864 | 2413 | free_extent_buffer(tmp); |
76a05b35 CM |
2414 | } |
2415 | return ret; | |
c8c42864 CM |
2416 | } |
2417 | ||
2418 | /* | |
2419 | * helper function for btrfs_search_slot. This does all of the checks | |
2420 | * for node-level blocks and does any balancing required based on | |
2421 | * the ins_len. | |
2422 | * | |
2423 | * If no extra work was required, zero is returned. If we had to | |
2424 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2425 | * start over | |
2426 | */ | |
2427 | static int | |
2428 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2429 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2430 | struct extent_buffer *b, int level, int ins_len, |
2431 | int *write_lock_level) | |
c8c42864 CM |
2432 | { |
2433 | int ret; | |
2434 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2435 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2436 | int sret; | |
2437 | ||
bd681513 CM |
2438 | if (*write_lock_level < level + 1) { |
2439 | *write_lock_level = level + 1; | |
2440 | btrfs_release_path(p); | |
2441 | goto again; | |
2442 | } | |
2443 | ||
c8c42864 CM |
2444 | sret = reada_for_balance(root, p, level); |
2445 | if (sret) | |
2446 | goto again; | |
2447 | ||
2448 | btrfs_set_path_blocking(p); | |
2449 | sret = split_node(trans, root, p, level); | |
bd681513 | 2450 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2451 | |
2452 | BUG_ON(sret > 0); | |
2453 | if (sret) { | |
2454 | ret = sret; | |
2455 | goto done; | |
2456 | } | |
2457 | b = p->nodes[level]; | |
2458 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2459 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2460 | int sret; |
2461 | ||
bd681513 CM |
2462 | if (*write_lock_level < level + 1) { |
2463 | *write_lock_level = level + 1; | |
2464 | btrfs_release_path(p); | |
2465 | goto again; | |
2466 | } | |
2467 | ||
c8c42864 CM |
2468 | sret = reada_for_balance(root, p, level); |
2469 | if (sret) | |
2470 | goto again; | |
2471 | ||
2472 | btrfs_set_path_blocking(p); | |
2473 | sret = balance_level(trans, root, p, level); | |
bd681513 | 2474 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2475 | |
2476 | if (sret) { | |
2477 | ret = sret; | |
2478 | goto done; | |
2479 | } | |
2480 | b = p->nodes[level]; | |
2481 | if (!b) { | |
b3b4aa74 | 2482 | btrfs_release_path(p); |
c8c42864 CM |
2483 | goto again; |
2484 | } | |
2485 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2486 | } | |
2487 | return 0; | |
2488 | ||
2489 | again: | |
2490 | ret = -EAGAIN; | |
2491 | done: | |
2492 | return ret; | |
2493 | } | |
2494 | ||
74123bd7 CM |
2495 | /* |
2496 | * look for key in the tree. path is filled in with nodes along the way | |
2497 | * if key is found, we return zero and you can find the item in the leaf | |
2498 | * level of the path (level 0) | |
2499 | * | |
2500 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2501 | * be inserted, and 1 is returned. If there are other errors during the |
2502 | * search a negative error number is returned. | |
97571fd0 CM |
2503 | * |
2504 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2505 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2506 | * possible) | |
74123bd7 | 2507 | */ |
e089f05c CM |
2508 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2509 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2510 | ins_len, int cow) | |
be0e5c09 | 2511 | { |
5f39d397 | 2512 | struct extent_buffer *b; |
be0e5c09 CM |
2513 | int slot; |
2514 | int ret; | |
33c66f43 | 2515 | int err; |
be0e5c09 | 2516 | int level; |
925baedd | 2517 | int lowest_unlock = 1; |
bd681513 CM |
2518 | int root_lock; |
2519 | /* everything at write_lock_level or lower must be write locked */ | |
2520 | int write_lock_level = 0; | |
9f3a7427 | 2521 | u8 lowest_level = 0; |
f7c79f30 | 2522 | int min_write_lock_level; |
9f3a7427 | 2523 | |
6702ed49 | 2524 | lowest_level = p->lowest_level; |
323ac95b | 2525 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2526 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2527 | |
bd681513 | 2528 | if (ins_len < 0) { |
925baedd | 2529 | lowest_unlock = 2; |
65b51a00 | 2530 | |
bd681513 CM |
2531 | /* when we are removing items, we might have to go up to level |
2532 | * two as we update tree pointers Make sure we keep write | |
2533 | * for those levels as well | |
2534 | */ | |
2535 | write_lock_level = 2; | |
2536 | } else if (ins_len > 0) { | |
2537 | /* | |
2538 | * for inserting items, make sure we have a write lock on | |
2539 | * level 1 so we can update keys | |
2540 | */ | |
2541 | write_lock_level = 1; | |
2542 | } | |
2543 | ||
2544 | if (!cow) | |
2545 | write_lock_level = -1; | |
2546 | ||
09a2a8f9 | 2547 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2548 | write_lock_level = BTRFS_MAX_LEVEL; |
2549 | ||
f7c79f30 CM |
2550 | min_write_lock_level = write_lock_level; |
2551 | ||
bb803951 | 2552 | again: |
bd681513 CM |
2553 | /* |
2554 | * we try very hard to do read locks on the root | |
2555 | */ | |
2556 | root_lock = BTRFS_READ_LOCK; | |
2557 | level = 0; | |
5d4f98a2 | 2558 | if (p->search_commit_root) { |
bd681513 CM |
2559 | /* |
2560 | * the commit roots are read only | |
2561 | * so we always do read locks | |
2562 | */ | |
5d4f98a2 YZ |
2563 | b = root->commit_root; |
2564 | extent_buffer_get(b); | |
bd681513 | 2565 | level = btrfs_header_level(b); |
5d4f98a2 | 2566 | if (!p->skip_locking) |
bd681513 | 2567 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2568 | } else { |
bd681513 | 2569 | if (p->skip_locking) { |
5d4f98a2 | 2570 | b = btrfs_root_node(root); |
bd681513 CM |
2571 | level = btrfs_header_level(b); |
2572 | } else { | |
2573 | /* we don't know the level of the root node | |
2574 | * until we actually have it read locked | |
2575 | */ | |
2576 | b = btrfs_read_lock_root_node(root); | |
2577 | level = btrfs_header_level(b); | |
2578 | if (level <= write_lock_level) { | |
2579 | /* whoops, must trade for write lock */ | |
2580 | btrfs_tree_read_unlock(b); | |
2581 | free_extent_buffer(b); | |
2582 | b = btrfs_lock_root_node(root); | |
2583 | root_lock = BTRFS_WRITE_LOCK; | |
2584 | ||
2585 | /* the level might have changed, check again */ | |
2586 | level = btrfs_header_level(b); | |
2587 | } | |
2588 | } | |
5d4f98a2 | 2589 | } |
bd681513 CM |
2590 | p->nodes[level] = b; |
2591 | if (!p->skip_locking) | |
2592 | p->locks[level] = root_lock; | |
925baedd | 2593 | |
eb60ceac | 2594 | while (b) { |
5f39d397 | 2595 | level = btrfs_header_level(b); |
65b51a00 CM |
2596 | |
2597 | /* | |
2598 | * setup the path here so we can release it under lock | |
2599 | * contention with the cow code | |
2600 | */ | |
02217ed2 | 2601 | if (cow) { |
c8c42864 CM |
2602 | /* |
2603 | * if we don't really need to cow this block | |
2604 | * then we don't want to set the path blocking, | |
2605 | * so we test it here | |
2606 | */ | |
5d4f98a2 | 2607 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2608 | goto cow_done; |
5d4f98a2 | 2609 | |
b4ce94de CM |
2610 | btrfs_set_path_blocking(p); |
2611 | ||
bd681513 CM |
2612 | /* |
2613 | * must have write locks on this node and the | |
2614 | * parent | |
2615 | */ | |
5124e00e JB |
2616 | if (level > write_lock_level || |
2617 | (level + 1 > write_lock_level && | |
2618 | level + 1 < BTRFS_MAX_LEVEL && | |
2619 | p->nodes[level + 1])) { | |
bd681513 CM |
2620 | write_lock_level = level + 1; |
2621 | btrfs_release_path(p); | |
2622 | goto again; | |
2623 | } | |
2624 | ||
33c66f43 YZ |
2625 | err = btrfs_cow_block(trans, root, b, |
2626 | p->nodes[level + 1], | |
2627 | p->slots[level + 1], &b); | |
2628 | if (err) { | |
33c66f43 | 2629 | ret = err; |
65b51a00 | 2630 | goto done; |
54aa1f4d | 2631 | } |
02217ed2 | 2632 | } |
65b51a00 | 2633 | cow_done: |
02217ed2 | 2634 | BUG_ON(!cow && ins_len); |
65b51a00 | 2635 | |
eb60ceac | 2636 | p->nodes[level] = b; |
bd681513 | 2637 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2638 | |
2639 | /* | |
2640 | * we have a lock on b and as long as we aren't changing | |
2641 | * the tree, there is no way to for the items in b to change. | |
2642 | * It is safe to drop the lock on our parent before we | |
2643 | * go through the expensive btree search on b. | |
2644 | * | |
2645 | * If cow is true, then we might be changing slot zero, | |
2646 | * which may require changing the parent. So, we can't | |
2647 | * drop the lock until after we know which slot we're | |
2648 | * operating on. | |
2649 | */ | |
2650 | if (!cow) | |
2651 | btrfs_unlock_up_safe(p, level + 1); | |
2652 | ||
5f39d397 | 2653 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2654 | |
5f39d397 | 2655 | if (level != 0) { |
33c66f43 YZ |
2656 | int dec = 0; |
2657 | if (ret && slot > 0) { | |
2658 | dec = 1; | |
be0e5c09 | 2659 | slot -= 1; |
33c66f43 | 2660 | } |
be0e5c09 | 2661 | p->slots[level] = slot; |
33c66f43 | 2662 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2663 | ins_len, &write_lock_level); |
33c66f43 | 2664 | if (err == -EAGAIN) |
c8c42864 | 2665 | goto again; |
33c66f43 YZ |
2666 | if (err) { |
2667 | ret = err; | |
c8c42864 | 2668 | goto done; |
33c66f43 | 2669 | } |
c8c42864 CM |
2670 | b = p->nodes[level]; |
2671 | slot = p->slots[level]; | |
b4ce94de | 2672 | |
bd681513 CM |
2673 | /* |
2674 | * slot 0 is special, if we change the key | |
2675 | * we have to update the parent pointer | |
2676 | * which means we must have a write lock | |
2677 | * on the parent | |
2678 | */ | |
2679 | if (slot == 0 && cow && | |
2680 | write_lock_level < level + 1) { | |
2681 | write_lock_level = level + 1; | |
2682 | btrfs_release_path(p); | |
2683 | goto again; | |
2684 | } | |
2685 | ||
f7c79f30 CM |
2686 | unlock_up(p, level, lowest_unlock, |
2687 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2688 | |
925baedd | 2689 | if (level == lowest_level) { |
33c66f43 YZ |
2690 | if (dec) |
2691 | p->slots[level]++; | |
5b21f2ed | 2692 | goto done; |
925baedd | 2693 | } |
ca7a79ad | 2694 | |
33c66f43 | 2695 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2696 | &b, level, slot, key, 0); |
33c66f43 | 2697 | if (err == -EAGAIN) |
c8c42864 | 2698 | goto again; |
33c66f43 YZ |
2699 | if (err) { |
2700 | ret = err; | |
76a05b35 | 2701 | goto done; |
33c66f43 | 2702 | } |
76a05b35 | 2703 | |
b4ce94de | 2704 | if (!p->skip_locking) { |
bd681513 CM |
2705 | level = btrfs_header_level(b); |
2706 | if (level <= write_lock_level) { | |
2707 | err = btrfs_try_tree_write_lock(b); | |
2708 | if (!err) { | |
2709 | btrfs_set_path_blocking(p); | |
2710 | btrfs_tree_lock(b); | |
2711 | btrfs_clear_path_blocking(p, b, | |
2712 | BTRFS_WRITE_LOCK); | |
2713 | } | |
2714 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2715 | } else { | |
2716 | err = btrfs_try_tree_read_lock(b); | |
2717 | if (!err) { | |
2718 | btrfs_set_path_blocking(p); | |
2719 | btrfs_tree_read_lock(b); | |
2720 | btrfs_clear_path_blocking(p, b, | |
2721 | BTRFS_READ_LOCK); | |
2722 | } | |
2723 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2724 | } |
bd681513 | 2725 | p->nodes[level] = b; |
b4ce94de | 2726 | } |
be0e5c09 CM |
2727 | } else { |
2728 | p->slots[level] = slot; | |
87b29b20 YZ |
2729 | if (ins_len > 0 && |
2730 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2731 | if (write_lock_level < 1) { |
2732 | write_lock_level = 1; | |
2733 | btrfs_release_path(p); | |
2734 | goto again; | |
2735 | } | |
2736 | ||
b4ce94de | 2737 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2738 | err = split_leaf(trans, root, key, |
2739 | p, ins_len, ret == 0); | |
bd681513 | 2740 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2741 | |
33c66f43 YZ |
2742 | BUG_ON(err > 0); |
2743 | if (err) { | |
2744 | ret = err; | |
65b51a00 CM |
2745 | goto done; |
2746 | } | |
5c680ed6 | 2747 | } |
459931ec | 2748 | if (!p->search_for_split) |
f7c79f30 CM |
2749 | unlock_up(p, level, lowest_unlock, |
2750 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2751 | goto done; |
be0e5c09 CM |
2752 | } |
2753 | } | |
65b51a00 CM |
2754 | ret = 1; |
2755 | done: | |
b4ce94de CM |
2756 | /* |
2757 | * we don't really know what they plan on doing with the path | |
2758 | * from here on, so for now just mark it as blocking | |
2759 | */ | |
b9473439 CM |
2760 | if (!p->leave_spinning) |
2761 | btrfs_set_path_blocking(p); | |
76a05b35 | 2762 | if (ret < 0) |
b3b4aa74 | 2763 | btrfs_release_path(p); |
65b51a00 | 2764 | return ret; |
be0e5c09 CM |
2765 | } |
2766 | ||
5d9e75c4 JS |
2767 | /* |
2768 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2769 | * current state of the tree together with the operations recorded in the tree | |
2770 | * modification log to search for the key in a previous version of this tree, as | |
2771 | * denoted by the time_seq parameter. | |
2772 | * | |
2773 | * Naturally, there is no support for insert, delete or cow operations. | |
2774 | * | |
2775 | * The resulting path and return value will be set up as if we called | |
2776 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2777 | */ | |
2778 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2779 | struct btrfs_path *p, u64 time_seq) | |
2780 | { | |
2781 | struct extent_buffer *b; | |
2782 | int slot; | |
2783 | int ret; | |
2784 | int err; | |
2785 | int level; | |
2786 | int lowest_unlock = 1; | |
2787 | u8 lowest_level = 0; | |
2788 | ||
2789 | lowest_level = p->lowest_level; | |
2790 | WARN_ON(p->nodes[0] != NULL); | |
2791 | ||
2792 | if (p->search_commit_root) { | |
2793 | BUG_ON(time_seq); | |
2794 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2795 | } | |
2796 | ||
2797 | again: | |
5d9e75c4 | 2798 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2799 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2800 | p->locks[level] = BTRFS_READ_LOCK; |
2801 | ||
2802 | while (b) { | |
2803 | level = btrfs_header_level(b); | |
2804 | p->nodes[level] = b; | |
2805 | btrfs_clear_path_blocking(p, NULL, 0); | |
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 | */ | |
2813 | btrfs_unlock_up_safe(p, level + 1); | |
2814 | ||
2815 | ret = bin_search(b, key, level, &slot); | |
2816 | ||
2817 | if (level != 0) { | |
2818 | int dec = 0; | |
2819 | if (ret && slot > 0) { | |
2820 | dec = 1; | |
2821 | slot -= 1; | |
2822 | } | |
2823 | p->slots[level] = slot; | |
2824 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2825 | ||
2826 | if (level == lowest_level) { | |
2827 | if (dec) | |
2828 | p->slots[level]++; | |
2829 | goto done; | |
2830 | } | |
2831 | ||
2832 | err = read_block_for_search(NULL, root, p, &b, level, | |
2833 | slot, key, time_seq); | |
2834 | if (err == -EAGAIN) | |
2835 | goto again; | |
2836 | if (err) { | |
2837 | ret = err; | |
2838 | goto done; | |
2839 | } | |
2840 | ||
2841 | level = btrfs_header_level(b); | |
2842 | err = btrfs_try_tree_read_lock(b); | |
2843 | if (!err) { | |
2844 | btrfs_set_path_blocking(p); | |
2845 | btrfs_tree_read_lock(b); | |
2846 | btrfs_clear_path_blocking(p, b, | |
2847 | BTRFS_READ_LOCK); | |
2848 | } | |
47fb091f | 2849 | b = tree_mod_log_rewind(root->fs_info, b, time_seq); |
5d9e75c4 JS |
2850 | p->locks[level] = BTRFS_READ_LOCK; |
2851 | p->nodes[level] = b; | |
5d9e75c4 JS |
2852 | } else { |
2853 | p->slots[level] = slot; | |
2854 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2855 | goto done; | |
