]>
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; | |
c8cc6341 JB |
487 | int ret = 0; |
488 | ||
489 | BUG_ON(!tm); | |
490 | ||
491 | tree_mod_log_write_lock(fs_info); | |
492 | if (list_empty(&fs_info->tree_mod_seq_list)) { | |
493 | tree_mod_log_write_unlock(fs_info); | |
494 | /* | |
495 | * Ok we no longer care about logging modifications, free up tm | |
496 | * and return 0. Any callers shouldn't be using tm after | |
497 | * calling tree_mod_log_insert, but if they do we can just | |
498 | * change this to return a special error code to let the callers | |
499 | * do their own thing. | |
500 | */ | |
501 | kfree(tm); | |
502 | return 0; | |
503 | } | |
bd989ba3 | 504 | |
c8cc6341 JB |
505 | spin_lock(&fs_info->tree_mod_seq_lock); |
506 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
507 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
bd989ba3 | 508 | |
bd989ba3 JS |
509 | tm_root = &fs_info->tree_mod_log; |
510 | new = &tm_root->rb_node; | |
511 | while (*new) { | |
512 | cur = container_of(*new, struct tree_mod_elem, node); | |
513 | parent = *new; | |
514 | if (cur->index < tm->index) | |
515 | new = &((*new)->rb_left); | |
516 | else if (cur->index > tm->index) | |
517 | new = &((*new)->rb_right); | |
097b8a7c | 518 | else if (cur->seq < tm->seq) |
bd989ba3 | 519 | new = &((*new)->rb_left); |
097b8a7c | 520 | else if (cur->seq > tm->seq) |
bd989ba3 JS |
521 | new = &((*new)->rb_right); |
522 | else { | |
c8cc6341 | 523 | ret = -EEXIST; |
bd989ba3 | 524 | kfree(tm); |
c8cc6341 | 525 | goto out; |
bd989ba3 JS |
526 | } |
527 | } | |
528 | ||
529 | rb_link_node(&tm->node, parent, new); | |
530 | rb_insert_color(&tm->node, tm_root); | |
c8cc6341 JB |
531 | out: |
532 | tree_mod_log_write_unlock(fs_info); | |
533 | return ret; | |
bd989ba3 JS |
534 | } |
535 | ||
097b8a7c JS |
536 | /* |
537 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
538 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
539 | * this until all tree mod log insertions are recorded in the rb tree and then | |
540 | * call tree_mod_log_write_unlock() to release. | |
541 | */ | |
e9b7fd4d JS |
542 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
543 | struct extent_buffer *eb) { | |
544 | smp_mb(); | |
545 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
546 | return 1; | |
097b8a7c JS |
547 | if (eb && btrfs_header_level(eb) == 0) |
548 | return 1; | |
e9b7fd4d JS |
549 | return 0; |
550 | } | |
551 | ||
097b8a7c JS |
552 | static inline int |
553 | __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, | |
554 | struct extent_buffer *eb, int slot, | |
555 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 556 | { |
097b8a7c | 557 | struct tree_mod_elem *tm; |
bd989ba3 | 558 | |
c8cc6341 JB |
559 | tm = kzalloc(sizeof(*tm), flags); |
560 | if (!tm) | |
561 | return -ENOMEM; | |
bd989ba3 JS |
562 | |
563 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
564 | if (op != MOD_LOG_KEY_ADD) { | |
565 | btrfs_node_key(eb, &tm->key, slot); | |
566 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
567 | } | |
568 | tm->op = op; | |
569 | tm->slot = slot; | |
570 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
571 | ||
097b8a7c JS |
572 | return __tree_mod_log_insert(fs_info, tm); |
573 | } | |
574 | ||
575 | static noinline int | |
c8cc6341 JB |
576 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
577 | struct extent_buffer *eb, int slot, | |
578 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 579 | { |
097b8a7c JS |
580 | if (tree_mod_dont_log(fs_info, eb)) |
581 | return 0; | |
582 | ||
c8cc6341 | 583 | return __tree_mod_log_insert_key(fs_info, eb, slot, op, flags); |
097b8a7c JS |
584 | } |
585 | ||
bd989ba3 JS |
586 | static noinline int |
587 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
588 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
589 | int nr_items, gfp_t flags) | |
590 | { | |
591 | struct tree_mod_elem *tm; | |
592 | int ret; | |
593 | int i; | |
594 | ||
f395694c JS |
595 | if (tree_mod_dont_log(fs_info, eb)) |
596 | return 0; | |
bd989ba3 | 597 | |
01763a2e JS |
598 | /* |
599 | * When we override something during the move, we log these removals. | |
600 | * This can only happen when we move towards the beginning of the | |
601 | * buffer, i.e. dst_slot < src_slot. | |
602 | */ | |
bd989ba3 | 603 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
c8cc6341 JB |
604 | ret = __tree_mod_log_insert_key(fs_info, eb, i + dst_slot, |
605 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); | |
bd989ba3 JS |
606 | BUG_ON(ret < 0); |
607 | } | |
608 | ||
c8cc6341 JB |
609 | tm = kzalloc(sizeof(*tm), flags); |
610 | if (!tm) | |
611 | return -ENOMEM; | |
f395694c | 612 | |
bd989ba3 JS |
613 | tm->index = eb->start >> PAGE_CACHE_SHIFT; |
614 | tm->slot = src_slot; | |
615 | tm->move.dst_slot = dst_slot; | |
616 | tm->move.nr_items = nr_items; | |
617 | tm->op = MOD_LOG_MOVE_KEYS; | |
618 | ||
c8cc6341 | 619 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
620 | } |
621 | ||
097b8a7c JS |
622 | static inline void |
623 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
624 | { | |
625 | int i; | |
626 | u32 nritems; | |
627 | int ret; | |
628 | ||
b12a3b1e CM |
629 | if (btrfs_header_level(eb) == 0) |
630 | return; | |
631 | ||
097b8a7c JS |
632 | nritems = btrfs_header_nritems(eb); |
633 | for (i = nritems - 1; i >= 0; i--) { | |
c8cc6341 JB |
634 | ret = __tree_mod_log_insert_key(fs_info, eb, i, |
635 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
097b8a7c JS |
636 | BUG_ON(ret < 0); |
637 | } | |
638 | } | |
639 | ||
bd989ba3 JS |
640 | static noinline int |
641 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
642 | struct extent_buffer *old_root, | |
90f8d62e JS |
643 | struct extent_buffer *new_root, gfp_t flags, |
644 | int log_removal) | |
bd989ba3 JS |
645 | { |
646 | struct tree_mod_elem *tm; | |
bd989ba3 | 647 | |
097b8a7c JS |
648 | if (tree_mod_dont_log(fs_info, NULL)) |
649 | return 0; | |
650 | ||
90f8d62e JS |
651 | if (log_removal) |
652 | __tree_mod_log_free_eb(fs_info, old_root); | |
d9abbf1c | 653 | |
c8cc6341 JB |
654 | tm = kzalloc(sizeof(*tm), flags); |
655 | if (!tm) | |
656 | return -ENOMEM; | |
bd989ba3 JS |
657 | |
658 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
659 | tm->old_root.logical = old_root->start; | |
660 | tm->old_root.level = btrfs_header_level(old_root); | |
661 | tm->generation = btrfs_header_generation(old_root); | |
662 | tm->op = MOD_LOG_ROOT_REPLACE; | |
663 | ||
c8cc6341 | 664 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
665 | } |
666 | ||
667 | static struct tree_mod_elem * | |
668 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
669 | int smallest) | |
670 | { | |
671 | struct rb_root *tm_root; | |
672 | struct rb_node *node; | |
673 | struct tree_mod_elem *cur = NULL; | |
674 | struct tree_mod_elem *found = NULL; | |
675 | u64 index = start >> PAGE_CACHE_SHIFT; | |
676 | ||
097b8a7c | 677 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
678 | tm_root = &fs_info->tree_mod_log; |
679 | node = tm_root->rb_node; | |
680 | while (node) { | |
681 | cur = container_of(node, struct tree_mod_elem, node); | |
682 | if (cur->index < index) { | |
683 | node = node->rb_left; | |
684 | } else if (cur->index > index) { | |
685 | node = node->rb_right; | |
097b8a7c | 686 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
687 | node = node->rb_left; |
688 | } else if (!smallest) { | |
689 | /* we want the node with the highest seq */ | |
690 | if (found) | |
097b8a7c | 691 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
692 | found = cur; |
693 | node = node->rb_left; | |
097b8a7c | 694 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
695 | /* we want the node with the smallest seq */ |
696 | if (found) | |
097b8a7c | 697 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
698 | found = cur; |
699 | node = node->rb_right; | |
700 | } else { | |
701 | found = cur; | |
702 | break; | |
703 | } | |
704 | } | |
097b8a7c | 705 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
706 | |
707 | return found; | |
708 | } | |
709 | ||
710 | /* | |
711 | * this returns the element from the log with the smallest time sequence | |
712 | * value that's in the log (the oldest log item). any element with a time | |
713 | * sequence lower than min_seq will be ignored. | |
714 | */ | |
715 | static struct tree_mod_elem * | |
716 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
717 | u64 min_seq) | |
718 | { | |
719 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
720 | } | |
721 | ||
722 | /* | |
723 | * this returns the element from the log with the largest time sequence | |
724 | * value that's in the log (the most recent log item). any element with | |
725 | * a time sequence lower than min_seq will be ignored. | |
726 | */ | |
727 | static struct tree_mod_elem * | |
728 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
729 | { | |
730 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
731 | } | |
732 | ||
097b8a7c | 733 | static noinline void |
bd989ba3 JS |
734 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
735 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 736 | unsigned long src_offset, int nr_items) |
bd989ba3 JS |
737 | { |
738 | int ret; | |
739 | int i; | |
740 | ||
e9b7fd4d | 741 | if (tree_mod_dont_log(fs_info, NULL)) |
bd989ba3 JS |
742 | return; |
743 | ||
c8cc6341 | 744 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
bd989ba3 JS |
745 | return; |
746 | ||
bd989ba3 | 747 | for (i = 0; i < nr_items; i++) { |
c8cc6341 | 748 | ret = __tree_mod_log_insert_key(fs_info, src, |
90f8d62e | 749 | i + src_offset, |
c8cc6341 | 750 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
90f8d62e | 751 | BUG_ON(ret < 0); |
c8cc6341 | 752 | ret = __tree_mod_log_insert_key(fs_info, dst, |
097b8a7c | 753 | i + dst_offset, |
c8cc6341 JB |
754 | MOD_LOG_KEY_ADD, |
755 | GFP_NOFS); | |
bd989ba3 JS |
756 | BUG_ON(ret < 0); |
757 | } | |
758 | } | |
759 | ||
760 | static inline void | |
761 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
762 | int dst_offset, int src_offset, int nr_items) | |
763 | { | |
764 | int ret; | |
765 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
766 | nr_items, GFP_NOFS); | |
767 | BUG_ON(ret < 0); | |
768 | } | |
769 | ||
097b8a7c | 770 | static noinline void |
bd989ba3 | 771 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 772 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
773 | { |
774 | int ret; | |
775 | ||
c8cc6341 JB |
776 | ret = __tree_mod_log_insert_key(fs_info, eb, slot, |
777 | MOD_LOG_KEY_REPLACE, | |
778 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
779 | BUG_ON(ret < 0); |
780 | } | |
781 | ||
097b8a7c JS |
782 | static noinline void |
783 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
bd989ba3 | 784 | { |
e9b7fd4d | 785 | if (tree_mod_dont_log(fs_info, eb)) |
bd989ba3 | 786 | return; |
097b8a7c | 787 | __tree_mod_log_free_eb(fs_info, eb); |
bd989ba3 JS |
788 | } |
789 | ||
097b8a7c | 790 | static noinline void |
bd989ba3 | 791 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
792 | struct extent_buffer *new_root_node, |
793 | int log_removal) | |
bd989ba3 JS |
794 | { |
795 | int ret; | |
bd989ba3 | 796 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 797 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
798 | BUG_ON(ret < 0); |
799 | } | |
800 | ||
5d4f98a2 YZ |
801 | /* |
802 | * check if the tree block can be shared by multiple trees | |
803 | */ | |
804 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
805 | struct extent_buffer *buf) | |
806 | { | |
807 | /* | |
808 | * Tree blocks not in refernece counted trees and tree roots | |
809 | * are never shared. If a block was allocated after the last | |
810 | * snapshot and the block was not allocated by tree relocation, | |
811 | * we know the block is not shared. | |
812 | */ | |
813 | if (root->ref_cows && | |
814 | buf != root->node && buf != root->commit_root && | |
815 | (btrfs_header_generation(buf) <= | |
816 | btrfs_root_last_snapshot(&root->root_item) || | |
817 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
818 | return 1; | |
819 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
820 | if (root->ref_cows && | |
821 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
822 | return 1; | |
823 | #endif | |
824 | return 0; | |
825 | } | |
826 | ||
827 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
828 | struct btrfs_root *root, | |
829 | struct extent_buffer *buf, | |
f0486c68 YZ |
830 | struct extent_buffer *cow, |
831 | int *last_ref) | |
5d4f98a2 YZ |
832 | { |
833 | u64 refs; | |
834 | u64 owner; | |
835 | u64 flags; | |
836 | u64 new_flags = 0; | |
837 | int ret; | |
838 | ||
839 | /* | |
840 | * Backrefs update rules: | |
841 | * | |
842 | * Always use full backrefs for extent pointers in tree block | |
843 | * allocated by tree relocation. | |
844 | * | |
845 | * If a shared tree block is no longer referenced by its owner | |
846 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
847 | * use full backrefs for extent pointers in tree block. | |
848 | * | |
849 | * If a tree block is been relocating | |
850 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
851 | * use full backrefs for extent pointers in tree block. | |
852 | * The reason for this is some operations (such as drop tree) | |
853 | * are only allowed for blocks use full backrefs. | |
854 | */ | |
855 | ||
856 | if (btrfs_block_can_be_shared(root, buf)) { | |
857 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
858 | btrfs_header_level(buf), 1, |
859 | &refs, &flags); | |
be1a5564 MF |
860 | if (ret) |
861 | return ret; | |
e5df9573 MF |
862 | if (refs == 0) { |
863 | ret = -EROFS; | |
864 | btrfs_std_error(root->fs_info, ret); | |
865 | return ret; | |
866 | } | |
5d4f98a2 YZ |
867 | } else { |
868 | refs = 1; | |
869 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
870 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
871 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
872 | else | |
873 | flags = 0; | |
874 | } | |
875 | ||
876 | owner = btrfs_header_owner(buf); | |
877 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
878 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
879 | ||
880 | if (refs > 1) { | |
881 | if ((owner == root->root_key.objectid || | |
882 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
883 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 884 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 885 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
886 | |
887 | if (root->root_key.objectid == | |
888 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 889 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 890 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 891 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 892 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
893 | } |
894 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
895 | } else { | |
896 | ||
897 | if (root->root_key.objectid == | |
898 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 899 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 900 | else |
66d7e7f0 | 901 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 902 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
903 | } |
904 | if (new_flags != 0) { | |
b1c79e09 JB |
905 | int level = btrfs_header_level(buf); |
906 | ||
5d4f98a2 YZ |
907 | ret = btrfs_set_disk_extent_flags(trans, root, |
908 | buf->start, | |
909 | buf->len, | |
b1c79e09 | 910 | new_flags, level, 0); |
be1a5564 MF |
911 | if (ret) |
912 | return ret; | |
5d4f98a2 YZ |
913 | } |
914 | } else { | |
915 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
916 | if (root->root_key.objectid == | |
917 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 918 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 919 | else |
66d7e7f0 | 920 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 921 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 922 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 923 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
924 | } |
925 | clean_tree_block(trans, root, buf); | |
f0486c68 | 926 | *last_ref = 1; |
5d4f98a2 YZ |
927 | } |
928 | return 0; | |
929 | } | |
930 | ||
d352ac68 | 931 | /* |
d397712b CM |
932 | * does the dirty work in cow of a single block. The parent block (if |