2856 | } | |
2857 | } | |
2858 | ret = 1; | |
2859 | done: | |
2860 | if (!p->leave_spinning) | |
2861 | btrfs_set_path_blocking(p); | |
2862 | if (ret < 0) | |
2863 | btrfs_release_path(p); | |
2864 | ||
2865 | return ret; | |
2866 | } | |
2867 | ||
2f38b3e1 AJ |
2868 | /* |
2869 | * helper to use instead of search slot if no exact match is needed but | |
2870 | * instead the next or previous item should be returned. | |
2871 | * When find_higher is true, the next higher item is returned, the next lower | |
2872 | * otherwise. | |
2873 | * When return_any and find_higher are both true, and no higher item is found, | |
2874 | * return the next lower instead. | |
2875 | * When return_any is true and find_higher is false, and no lower item is found, | |
2876 | * return the next higher instead. | |
2877 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2878 | * < 0 on error | |
2879 | */ | |
2880 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
2881 | struct btrfs_key *key, struct btrfs_path *p, | |
2882 | int find_higher, int return_any) | |
2883 | { | |
2884 | int ret; | |
2885 | struct extent_buffer *leaf; | |
2886 | ||
2887 | again: | |
2888 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2889 | if (ret <= 0) | |
2890 | return ret; | |
2891 | /* | |
2892 | * a return value of 1 means the path is at the position where the | |
2893 | * item should be inserted. Normally this is the next bigger item, | |
2894 | * but in case the previous item is the last in a leaf, path points | |
2895 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2896 | * item. | |
2897 | */ | |
2898 | leaf = p->nodes[0]; | |
2899 | ||
2900 | if (find_higher) { | |
2901 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2902 | ret = btrfs_next_leaf(root, p); | |
2903 | if (ret <= 0) | |
2904 | return ret; | |
2905 | if (!return_any) | |
2906 | return 1; | |
2907 | /* | |
2908 | * no higher item found, return the next | |
2909 | * lower instead | |
2910 | */ | |
2911 | return_any = 0; | |
2912 | find_higher = 0; | |
2913 | btrfs_release_path(p); | |
2914 | goto again; | |
2915 | } | |
2916 | } else { | |
e6793769 AJ |
2917 | if (p->slots[0] == 0) { |
2918 | ret = btrfs_prev_leaf(root, p); | |
2919 | if (ret < 0) | |
2920 | return ret; | |
2921 | if (!ret) { | |
2922 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
2923 | return 0; | |
2f38b3e1 | 2924 | } |
e6793769 AJ |
2925 | if (!return_any) |
2926 | return 1; | |
2927 | /* | |
2928 | * no lower item found, return the next | |
2929 | * higher instead | |
2930 | */ | |
2931 | return_any = 0; | |
2932 | find_higher = 1; | |
2933 | btrfs_release_path(p); | |
2934 | goto again; | |
2935 | } else { | |
2f38b3e1 AJ |
2936 | --p->slots[0]; |
2937 | } | |
2938 | } | |
2939 | return 0; | |
2940 | } | |
2941 | ||
74123bd7 CM |
2942 | /* |
2943 | * adjust the pointers going up the tree, starting at level | |
2944 | * making sure the right key of each node is points to 'key'. | |
2945 | * This is used after shifting pointers to the left, so it stops | |
2946 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2947 | * higher levels | |
aa5d6bed | 2948 | * |
74123bd7 | 2949 | */ |
d6a0a126 | 2950 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2951 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2952 | { |
2953 | int i; | |
5f39d397 CM |
2954 | struct extent_buffer *t; |
2955 | ||
234b63a0 | 2956 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2957 | int tslot = path->slots[i]; |
eb60ceac | 2958 | if (!path->nodes[i]) |
be0e5c09 | 2959 | break; |
5f39d397 | 2960 | t = path->nodes[i]; |
32adf090 | 2961 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 2962 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2963 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2964 | if (tslot != 0) |
2965 | break; | |
2966 | } | |
2967 | } | |
2968 | ||
31840ae1 ZY |
2969 | /* |
2970 | * update item key. | |
2971 | * | |
2972 | * This function isn't completely safe. It's the caller's responsibility | |
2973 | * that the new key won't break the order | |
2974 | */ | |
afe5fea7 | 2975 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2976 | struct btrfs_key *new_key) |
31840ae1 ZY |
2977 | { |
2978 | struct btrfs_disk_key disk_key; | |
2979 | struct extent_buffer *eb; | |
2980 | int slot; | |
2981 | ||
2982 | eb = path->nodes[0]; | |
2983 | slot = path->slots[0]; | |
2984 | if (slot > 0) { | |
2985 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2986 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2987 | } |
2988 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2989 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2990 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2991 | } |
2992 | ||
2993 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2994 | btrfs_set_item_key(eb, &disk_key, slot); | |
2995 | btrfs_mark_buffer_dirty(eb); | |
2996 | if (slot == 0) | |
d6a0a126 | 2997 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
2998 | } |
2999 | ||
74123bd7 CM |
3000 | /* |
3001 | * try to push data from one node into the next node left in the | |
79f95c82 | 3002 | * tree. |
aa5d6bed CM |
3003 | * |
3004 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3005 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3006 | */ |
98ed5174 CM |
3007 | static int push_node_left(struct btrfs_trans_handle *trans, |
3008 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3009 | struct extent_buffer *src, int empty) |
be0e5c09 | 3010 | { |
be0e5c09 | 3011 | int push_items = 0; |
bb803951 CM |
3012 | int src_nritems; |
3013 | int dst_nritems; | |
aa5d6bed | 3014 | int ret = 0; |
be0e5c09 | 3015 | |
5f39d397 CM |
3016 | src_nritems = btrfs_header_nritems(src); |
3017 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3018 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
3019 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3020 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3021 | |
bce4eae9 | 3022 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3023 | return 1; |
3024 | ||
d397712b | 3025 | if (push_items <= 0) |
be0e5c09 CM |
3026 | return 1; |
3027 | ||
bce4eae9 | 3028 | if (empty) { |
971a1f66 | 3029 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3030 | if (push_items < src_nritems) { |
3031 | /* leave at least 8 pointers in the node if | |
3032 | * we aren't going to empty it | |
3033 | */ | |
3034 | if (src_nritems - push_items < 8) { | |
3035 | if (push_items <= 8) | |
3036 | return 1; | |
3037 | push_items -= 8; | |
3038 | } | |
3039 | } | |
3040 | } else | |
3041 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3042 | |
f230475e | 3043 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
90f8d62e | 3044 | push_items); |
5f39d397 CM |
3045 | copy_extent_buffer(dst, src, |
3046 | btrfs_node_key_ptr_offset(dst_nritems), | |
3047 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3048 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3049 | |
bb803951 | 3050 | if (push_items < src_nritems) { |
57911b8b JS |
3051 | /* |
3052 | * don't call tree_mod_log_eb_move here, key removal was already | |
3053 | * fully logged by tree_mod_log_eb_copy above. | |
3054 | */ | |
5f39d397 CM |
3055 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3056 | btrfs_node_key_ptr_offset(push_items), | |
3057 | (src_nritems - push_items) * | |
3058 | sizeof(struct btrfs_key_ptr)); | |
3059 | } | |
3060 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3061 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3062 | btrfs_mark_buffer_dirty(src); | |
3063 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3064 | |
79f95c82 CM |
3065 | return ret; |
3066 | } | |
3067 | ||
3068 | /* | |
3069 | * try to push data from one node into the next node right in the | |
3070 | * tree. | |
3071 | * | |
3072 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3073 | * error, and > 0 if there was no room in the right hand block. | |
3074 | * | |
3075 | * this will only push up to 1/2 the contents of the left node over | |
3076 | */ | |
5f39d397 CM |
3077 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3078 | struct btrfs_root *root, | |
3079 | struct extent_buffer *dst, | |
3080 | struct extent_buffer *src) | |
79f95c82 | 3081 | { |
79f95c82 CM |
3082 | int push_items = 0; |
3083 | int max_push; | |
3084 | int src_nritems; | |
3085 | int dst_nritems; | |
3086 | int ret = 0; | |
79f95c82 | 3087 | |
7bb86316 CM |
3088 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3089 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3090 | ||
5f39d397 CM |
3091 | src_nritems = btrfs_header_nritems(src); |
3092 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3093 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3094 | if (push_items <= 0) |
79f95c82 | 3095 | return 1; |
bce4eae9 | 3096 | |
d397712b | 3097 | if (src_nritems < 4) |
bce4eae9 | 3098 | return 1; |
79f95c82 CM |
3099 | |
3100 | max_push = src_nritems / 2 + 1; | |
3101 | /* don't try to empty the node */ | |
d397712b | 3102 | if (max_push >= src_nritems) |
79f95c82 | 3103 | return 1; |
252c38f0 | 3104 | |
79f95c82 CM |
3105 | if (max_push < push_items) |
3106 | push_items = max_push; | |
3107 | ||
f230475e | 3108 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3109 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3110 | btrfs_node_key_ptr_offset(0), | |
3111 | (dst_nritems) * | |
3112 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3113 | |
f230475e | 3114 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
90f8d62e | 3115 | src_nritems - push_items, push_items); |
5f39d397 CM |
3116 | copy_extent_buffer(dst, src, |
3117 | btrfs_node_key_ptr_offset(0), | |
3118 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3119 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3120 | |
5f39d397 CM |
3121 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3122 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3123 | |
5f39d397 CM |
3124 | btrfs_mark_buffer_dirty(src); |
3125 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3126 | |
aa5d6bed | 3127 | return ret; |
be0e5c09 CM |
3128 | } |
3129 | ||
97571fd0 CM |
3130 | /* |
3131 | * helper function to insert a new root level in the tree. | |
3132 | * A new node is allocated, and a single item is inserted to | |
3133 | * point to the existing root | |
aa5d6bed CM |
3134 | * |
3135 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3136 | */ |
d397712b | 3137 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3138 | struct btrfs_root *root, |
fdd99c72 | 3139 | struct btrfs_path *path, int level) |
5c680ed6 | 3140 | { |
7bb86316 | 3141 | u64 lower_gen; |
5f39d397 CM |
3142 | struct extent_buffer *lower; |
3143 | struct extent_buffer *c; | |
925baedd | 3144 | struct extent_buffer *old; |
5f39d397 | 3145 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3146 | |
3147 | BUG_ON(path->nodes[level]); | |
3148 | BUG_ON(path->nodes[level-1] != root->node); | |
3149 | ||
7bb86316 CM |
3150 | lower = path->nodes[level-1]; |
3151 | if (level == 1) | |
3152 | btrfs_item_key(lower, &lower_key, 0); | |
3153 | else | |
3154 | btrfs_node_key(lower, &lower_key, 0); | |
3155 | ||
31840ae1 | 3156 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3157 | root->root_key.objectid, &lower_key, |
5581a51a | 3158 | level, root->node->start, 0); |
5f39d397 CM |
3159 | if (IS_ERR(c)) |
3160 | return PTR_ERR(c); | |
925baedd | 3161 | |
f0486c68 YZ |
3162 | root_add_used(root, root->nodesize); |
3163 | ||
5d4f98a2 | 3164 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3165 | btrfs_set_header_nritems(c, 1); |
3166 | btrfs_set_header_level(c, level); | |
db94535d | 3167 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3168 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3169 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3170 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
3171 | |
3172 | write_extent_buffer(c, root->fs_info->fsid, | |
3173 | (unsigned long)btrfs_header_fsid(c), | |
3174 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3175 | |
3176 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
3177 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
3178 | BTRFS_UUID_SIZE); | |
3179 | ||
5f39d397 | 3180 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3181 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3182 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3183 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3184 | |
3185 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3186 | |
5f39d397 | 3187 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3188 | |
925baedd | 3189 | old = root->node; |
fdd99c72 | 3190 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3191 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3192 | |
3193 | /* the super has an extra ref to root->node */ | |
3194 | free_extent_buffer(old); | |
3195 | ||
0b86a832 | 3196 | add_root_to_dirty_list(root); |
5f39d397 CM |
3197 | extent_buffer_get(c); |
3198 | path->nodes[level] = c; | |
bd681513 | 3199 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3200 | path->slots[level] = 0; |
3201 | return 0; | |
3202 | } | |
3203 | ||
74123bd7 CM |
3204 | /* |
3205 | * worker function to insert a single pointer in a node. | |
3206 | * the node should have enough room for the pointer already | |
97571fd0 | 3207 | * |
74123bd7 CM |
3208 | * slot and level indicate where you want the key to go, and |
3209 | * blocknr is the block the key points to. | |
3210 | */ | |
143bede5 JM |
3211 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3212 | struct btrfs_root *root, struct btrfs_path *path, | |
3213 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3214 | int slot, int level) |
74123bd7 | 3215 | { |
5f39d397 | 3216 | struct extent_buffer *lower; |
74123bd7 | 3217 | int nritems; |
f3ea38da | 3218 | int ret; |
5c680ed6 CM |
3219 | |
3220 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3221 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3222 | lower = path->nodes[level]; |
3223 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3224 | BUG_ON(slot > nritems); |
143bede5 | 3225 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3226 | if (slot != nritems) { |
c3e06965 | 3227 | if (level) |
f3ea38da JS |
3228 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3229 | slot, nritems - slot); | |
5f39d397 CM |
3230 | memmove_extent_buffer(lower, |
3231 | btrfs_node_key_ptr_offset(slot + 1), | |
3232 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3233 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3234 | } |
c3e06965 | 3235 | if (level) { |
f3ea38da JS |
3236 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
3237 | MOD_LOG_KEY_ADD); | |
3238 | BUG_ON(ret < 0); | |
3239 | } | |
5f39d397 | 3240 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3241 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3242 | WARN_ON(trans->transid == 0); |
3243 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3244 | btrfs_set_header_nritems(lower, nritems + 1); |
3245 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3246 | } |
3247 | ||
97571fd0 CM |
3248 | /* |
3249 | * split the node at the specified level in path in two. | |
3250 | * The path is corrected to point to the appropriate node after the split | |