933 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
934 | * dirty and returned locked. If you modify the block it needs to be marked | |
935 | * dirty again. | |
d352ac68 CM |
936 | * |
937 | * search_start -- an allocation hint for the new block | |
938 | * | |
d397712b CM |
939 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
940 | * bytes the allocator should try to find free next to the block it returns. | |
941 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 942 | */ |
d397712b | 943 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
944 | struct btrfs_root *root, |
945 | struct extent_buffer *buf, | |
946 | struct extent_buffer *parent, int parent_slot, | |
947 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 948 | u64 search_start, u64 empty_size) |
02217ed2 | 949 | { |
5d4f98a2 | 950 | struct btrfs_disk_key disk_key; |
5f39d397 | 951 | struct extent_buffer *cow; |
be1a5564 | 952 | int level, ret; |
f0486c68 | 953 | int last_ref = 0; |
925baedd | 954 | int unlock_orig = 0; |
5d4f98a2 | 955 | u64 parent_start; |
7bb86316 | 956 | |
925baedd CM |
957 | if (*cow_ret == buf) |
958 | unlock_orig = 1; | |
959 | ||
b9447ef8 | 960 | btrfs_assert_tree_locked(buf); |
925baedd | 961 | |
7bb86316 CM |
962 | WARN_ON(root->ref_cows && trans->transid != |
963 | root->fs_info->running_transaction->transid); | |
6702ed49 | 964 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 965 | |
7bb86316 | 966 | level = btrfs_header_level(buf); |
31840ae1 | 967 | |
5d4f98a2 YZ |
968 | if (level == 0) |
969 | btrfs_item_key(buf, &disk_key, 0); | |
970 | else | |
971 | btrfs_node_key(buf, &disk_key, 0); | |
972 | ||
973 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
974 | if (parent) | |
975 | parent_start = parent->start; | |
976 | else | |
977 | parent_start = 0; | |
978 | } else | |
979 | parent_start = 0; | |
980 | ||
981 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
982 | root->root_key.objectid, &disk_key, | |
5581a51a | 983 | level, search_start, empty_size); |
54aa1f4d CM |
984 | if (IS_ERR(cow)) |
985 | return PTR_ERR(cow); | |
6702ed49 | 986 | |
b4ce94de CM |
987 | /* cow is set to blocking by btrfs_init_new_buffer */ |
988 | ||
5f39d397 | 989 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 990 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 991 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
992 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
993 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
994 | BTRFS_HEADER_FLAG_RELOC); | |
995 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
996 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
997 | else | |
998 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 999 | |
2b82032c YZ |
1000 | write_extent_buffer(cow, root->fs_info->fsid, |
1001 | (unsigned long)btrfs_header_fsid(cow), | |
1002 | BTRFS_FSID_SIZE); | |
1003 | ||
be1a5564 | 1004 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1005 | if (ret) { |
79787eaa | 1006 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1007 | return ret; |
1008 | } | |
1a40e23b | 1009 | |
3fd0a558 YZ |
1010 | if (root->ref_cows) |
1011 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
1012 | ||
02217ed2 | 1013 | if (buf == root->node) { |
925baedd | 1014 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1015 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1016 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1017 | parent_start = buf->start; | |
1018 | else | |
1019 | parent_start = 0; | |
925baedd | 1020 | |
5f39d397 | 1021 | extent_buffer_get(cow); |
90f8d62e | 1022 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1023 | rcu_assign_pointer(root->node, cow); |
925baedd | 1024 | |
f0486c68 | 1025 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1026 | last_ref); |
5f39d397 | 1027 | free_extent_buffer(buf); |
0b86a832 | 1028 | add_root_to_dirty_list(root); |
02217ed2 | 1029 | } else { |
5d4f98a2 YZ |
1030 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1031 | parent_start = parent->start; | |
1032 | else | |
1033 | parent_start = 0; | |
1034 | ||
1035 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1036 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1037 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1038 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1039 | cow->start); |
74493f7a CM |
1040 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1041 | trans->transid); | |
d6025579 | 1042 | btrfs_mark_buffer_dirty(parent); |
7fb7d76f JB |
1043 | if (last_ref) |
1044 | tree_mod_log_free_eb(root->fs_info, buf); | |
f0486c68 | 1045 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1046 | last_ref); |
02217ed2 | 1047 | } |
925baedd CM |
1048 | if (unlock_orig) |
1049 | btrfs_tree_unlock(buf); | |
3083ee2e | 1050 | free_extent_buffer_stale(buf); |
ccd467d6 | 1051 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1052 | *cow_ret = cow; |
02217ed2 CM |
1053 | return 0; |
1054 | } | |
1055 | ||
5d9e75c4 JS |
1056 | /* |
1057 | * returns the logical address of the oldest predecessor of the given root. | |
1058 | * entries older than time_seq are ignored. | |
1059 | */ | |
1060 | static struct tree_mod_elem * | |
1061 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1062 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1063 | { |
1064 | struct tree_mod_elem *tm; | |
1065 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1066 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1067 | int looped = 0; |
1068 | ||
1069 | if (!time_seq) | |
35a3621b | 1070 | return NULL; |
5d9e75c4 JS |
1071 | |
1072 | /* | |
1073 | * the very last operation that's logged for a root is the replacement | |
1074 | * operation (if it is replaced at all). this has the index of the *new* | |
1075 | * root, making it the very first operation that's logged for this root. | |
1076 | */ | |
1077 | while (1) { | |
1078 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1079 | time_seq); | |
1080 | if (!looped && !tm) | |
35a3621b | 1081 | return NULL; |
5d9e75c4 | 1082 | /* |
28da9fb4 JS |
1083 | * if there are no tree operation for the oldest root, we simply |
1084 | * return it. this should only happen if that (old) root is at | |
1085 | * level 0. | |
5d9e75c4 | 1086 | */ |
28da9fb4 JS |
1087 | if (!tm) |
1088 | break; | |
5d9e75c4 | 1089 | |
28da9fb4 JS |
1090 | /* |
1091 | * if there's an operation that's not a root replacement, we | |
1092 | * found the oldest version of our root. normally, we'll find a | |
1093 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1094 | */ | |
5d9e75c4 JS |
1095 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1096 | break; | |
1097 | ||
1098 | found = tm; | |
1099 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1100 | looped = 1; |
1101 | } | |
1102 | ||
a95236d9 JS |
1103 | /* if there's no old root to return, return what we found instead */ |
1104 | if (!found) | |
1105 | found = tm; | |
1106 | ||
5d9e75c4 JS |
1107 | return found; |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1112 | * previous operations will be rewinded (until we reach something older than | |
1113 | * time_seq). | |
1114 | */ | |
1115 | static void | |
f1ca7e98 JB |
1116 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1117 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1118 | { |
1119 | u32 n; | |
1120 | struct rb_node *next; | |
1121 | struct tree_mod_elem *tm = first_tm; | |
1122 | unsigned long o_dst; | |
1123 | unsigned long o_src; | |
1124 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1125 | ||
1126 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1127 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1128 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1129 | /* |
1130 | * all the operations are recorded with the operator used for | |
1131 | * the modification. as we're going backwards, we do the | |
1132 | * opposite of each operation here. | |
1133 | */ | |
1134 | switch (tm->op) { | |
1135 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1136 | BUG_ON(tm->slot < n); | |
1c697d4a | 1137 | /* Fallthrough */ |
95c80bb1 | 1138 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1139 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1140 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1141 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1142 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1143 | tm->generation); | |
4c3e6969 | 1144 | n++; |
5d9e75c4 JS |
1145 | break; |
1146 | case MOD_LOG_KEY_REPLACE: | |
1147 | BUG_ON(tm->slot >= n); | |
1148 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1149 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1150 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1151 | tm->generation); | |
1152 | break; | |
1153 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1154 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1155 | n--; |
1156 | break; | |
1157 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1158 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1159 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1160 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1161 | tm->move.nr_items * p_size); |
1162 | break; | |
1163 | case MOD_LOG_ROOT_REPLACE: | |
1164 | /* | |
1165 | * this operation is special. for roots, this must be | |
1166 | * handled explicitly before rewinding. | |
1167 | * for non-roots, this operation may exist if the node | |
1168 | * was a root: root A -> child B; then A gets empty and | |
1169 | * B is promoted to the new root. in the mod log, we'll | |
1170 | * have a root-replace operation for B, a tree block | |
1171 | * that is no root. we simply ignore that operation. | |
1172 | */ | |
1173 | break; | |
1174 | } | |
1175 | next = rb_next(&tm->node); | |
1176 | if (!next) | |
1177 | break; | |
1178 | tm = container_of(next, struct tree_mod_elem, node); | |
1179 | if (tm->index != first_tm->index) | |
1180 | break; | |
1181 | } | |
f1ca7e98 | 1182 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1183 | btrfs_set_header_nritems(eb, n); |
1184 | } | |
1185 | ||
47fb091f JS |
1186 | /* |
1187 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1188 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1189 | * returned buffer is always read-locked. If the returned buffer is not the | |
1190 | * input buffer, the lock on the input buffer is released and the input buffer | |
1191 | * is freed (its refcount is decremented). | |
1192 | */ | |
5d9e75c4 | 1193 | static struct extent_buffer * |
9ec72677 JB |
1194 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1195 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1196 | { |
1197 | struct extent_buffer *eb_rewin; | |
1198 | struct tree_mod_elem *tm; | |
1199 | ||
1200 | if (!time_seq) | |
1201 | return eb; | |
1202 | ||
1203 | if (btrfs_header_level(eb) == 0) | |
1204 | return eb; | |
1205 | ||
1206 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1207 | if (!tm) | |
1208 | return eb; | |
1209 | ||
9ec72677 JB |
1210 | btrfs_set_path_blocking(path); |
1211 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1212 | ||
5d9e75c4 JS |
1213 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1214 | BUG_ON(tm->slot != 0); | |
1215 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1216 | fs_info->tree_root->nodesize); | |
db7f3436 | 1217 | if (!eb_rewin) { |
9ec72677 | 1218 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1219 | free_extent_buffer(eb); |
1220 | return NULL; | |
1221 | } | |
5d9e75c4 JS |
1222 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1223 | btrfs_set_header_backref_rev(eb_rewin, | |
1224 | btrfs_header_backref_rev(eb)); | |
1225 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1226 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1227 | } else { |
1228 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1229 | if (!eb_rewin) { |
9ec72677 | 1230 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1231 | free_extent_buffer(eb); |
1232 | return NULL; | |
1233 | } | |
5d9e75c4 JS |
1234 | } |
1235 | ||
9ec72677 JB |
1236 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1237 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1238 | free_extent_buffer(eb); |
1239 | ||
47fb091f JS |
1240 | extent_buffer_get(eb_rewin); |
1241 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1242 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1243 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1244 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1245 | |
1246 | return eb_rewin; | |
1247 | } | |
1248 | ||
8ba97a15 JS |
1249 | /* |
1250 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1251 | * value. If there are no changes, the current root->root_node is returned. If | |
1252 | * anything changed in between, there's a fresh buffer allocated on which the | |
1253 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1254 | * Returns NULL on error (with no locks held). | |
1255 | */ | |
5d9e75c4 JS |
1256 | static inline struct extent_buffer * |
1257 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1258 | { | |
1259 | struct tree_mod_elem *tm; | |
30b0463a JS |
1260 | struct extent_buffer *eb = NULL; |
1261 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1262 | struct extent_buffer *old; |
a95236d9 | 1263 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1264 | u64 old_generation = 0; |
a95236d9 | 1265 | u64 logical; |
834328a8 | 1266 | u32 blocksize; |
5d9e75c4 | 1267 | |
30b0463a JS |
1268 | eb_root = btrfs_read_lock_root_node(root); |
1269 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1270 | if (!tm) |
30b0463a | 1271 | return eb_root; |
5d9e75c4 | 1272 | |
a95236d9 JS |
1273 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1274 | old_root = &tm->old_root; | |
1275 | old_generation = tm->generation; | |
1276 | logical = old_root->logical; | |
1277 | } else { | |
30b0463a | 1278 | logical = eb_root->start; |
a95236d9 | 1279 | } |
5d9e75c4 | 1280 | |
a95236d9 | 1281 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1282 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1283 | btrfs_tree_read_unlock(eb_root); |
1284 | free_extent_buffer(eb_root); | |
834328a8 | 1285 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1286 | old = read_tree_block(root, logical, blocksize, 0); |
416bc658 JB |
1287 | if (!old || !extent_buffer_uptodate(old)) { |
1288 | free_extent_buffer(old); | |
834328a8 JS |
1289 | pr_warn("btrfs: failed to read tree block %llu from get_old_root\n", |
1290 | logical); | |
1291 | WARN_ON(1); | |
1292 | } else { | |
7bfdcf7f LB |
1293 | eb = btrfs_clone_extent_buffer(old); |
1294 | free_extent_buffer(old); | |
834328a8 JS |
1295 | } |
1296 | } else if (old_root) { | |
30b0463a JS |
1297 | btrfs_tree_read_unlock(eb_root); |
1298 | free_extent_buffer(eb_root); | |
28da9fb4 | 1299 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1300 | } else { |
9ec72677 | 1301 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1302 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1303 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1304 | free_extent_buffer(eb_root); |
834328a8 JS |
1305 | } |
1306 | ||
8ba97a15 JS |
1307 | if (!eb) |
1308 | return NULL; | |
d6381084 | 1309 | extent_buffer_get(eb); |
8ba97a15 | 1310 | btrfs_tree_read_lock(eb); |
a95236d9 | 1311 | if (old_root) { |
5d9e75c4 JS |
1312 | btrfs_set_header_bytenr(eb, eb->start); |
1313 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1314 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1315 | btrfs_set_header_level(eb, old_root->level); |
1316 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1317 | } |
28da9fb4 | 1318 | if (tm) |
f1ca7e98 | 1319 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1320 | else |
1321 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1322 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1323 | |
1324 | return eb; | |
1325 | } | |
1326 | ||
5b6602e7 JS |
1327 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1328 | { | |
1329 | struct tree_mod_elem *tm; | |
1330 | int level; | |
30b0463a | 1331 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1332 | |
30b0463a | 1333 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1334 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1335 | level = tm->old_root.level; | |
1336 | } else { | |
30b0463a | 1337 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1338 | } |