3251 | * | |
3252 | * Before splitting this tries to make some room in the node by pushing | |
3253 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3254 | * |
3255 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3256 | */ |
e02119d5 CM |
3257 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3258 | struct btrfs_root *root, | |
3259 | struct btrfs_path *path, int level) | |
be0e5c09 | 3260 | { |
5f39d397 CM |
3261 | struct extent_buffer *c; |
3262 | struct extent_buffer *split; | |
3263 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3264 | int mid; |
5c680ed6 | 3265 | int ret; |
7518a238 | 3266 | u32 c_nritems; |
eb60ceac | 3267 | |
5f39d397 | 3268 | c = path->nodes[level]; |
7bb86316 | 3269 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3270 | if (c == root->node) { |
d9abbf1c | 3271 | /* |
90f8d62e JS |
3272 | * trying to split the root, lets make a new one |
3273 | * | |
fdd99c72 | 3274 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3275 | * insert_new_root, because that root buffer will be kept as a |
3276 | * normal node. We are going to log removal of half of the | |
3277 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3278 | * tree lock on the buffer, which is why we cannot race with | |
3279 | * other tree_mod_log users. | |
d9abbf1c | 3280 | */ |
fdd99c72 | 3281 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3282 | if (ret) |
3283 | return ret; | |
b3612421 | 3284 | } else { |
e66f709b | 3285 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3286 | c = path->nodes[level]; |
3287 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3288 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3289 | return 0; |
54aa1f4d CM |
3290 | if (ret < 0) |
3291 | return ret; | |
be0e5c09 | 3292 | } |
e66f709b | 3293 | |
5f39d397 | 3294 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3295 | mid = (c_nritems + 1) / 2; |
3296 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3297 | |
5d4f98a2 | 3298 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3299 | root->root_key.objectid, |
5581a51a | 3300 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3301 | if (IS_ERR(split)) |
3302 | return PTR_ERR(split); | |
3303 | ||
f0486c68 YZ |
3304 | root_add_used(root, root->nodesize); |
3305 | ||
5d4f98a2 | 3306 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3307 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3308 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3309 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3310 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3311 | btrfs_set_header_owner(split, root->root_key.objectid); |
3312 | write_extent_buffer(split, root->fs_info->fsid, | |
3313 | (unsigned long)btrfs_header_fsid(split), | |
3314 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3315 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
3316 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
3317 | BTRFS_UUID_SIZE); | |
54aa1f4d | 3318 | |
90f8d62e | 3319 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
3320 | copy_extent_buffer(split, c, |
3321 | btrfs_node_key_ptr_offset(0), | |
3322 | btrfs_node_key_ptr_offset(mid), | |
3323 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3324 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3325 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3326 | ret = 0; |
3327 | ||
5f39d397 CM |
3328 | btrfs_mark_buffer_dirty(c); |
3329 | btrfs_mark_buffer_dirty(split); | |
3330 | ||
143bede5 | 3331 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3332 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3333 | |
5de08d7d | 3334 | if (path->slots[level] >= mid) { |
5c680ed6 | 3335 | path->slots[level] -= mid; |
925baedd | 3336 | btrfs_tree_unlock(c); |
5f39d397 CM |
3337 | free_extent_buffer(c); |
3338 | path->nodes[level] = split; | |
5c680ed6 CM |
3339 | path->slots[level + 1] += 1; |
3340 | } else { | |
925baedd | 3341 | btrfs_tree_unlock(split); |
5f39d397 | 3342 | free_extent_buffer(split); |
be0e5c09 | 3343 | } |
aa5d6bed | 3344 | return ret; |
be0e5c09 CM |
3345 | } |
3346 | ||
74123bd7 CM |
3347 | /* |
3348 | * how many bytes are required to store the items in a leaf. start | |
3349 | * and nr indicate which items in the leaf to check. This totals up the | |
3350 | * space used both by the item structs and the item data | |
3351 | */ | |
5f39d397 | 3352 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3353 | { |
41be1f3b JB |
3354 | struct btrfs_item *start_item; |
3355 | struct btrfs_item *end_item; | |
3356 | struct btrfs_map_token token; | |
be0e5c09 | 3357 | int data_len; |
5f39d397 | 3358 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3359 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3360 | |
3361 | if (!nr) | |
3362 | return 0; | |
41be1f3b JB |
3363 | btrfs_init_map_token(&token); |
3364 | start_item = btrfs_item_nr(l, start); | |
3365 | end_item = btrfs_item_nr(l, end); | |
3366 | data_len = btrfs_token_item_offset(l, start_item, &token) + | |
3367 | btrfs_token_item_size(l, start_item, &token); | |
3368 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3369 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3370 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3371 | return data_len; |
3372 | } | |
3373 | ||
d4dbff95 CM |
3374 | /* |
3375 | * The space between the end of the leaf items and | |
3376 | * the start of the leaf data. IOW, how much room | |
3377 | * the leaf has left for both items and data | |
3378 | */ | |
d397712b | 3379 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3380 | struct extent_buffer *leaf) |
d4dbff95 | 3381 | { |
5f39d397 CM |
3382 | int nritems = btrfs_header_nritems(leaf); |
3383 | int ret; | |
3384 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3385 | if (ret < 0) { | |
d397712b CM |
3386 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
3387 | "used %d nritems %d\n", | |
ae2f5411 | 3388 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3389 | leaf_space_used(leaf, 0, nritems), nritems); |
3390 | } | |
3391 | return ret; | |
d4dbff95 CM |
3392 | } |
3393 | ||
99d8f83c CM |
3394 | /* |
3395 | * min slot controls the lowest index we're willing to push to the | |
3396 | * right. We'll push up to and including min_slot, but no lower | |
3397 | */ | |
44871b1b CM |
3398 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3399 | struct btrfs_root *root, | |
3400 | struct btrfs_path *path, | |
3401 | int data_size, int empty, | |
3402 | struct extent_buffer *right, | |
99d8f83c CM |
3403 | int free_space, u32 left_nritems, |
3404 | u32 min_slot) | |
00ec4c51 | 3405 | { |
5f39d397 | 3406 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3407 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3408 | struct btrfs_map_token token; |
5f39d397 | 3409 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3410 | int slot; |
34a38218 | 3411 | u32 i; |
00ec4c51 CM |
3412 | int push_space = 0; |
3413 | int push_items = 0; | |
0783fcfc | 3414 | struct btrfs_item *item; |
34a38218 | 3415 | u32 nr; |
7518a238 | 3416 | u32 right_nritems; |
5f39d397 | 3417 | u32 data_end; |
db94535d | 3418 | u32 this_item_size; |
00ec4c51 | 3419 | |
cfed81a0 CM |
3420 | btrfs_init_map_token(&token); |
3421 | ||
34a38218 CM |
3422 | if (empty) |
3423 | nr = 0; | |
3424 | else | |
99d8f83c | 3425 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3426 | |
31840ae1 | 3427 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3428 | push_space += data_size; |
31840ae1 | 3429 | |
44871b1b | 3430 | slot = path->slots[1]; |
34a38218 CM |
3431 | i = left_nritems - 1; |
3432 | while (i >= nr) { | |
5f39d397 | 3433 | item = btrfs_item_nr(left, i); |
db94535d | 3434 | |
31840ae1 ZY |
3435 | if (!empty && push_items > 0) { |
3436 | if (path->slots[0] > i) | |
3437 | break; | |
3438 | if (path->slots[0] == i) { | |
3439 | int space = btrfs_leaf_free_space(root, left); | |
3440 | if (space + push_space * 2 > free_space) | |
3441 | break; | |
3442 | } | |
3443 | } | |
3444 | ||
00ec4c51 | 3445 | if (path->slots[0] == i) |
87b29b20 | 3446 | push_space += data_size; |
db94535d | 3447 | |
db94535d CM |
3448 | this_item_size = btrfs_item_size(left, item); |
3449 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3450 | break; |
31840ae1 | 3451 | |
00ec4c51 | 3452 | push_items++; |
db94535d | 3453 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3454 | if (i == 0) |
3455 | break; | |
3456 | i--; | |
db94535d | 3457 | } |
5f39d397 | 3458 | |
925baedd CM |
3459 | if (push_items == 0) |
3460 | goto out_unlock; | |
5f39d397 | 3461 | |
6c1500f2 | 3462 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3463 | |
00ec4c51 | 3464 | /* push left to right */ |
5f39d397 | 3465 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3466 | |
5f39d397 | 3467 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3468 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3469 | |
00ec4c51 | 3470 | /* make room in the right data area */ |
5f39d397 CM |
3471 | data_end = leaf_data_end(root, right); |
3472 | memmove_extent_buffer(right, | |
3473 | btrfs_leaf_data(right) + data_end - push_space, | |
3474 | btrfs_leaf_data(right) + data_end, | |
3475 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3476 | ||
00ec4c51 | 3477 | /* copy from the left data area */ |
5f39d397 | 3478 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3479 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3480 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3481 | push_space); | |
5f39d397 CM |
3482 | |
3483 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3484 | btrfs_item_nr_offset(0), | |
3485 | right_nritems * sizeof(struct btrfs_item)); | |
3486 | ||
00ec4c51 | 3487 | /* copy the items from left to right */ |
5f39d397 CM |
3488 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3489 | btrfs_item_nr_offset(left_nritems - push_items), | |
3490 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3491 | |
3492 | /* update the item pointers */ | |
7518a238 | 3493 | right_nritems += push_items; |
5f39d397 | 3494 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3495 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3496 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 3497 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
3498 | push_space -= btrfs_token_item_size(right, item, &token); |
3499 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3500 | } |
3501 | ||
7518a238 | 3502 | left_nritems -= push_items; |
5f39d397 | 3503 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3504 | |
34a38218 CM |
3505 | if (left_nritems) |
3506 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3507 | else |
3508 | clean_tree_block(trans, root, left); | |
3509 | ||
5f39d397 | 3510 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3511 | |
5f39d397 CM |
3512 | btrfs_item_key(right, &disk_key, 0); |
3513 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3514 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3515 | |
00ec4c51 | 3516 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3517 | if (path->slots[0] >= left_nritems) { |
3518 | path->slots[0] -= left_nritems; | |
925baedd CM |
3519 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3520 | clean_tree_block(trans, root, path->nodes[0]); | |
3521 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3522 | free_extent_buffer(path->nodes[0]); |
3523 | path->nodes[0] = right; | |
00ec4c51 CM |
3524 | path->slots[1] += 1; |
3525 | } else { | |
925baedd | 3526 | btrfs_tree_unlock(right); |
5f39d397 | 3527 | free_extent_buffer(right); |
00ec4c51 CM |
3528 | } |
3529 | return 0; | |
925baedd CM |
3530 | |
3531 | out_unlock: | |
3532 | btrfs_tree_unlock(right); | |
3533 | free_extent_buffer(right); | |
3534 | return 1; | |
00ec4c51 | 3535 | } |
925baedd | 3536 | |
44871b1b CM |
3537 | /* |
3538 | * push some data in the path leaf to the right, trying to free up at | |
3539 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3540 | * | |
3541 | * returns 1 if the push failed because the other node didn't have enough | |
3542 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3543 | * |
3544 | * this will push starting from min_slot to the end of the leaf. It won't | |
3545 | * push any slot lower than min_slot | |
44871b1b CM |
3546 | */ |
3547 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3548 | *root, struct btrfs_path *path, |
3549 | int min_data_size, int data_size, | |
3550 | int empty, u32 min_slot) | |
44871b1b CM |
3551 | { |
3552 | struct extent_buffer *left = path->nodes[0]; | |
3553 | struct extent_buffer *right; | |
3554 | struct extent_buffer *upper; | |
3555 | int slot; | |
3556 | int free_space; | |
3557 | u32 left_nritems; | |
3558 | int ret; | |
3559 | ||
3560 | if (!path->nodes[1]) | |
3561 | return 1; | |
3562 | ||
3563 | slot = path->slots[1]; | |
3564 | upper = path->nodes[1]; | |
3565 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3566 | return 1; | |
3567 | ||
3568 | btrfs_assert_tree_locked(path->nodes[1]); | |
3569 | ||
3570 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3571 | if (right == NULL) |
3572 | return 1; | |
3573 | ||
44871b1b CM |
3574 | btrfs_tree_lock(right); |
3575 | btrfs_set_lock_blocking(right); | |
3576 | ||
3577 | free_space = btrfs_leaf_free_space(root, right); | |
3578 | if (free_space < data_size) | |
3579 | goto out_unlock; | |
3580 | ||
3581 | /* cow and double check */ | |
3582 | ret = btrfs_cow_block(trans, root, right, upper, | |
3583 | slot + 1, &right); | |
3584 | if (ret) | |
3585 | goto out_unlock; | |
3586 | ||
3587 | free_space = btrfs_leaf_free_space(root, right); | |
3588 | if (free_space < data_size) | |
3589 | goto out_unlock; | |
3590 | ||
3591 | left_nritems = btrfs_header_nritems(left); | |
3592 | if (left_nritems == 0) | |
3593 | goto out_unlock; | |
3594 | ||
99d8f83c CM |
3595 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3596 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3597 | out_unlock: |
3598 | btrfs_tree_unlock(right); | |
3599 | free_extent_buffer(right); | |
3600 | return 1; | |
3601 | } | |
3602 | ||
74123bd7 CM |
3603 | /* |
3604 | * push some data in the path leaf to the left, trying to free up at | |
3605 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3606 | * |
3607 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3608 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3609 | * items | |
74123bd7 | 3610 | */ |
44871b1b CM |
3611 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3612 | struct btrfs_root *root, | |
3613 | struct btrfs_path *path, int data_size, | |
3614 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3615 | int free_space, u32 right_nritems, |
3616 | u32 max_slot) | |
be0e5c09 | 3617 | { |
5f39d397 CM |
3618 | struct btrfs_disk_key disk_key; |
3619 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3620 | int i; |
be0e5c09 CM |
3621 | int push_space = 0; |
3622 | int push_items = 0; | |
0783fcfc | 3623 | struct btrfs_item *item; |
7518a238 | 3624 | u32 old_left_nritems; |
34a38218 | 3625 | u32 nr; |
aa5d6bed | 3626 | int ret = 0; |
db94535d CM |
3627 | u32 this_item_size; |
3628 | u32 old_left_item_size; | |
cfed81a0 CM |
3629 | struct btrfs_map_token token; |
3630 | ||
3631 | btrfs_init_map_token(&token); | |
be0e5c09 | 3632 | |
34a38218 | 3633 | if (empty) |
99d8f83c | 3634 | nr = min(right_nritems, max_slot); |
34a38218 | 3635 | else |
99d8f83c | 3636 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3637 | |
3638 | for (i = 0; i < nr; i++) { | |
5f39d397 | 3639 | item = btrfs_item_nr(right, i); |
db94535d | 3640 | |
31840ae1 ZY |
3641 | if (!empty && push_items > 0) { |
3642 | if (path->slots[0] < i) | |
3643 | break; | |
3644 | if (path->slots[0] == i) { | |
3645 | int space = btrfs_leaf_free_space(root, right); | |
3646 | if (space + push_space * 2 > free_space) | |
3647 | break; | |
3648 | } | |
3649 | } | |
3650 | ||
be0e5c09 | 3651 | if (path->slots[0] == i) |
87b29b20 | 3652 | push_space += data_size; |
db94535d CM |
3653 | |
3654 | this_item_size = btrfs_item_size(right, item); | |
3655 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3656 | break; |
db94535d | 3657 | |
be0e5c09 | 3658 | push_items++; |
db94535d CM |
3659 | push_space += this_item_size + sizeof(*item); |
3660 | } | |
3661 | ||
be0e5c09 | 3662 | if (push_items == 0) { |
925baedd CM |
3663 | ret = 1; |
3664 | goto out; | |
be0e5c09 | 3665 | } |
34a38218 | 3666 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3667 | WARN_ON(1); |
5f39d397 | 3668 | |
be0e5c09 | 3669 | /* push data from right to left */ |
5f39d397 CM |
3670 | copy_extent_buffer(left, right, |
3671 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3672 | btrfs_item_nr_offset(0), | |
3673 | push_items * sizeof(struct btrfs_item)); | |
3674 | ||
123abc88 | 3675 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3676 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3677 | |
3678 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3679 | leaf_data_end(root, left) - push_space, |
3680 | btrfs_leaf_data(right) + | |
5f39d397 | 3681 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3682 | push_space); |
5f39d397 | 3683 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3684 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3685 | |
db94535d | 3686 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3687 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3688 | u32 ioff; |
db94535d | 3689 | |
5f39d397 | 3690 | item = btrfs_item_nr(left, i); |
db94535d | 3691 | |
cfed81a0 CM |
3692 | ioff = btrfs_token_item_offset(left, item, &token); |
3693 | btrfs_set_token_item_offset(left, item, | |
3694 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3695 | &token); | |
be0e5c09 | 3696 | } |
5f39d397 | 3697 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3698 | |
3699 | /* fixup right node */ | |
31b1a2bd JL |
3700 | if (push_items > right_nritems) |
3701 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3702 | right_nritems); |
34a38218 CM |
3703 | |
3704 | if (push_items < right_nritems) { | |
3705 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3706 | leaf_data_end(root, right); | |
3707 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3708 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3709 | btrfs_leaf_data(right) + | |
3710 | leaf_data_end(root, right), push_space); | |
3711 | ||
3712 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3713 | btrfs_item_nr_offset(push_items), |
3714 | (btrfs_header_nritems(right) - push_items) * | |
3715 | sizeof(struct btrfs_item)); | |
34a38218 | 3716 | } |
eef1c494 Y |
3717 | right_nritems -= push_items; |
3718 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3719 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3720 | for (i = 0; i < right_nritems; i++) { |
3721 | item = btrfs_item_nr(right, i); | |
db94535d | 3722 | |
cfed81a0 CM |
3723 | push_space = push_space - btrfs_token_item_size(right, |
3724 | item, &token); | |
3725 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3726 | } |
eb60ceac | 3727 | |
5f39d397 | 3728 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3729 | if (right_nritems) |
3730 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3731 | else |
3732 | clean_tree_block(trans, root, right); | |
098f59c2 | 3733 | |
5f39d397 | 3734 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3735 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3736 | |
3737 | /* then fixup the leaf pointer in the path */ | |
3738 | if (path->slots[0] < push_items) { | |
3739 | path->slots[0] += old_left_nritems; | |
925baedd | 3740 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3741 | free_extent_buffer(path->nodes[0]); |
3742 | path->nodes[0] = left; | |
be0e5c09 CM |
3743 | path->slots[1] -= 1; |
3744 | } else { | |
925baedd | 3745 | btrfs_tree_unlock(left); |
5f39d397 | 3746 | free_extent_buffer(left); |
be0e5c09 CM |
3747 | path->slots[0] -= push_items; |
3748 | } | |
eb60ceac | 3749 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3750 | return ret; |
925baedd CM |
3751 | out: |
3752 | btrfs_tree_unlock(left); | |
3753 | free_extent_buffer(left); | |
3754 | return ret; | |
be0e5c09 CM |
3755 | } |
3756 | ||
44871b1b CM |
3757 | /* |
3758 | * push some data in the path leaf to the left, trying to free up at | |
3759 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3760 | * |
3761 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3762 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3763 | * items | |
44871b1b CM |
3764 | */ |
3765 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3766 | *root, struct btrfs_path *path, int min_data_size, |
3767 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3768 | { |
3769 | struct extent_buffer *right = path->nodes[0]; | |
3770 | struct extent_buffer *left; | |
3771 | int slot; | |
3772 | int free_space; | |
3773 | u32 right_nritems; | |
3774 | int ret = 0; | |
3775 | ||
3776 | slot = path->slots[1]; | |
3777 | if (slot == 0) | |
3778 | return 1; | |
3779 | if (!path->nodes[1]) | |
3780 | return 1; | |
3781 | ||
3782 | right_nritems = btrfs_header_nritems(right); | |
3783 | if (right_nritems == 0) | |
3784 | return 1; | |
3785 | ||
3786 | btrfs_assert_tree_locked(path->nodes[1]); | |
3787 | ||
3788 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3789 | if (left == NULL) |
3790 | return 1; | |
3791 | ||
44871b1b CM |
3792 | btrfs_tree_lock(left); |
3793 | btrfs_set_lock_blocking(left); | |
3794 | ||
3795 | free_space = btrfs_leaf_free_space(root, left); | |
3796 | if (free_space < data_size) { | |
3797 | ret = 1; | |
3798 | goto out; | |
3799 | } | |
3800 | ||
3801 | /* cow and double check */ | |
3802 | ret = btrfs_cow_block(trans, root, left, | |
3803 | path->nodes[1], slot - 1, &left); | |
3804 | if (ret) { | |
3805 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3806 | if (ret == -ENOSPC) |
3807 | ret = 1; | |
44871b1b CM |
3808 | goto out; |
3809 | } | |
3810 | ||
3811 | free_space = btrfs_leaf_free_space(root, left); | |
3812 | if (free_space < data_size) { | |
3813 | ret = 1; | |
3814 | goto out; | |
3815 | } | |
3816 | ||
99d8f83c CM |
3817 | return __push_leaf_left(trans, root, path, min_data_size, |
3818 | empty, left, free_space, right_nritems, | |
3819 | max_slot); | |
44871b1b CM |
3820 | out: |
3821 | btrfs_tree_unlock(left); | |
3822 | free_extent_buffer(left); | |
3823 | return ret; | |
3824 | } | |
3825 | ||
3826 | /* | |
3827 | * split the path's leaf in two, making sure there is at least data_size | |
3828 | * available for the resulting leaf level of the path. | |
44871b1b | 3829 | */ |
143bede5 JM |
3830 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3831 | struct btrfs_root *root, | |
3832 | struct btrfs_path *path, | |
3833 | struct extent_buffer *l, | |
3834 | struct extent_buffer *right, | |
3835 | int slot, int mid, int nritems) | |
44871b1b CM |
3836 | { |
3837 | int data_copy_size; | |
3838 | int rt_data_off; | |
3839 | int i; | |
44871b1b | 3840 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3841 | struct btrfs_map_token token; |
3842 | ||
3843 | btrfs_init_map_token(&token); | |
44871b1b CM |
3844 | |
3845 | nritems = nritems - mid; | |
3846 | btrfs_set_header_nritems(right, nritems); | |
3847 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3848 | ||
3849 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3850 | btrfs_item_nr_offset(mid), | |
3851 | nritems * sizeof(struct btrfs_item)); | |
3852 | ||
3853 | copy_extent_buffer(right, l, | |
3854 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3855 | data_copy_size, btrfs_leaf_data(l) + | |
3856 | leaf_data_end(root, l), data_copy_size); | |
3857 | ||
3858 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3859 | btrfs_item_end_nr(l, mid); | |
3860 | ||
3861 | for (i = 0; i < nritems; i++) { | |
3862 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3863 | u32 ioff; | |
3864 | ||
cfed81a0 CM |
3865 | ioff = btrfs_token_item_offset(right, item, &token); |
3866 | btrfs_set_token_item_offset(right, item, | |
3867 | ioff + rt_data_off, &token); | |
44871b1b CM |
3868 | } |
3869 | ||
44871b1b | 3870 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3871 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3872 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 3873 | path->slots[1] + 1, 1); |
44871b1b CM |
3874 | |
3875 | btrfs_mark_buffer_dirty(right); | |
3876 | btrfs_mark_buffer_dirty(l); | |
3877 | BUG_ON(path->slots[0] != slot); | |
3878 | ||
44871b1b CM |
3879 | if (mid <= slot) { |
3880 | btrfs_tree_unlock(path->nodes[0]); | |
3881 | free_extent_buffer(path->nodes[0]); | |
3882 | path->nodes[0] = right; | |
3883 | path->slots[0] -= mid; | |
3884 | path->slots[1] += 1; | |
3885 | } else { | |
3886 | btrfs_tree_unlock(right); | |
3887 | free_extent_buffer(right); | |
3888 | } | |
3889 | ||
3890 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3891 | } |
3892 | ||
99d8f83c CM |
3893 | /* |
3894 | * double splits happen when we need to insert a big item in the middle | |
3895 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3896 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3897 | * A B C | |
3898 | * | |
3899 | * We avoid this by trying to push the items on either side of our target | |
3900 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3901 | * completely. | |
3902 | */ | |
3903 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3904 | struct btrfs_root *root, | |
3905 | struct btrfs_path *path, | |
3906 | int data_size) | |
3907 | { | |
3908 | int ret; | |
3909 | int progress = 0; | |
3910 | int slot; | |
3911 | u32 nritems; | |
3912 | ||
3913 | slot = path->slots[0]; | |
3914 | ||
3915 | /* | |
3916 | * try to push all the items after our slot into the | |
3917 | * right leaf | |
3918 | */ | |
3919 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3920 | if (ret < 0) | |
3921 | return ret; | |
3922 | ||
3923 | if (ret == 0) | |
3924 | progress++; | |
3925 | ||
3926 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3927 | /* | |
3928 | * our goal is to get our slot at the start or end of a leaf. If | |
3929 | * we've done so we're done | |
3930 | */ | |
3931 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3932 | return 0; | |
3933 | ||
3934 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3935 | return 0; | |
3936 | ||
3937 | /* try to push all the items before our slot into the next leaf */ | |
3938 | slot = path->slots[0]; | |
3939 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3940 | if (ret < 0) | |
3941 | return ret; | |
3942 | ||
3943 | if (ret == 0) | |
3944 | progress++; | |
3945 | ||
3946 | if (progress) | |
3947 | return 0; | |
3948 | return 1; | |
3949 | } | |
3950 | ||
74123bd7 CM |
3951 | /* |
3952 | * split the path's leaf in two, making sure there is at least data_size | |
3953 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3954 | * |
3955 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3956 | */ |
e02119d5 CM |
3957 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3958 | struct btrfs_root *root, | |
3959 | struct btrfs_key *ins_key, | |
3960 | struct btrfs_path *path, int data_size, | |
3961 | int extend) | |
be0e5c09 | 3962 | { |
5d4f98a2 | 3963 | struct btrfs_disk_key disk_key; |
5f39d397 | 3964 | struct extent_buffer *l; |
7518a238 | 3965 | u32 nritems; |
eb60ceac CM |
3966 | int mid; |
3967 | int slot; | |
5f39d397 | 3968 | struct extent_buffer *right; |
d4dbff95 | 3969 | int ret = 0; |
aa5d6bed | 3970 | int wret; |
5d4f98a2 | 3971 | int split; |
cc0c5538 | 3972 | int num_doubles = 0; |
99d8f83c | 3973 | int tried_avoid_double = 0; |
aa5d6bed | 3974 | |
a5719521 YZ |
3975 | l = path->nodes[0]; |
3976 | slot = path->slots[0]; | |
3977 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3978 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3979 | return -EOVERFLOW; | |
3980 | ||
40689478 | 3981 | /* first try to make some room by pushing left and right */ |
33157e05 | 3982 | if (data_size && path->nodes[1]) { |
99d8f83c CM |
3983 | wret = push_leaf_right(trans, root, path, data_size, |
3984 | data_size, 0, 0); | |
d397712b | 3985 | if (wret < 0) |
eaee50e8 | 3986 | return wret; |
3685f791 | 3987 | if (wret) { |
99d8f83c CM |
3988 | wret = push_leaf_left(trans, root, path, data_size, |
3989 | data_size, 0, (u32)-1); | |
3685f791 CM |
3990 | if (wret < 0) |
3991 | return wret; | |
3992 | } | |
3993 | l = path->nodes[0]; | |
aa5d6bed | 3994 | |
3685f791 | 3995 | /* did the pushes work? */ |
87b29b20 | 3996 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3997 | return 0; |
3326d1b0 | 3998 | } |
aa5d6bed | 3999 | |
5c680ed6 | 4000 | if (!path->nodes[1]) { |
fdd99c72 | 4001 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4002 | if (ret) |
4003 | return ret; | |
4004 | } | |
cc0c5538 | 4005 | again: |
5d4f98a2 | 4006 | split = 1; |
cc0c5538 | 4007 | l = path->nodes[0]; |
eb60ceac | 4008 | slot = path->slots[0]; |
5f39d397 | 4009 | nritems = btrfs_header_nritems(l); |
d397712b | 4010 | mid = (nritems + 1) / 2; |
54aa1f4d | 4011 | |
5d4f98a2 YZ |
4012 | if (mid <= slot) { |
4013 | if (nritems == 1 || | |
4014 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
4015 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4016 | if (slot >= nritems) { | |
4017 | split = 0; | |
4018 | } else { | |
4019 | mid = slot; | |
4020 | if (mid != nritems && | |
4021 | leaf_space_used(l, mid, nritems - mid) + | |
4022 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4023 | if (data_size && !tried_avoid_double) |
4024 | goto push_for_double; | |
5d4f98a2 YZ |
4025 | split = 2; |
4026 | } | |
4027 | } | |
4028 | } | |
4029 | } else { | |
4030 | if (leaf_space_used(l, 0, mid) + data_size > | |
4031 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4032 | if (!extend && data_size && slot == 0) { | |
4033 | split = 0; | |
4034 | } else if ((extend || !data_size) && slot == 0) { | |
4035 | mid = 1; | |
4036 | } else { | |
4037 | mid = slot; | |
4038 | if (mid != nritems && | |
4039 | leaf_space_used(l, mid, nritems - mid) + | |
4040 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4041 | if (data_size && !tried_avoid_double) |
4042 | goto push_for_double; | |
5d4f98a2 YZ |
4043 | split = 2 ; |
4044 | } | |
4045 | } | |
4046 | } | |
4047 | } | |
4048 | ||
4049 | if (split == 0) | |
4050 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4051 | else | |
4052 | btrfs_item_key(l, &disk_key, mid); | |
4053 | ||
4054 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 4055 | root->root_key.objectid, |
5581a51a | 4056 | &disk_key, 0, l->start, 0); |