30b0463a | 1339 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1340 | |
1341 | return level; | |
1342 | } | |
1343 | ||
5d4f98a2 YZ |
1344 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1345 | struct btrfs_root *root, | |
1346 | struct extent_buffer *buf) | |
1347 | { | |
f1ebcc74 LB |
1348 | /* ensure we can see the force_cow */ |
1349 | smp_rmb(); | |
1350 | ||
1351 | /* | |
1352 | * We do not need to cow a block if | |
1353 | * 1) this block is not created or changed in this transaction; | |
1354 | * 2) this block does not belong to TREE_RELOC tree; | |
1355 | * 3) the root is not forced COW. | |
1356 | * | |
1357 | * What is forced COW: | |
1358 | * when we create snapshot during commiting the transaction, | |
1359 | * after we've finished coping src root, we must COW the shared | |
1360 | * block to ensure the metadata consistency. | |
1361 | */ | |
5d4f98a2 YZ |
1362 | if (btrfs_header_generation(buf) == trans->transid && |
1363 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1364 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1365 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1366 | !root->force_cow) | |
5d4f98a2 YZ |
1367 | return 0; |
1368 | return 1; | |
1369 | } | |
1370 | ||
d352ac68 CM |
1371 | /* |
1372 | * cows a single block, see __btrfs_cow_block for the real work. | |
1373 | * This version of it has extra checks so that a block isn't cow'd more than | |
1374 | * once per transaction, as long as it hasn't been written yet | |
1375 | */ | |
d397712b | 1376 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1377 | struct btrfs_root *root, struct extent_buffer *buf, |
1378 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1379 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1380 | { |
1381 | u64 search_start; | |
f510cfec | 1382 | int ret; |
dc17ff8f | 1383 | |
31b1a2bd JL |
1384 | if (trans->transaction != root->fs_info->running_transaction) |
1385 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1386 | trans->transid, |
6702ed49 | 1387 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1388 | |
1389 | if (trans->transid != root->fs_info->generation) | |
1390 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1391 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1392 | |
5d4f98a2 | 1393 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1394 | *cow_ret = buf; |
1395 | return 0; | |
1396 | } | |
c487685d | 1397 | |
0b86a832 | 1398 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1399 | |
1400 | if (parent) | |
1401 | btrfs_set_lock_blocking(parent); | |
1402 | btrfs_set_lock_blocking(buf); | |
1403 | ||
f510cfec | 1404 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1405 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1406 | |
1407 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1408 | ||
f510cfec | 1409 | return ret; |
6702ed49 CM |
1410 | } |
1411 | ||
d352ac68 CM |
1412 | /* |
1413 | * helper function for defrag to decide if two blocks pointed to by a | |
1414 | * node are actually close by | |
1415 | */ | |
6b80053d | 1416 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1417 | { |
6b80053d | 1418 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1419 | return 1; |
6b80053d | 1420 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1421 | return 1; |
1422 | return 0; | |
1423 | } | |
1424 | ||
081e9573 CM |
1425 | /* |
1426 | * compare two keys in a memcmp fashion | |
1427 | */ | |
1428 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1429 | { | |
1430 | struct btrfs_key k1; | |
1431 | ||
1432 | btrfs_disk_key_to_cpu(&k1, disk); | |
1433 | ||
20736aba | 1434 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1435 | } |
1436 | ||
f3465ca4 JB |
1437 | /* |
1438 | * same as comp_keys only with two btrfs_key's | |
1439 | */ | |
5d4f98a2 | 1440 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1441 | { |
1442 | if (k1->objectid > k2->objectid) | |
1443 | return 1; | |
1444 | if (k1->objectid < k2->objectid) | |
1445 | return -1; | |
1446 | if (k1->type > k2->type) | |
1447 | return 1; | |
1448 | if (k1->type < k2->type) | |
1449 | return -1; | |
1450 | if (k1->offset > k2->offset) | |
1451 | return 1; | |
1452 | if (k1->offset < k2->offset) | |
1453 | return -1; | |
1454 | return 0; | |
1455 | } | |
081e9573 | 1456 | |
d352ac68 CM |
1457 | /* |
1458 | * this is used by the defrag code to go through all the | |
1459 | * leaves pointed to by a node and reallocate them so that | |
1460 | * disk order is close to key order | |
1461 | */ | |
6702ed49 | 1462 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1463 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1464 | int start_slot, u64 *last_ret, |
a6b6e75e | 1465 | struct btrfs_key *progress) |
6702ed49 | 1466 | { |
6b80053d | 1467 | struct extent_buffer *cur; |
6702ed49 | 1468 | u64 blocknr; |
ca7a79ad | 1469 | u64 gen; |
e9d0b13b CM |
1470 | u64 search_start = *last_ret; |
1471 | u64 last_block = 0; | |
6702ed49 CM |
1472 | u64 other; |
1473 | u32 parent_nritems; | |
6702ed49 CM |
1474 | int end_slot; |
1475 | int i; | |
1476 | int err = 0; | |
f2183bde | 1477 | int parent_level; |
6b80053d CM |
1478 | int uptodate; |
1479 | u32 blocksize; | |
081e9573 CM |
1480 | int progress_passed = 0; |
1481 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1482 | |
5708b959 | 1483 | parent_level = btrfs_header_level(parent); |
5708b959 | 1484 | |
6c1500f2 JL |
1485 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1486 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1487 | |
6b80053d | 1488 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1489 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1490 | end_slot = parent_nritems; |
1491 | ||
1492 | if (parent_nritems == 1) | |
1493 | return 0; | |
1494 | ||
b4ce94de CM |
1495 | btrfs_set_lock_blocking(parent); |
1496 | ||
6702ed49 CM |
1497 | for (i = start_slot; i < end_slot; i++) { |
1498 | int close = 1; | |
a6b6e75e | 1499 | |
081e9573 CM |
1500 | btrfs_node_key(parent, &disk_key, i); |
1501 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1502 | continue; | |
1503 | ||
1504 | progress_passed = 1; | |
6b80053d | 1505 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1506 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1507 | if (last_block == 0) |
1508 | last_block = blocknr; | |
5708b959 | 1509 | |
6702ed49 | 1510 | if (i > 0) { |
6b80053d CM |
1511 | other = btrfs_node_blockptr(parent, i - 1); |
1512 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1513 | } |
0ef3e66b | 1514 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1515 | other = btrfs_node_blockptr(parent, i + 1); |
1516 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1517 | } |
e9d0b13b CM |
1518 | if (close) { |
1519 | last_block = blocknr; | |
6702ed49 | 1520 | continue; |
e9d0b13b | 1521 | } |
6702ed49 | 1522 | |
6b80053d CM |
1523 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1524 | if (cur) | |
b9fab919 | 1525 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1526 | else |
1527 | uptodate = 0; | |
5708b959 | 1528 | if (!cur || !uptodate) { |
6b80053d CM |
1529 | if (!cur) { |
1530 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1531 | blocksize, gen); |
416bc658 JB |
1532 | if (!cur || !extent_buffer_uptodate(cur)) { |
1533 | free_extent_buffer(cur); | |
97d9a8a4 | 1534 | return -EIO; |
416bc658 | 1535 | } |
6b80053d | 1536 | } else if (!uptodate) { |
018642a1 TI |
1537 | err = btrfs_read_buffer(cur, gen); |
1538 | if (err) { | |
1539 | free_extent_buffer(cur); | |
1540 | return err; | |
1541 | } | |
f2183bde | 1542 | } |
6702ed49 | 1543 | } |
e9d0b13b | 1544 | if (search_start == 0) |
6b80053d | 1545 | search_start = last_block; |
e9d0b13b | 1546 | |
e7a84565 | 1547 | btrfs_tree_lock(cur); |
b4ce94de | 1548 | btrfs_set_lock_blocking(cur); |
6b80053d | 1549 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1550 | &cur, search_start, |
6b80053d | 1551 | min(16 * blocksize, |
9fa8cfe7 | 1552 | (end_slot - i) * blocksize)); |
252c38f0 | 1553 | if (err) { |
e7a84565 | 1554 | btrfs_tree_unlock(cur); |
6b80053d | 1555 | free_extent_buffer(cur); |
6702ed49 | 1556 | break; |
252c38f0 | 1557 | } |
e7a84565 CM |
1558 | search_start = cur->start; |