f0486c68 | 4057 | if (IS_ERR(right)) |
5f39d397 | 4058 | return PTR_ERR(right); |
f0486c68 YZ |
4059 | |
4060 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4061 | |
4062 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4063 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4064 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4065 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4066 | btrfs_set_header_owner(right, root->root_key.objectid); |
4067 | btrfs_set_header_level(right, 0); | |
4068 | write_extent_buffer(right, root->fs_info->fsid, | |
4069 | (unsigned long)btrfs_header_fsid(right), | |
4070 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
4071 | |
4072 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
4073 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
4074 | BTRFS_UUID_SIZE); | |
44871b1b | 4075 | |
5d4f98a2 YZ |
4076 | if (split == 0) { |
4077 | if (mid <= slot) { | |
4078 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4079 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4080 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4081 | btrfs_tree_unlock(path->nodes[0]); |
4082 | free_extent_buffer(path->nodes[0]); | |
4083 | path->nodes[0] = right; | |
4084 | path->slots[0] = 0; | |
4085 | path->slots[1] += 1; | |
4086 | } else { | |
4087 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4088 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4089 | path->slots[1], 1); |
5d4f98a2 YZ |
4090 | btrfs_tree_unlock(path->nodes[0]); |
4091 | free_extent_buffer(path->nodes[0]); | |
4092 | path->nodes[0] = right; | |
4093 | path->slots[0] = 0; | |
143bede5 | 4094 | if (path->slots[1] == 0) |
d6a0a126 | 4095 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4096 | } |
5d4f98a2 YZ |
4097 | btrfs_mark_buffer_dirty(right); |
4098 | return ret; | |
d4dbff95 | 4099 | } |
74123bd7 | 4100 | |
143bede5 | 4101 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4102 | |
5d4f98a2 | 4103 | if (split == 2) { |
cc0c5538 CM |
4104 | BUG_ON(num_doubles != 0); |
4105 | num_doubles++; | |
4106 | goto again; | |
a429e513 | 4107 | } |
44871b1b | 4108 | |
143bede5 | 4109 | return 0; |
99d8f83c CM |
4110 | |
4111 | push_for_double: | |
4112 | push_for_double_split(trans, root, path, data_size); | |
4113 | tried_avoid_double = 1; | |
4114 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4115 | return 0; | |
4116 | goto again; | |
be0e5c09 CM |
4117 | } |
4118 | ||
ad48fd75 YZ |
4119 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4120 | struct btrfs_root *root, | |
4121 | struct btrfs_path *path, int ins_len) | |
459931ec | 4122 | { |
ad48fd75 | 4123 | struct btrfs_key key; |
459931ec | 4124 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4125 | struct btrfs_file_extent_item *fi; |
4126 | u64 extent_len = 0; | |
4127 | u32 item_size; | |
4128 | int ret; | |
459931ec CM |
4129 | |
4130 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4131 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4132 | ||
4133 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4134 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4135 | ||
4136 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4137 | return 0; | |
459931ec CM |
4138 | |
4139 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4140 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4141 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4142 | struct btrfs_file_extent_item); | |
4143 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4144 | } | |
b3b4aa74 | 4145 | btrfs_release_path(path); |
459931ec | 4146 | |
459931ec | 4147 | path->keep_locks = 1; |
ad48fd75 YZ |
4148 | path->search_for_split = 1; |
4149 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4150 | path->search_for_split = 0; |
ad48fd75 YZ |
4151 | if (ret < 0) |
4152 | goto err; | |
459931ec | 4153 | |
ad48fd75 YZ |
4154 | ret = -EAGAIN; |
4155 | leaf = path->nodes[0]; | |
459931ec | 4156 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4157 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4158 | goto err; | |
4159 | ||
109f6aef CM |
4160 | /* the leaf has changed, it now has room. return now */ |
4161 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4162 | goto err; | |
4163 | ||
ad48fd75 YZ |
4164 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4165 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4166 | struct btrfs_file_extent_item); | |
4167 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4168 | goto err; | |
459931ec CM |
4169 | } |
4170 | ||
b9473439 | 4171 | btrfs_set_path_blocking(path); |
ad48fd75 | 4172 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4173 | if (ret) |
4174 | goto err; | |
459931ec | 4175 | |
ad48fd75 | 4176 | path->keep_locks = 0; |
b9473439 | 4177 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4178 | return 0; |
4179 | err: | |
4180 | path->keep_locks = 0; | |
4181 | return ret; | |
4182 | } | |
4183 | ||
4184 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4185 | struct btrfs_root *root, | |
4186 | struct btrfs_path *path, | |
4187 | struct btrfs_key *new_key, | |
4188 | unsigned long split_offset) | |
4189 | { | |
4190 | struct extent_buffer *leaf; | |
4191 | struct btrfs_item *item; | |
4192 | struct btrfs_item *new_item; | |
4193 | int slot; | |
4194 | char *buf; | |
4195 | u32 nritems; | |
4196 | u32 item_size; | |
4197 | u32 orig_offset; | |
4198 | struct btrfs_disk_key disk_key; | |
4199 | ||
b9473439 CM |
4200 | leaf = path->nodes[0]; |
4201 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4202 | ||
b4ce94de CM |
4203 | btrfs_set_path_blocking(path); |
4204 | ||
459931ec CM |
4205 | item = btrfs_item_nr(leaf, path->slots[0]); |
4206 | orig_offset = btrfs_item_offset(leaf, item); | |
4207 | item_size = btrfs_item_size(leaf, item); | |
4208 | ||
459931ec | 4209 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4210 | if (!buf) |
4211 | return -ENOMEM; | |
4212 | ||
459931ec CM |
4213 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4214 | path->slots[0]), item_size); | |
459931ec | 4215 | |
ad48fd75 | 4216 | slot = path->slots[0] + 1; |
459931ec | 4217 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4218 | if (slot != nritems) { |
4219 | /* shift the items */ | |
4220 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4221 | btrfs_item_nr_offset(slot), |
4222 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4223 | } |
4224 | ||
4225 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4226 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4227 | ||
4228 | new_item = btrfs_item_nr(leaf, slot); | |
4229 | ||
4230 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4231 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4232 | ||
4233 | btrfs_set_item_offset(leaf, item, | |
4234 | orig_offset + item_size - split_offset); | |
4235 | btrfs_set_item_size(leaf, item, split_offset); | |
4236 | ||
4237 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4238 | ||
4239 | /* write the data for the start of the original item */ | |
4240 | write_extent_buffer(leaf, buf, | |
4241 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4242 | split_offset); | |
4243 | ||
4244 | /* write the data for the new item */ | |
4245 | write_extent_buffer(leaf, buf + split_offset, | |
4246 | btrfs_item_ptr_offset(leaf, slot), | |
4247 | item_size - split_offset); | |
4248 | btrfs_mark_buffer_dirty(leaf); | |
4249 | ||
ad48fd75 | 4250 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4251 | kfree(buf); |
ad48fd75 YZ |
4252 | return 0; |
4253 | } | |
4254 | ||
4255 | /* | |
4256 | * This function splits a single item into two items, | |
4257 | * giving 'new_key' to the new item and splitting the | |
4258 | * old one at split_offset (from the start of the item). | |
4259 | * | |
4260 | * The path may be released by this operation. After | |
4261 | * the split, the path is pointing to the old item. The | |
4262 | * new item is going to be in the same node as the old one. | |
4263 | * | |
4264 | * Note, the item being split must be smaller enough to live alone on | |
4265 | * a tree block with room for one extra struct btrfs_item | |
4266 | * | |
4267 | * This allows us to split the item in place, keeping a lock on the | |
4268 | * leaf the entire time. | |
4269 | */ | |
4270 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4271 | struct btrfs_root *root, | |
4272 | struct btrfs_path *path, | |
4273 | struct btrfs_key *new_key, | |
4274 | unsigned long split_offset) | |
4275 | { | |
4276 | int ret; | |
4277 | ret = setup_leaf_for_split(trans, root, path, | |
4278 | sizeof(struct btrfs_item)); | |
4279 | if (ret) | |
4280 | return ret; | |
4281 | ||
4282 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4283 | return ret; |
4284 | } | |
4285 | ||
ad48fd75 YZ |
4286 | /* |
4287 | * This function duplicate a item, giving 'new_key' to the new item. | |
4288 | * It guarantees both items live in the same tree leaf and the new item | |
4289 | * is contiguous with the original item. | |
4290 | * | |
4291 | * This allows us to split file extent in place, keeping a lock on the | |
4292 | * leaf the entire time. | |
4293 | */ | |
4294 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4295 | struct btrfs_root *root, | |
4296 | struct btrfs_path *path, | |
4297 | struct btrfs_key *new_key) | |
4298 | { | |
4299 | struct extent_buffer *leaf; | |
4300 | int ret; | |
4301 | u32 item_size; | |
4302 | ||
4303 | leaf = path->nodes[0]; | |
4304 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4305 | ret = setup_leaf_for_split(trans, root, path, | |
4306 | item_size + sizeof(struct btrfs_item)); | |
4307 | if (ret) | |
4308 | return ret; | |
4309 | ||
4310 | path->slots[0]++; | |
afe5fea7 | 4311 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4312 | item_size, item_size + |
4313 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4314 | leaf = path->nodes[0]; |
4315 | memcpy_extent_buffer(leaf, | |
4316 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4317 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4318 | item_size); | |
4319 | return 0; | |
4320 | } | |
4321 | ||
d352ac68 CM |
4322 | /* |
4323 | * make the item pointed to by the path smaller. new_size indicates | |
4324 | * how small to make it, and from_end tells us if we just chop bytes | |
4325 | * off the end of the item or if we shift the item to chop bytes off | |
4326 | * the front. | |
4327 | */ | |
afe5fea7 | 4328 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4329 | u32 new_size, int from_end) |
b18c6685 | 4330 | { |
b18c6685 | 4331 | int slot; |
5f39d397 CM |
4332 | struct extent_buffer *leaf; |
4333 | struct btrfs_item *item; | |
b18c6685 CM |
4334 | u32 nritems; |
4335 | unsigned int data_end; | |
4336 | unsigned int old_data_start; | |
4337 | unsigned int old_size; | |
4338 | unsigned int size_diff; | |
4339 | int i; | |
cfed81a0 CM |
4340 | struct btrfs_map_token token; |
4341 | ||
4342 | btrfs_init_map_token(&token); | |
b18c6685 | 4343 | |
5f39d397 | 4344 | leaf = path->nodes[0]; |
179e29e4 CM |
4345 | slot = path->slots[0]; |
4346 | ||
4347 | old_size = btrfs_item_size_nr(leaf, slot); | |
4348 | if (old_size == new_size) | |
143bede5 | 4349 | return; |
b18c6685 | 4350 | |
5f39d397 | 4351 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4352 | data_end = leaf_data_end(root, leaf); |
4353 | ||
5f39d397 | 4354 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4355 | |
b18c6685 CM |
4356 | size_diff = old_size - new_size; |
4357 | ||
4358 | BUG_ON(slot < 0); | |
4359 | BUG_ON(slot >= nritems); | |
4360 | ||
4361 | /* | |
4362 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4363 | */ | |
4364 | /* first correct the data pointers */ | |
4365 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4366 | u32 ioff; |
4367 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4368 | |
cfed81a0 CM |
4369 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4370 | btrfs_set_token_item_offset(leaf, item, | |
4371 | ioff + size_diff, &token); | |
b18c6685 | 4372 | } |
db94535d | 4373 | |
b18c6685 | 4374 | /* shift the data */ |
179e29e4 CM |
4375 | if (from_end) { |
4376 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4377 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4378 | data_end, old_data_start + new_size - data_end); | |
4379 | } else { | |
4380 | struct btrfs_disk_key disk_key; | |
4381 | u64 offset; | |
4382 | ||
4383 | btrfs_item_key(leaf, &disk_key, slot); | |
4384 | ||
4385 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4386 | unsigned long ptr; | |
4387 | struct btrfs_file_extent_item *fi; | |
4388 | ||
4389 | fi = btrfs_item_ptr(leaf, slot, | |
4390 | struct btrfs_file_extent_item); | |
4391 | fi = (struct btrfs_file_extent_item *)( | |
4392 | (unsigned long)fi - size_diff); | |
4393 | ||
4394 | if (btrfs_file_extent_type(leaf, fi) == | |
4395 | BTRFS_FILE_EXTENT_INLINE) { | |
4396 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4397 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4398 | (unsigned long)fi, |
4399 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4400 | disk_bytenr)); |
4401 | } | |
4402 | } | |
4403 | ||
4404 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4405 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4406 | data_end, old_data_start - data_end); | |
4407 | ||
4408 | offset = btrfs_disk_key_offset(&disk_key); | |
4409 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4410 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4411 | if (slot == 0) | |
d6a0a126 | 4412 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4413 | } |
5f39d397 CM |
4414 | |
4415 | item = btrfs_item_nr(leaf, slot); | |
4416 | btrfs_set_item_size(leaf, item, new_size); | |
4417 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4418 | |
5f39d397 CM |
4419 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4420 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4421 | BUG(); |
5f39d397 | 4422 | } |
b18c6685 CM |
4423 | } |
4424 | ||
d352ac68 | 4425 | /* |
8f69dbd2 | 4426 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4427 | */ |
4b90c680 | 4428 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4429 | u32 data_size) |
6567e837 | 4430 | { |
6567e837 | 4431 | int slot; |
5f39d397 CM |
4432 | struct extent_buffer *leaf; |
4433 | struct btrfs_item *item; | |
6567e837 CM |
4434 | u32 nritems; |
4435 | unsigned int data_end; | |
4436 | unsigned int old_data; | |
4437 | unsigned int old_size; | |
4438 | int i; | |
cfed81a0 CM |
4439 | struct btrfs_map_token token; |
4440 | ||
4441 | btrfs_init_map_token(&token); | |
6567e837 | 4442 | |
5f39d397 | 4443 | leaf = path->nodes[0]; |
6567e837 | 4444 | |
5f39d397 | 4445 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4446 | data_end = leaf_data_end(root, leaf); |
4447 | ||
5f39d397 CM |
4448 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4449 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4450 | BUG(); |
5f39d397 | 4451 | } |
6567e837 | 4452 | slot = path->slots[0]; |
5f39d397 | 4453 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4454 | |
4455 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4456 | if (slot >= nritems) { |
4457 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
4458 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