1559 | last_block = cur->start; | |
f2183bde | 1560 | *last_ret = search_start; |
e7a84565 CM |
1561 | btrfs_tree_unlock(cur); |
1562 | free_extent_buffer(cur); | |
6702ed49 CM |
1563 | } |
1564 | return err; | |
1565 | } | |
1566 | ||
74123bd7 CM |
1567 | /* |
1568 | * The leaf data grows from end-to-front in the node. | |
1569 | * this returns the address of the start of the last item, | |
1570 | * which is the stop of the leaf data stack | |
1571 | */ | |
123abc88 | 1572 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1573 | struct extent_buffer *leaf) |
be0e5c09 | 1574 | { |
5f39d397 | 1575 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1576 | if (nr == 0) |
123abc88 | 1577 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1578 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1579 | } |
1580 | ||
aa5d6bed | 1581 | |
74123bd7 | 1582 | /* |
5f39d397 CM |
1583 | * search for key in the extent_buffer. The items start at offset p, |
1584 | * and they are item_size apart. There are 'max' items in p. | |
1585 | * | |
74123bd7 CM |
1586 | * the slot in the array is returned via slot, and it points to |
1587 | * the place where you would insert key if it is not found in | |
1588 | * the array. | |
1589 | * | |
1590 | * slot may point to max if the key is bigger than all of the keys | |
1591 | */ | |
e02119d5 CM |
1592 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1593 | unsigned long p, | |
1594 | int item_size, struct btrfs_key *key, | |
1595 | int max, int *slot) | |
be0e5c09 CM |
1596 | { |
1597 | int low = 0; | |
1598 | int high = max; | |
1599 | int mid; | |
1600 | int ret; | |
479965d6 | 1601 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1602 | struct btrfs_disk_key unaligned; |
1603 | unsigned long offset; | |
5f39d397 CM |
1604 | char *kaddr = NULL; |
1605 | unsigned long map_start = 0; | |
1606 | unsigned long map_len = 0; | |
479965d6 | 1607 | int err; |
be0e5c09 | 1608 | |
d397712b | 1609 | while (low < high) { |
be0e5c09 | 1610 | mid = (low + high) / 2; |
5f39d397 CM |
1611 | offset = p + mid * item_size; |
1612 | ||
a6591715 | 1613 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1614 | (offset + sizeof(struct btrfs_disk_key)) > |
1615 | map_start + map_len) { | |
934d375b CM |
1616 | |
1617 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1618 | sizeof(struct btrfs_disk_key), |
a6591715 | 1619 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1620 | |
1621 | if (!err) { | |
1622 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1623 | map_start); | |
1624 | } else { | |
1625 | read_extent_buffer(eb, &unaligned, | |
1626 | offset, sizeof(unaligned)); | |
1627 | tmp = &unaligned; | |
1628 | } | |
5f39d397 | 1629 | |
5f39d397 CM |
1630 | } else { |
1631 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1632 | map_start); | |
1633 | } | |
be0e5c09 CM |
1634 | ret = comp_keys(tmp, key); |
1635 | ||
1636 | if (ret < 0) | |
1637 | low = mid + 1; | |
1638 | else if (ret > 0) | |
1639 | high = mid; | |
1640 | else { | |
1641 | *slot = mid; | |
1642 | return 0; | |
1643 | } | |
1644 | } | |
1645 | *slot = low; | |
1646 | return 1; | |
1647 | } | |
1648 | ||
97571fd0 CM |
1649 | /* |
1650 | * simple bin_search frontend that does the right thing for | |
1651 | * leaves vs nodes | |
1652 | */ | |
5f39d397 CM |
1653 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1654 | int level, int *slot) | |
be0e5c09 | 1655 | { |
f775738f | 1656 | if (level == 0) |
5f39d397 CM |
1657 | return generic_bin_search(eb, |
1658 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1659 | sizeof(struct btrfs_item), |
5f39d397 | 1660 | key, btrfs_header_nritems(eb), |
7518a238 | 1661 | slot); |
f775738f | 1662 | else |
5f39d397 CM |
1663 | return generic_bin_search(eb, |
1664 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1665 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1666 | key, btrfs_header_nritems(eb), |
7518a238 | 1667 | slot); |
be0e5c09 CM |
1668 | } |
1669 | ||
5d4f98a2 YZ |
1670 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1671 | int level, int *slot) | |
1672 | { | |
1673 | return bin_search(eb, key, level, slot); | |
1674 | } | |
1675 | ||
f0486c68 YZ |
1676 | static void root_add_used(struct btrfs_root *root, u32 size) |
1677 | { | |
1678 | spin_lock(&root->accounting_lock); | |
1679 | btrfs_set_root_used(&root->root_item, | |
1680 | btrfs_root_used(&root->root_item) + size); | |
1681 | spin_unlock(&root->accounting_lock); | |
1682 | } | |
1683 | ||
1684 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1685 | { | |
1686 | spin_lock(&root->accounting_lock); | |
1687 | btrfs_set_root_used(&root->root_item, | |
1688 | btrfs_root_used(&root->root_item) - size); | |
1689 | spin_unlock(&root->accounting_lock); | |
1690 | } | |
1691 | ||
d352ac68 CM |
1692 | /* given a node and slot number, this reads the blocks it points to. The |
1693 | * extent buffer is returned with a reference taken (but unlocked). | |
1694 | * NULL is returned on error. | |
1695 | */ | |
e02119d5 | 1696 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1697 | struct extent_buffer *parent, int slot) |
bb803951 | 1698 | { |
ca7a79ad | 1699 | int level = btrfs_header_level(parent); |
416bc658 JB |
1700 | struct extent_buffer *eb; |
1701 | ||
bb803951 CM |
1702 | if (slot < 0) |
1703 | return NULL; | |
5f39d397 | 1704 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1705 | return NULL; |
ca7a79ad CM |
1706 | |
1707 | BUG_ON(level == 0); | |
1708 | ||
416bc658 JB |
1709 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1710 | btrfs_level_size(root, level - 1), | |
1711 | btrfs_node_ptr_generation(parent, slot)); | |
1712 | if (eb && !extent_buffer_uptodate(eb)) { | |
1713 | free_extent_buffer(eb); | |
1714 | eb = NULL; | |
1715 | } | |
1716 | ||
1717 | return eb; | |
bb803951 CM |
1718 | } |
1719 | ||
d352ac68 CM |
1720 | /* |
1721 | * node level balancing, used to make sure nodes are in proper order for | |
1722 | * item deletion. We balance from the top down, so we have to make sure | |
1723 | * that a deletion won't leave an node completely empty later on. | |
1724 | */ | |
e02119d5 | 1725 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1726 | struct btrfs_root *root, |
1727 | struct btrfs_path *path, int level) | |
bb803951 | 1728 | { |
5f39d397 CM |
1729 | struct extent_buffer *right = NULL; |
1730 | struct extent_buffer *mid; | |
1731 | struct extent_buffer *left = NULL; | |
1732 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1733 | int ret = 0; |
1734 | int wret; | |
1735 | int pslot; | |
bb803951 | 1736 | int orig_slot = path->slots[level]; |
79f95c82 | 1737 | u64 orig_ptr; |
bb803951 CM |
1738 | |
1739 | if (level == 0) | |
1740 | return 0; | |
1741 | ||
5f39d397 | 1742 | mid = path->nodes[level]; |
b4ce94de | 1743 | |
bd681513 CM |
1744 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1745 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1746 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1747 | ||
1d4f8a0c | 1748 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1749 | |
a05a9bb1 | 1750 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1751 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1752 | pslot = path->slots[level + 1]; |
1753 | } | |
bb803951 | 1754 | |
40689478 CM |
1755 | /* |
1756 | * deal with the case where there is only one pointer in the root | |
1757 | * by promoting the node below to a root | |
1758 | */ | |
5f39d397 CM |
1759 | if (!parent) { |
1760 | struct extent_buffer *child; | |
bb803951 | 1761 | |
5f39d397 | 1762 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1763 | return 0; |
1764 | ||
1765 | /* promote the child to a root */ | |
5f39d397 | 1766 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1767 | if (!child) { |
1768 | ret = -EROFS; | |
1769 | btrfs_std_error(root->fs_info, ret); | |
1770 | goto enospc; | |
1771 | } | |
1772 | ||
925baedd | 1773 | btrfs_tree_lock(child); |
b4ce94de | 1774 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1775 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1776 | if (ret) { |
1777 | btrfs_tree_unlock(child); | |
1778 | free_extent_buffer(child); | |
1779 | goto enospc; | |
1780 | } | |
2f375ab9 | 1781 | |
90f8d62e | 1782 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1783 | rcu_assign_pointer(root->node, child); |
925baedd | 1784 | |
0b86a832 | 1785 | add_root_to_dirty_list(root); |
925baedd | 1786 | btrfs_tree_unlock(child); |
b4ce94de | 1787 | |
925baedd | 1788 | path->locks[level] = 0; |
bb803951 | 1789 | path->nodes[level] = NULL; |
5f39d397 | 1790 | clean_tree_block(trans, root, mid); |
925baedd | 1791 | btrfs_tree_unlock(mid); |
bb803951 | 1792 | /* once for the path */ |
5f39d397 | 1793 | free_extent_buffer(mid); |
f0486c68 YZ |
1794 | |
1795 | root_sub_used(root, mid->len); | |
5581a51a | 1796 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1797 | /* once for the root ptr */ |
3083ee2e | 1798 | free_extent_buffer_stale(mid); |
f0486c68 | 1799 | return 0; |
bb803951 | 1800 | } |
5f39d397 | 1801 | if (btrfs_header_nritems(mid) > |
123abc88 | 1802 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1803 | return 0; |
1804 | ||
5f39d397 CM |
1805 | left = read_node_slot(root, parent, pslot - 1); |
1806 | if (left) { | |
925baedd | 1807 | btrfs_tree_lock(left); |
b4ce94de | 1808 | btrfs_set_lock_blocking(left); |
5f39d397 | 1809 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1810 | parent, pslot - 1, &left); |
54aa1f4d CM |
1811 | if (wret) { |
1812 | ret = wret; | |
1813 | goto enospc; | |
1814 | } | |
2cc58cf2 | 1815 | } |
5f39d397 CM |
1816 | right = read_node_slot(root, parent, pslot + 1); |
1817 | if (right) { | |
925baedd | 1818 | btrfs_tree_lock(right); |
b4ce94de | 1819 | btrfs_set_lock_blocking(right); |
5f39d397 | 1820 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1821 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1822 | if (wret) { |
1823 | ret = wret; | |
1824 | goto enospc; | |
1825 | } | |
1826 | } | |
1827 | ||
1828 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1829 | if (left) { |
1830 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1831 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1832 | if (wret < 0) |
1833 | ret = wret; | |
bb803951 | 1834 | } |
79f95c82 CM |
1835 | |
1836 | /* | |
1837 | * then try to empty the right most buffer into the middle | |
1838 | */ | |
5f39d397 | 1839 | if (right) { |
971a1f66 | 1840 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1841 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1842 | ret = wret; |
5f39d397 | 1843 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1844 | clean_tree_block(trans, root, right); |
925baedd | 1845 | btrfs_tree_unlock(right); |
afe5fea7 | 1846 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1847 | root_sub_used(root, right->len); |
5581a51a | 1848 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1849 | free_extent_buffer_stale(right); |
f0486c68 | 1850 | right = NULL; |
bb803951 | 1851 | } else { |
5f39d397 CM |
1852 | struct btrfs_disk_key right_key; |
1853 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 1854 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1855 | pslot + 1, 0); |
5f39d397 CM |
1856 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1857 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1858 | } |
1859 | } | |
5f39d397 | 1860 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1861 | /* |
1862 | * we're not allowed to leave a node with one item in the | |
1863 | * tree during a delete. A deletion from lower in the tree | |
1864 | * could try to delete the only pointer in this node. | |
1865 | * So, pull some keys from the left. | |
1866 | * There has to be a left pointer at this point because | |
1867 | * otherwise we would have pulled some pointers from the | |
1868 | * right | |
1869 | */ | |
305a26af MF |
1870 | if (!left) { |
1871 | ret = -EROFS; | |
1872 | btrfs_std_error(root->fs_info, ret); | |
1873 | goto enospc; | |
1874 | } | |
5f39d397 | 1875 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1876 | if (wret < 0) { |
79f95c82 | 1877 | ret = wret; |
54aa1f4d CM |
1878 | goto enospc; |
1879 | } | |
bce4eae9 CM |
1880 | if (wret == 1) { |
1881 | wret = push_node_left(trans, root, left, mid, 1); | |
1882 | if (wret < 0) | |
1883 | ret = wret; | |
1884 | } | |
79f95c82 CM |
1885 | BUG_ON(wret == 1); |
1886 | } | |
5f39d397 | 1887 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1888 | clean_tree_block(trans, root, mid); |
925baedd | 1889 | btrfs_tree_unlock(mid); |
afe5fea7 | 1890 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1891 | root_sub_used(root, mid->len); |
5581a51a | 1892 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1893 | free_extent_buffer_stale(mid); |
f0486c68 | 1894 | mid = NULL; |
79f95c82 CM |
1895 | } else { |
1896 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1897 | struct btrfs_disk_key mid_key; |
1898 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 1899 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 1900 | pslot, 0); |
5f39d397 CM |
1901 | btrfs_set_node_key(parent, &mid_key, pslot); |
1902 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1903 | } |
bb803951 | 1904 | |
79f95c82 | 1905 | /* update the path */ |
5f39d397 CM |
1906 | if (left) { |
1907 | if (btrfs_header_nritems(left) > orig_slot) { | |
1908 | extent_buffer_get(left); | |
925baedd | 1909 | /* left was locked after cow */ |
5f39d397 | 1910 | path->nodes[level] = left; |
bb803951 CM |
1911 | path->slots[level + 1] -= 1; |
1912 | path->slots[level] = orig_slot; | |
925baedd CM |
1913 | if (mid) { |
1914 | btrfs_tree_unlock(mid); | |
5f39d397 | 1915 | free_extent_buffer(mid); |
925baedd | 1916 | } |
bb803951 | 1917 | } else { |
5f39d397 | 1918 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1919 | path->slots[level] = orig_slot; |
1920 | } | |
1921 | } | |
79f95c82 | 1922 | /* double check we haven't messed things up */ |
e20d96d6 | 1923 | if (orig_ptr != |
5f39d397 | 1924 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1925 | BUG(); |
54aa1f4d | 1926 | enospc: |
925baedd CM |
1927 | if (right) { |
1928 | btrfs_tree_unlock(right); | |
5f39d397 | 1929 | free_extent_buffer(right); |
925baedd CM |
1930 | } |
1931 | if (left) { | |
1932 | if (path->nodes[level] != left) | |
1933 | btrfs_tree_unlock(left); | |
5f39d397 | 1934 | free_extent_buffer(left); |
925baedd | 1935 | } |
bb803951 CM |
1936 | return ret; |
1937 | } | |
1938 | ||
d352ac68 CM |
1939 | /* Node balancing for insertion. Here we only split or push nodes around |
1940 | * when they are completely full. This is also done top down, so we | |
1941 | * have to be pessimistic. | |
1942 | */ | |
d397712b | 1943 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1944 | struct btrfs_root *root, |
1945 | struct btrfs_path *path, int level) | |
e66f709b | 1946 | { |
5f39d397 CM |
1947 | struct extent_buffer *right = NULL; |
1948 | struct extent_buffer *mid; | |
1949 | struct extent_buffer *left = NULL; | |
1950 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1951 | int ret = 0; |
1952 | int wret; | |
1953 | int pslot; | |
1954 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1955 | |
1956 | if (level == 0) | |
1957 | return 1; | |
1958 | ||
5f39d397 | 1959 | mid = path->nodes[level]; |
7bb86316 | 1960 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1961 | |
a05a9bb1 | 1962 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1963 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1964 | pslot = path->slots[level + 1]; |
1965 | } | |
e66f709b | 1966 | |
5f39d397 | 1967 | if (!parent) |
e66f709b | 1968 | return 1; |
e66f709b | 1969 | |
5f39d397 | 1970 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1971 | |
1972 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1973 | if (left) { |
e66f709b | 1974 | u32 left_nr; |
925baedd CM |
1975 | |
1976 | btrfs_tree_lock(left); | |
b4ce94de CM |
1977 | btrfs_set_lock_blocking(left); |
1978 | ||
5f39d397 | 1979 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1980 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1981 | wret = 1; | |
1982 | } else { | |
5f39d397 | 1983 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1984 | pslot - 1, &left); |
54aa1f4d CM |
1985 | if (ret) |
1986 | wret = 1; | |
1987 | else { | |
54aa1f4d | 1988 | wret = push_node_left(trans, root, |
971a1f66 | 1989 | left, mid, 0); |