4459 | slot, nritems); | |
3326d1b0 CM |
4460 | BUG_ON(1); |
4461 | } | |
6567e837 CM |
4462 | |
4463 | /* | |
4464 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4465 | */ | |
4466 | /* first correct the data pointers */ | |
4467 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4468 | u32 ioff; |
4469 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4470 | |
cfed81a0 CM |
4471 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4472 | btrfs_set_token_item_offset(leaf, item, | |
4473 | ioff - data_size, &token); | |
6567e837 | 4474 | } |
5f39d397 | 4475 | |
6567e837 | 4476 | /* shift the data */ |
5f39d397 | 4477 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4478 | data_end - data_size, btrfs_leaf_data(leaf) + |
4479 | data_end, old_data - data_end); | |
5f39d397 | 4480 | |
6567e837 | 4481 | data_end = old_data; |
5f39d397 CM |
4482 | old_size = btrfs_item_size_nr(leaf, slot); |
4483 | item = btrfs_item_nr(leaf, slot); | |
4484 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
4485 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4486 | |
5f39d397 CM |
4487 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4488 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4489 | BUG(); |
5f39d397 | 4490 | } |
6567e837 CM |
4491 | } |
4492 | ||
74123bd7 | 4493 | /* |
44871b1b CM |
4494 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4495 | * to save stack depth by doing the bulk of the work in a function | |
4496 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4497 | */ |
afe5fea7 | 4498 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4499 | struct btrfs_key *cpu_key, u32 *data_size, |
4500 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4501 | { |
5f39d397 | 4502 | struct btrfs_item *item; |
9c58309d | 4503 | int i; |
7518a238 | 4504 | u32 nritems; |
be0e5c09 | 4505 | unsigned int data_end; |
e2fa7227 | 4506 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4507 | struct extent_buffer *leaf; |
4508 | int slot; | |
cfed81a0 CM |
4509 | struct btrfs_map_token token; |
4510 | ||
4511 | btrfs_init_map_token(&token); | |
e2fa7227 | 4512 | |
5f39d397 | 4513 | leaf = path->nodes[0]; |
44871b1b | 4514 | slot = path->slots[0]; |
74123bd7 | 4515 | |
5f39d397 | 4516 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4517 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4518 | |
f25956cc | 4519 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4520 | btrfs_print_leaf(root, leaf); |
d397712b | 4521 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4522 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4523 | BUG(); |
d4dbff95 | 4524 | } |
5f39d397 | 4525 | |
be0e5c09 | 4526 | if (slot != nritems) { |
5f39d397 | 4527 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4528 | |
5f39d397 CM |
4529 | if (old_data < data_end) { |
4530 | btrfs_print_leaf(root, leaf); | |
d397712b | 4531 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4532 | slot, old_data, data_end); |
4533 | BUG_ON(1); | |
4534 | } | |
be0e5c09 CM |
4535 | /* |
4536 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4537 | */ | |
4538 | /* first correct the data pointers */ | |
0783fcfc | 4539 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4540 | u32 ioff; |
db94535d | 4541 | |
5f39d397 | 4542 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4543 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4544 | btrfs_set_token_item_offset(leaf, item, | |
4545 | ioff - total_data, &token); | |
0783fcfc | 4546 | } |
be0e5c09 | 4547 | /* shift the items */ |
9c58309d | 4548 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4549 | btrfs_item_nr_offset(slot), |
d6025579 | 4550 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4551 | |
4552 | /* shift the data */ | |
5f39d397 | 4553 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4554 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4555 | data_end, old_data - data_end); |
be0e5c09 CM |
4556 | data_end = old_data; |
4557 | } | |
5f39d397 | 4558 | |
62e2749e | 4559 | /* setup the item for the new data */ |
9c58309d CM |
4560 | for (i = 0; i < nr; i++) { |
4561 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4562 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4563 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4564 | btrfs_set_token_item_offset(leaf, item, |
4565 | data_end - data_size[i], &token); | |
9c58309d | 4566 | data_end -= data_size[i]; |
cfed81a0 | 4567 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4568 | } |
44871b1b | 4569 | |
9c58309d | 4570 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4571 | |
5a01a2e3 CM |
4572 | if (slot == 0) { |
4573 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4574 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4575 | } |
b9473439 CM |
4576 | btrfs_unlock_up_safe(path, 1); |
4577 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4578 | |
5f39d397 CM |
4579 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4580 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4581 | BUG(); |
5f39d397 | 4582 | } |
44871b1b CM |
4583 | } |
4584 | ||
4585 | /* | |
4586 | * Given a key and some data, insert items into the tree. | |
4587 | * This does all the path init required, making room in the tree if needed. | |
4588 | */ | |
4589 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4590 | struct btrfs_root *root, | |
4591 | struct btrfs_path *path, | |
4592 | struct btrfs_key *cpu_key, u32 *data_size, | |
4593 | int nr) | |
4594 | { | |
44871b1b CM |
4595 | int ret = 0; |
4596 | int slot; | |
4597 | int i; | |
4598 | u32 total_size = 0; | |
4599 | u32 total_data = 0; | |
4600 | ||
4601 | for (i = 0; i < nr; i++) | |
4602 | total_data += data_size[i]; | |
4603 | ||
4604 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4605 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4606 | if (ret == 0) | |
4607 | return -EEXIST; | |
4608 | if (ret < 0) | |
143bede5 | 4609 | return ret; |
44871b1b | 4610 | |
44871b1b CM |
4611 | slot = path->slots[0]; |
4612 | BUG_ON(slot < 0); | |
4613 | ||
afe5fea7 | 4614 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4615 | total_data, total_size, nr); |
143bede5 | 4616 | return 0; |
62e2749e CM |
4617 | } |
4618 | ||
4619 | /* | |
4620 | * Given a key and some data, insert an item into the tree. | |
4621 | * This does all the path init required, making room in the tree if needed. | |
4622 | */ | |
e089f05c CM |
4623 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4624 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4625 | data_size) | |
62e2749e CM |
4626 | { |
4627 | int ret = 0; | |
2c90e5d6 | 4628 | struct btrfs_path *path; |
5f39d397 CM |
4629 | struct extent_buffer *leaf; |
4630 | unsigned long ptr; | |
62e2749e | 4631 | |
2c90e5d6 | 4632 | path = btrfs_alloc_path(); |
db5b493a TI |
4633 | if (!path) |
4634 | return -ENOMEM; | |
2c90e5d6 | 4635 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4636 | if (!ret) { |
5f39d397 CM |
4637 | leaf = path->nodes[0]; |
4638 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4639 | write_extent_buffer(leaf, data, ptr, data_size); | |
4640 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4641 | } |
2c90e5d6 | 4642 | btrfs_free_path(path); |
aa5d6bed | 4643 | return ret; |
be0e5c09 CM |
4644 | } |
4645 | ||
74123bd7 | 4646 | /* |
5de08d7d | 4647 | * delete the pointer from a given node. |
74123bd7 | 4648 | * |
d352ac68 CM |
4649 | * the tree should have been previously balanced so the deletion does not |
4650 | * empty a node. | |
74123bd7 | 4651 | */ |
afe5fea7 TI |
4652 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4653 | int level, int slot) | |
be0e5c09 | 4654 | { |
5f39d397 | 4655 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4656 | u32 nritems; |
f3ea38da | 4657 | int ret; |
be0e5c09 | 4658 | |
5f39d397 | 4659 | nritems = btrfs_header_nritems(parent); |
d397712b | 4660 | if (slot != nritems - 1) { |
0e411ece | 4661 | if (level) |
f3ea38da JS |
4662 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4663 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4664 | memmove_extent_buffer(parent, |
4665 | btrfs_node_key_ptr_offset(slot), | |
4666 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4667 | sizeof(struct btrfs_key_ptr) * |
4668 | (nritems - slot - 1)); | |
57ba86c0 CM |
4669 | } else if (level) { |
4670 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
4671 | MOD_LOG_KEY_REMOVE); | |
4672 | BUG_ON(ret < 0); | |
bb803951 | 4673 | } |
f3ea38da | 4674 | |
7518a238 | 4675 | nritems--; |
5f39d397 | 4676 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4677 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4678 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4679 | /* just turn the root into a leaf and break */ |
5f39d397 | 4680 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4681 | } else if (slot == 0) { |
5f39d397 CM |
4682 | struct btrfs_disk_key disk_key; |
4683 | ||
4684 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4685 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4686 | } |
d6025579 | 4687 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4688 | } |
4689 | ||
323ac95b CM |
4690 | /* |
4691 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4692 | * path->nodes[1]. |
323ac95b CM |
4693 | * |
4694 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4695 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4696 | * | |
4697 | * The path must have already been setup for deleting the leaf, including | |
4698 | * all the proper balancing. path->nodes[1] must be locked. | |
4699 | */ | |
143bede5 JM |
4700 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4701 | struct btrfs_root *root, | |
4702 | struct btrfs_path *path, | |
4703 | struct extent_buffer *leaf) | |
323ac95b | 4704 | { |
5d4f98a2 | 4705 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4706 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4707 | |
4d081c41 CM |
4708 | /* |
4709 | * btrfs_free_extent is expensive, we want to make sure we | |
4710 | * aren't holding any locks when we call it | |
4711 | */ | |
4712 | btrfs_unlock_up_safe(path, 0); | |
4713 | ||
f0486c68 YZ |
4714 | root_sub_used(root, leaf->len); |
4715 | ||
3083ee2e | 4716 | extent_buffer_get(leaf); |
5581a51a | 4717 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4718 | free_extent_buffer_stale(leaf); |
323ac95b | 4719 | } |
74123bd7 CM |
4720 | /* |
4721 | * delete the item at the leaf level in path. If that empties | |
4722 | * the leaf, remove it from the tree | |
4723 | */ | |
85e21bac CM |
4724 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4725 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4726 | { |
5f39d397 CM |
4727 | struct extent_buffer *leaf; |
4728 | struct btrfs_item *item; | |
85e21bac CM |
4729 | int last_off; |
4730 | int dsize = 0; | |
aa5d6bed CM |
4731 | int ret = 0; |
4732 | int wret; | |
85e21bac | 4733 | int i; |
7518a238 | 4734 | u32 nritems; |
cfed81a0 CM |
4735 | struct btrfs_map_token token; |
4736 | ||
4737 | btrfs_init_map_token(&token); | |
be0e5c09 | 4738 | |
5f39d397 | 4739 | leaf = path->nodes[0]; |
85e21bac CM |
4740 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4741 | ||
4742 | for (i = 0; i < nr; i++) | |
4743 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4744 | ||
5f39d397 | 4745 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4746 | |
85e21bac | 4747 | if (slot + nr != nritems) { |
123abc88 | 4748 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4749 | |
4750 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4751 | data_end + dsize, |
4752 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4753 | last_off - data_end); |
5f39d397 | 4754 | |
85e21bac | 4755 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4756 | u32 ioff; |
db94535d | 4757 | |
5f39d397 | 4758 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4759 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4760 | btrfs_set_token_item_offset(leaf, item, | |
4761 | ioff + dsize, &token); | |
0783fcfc | 4762 | } |
db94535d | 4763 | |
5f39d397 | 4764 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4765 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4766 | sizeof(struct btrfs_item) * |
85e21bac | 4767 | (nritems - slot - nr)); |
be0e5c09 | 4768 | } |
85e21bac CM |
4769 | btrfs_set_header_nritems(leaf, nritems - nr); |
4770 | nritems -= nr; | |
5f39d397 | 4771 | |
74123bd7 | 4772 | /* delete the leaf if we've emptied it */ |
7518a238 | 4773 | if (nritems == 0) { |
5f39d397 CM |
4774 | if (leaf == root->node) { |
4775 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4776 | } else { |
f0486c68 YZ |
4777 | btrfs_set_path_blocking(path); |
4778 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4779 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4780 | } |
be0e5c09 | 4781 | } else { |
7518a238 | 4782 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4783 | if (slot == 0) { |
5f39d397 CM |
4784 | struct btrfs_disk_key disk_key; |
4785 | ||
4786 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 4787 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 4788 | } |
aa5d6bed | 4789 | |
74123bd7 | 4790 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4791 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4792 | /* push_leaf_left fixes the path. |
4793 | * make sure the path still points to our leaf | |
4794 | * for possible call to del_ptr below | |
4795 | */ | |
4920c9ac | 4796 | slot = path->slots[1]; |
5f39d397 CM |
4797 | extent_buffer_get(leaf); |
4798 | ||
b9473439 | 4799 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4800 | wret = push_leaf_left(trans, root, path, 1, 1, |
4801 | 1, (u32)-1); | |
54aa1f4d | 4802 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4803 | ret = wret; |
5f39d397 CM |
4804 | |
4805 | if (path->nodes[0] == leaf && | |
4806 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4807 | wret = push_leaf_right(trans, root, path, 1, |
4808 | 1, 1, 0); | |
54aa1f4d | 4809 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4810 | ret = wret; |
4811 | } | |
5f39d397 CM |
4812 | |
4813 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4814 | path->slots[1] = slot; |
143bede5 | 4815 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4816 | free_extent_buffer(leaf); |
143bede5 | 4817 | ret = 0; |
5de08d7d | 4818 | } else { |
925baedd CM |
4819 | /* if we're still in the path, make sure |
4820 | * we're dirty. Otherwise, one of the | |
4821 | * push_leaf functions must have already | |
4822 | * dirtied this buffer | |
4823 | */ | |
4824 | if (path->nodes[0] == leaf) | |
4825 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4826 | free_extent_buffer(leaf); |
be0e5c09 | 4827 | } |
d5719762 | 4828 | } else { |
5f39d397 | 4829 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4830 | } |
4831 | } | |
aa5d6bed | 4832 | return ret; |
be0e5c09 CM |
4833 | } |
4834 | ||
7bb86316 | 4835 | /* |
925baedd | 4836 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4837 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4838 | * returns < 0 on io errors. | |
d352ac68 CM |
4839 | * |
4840 | * This may release the path, and so you may lose any locks held at the | |
4841 | * time you call it. | |
7bb86316 CM |
4842 | */ |
4843 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
4844 | { | |
925baedd CM |
4845 | struct btrfs_key key; |
4846 | struct btrfs_disk_key found_key; | |
4847 | int ret; | |
7bb86316 | 4848 | |
925baedd | 4849 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4850 | |
925baedd CM |
4851 | if (key.offset > 0) |
4852 | key.offset--; | |
4853 | else if (key.type > 0) | |
4854 | key.type--; | |
4855 | else if (key.objectid > 0) | |
4856 | key.objectid--; | |
4857 | else | |
4858 | return 1; | |
7bb86316 | 4859 | |
b3b4aa74 | 4860 | btrfs_release_path(path); |
925baedd CM |
4861 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4862 | if (ret < 0) | |
4863 | return ret; | |
4864 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4865 | ret = comp_keys(&found_key, &key); | |
4866 | if (ret < 0) | |
4867 | return 0; | |
4868 | return 1; | |
7bb86316 CM |
4869 | } |
4870 | ||
3f157a2f CM |
4871 | /* |
4872 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4873 | * for nodes or leaves that are have a minimum transaction id. |
4874 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4875 | * |
4876 | * This does not cow, but it does stuff the starting key it finds back | |
4877 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4878 | * key and get a writable path. | |
4879 | * | |
4880 | * This does lock as it descends, and path->keep_locks should be set | |
4881 | * to 1 by the caller. | |
4882 | * | |
4883 | * This honors path->lowest_level to prevent descent past a given level | |
4884 | * of the tree. | |
4885 | * | |
d352ac68 CM |
4886 | * min_trans indicates the oldest transaction that you are interested |
4887 | * in walking through. Any nodes or leaves older than min_trans are | |
4888 | * skipped over (without reading them). | |
4889 | * | |
3f157a2f CM |
4890 | * returns zero if something useful was found, < 0 on error and 1 if there |
4891 | * was nothing in the tree that matched the search criteria. | |
4892 | */ | |
4893 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4894 | struct btrfs_key *max_key, |
de78b51a | 4895 | struct btrfs_path *path, |
3f157a2f CM |
4896 | u64 min_trans) |
4897 | { | |
4898 | struct extent_buffer *cur; | |
4899 | struct btrfs_key found_key; | |
4900 | int slot; | |
9652480b | 4901 | int sret; |
3f157a2f CM |
4902 | u32 nritems; |
4903 | int level; | |
4904 | int ret = 1; | |
4905 | ||
934d375b | 4906 | WARN_ON(!path->keep_locks); |
3f157a2f | 4907 | again: |
bd681513 | 4908 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4909 | level = btrfs_header_level(cur); |
e02119d5 | 4910 | WARN_ON(path->nodes[level]); |
3f157a2f | 4911 | path->nodes[level] = cur; |
bd681513 | 4912 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4913 | |
4914 | if (btrfs_header_generation(cur) < min_trans) { | |
4915 | ret = 1; | |
4916 | goto out; | |
4917 | } | |
d397712b | 4918 | while (1) { |
3f157a2f CM |
4919 | nritems = btrfs_header_nritems(cur); |
4920 | level = btrfs_header_level(cur); | |
9652480b | 4921 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4922 | |
323ac95b CM |
4923 | /* at the lowest level, we're done, setup the path and exit */ |
4924 | if (level == path->lowest_level) { | |
e02119d5 CM |
4925 | if (slot >= nritems) |
4926 | goto find_next_key; | |
3f157a2f CM |
4927 | ret = 0; |
4928 | path->slots[level] = slot; | |
4929 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4930 | goto out; | |
4931 | } | |
9652480b Y |
4932 | if (sret && slot > 0) |
4933 | slot--; | |
3f157a2f | 4934 | /* |
de78b51a ES |
4935 | * check this node pointer against the min_trans parameters. |
4936 | * If it is too old, old, skip to the next one. | |
3f157a2f | 4937 | */ |
d397712b | 4938 | while (slot < nritems) { |
3f157a2f CM |
4939 | u64 blockptr; |
4940 | u64 gen; | |
e02119d5 | 4941 | |
3f157a2f CM |
4942 | blockptr = btrfs_node_blockptr(cur, slot); |
4943 | gen = btrfs_node_ptr_generation(cur, slot); | |
4944 | if (gen < min_trans) { | |
4945 | slot++; | |
4946 | continue; | |
4947 | } | |
de78b51a | 4948 | break; |
3f157a2f | 4949 | } |
e02119d5 | 4950 | find_next_key: |
3f157a2f CM |
4951 | /* |
4952 | * we didn't find a candidate key in this node, walk forward | |
4953 | * and find another one | |
4954 | */ | |
4955 | if (slot >= nritems) { | |
e02119d5 | 4956 | path->slots[level] = slot; |
b4ce94de | 4957 | btrfs_set_path_blocking(path); |
e02119d5 | 4958 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 4959 | min_trans); |
e02119d5 | 4960 | if (sret == 0) { |
b3b4aa74 | 4961 | btrfs_release_path(path); |
3f157a2f CM |
4962 | goto again; |
4963 | } else { | |
4964 | goto out; | |
4965 | } | |
4966 | } | |
4967 | /* save our key for returning back */ | |
4968 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4969 | path->slots[level] = slot; | |
4970 | if (level == path->lowest_level) { | |
4971 | ret = 0; | |
f7c79f30 | 4972 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4973 | goto out; |
4974 | } | |
b4ce94de | 4975 | btrfs_set_path_blocking(path); |
3f157a2f | 4976 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4977 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4978 | |
bd681513 | 4979 | btrfs_tree_read_lock(cur); |
b4ce94de | 4980 | |
bd681513 | 4981 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4982 | path->nodes[level - 1] = cur; |
f7c79f30 | 4983 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4984 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4985 | } |
4986 | out: | |
4987 | if (ret == 0) | |
4988 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4989 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4990 | return ret; |
4991 | } | |
4992 | ||
7069830a AB |
4993 | static void tree_move_down(struct btrfs_root *root, |
4994 | struct btrfs_path *path, | |
4995 | int *level, int root_level) | |
4996 | { | |
74dd17fb | 4997 | BUG_ON(*level == 0); |
7069830a AB |
4998 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
4999 | path->slots[*level]); | |
5000 | path->slots[*level - 1] = 0; | |
5001 | (*level)--; | |
5002 | } | |
5003 | ||
5004 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5005 | struct btrfs_path *path, | |
5006 | int *level, int root_level) | |
5007 | { | |
5008 | int ret = 0; | |
5009 | int nritems; | |
5010 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5011 | ||
5012 | path->slots[*level]++; | |
5013 | ||
74dd17fb | 5014 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5015 | if (*level == root_level) |
5016 | return -1; | |
5017 | ||
5018 | /* move upnext */ | |
5019 | path->slots[*level] = 0; | |
5020 | free_extent_buffer(path->nodes[*level]); | |
5021 | path->nodes[*level] = NULL; | |
5022 | (*level)++; | |
5023 | path->slots[*level]++; | |
5024 | ||
5025 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5026 | ret = 1; | |
5027 | } | |
5028 | return ret; | |
5029 | } | |
5030 | ||
5031 | /* | |
5032 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5033 | * or down. | |
5034 | */ | |
5035 | static int tree_advance(struct btrfs_root *root, | |
5036 | struct btrfs_path *path, | |
5037 | int *level, int root_level, | |
5038 | int allow_down, | |
5039 | struct btrfs_key *key) | |
5040 | { | |
5041 | int ret; | |
5042 | ||
5043 | if (*level == 0 || !allow_down) { | |
5044 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5045 | } else { | |
5046 | tree_move_down(root, path, level, root_level); | |
5047 | ret = 0; | |
5048 | } | |
5049 | if (ret >= 0) { | |
5050 | if (*level == 0) | |
5051 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5052 | path->slots[*level]); | |
5053 | else | |
5054 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5055 | path->slots[*level]); | |
5056 | } | |
5057 | return ret; | |
5058 | } | |
5059 | ||
5060 | static int tree_compare_item(struct btrfs_root *left_root, | |
5061 | struct btrfs_path *left_path, | |
5062 | struct btrfs_path *right_path, | |
5063 | char *tmp_buf) | |
5064 | { | |
5065 | int cmp; | |
5066 | int len1, len2; | |
5067 | unsigned long off1, off2; | |
5068 | ||
5069 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5070 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5071 | if (len1 != len2) | |
5072 | return 1; | |
5073 | ||
5074 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5075 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5076 | right_path->slots[0]); | |
5077 | ||
5078 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5079 | ||
5080 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5081 | if (cmp) | |
5082 | return 1; | |
5083 | return 0; | |
5084 | } | |
5085 | ||
5086 | #define ADVANCE 1 | |
5087 | #define ADVANCE_ONLY_NEXT -1 | |
5088 | ||
5089 | /* | |
5090 | * This function compares two trees and calls the provided callback for | |
5091 | * every changed/new/deleted item it finds. | |
5092 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5093 | * the compare pretty fast on snapshotted subvolumes. | |
5094 | * | |
5095 | * This currently works on commit roots only. As commit roots are read only, | |
5096 | * we don't do any locking. The commit roots are protected with transactions. | |
5097 | * Transactions are ended and rejoined when a commit is tried in between. | |
5098 | * | |
5099 | * This function checks for modifications done to the trees while comparing. | |
5100 | * If it detects a change, it aborts immediately. | |
5101 | */ | |
5102 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5103 | struct btrfs_root *right_root, | |
5104 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5105 | { | |
5106 | int ret; | |
5107 | int cmp; | |
5108 | struct btrfs_trans_handle *trans = NULL; | |
5109 | struct btrfs_path *left_path = NULL; | |
5110 | struct btrfs_path *right_path = NULL; | |
5111 | struct btrfs_key left_key; | |
5112 | struct btrfs_key right_key; | |
5113 | char *tmp_buf = NULL; | |
5114 | int left_root_level; | |
5115 | int right_root_level; | |
5116 | int left_level; | |
5117 | int right_level; | |
5118 | int left_end_reached; | |
5119 | int right_end_reached; | |
5120 | int advance_left; | |
5121 | int advance_right; | |
5122 | u64 left_blockptr; | |
5123 | u64 right_blockptr; | |
5124 | u64 left_start_ctransid; | |
5125 | u64 right_start_ctransid; | |
5126 | u64 ctransid; | |
5127 | ||
5128 | left_path = btrfs_alloc_path(); | |
5129 | if (!left_path) { | |
5130 | ret = -ENOMEM; | |
5131 | goto out; | |
5132 | } | |
5133 | right_path = btrfs_alloc_path(); | |
5134 | if (!right_path) { | |
5135 | ret = -ENOMEM; | |
5136 | goto out; | |
5137 | } | |
5138 | ||
5139 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5140 | if (!tmp_buf) { | |
5141 | ret = -ENOMEM; | |
5142 | goto out; | |
5143 | } | |
5144 | ||
5145 | left_path->search_commit_root = 1; | |
5146 | left_path->skip_locking = 1; | |
5147 | right_path->search_commit_root = 1; | |
5148 | right_path->skip_locking = 1; | |
5149 | ||
5f3ab90a | 5150 | spin_lock(&left_root->root_item_lock); |
7069830a | 5151 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5152 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5153 | |
5f3ab90a | 5154 | spin_lock(&right_root->root_item_lock); |
7069830a | 5155 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5156 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5157 | |
5158 | trans = btrfs_join_transaction(left_root); | |
5159 | if (IS_ERR(trans)) { | |
5160 | ret = PTR_ERR(trans); | |
5161 | trans = NULL; | |
5162 | goto out; | |
5163 | } | |
5164 | ||
5165 | /* | |
5166 | * Strategy: Go to the first items of both trees. Then do | |
5167 | * | |
5168 | * If both trees are at level 0 | |
5169 | * Compare keys of current items | |
5170 | * If left < right treat left item as new, advance left tree | |
5171 | * and repeat | |
5172 | * If left > right treat right item as deleted, advance right tree | |
5173 | * and repeat | |
5174 | * If left == right do deep compare of items, treat as changed if | |
5175 | * needed, advance both trees and repeat | |
5176 | * If both trees are at the same level but not at level 0 | |
5177 | * Compare keys of current nodes/leafs | |
5178 | * If left < right advance left tree and repeat | |
5179 | * If left > right advance right tree and repeat | |
5180 | * If left == right compare blockptrs of the next nodes/leafs | |
5181 | * If they match advance both trees but stay at the same level | |
5182 | * and repeat | |
5183 | * If they don't match advance both trees while allowing to go | |
5184 | * deeper and repeat | |
5185 | * If tree levels are different | |
5186 | * Advance the tree that needs it and repeat | |
5187 | * | |
5188 | * Advancing a tree means: | |
5189 | * If we are at level 0, try to go to the next slot. If that's not | |
5190 | * possible, go one level up and repeat. Stop when we found a level | |
5191 | * where we could go to the next slot. We may at this point be on a | |
5192 | * node or a leaf. | |
5193 | * | |
5194 | * If we are not at level 0 and not on shared tree blocks, go one | |
5195 | * level deeper. | |
5196 | * | |
5197 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5198 | * the right if possible or go up and right. | |
5199 | */ | |
5200 | ||
5201 | left_level = btrfs_header_level(left_root->commit_root); | |
5202 | left_root_level = left_level; | |
5203 | left_path->nodes[left_level] = left_root->commit_root; | |
5204 | extent_buffer_get(left_path->nodes[left_level]); | |
5205 | ||
5206 | right_level = btrfs_header_level(right_root->commit_root); | |
5207 | right_root_level = right_level; | |
5208 | right_path->nodes[right_level] = right_root->commit_root; | |
5209 | extent_buffer_get(right_path->nodes[right_level]); | |
5210 | ||
5211 | if (left_level == 0) | |
5212 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5213 | &left_key, left_path->slots[left_level]); | |
5214 | else | |
5215 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5216 | &left_key, left_path->slots[left_level]); | |
5217 | if (right_level == 0) | |
5218 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5219 | &right_key, right_path->slots[right_level]); | |
5220 | else | |
5221 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5222 | &right_key, right_path->slots[right_level]); | |
5223 | ||
5224 | left_end_reached = right_end_reached = 0; | |
5225 | advance_left = advance_right = 0; | |
5226 | ||
5227 | while (1) { | |
5228 | /* | |
5229 | * We need to make sure the transaction does not get committed | |
5230 | * while we do anything on commit roots. This means, we need to | |
5231 | * join and leave transactions for every item that we process. | |
5232 | */ | |
5233 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5234 | btrfs_release_path(left_path); | |
5235 | btrfs_release_path(right_path); | |
5236 | ||
5237 | ret = btrfs_end_transaction(trans, left_root); | |
5238 | trans = NULL; | |
5239 | if (ret < 0) | |
5240 | goto out; | |
5241 | } | |
5242 | /* now rejoin the transaction */ | |
5243 | if (!trans) { | |
5244 | trans = btrfs_join_transaction(left_root); | |
5245 | if (IS_ERR(trans)) { | |
5246 | ret = PTR_ERR(trans); | |
5247 | trans = NULL; | |
5248 | goto out; | |
5249 | } | |
5250 | ||
5f3ab90a | 5251 | spin_lock(&left_root->root_item_lock); |
7069830a | 5252 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5253 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5254 | if (ctransid != left_start_ctransid) |
5255 | left_start_ctransid = 0; | |
5256 | ||
5f3ab90a | 5257 | spin_lock(&right_root->root_item_lock); |
7069830a | 5258 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5259 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5260 | if (ctransid != right_start_ctransid) |
5261 | right_start_ctransid = 0; | |
5262 | ||
5263 | if (!left_start_ctransid || !right_start_ctransid) { | |
5264 | WARN(1, KERN_WARNING | |
5265 | "btrfs: btrfs_compare_tree detected " | |
5266 | "a change in one of the trees while " | |
5267 | "iterating. This is probably a " | |
5268 | "bug.\n"); | |
5269 | ret = -EIO; | |
5270 | goto out; | |
5271 | } | |
5272 | ||
5273 | /* | |
5274 | * the commit root may have changed, so start again | |
5275 | * where we stopped | |
5276 | */ | |
5277 | left_path->lowest_level = left_level; | |
5278 | right_path->lowest_level = right_level; | |
5279 | ret = btrfs_search_slot(NULL, left_root, | |
5280 | &left_key, left_path, 0, 0); | |
5281 | if (ret < 0) | |
5282 | goto out; | |
5283 | ret = btrfs_search_slot(NULL, right_root, | |
5284 | &right_key, right_path, 0, 0); | |
5285 | if (ret < 0) | |
5286 | goto out; | |
5287 | } | |
5288 | ||
5289 | if (advance_left && !left_end_reached) { | |
5290 | ret = tree_advance(left_root, left_path, &left_level, | |
5291 | left_root_level, | |
5292 | advance_left != ADVANCE_ONLY_NEXT, | |
5293 | &left_key); | |
5294 | if (ret < 0) | |
5295 | left_end_reached = ADVANCE; | |
5296 | advance_left = 0; | |
5297 | } | |
5298 | if (advance_right && !right_end_reached) { | |
5299 | ret = tree_advance(right_root, right_path, &right_level, | |
5300 | right_root_level, | |
5301 | advance_right != ADVANCE_ONLY_NEXT, | |
5302 | &right_key); | |
5303 | if (ret < 0) | |
5304 | right_end_reached = ADVANCE; | |
5305 | advance_right = 0; | |
5306 | } | |
5307 | ||
5308 | if (left_end_reached && right_end_reached) { | |
5309 | ret = 0; | |
5310 | goto out; | |
5311 | } else if (left_end_reached) { | |
5312 | if (right_level == 0) { | |
5313 | ret = changed_cb(left_root, right_root, | |
5314 | left_path, right_path, | |
5315 | &right_key, | |
5316 | BTRFS_COMPARE_TREE_DELETED, | |
5317 | ctx); | |
5318 | if (ret < 0) | |
5319 | goto out; | |
5320 | } | |
5321 | advance_right = ADVANCE; | |
5322 | continue; | |
5323 | } else if (right_end_reached) { | |
5324 | if (left_level == 0) { | |
5325 | ret = changed_cb(left_root, right_root, | |
5326 | left_path, right_path, | |
5327 | &left_key, | |
5328 | BTRFS_COMPARE_TREE_NEW, | |
5329 | ctx); | |
5330 | if (ret < 0) | |
5331 | goto out; | |
5332 | } | |
5333 | advance_left = ADVANCE; | |
5334 | continue; | |
5335 | } | |
5336 | ||
5337 | if (left_level == 0 && right_level == 0) { | |
5338 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5339 | if (cmp < 0) { | |
5340 | ret = changed_cb(left_root, right_root, | |
5341 | left_path, right_path, | |
5342 | &left_key, | |
5343 | BTRFS_COMPARE_TREE_NEW, | |
5344 | ctx); | |
5345 | if (ret < 0) | |
5346 | goto out; | |
5347 | advance_left = ADVANCE; | |
5348 | } else if (cmp > 0) { | |
5349 | ret = changed_cb(left_root, right_root, | |
5350 | left_path, right_path, | |
5351 | &right_key, | |
5352 | BTRFS_COMPARE_TREE_DELETED, | |
5353 | ctx); | |
5354 | if (ret < 0) | |
5355 | goto out; | |
5356 | advance_right = ADVANCE; | |
5357 | } else { | |
74dd17fb | 5358 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5359 | ret = tree_compare_item(left_root, left_path, |
5360 | right_path, tmp_buf); | |
5361 | if (ret) { | |
74dd17fb | 5362 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5363 | ret = changed_cb(left_root, right_root, |
5364 | left_path, right_path, | |
5365 | &left_key, | |
5366 | BTRFS_COMPARE_TREE_CHANGED, | |
5367 | ctx); | |
5368 | if (ret < 0) | |
5369 | goto out; | |
5370 | } | |
5371 | advance_left = ADVANCE; | |
5372 | advance_right = ADVANCE; | |
5373 | } | |
5374 | } else if (left_level == right_level) { | |
5375 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5376 | if (cmp < 0) { | |
5377 | advance_left = ADVANCE; | |
5378 | } else if (cmp > 0) { | |
5379 | advance_right = ADVANCE; | |
5380 | } else { | |
5381 | left_blockptr = btrfs_node_blockptr( | |
5382 | left_path->nodes[left_level], | |
5383 | left_path->slots[left_level]); | |
5384 | right_blockptr = btrfs_node_blockptr( | |
5385 | right_path->nodes[right_level], | |
5386 | right_path->slots[right_level]); | |
5387 | if (left_blockptr == right_blockptr) { | |
5388 | /* | |
5389 | * As we're on a shared block, don't | |
5390 | * allow to go deeper. | |
5391 | */ | |
5392 | advance_left = ADVANCE_ONLY_NEXT; | |
5393 | advance_right = ADVANCE_ONLY_NEXT; | |
5394 | } else { | |
5395 | advance_left = ADVANCE; | |
5396 | advance_right = ADVANCE; | |
5397 | } | |
5398 | } | |
5399 | } else if (left_level < right_level) { | |
5400 | advance_right = ADVANCE; | |
5401 | } else { | |
5402 | advance_left = ADVANCE; | |
5403 | } | |
5404 | } | |
5405 | ||
5406 | out: | |
5407 | btrfs_free_path(left_path); | |
5408 | btrfs_free_path(right_path); | |
5409 | kfree(tmp_buf); | |
5410 | ||
5411 | if (trans) { | |
5412 | if (!ret) | |
5413 | ret = btrfs_end_transaction(trans, left_root); | |
5414 | else | |
5415 | btrfs_end_transaction(trans, left_root); | |
5416 | } | |
5417 | ||
5418 | return ret; | |
5419 | } | |
5420 | ||
3f157a2f CM |
5421 | /* |
5422 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5423 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5424 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5425 | * |
5426 | * 0 is returned if another key is found, < 0 if there are any errors | |
5427 | * and 1 is returned if there are no higher keys in the tree | |
5428 | * | |
5429 | * path->keep_locks should be set to 1 on the search made before | |
5430 | * calling this function. | |
5431 | */ | |
e7a84565 | 5432 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5433 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5434 | { |
e7a84565 CM |
5435 | int slot; |
5436 | struct extent_buffer *c; | |
5437 | ||
934d375b | 5438 | WARN_ON(!path->keep_locks); |
d397712b | 5439 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5440 | if (!path->nodes[level]) |
5441 | return 1; | |
5442 | ||
5443 | slot = path->slots[level] + 1; | |
5444 | c = path->nodes[level]; | |
3f157a2f | 5445 | next: |
e7a84565 | 5446 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5447 | int ret; |
5448 | int orig_lowest; | |
5449 | struct btrfs_key cur_key; | |
5450 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5451 | !path->nodes[level + 1]) | |
e7a84565 | 5452 | return 1; |
33c66f43 YZ |
5453 | |
5454 | if (path->locks[level + 1]) { | |
5455 | level++; | |
5456 | continue; | |
5457 | } | |
5458 | ||
5459 | slot = btrfs_header_nritems(c) - 1; | |
5460 | if (level == 0) | |
5461 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5462 | else | |
5463 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5464 | ||
5465 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5466 | btrfs_release_path(path); |
33c66f43 YZ |
5467 | path->lowest_level = level; |
5468 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5469 | 0, 0); | |
5470 | path->lowest_level = orig_lowest; | |
5471 | if (ret < 0) | |
5472 | return ret; | |
5473 | ||
5474 | c = path->nodes[level]; | |
5475 | slot = path->slots[level]; | |
5476 | if (ret == 0) | |
5477 | slot++; | |
5478 | goto next; | |
e7a84565 | 5479 | } |
33c66f43 | 5480 | |
e7a84565 CM |
5481 | if (level == 0) |
5482 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5483 | else { |
3f157a2f CM |
5484 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5485 | ||
3f157a2f CM |
5486 | if (gen < min_trans) { |
5487 | slot++; | |
5488 | goto next; | |
5489 | } | |
e7a84565 | 5490 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5491 | } |
e7a84565 CM |
5492 | return 0; |
5493 | } | |
5494 | return 1; | |
5495 | } | |
5496 | ||
97571fd0 | 5497 | /* |
925baedd | 5498 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5499 | * returns 0 if it found something or 1 if there are no greater leaves. |
5500 | * returns < 0 on io errors. | |
97571fd0 | 5501 | */ |
234b63a0 | 5502 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5503 | { |
5504 | return btrfs_next_old_leaf(root, path, 0); | |
5505 | } | |
5506 | ||
5507 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5508 | u64 time_seq) | |
d97e63b6 CM |
5509 | { |
5510 | int slot; | |
8e73f275 | 5511 | int level; |
5f39d397 | 5512 | struct extent_buffer *c; |
8e73f275 | 5513 | struct extent_buffer *next; |
925baedd CM |
5514 | struct btrfs_key key; |
5515 | u32 nritems; | |
5516 | int ret; | |
8e73f275 | 5517 | int old_spinning = path->leave_spinning; |
bd681513 | 5518 | int next_rw_lock = 0; |
925baedd CM |
5519 | |
5520 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5521 | if (nritems == 0) |
925baedd | 5522 | return 1; |
925baedd | 5523 | |
8e73f275 CM |
5524 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5525 | again: | |
5526 | level = 1; | |
5527 | next = NULL; | |
bd681513 | 5528 | next_rw_lock = 0; |
b3b4aa74 | 5529 | btrfs_release_path(path); |
8e73f275 | 5530 | |
a2135011 | 5531 | path->keep_locks = 1; |
31533fb2 | 5532 | path->leave_spinning = 1; |
8e73f275 | 5533 | |
3d7806ec JS |
5534 | if (time_seq) |
5535 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5536 | else | |
5537 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5538 | path->keep_locks = 0; |
5539 | ||
5540 | if (ret < 0) | |
5541 | return ret; | |
5542 | ||
a2135011 | 5543 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5544 | /* |
5545 | * by releasing the path above we dropped all our locks. A balance | |
5546 | * could have added more items next to the key that used to be | |
5547 | * at the very end of the block. So, check again here and | |
5548 | * advance the path if there are now more items available. | |
5549 | */ | |
a2135011 | 5550 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5551 | if (ret == 0) |
5552 | path->slots[0]++; | |
8e73f275 | 5553 | ret = 0; |
925baedd CM |
5554 | goto done; |
5555 | } | |
d97e63b6 | 5556 | |
d397712b | 5557 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5558 | if (!path->nodes[level]) { |
5559 | ret = 1; | |
5560 | goto done; | |
5561 | } | |
5f39d397 | 5562 | |
d97e63b6 CM |
5563 | slot = path->slots[level] + 1; |
5564 | c = path->nodes[level]; | |
5f39d397 | 5565 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5566 | level++; |
8e73f275 CM |
5567 | if (level == BTRFS_MAX_LEVEL) { |
5568 | ret = 1; | |
5569 | goto done; | |
5570 | } | |
d97e63b6 CM |
5571 | continue; |
5572 | } | |
5f39d397 | 5573 | |
925baedd | 5574 | if (next) { |
bd681513 | 5575 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5576 | free_extent_buffer(next); |
925baedd | 5577 | } |
5f39d397 | 5578 | |
8e73f275 | 5579 | next = c; |
bd681513 | 5580 | next_rw_lock = path->locks[level]; |
8e73f275 | 5581 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5582 | slot, &key, 0); |
8e73f275 CM |
5583 | if (ret == -EAGAIN) |
5584 | goto again; | |
5f39d397 | 5585 | |
76a05b35 | 5586 | if (ret < 0) { |
b3b4aa74 | 5587 | btrfs_release_path(path); |
76a05b35 CM |
5588 | goto done; |
5589 | } | |
5590 | ||
5cd57b2c | 5591 | if (!path->skip_locking) { |
bd681513 | 5592 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5593 | if (!ret && time_seq) { |
5594 | /* | |
5595 | * If we don't get the lock, we may be racing | |
5596 | * with push_leaf_left, holding that lock while | |
5597 | * itself waiting for the leaf we've currently | |
5598 | * locked. To solve this situation, we give up | |
5599 | * on our lock and cycle. | |
5600 | */ | |
cf538830 | 5601 | free_extent_buffer(next); |
d42244a0 JS |
5602 | btrfs_release_path(path); |
5603 | cond_resched(); | |
5604 | goto again; | |
5605 | } | |
8e73f275 CM |
5606 | if (!ret) { |
5607 | btrfs_set_path_blocking(path); | |
bd681513 | 5608 | btrfs_tree_read_lock(next); |
31533fb2 | 5609 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5610 | BTRFS_READ_LOCK); |
8e73f275 | 5611 | } |
31533fb2 | 5612 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5613 | } |
d97e63b6 CM |
5614 | break; |
5615 | } | |
5616 | path->slots[level] = slot; | |
d397712b | 5617 | while (1) { |
d97e63b6 CM |
5618 | level--; |
5619 | c = path->nodes[level]; | |
925baedd | 5620 | if (path->locks[level]) |
bd681513 | 5621 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5622 | |
5f39d397 | 5623 | free_extent_buffer(c); |
d97e63b6 CM |
5624 | path->nodes[level] = next; |
5625 | path->slots[level] = 0; | |
a74a4b97 | 5626 | if (!path->skip_locking) |
bd681513 | 5627 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5628 | if (!level) |
5629 | break; | |
b4ce94de | 5630 | |
8e73f275 | 5631 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5632 | 0, &key, 0); |
8e73f275 CM |
5633 | if (ret == -EAGAIN) |
5634 | goto again; | |
5635 | ||
76a05b35 | 5636 | if (ret < 0) { |
b3b4aa74 | 5637 | btrfs_release_path(path); |
76a05b35 CM |
5638 | goto done; |
5639 | } | |
5640 | ||
5cd57b2c | 5641 | if (!path->skip_locking) { |
bd681513 | 5642 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5643 | if (!ret) { |
5644 | btrfs_set_path_blocking(path); | |
bd681513 | 5645 | btrfs_tree_read_lock(next); |
31533fb2 | 5646 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5647 | BTRFS_READ_LOCK); |
5648 | } | |
31533fb2 | 5649 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5650 | } |
d97e63b6 | 5651 | } |
8e73f275 | 5652 | ret = 0; |
925baedd | 5653 | done: |
f7c79f30 | 5654 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5655 | path->leave_spinning = old_spinning; |
5656 | if (!old_spinning) | |
5657 | btrfs_set_path_blocking(path); | |
5658 | ||
5659 | return ret; | |
d97e63b6 | 5660 | } |
0b86a832 | 5661 | |
3f157a2f CM |
5662 | /* |
5663 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5664 | * searching until it gets past min_objectid or finds an item of 'type' | |
5665 | * | |
5666 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5667 | */ | |
0b86a832 CM |
5668 | int btrfs_previous_item(struct btrfs_root *root, |
5669 | struct btrfs_path *path, u64 min_objectid, | |
5670 | int type) | |
5671 | { | |
5672 | struct btrfs_key found_key; | |
5673 | struct extent_buffer *leaf; | |
e02119d5 | 5674 | u32 nritems; |
0b86a832 CM |
5675 | int ret; |
5676 | ||
d397712b | 5677 | while (1) { |
0b86a832 | 5678 | if (path->slots[0] == 0) { |
b4ce94de | 5679 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5680 | ret = btrfs_prev_leaf(root, path); |
5681 | if (ret != 0) | |
5682 | return ret; | |
5683 | } else { | |
5684 | path->slots[0]--; | |
5685 | } | |
5686 | leaf = path->nodes[0]; | |
e02119d5 CM |
5687 | nritems = btrfs_header_nritems(leaf); |
5688 | if (nritems == 0) | |
5689 | return 1; | |
5690 | if (path->slots[0] == nritems) | |
5691 | path->slots[0]--; | |
5692 | ||
0b86a832 | 5693 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5694 | if (found_key.objectid < min_objectid) |
5695 | break; | |
0a4eefbb YZ |
5696 | if (found_key.type == type) |
5697 | return 0; | |
e02119d5 CM |
5698 | if (found_key.objectid == min_objectid && |
5699 | found_key.type < type) | |
5700 | break; | |
0b86a832 CM |
5701 | } |
5702 | return 1; | |
5703 | } |