54aa1f4d | 1990 | } |
33ade1f8 | 1991 | } |
e66f709b CM |
1992 | if (wret < 0) |
1993 | ret = wret; | |
1994 | if (wret == 0) { | |
5f39d397 | 1995 | struct btrfs_disk_key disk_key; |
e66f709b | 1996 | orig_slot += left_nr; |
5f39d397 | 1997 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 1998 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1999 | pslot, 0); |
5f39d397 CM |
2000 | btrfs_set_node_key(parent, &disk_key, pslot); |
2001 | btrfs_mark_buffer_dirty(parent); | |
2002 | if (btrfs_header_nritems(left) > orig_slot) { | |
2003 | path->nodes[level] = left; | |
e66f709b CM |
2004 | path->slots[level + 1] -= 1; |
2005 | path->slots[level] = orig_slot; | |
925baedd | 2006 | btrfs_tree_unlock(mid); |
5f39d397 | 2007 | free_extent_buffer(mid); |
e66f709b CM |
2008 | } else { |
2009 | orig_slot -= | |
5f39d397 | 2010 | btrfs_header_nritems(left); |
e66f709b | 2011 | path->slots[level] = orig_slot; |
925baedd | 2012 | btrfs_tree_unlock(left); |
5f39d397 | 2013 | free_extent_buffer(left); |
e66f709b | 2014 | } |
e66f709b CM |
2015 | return 0; |
2016 | } | |
925baedd | 2017 | btrfs_tree_unlock(left); |
5f39d397 | 2018 | free_extent_buffer(left); |
e66f709b | 2019 | } |
925baedd | 2020 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2021 | |
2022 | /* | |
2023 | * then try to empty the right most buffer into the middle | |
2024 | */ | |
5f39d397 | 2025 | if (right) { |
33ade1f8 | 2026 | u32 right_nr; |
b4ce94de | 2027 | |
925baedd | 2028 | btrfs_tree_lock(right); |
b4ce94de CM |
2029 | btrfs_set_lock_blocking(right); |
2030 | ||
5f39d397 | 2031 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2032 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2033 | wret = 1; | |
2034 | } else { | |
5f39d397 CM |
2035 | ret = btrfs_cow_block(trans, root, right, |
2036 | parent, pslot + 1, | |
9fa8cfe7 | 2037 | &right); |
54aa1f4d CM |
2038 | if (ret) |
2039 | wret = 1; | |
2040 | else { | |
54aa1f4d | 2041 | wret = balance_node_right(trans, root, |
5f39d397 | 2042 | right, mid); |
54aa1f4d | 2043 | } |
33ade1f8 | 2044 | } |
e66f709b CM |
2045 | if (wret < 0) |
2046 | ret = wret; | |
2047 | if (wret == 0) { | |
5f39d397 CM |
2048 | struct btrfs_disk_key disk_key; |
2049 | ||
2050 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2051 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2052 | pslot + 1, 0); |
5f39d397 CM |
2053 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2054 | btrfs_mark_buffer_dirty(parent); | |
2055 | ||
2056 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2057 | path->nodes[level] = right; | |
e66f709b CM |
2058 | path->slots[level + 1] += 1; |
2059 | path->slots[level] = orig_slot - | |
5f39d397 | 2060 | btrfs_header_nritems(mid); |
925baedd | 2061 | btrfs_tree_unlock(mid); |
5f39d397 | 2062 | free_extent_buffer(mid); |
e66f709b | 2063 | } else { |
925baedd | 2064 | btrfs_tree_unlock(right); |
5f39d397 | 2065 | free_extent_buffer(right); |
e66f709b | 2066 | } |
e66f709b CM |
2067 | return 0; |
2068 | } | |
925baedd | 2069 | btrfs_tree_unlock(right); |
5f39d397 | 2070 | free_extent_buffer(right); |
e66f709b | 2071 | } |
e66f709b CM |
2072 | return 1; |
2073 | } | |
2074 | ||
3c69faec | 2075 | /* |
d352ac68 CM |
2076 | * readahead one full node of leaves, finding things that are close |
2077 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2078 | */ |
c8c42864 CM |
2079 | static void reada_for_search(struct btrfs_root *root, |
2080 | struct btrfs_path *path, | |
2081 | int level, int slot, u64 objectid) | |
3c69faec | 2082 | { |
5f39d397 | 2083 | struct extent_buffer *node; |
01f46658 | 2084 | struct btrfs_disk_key disk_key; |
3c69faec | 2085 | u32 nritems; |
3c69faec | 2086 | u64 search; |
a7175319 | 2087 | u64 target; |
6b80053d | 2088 | u64 nread = 0; |
cb25c2ea | 2089 | u64 gen; |
3c69faec | 2090 | int direction = path->reada; |
5f39d397 | 2091 | struct extent_buffer *eb; |
6b80053d CM |
2092 | u32 nr; |
2093 | u32 blocksize; | |
2094 | u32 nscan = 0; | |
db94535d | 2095 | |
a6b6e75e | 2096 | if (level != 1) |
6702ed49 CM |
2097 | return; |
2098 | ||
2099 | if (!path->nodes[level]) | |
3c69faec CM |
2100 | return; |
2101 | ||
5f39d397 | 2102 | node = path->nodes[level]; |
925baedd | 2103 | |
3c69faec | 2104 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2105 | blocksize = btrfs_level_size(root, level - 1); |
2106 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2107 | if (eb) { |
2108 | free_extent_buffer(eb); | |
3c69faec CM |
2109 | return; |
2110 | } | |
2111 | ||
a7175319 | 2112 | target = search; |
6b80053d | 2113 | |
5f39d397 | 2114 | nritems = btrfs_header_nritems(node); |
6b80053d | 2115 | nr = slot; |
25b8b936 | 2116 | |
d397712b | 2117 | while (1) { |
6b80053d CM |
2118 | if (direction < 0) { |
2119 | if (nr == 0) | |
2120 | break; | |
2121 | nr--; | |
2122 | } else if (direction > 0) { | |
2123 | nr++; | |
2124 | if (nr >= nritems) | |
2125 | break; | |
3c69faec | 2126 | } |
01f46658 CM |
2127 | if (path->reada < 0 && objectid) { |
2128 | btrfs_node_key(node, &disk_key, nr); | |
2129 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2130 | break; | |
2131 | } | |
6b80053d | 2132 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2133 | if ((search <= target && target - search <= 65536) || |
2134 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2135 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2136 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2137 | nread += blocksize; |
2138 | } | |
2139 | nscan++; | |
a7175319 | 2140 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2141 | break; |
3c69faec CM |
2142 | } |
2143 | } | |
925baedd | 2144 | |
0b08851f JB |
2145 | static noinline void reada_for_balance(struct btrfs_root *root, |
2146 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2147 | { |
2148 | int slot; | |
2149 | int nritems; | |
2150 | struct extent_buffer *parent; | |
2151 | struct extent_buffer *eb; | |
2152 | u64 gen; | |
2153 | u64 block1 = 0; | |
2154 | u64 block2 = 0; | |
b4ce94de CM |
2155 | int blocksize; |
2156 | ||
8c594ea8 | 2157 | parent = path->nodes[level + 1]; |
b4ce94de | 2158 | if (!parent) |
0b08851f | 2159 | return; |
b4ce94de CM |
2160 | |
2161 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2162 | slot = path->slots[level + 1]; |
b4ce94de CM |
2163 | blocksize = btrfs_level_size(root, level); |
2164 | ||
2165 | if (slot > 0) { | |
2166 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2167 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2168 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2169 | /* |
2170 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2171 | * don't want to return eagain here. That will loop | |
2172 | * forever | |
2173 | */ | |
2174 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2175 | block1 = 0; |
2176 | free_extent_buffer(eb); | |
2177 | } | |
8c594ea8 | 2178 | if (slot + 1 < nritems) { |
b4ce94de CM |
2179 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2180 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2181 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2182 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2183 | block2 = 0; |
2184 | free_extent_buffer(eb); | |
2185 | } | |
8c594ea8 | 2186 | |
0b08851f JB |
2187 | if (block1) |
2188 | readahead_tree_block(root, block1, blocksize, 0); | |
2189 | if (block2) | |
2190 | readahead_tree_block(root, block2, blocksize, 0); | |
b4ce94de CM |
2191 | } |
2192 | ||
2193 | ||
d352ac68 | 2194 | /* |
d397712b CM |
2195 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2196 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2197 | * operations on the tree might require changing key pointers higher up in the | |
2198 | * tree. | |
d352ac68 | 2199 | * |
d397712b CM |
2200 | * callers might also have set path->keep_locks, which tells this code to keep |
2201 | * the lock if the path points to the last slot in the block. This is part of | |
2202 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2203 | * |
d397712b CM |
2204 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2205 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2206 | */ |
e02119d5 | 2207 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2208 | int lowest_unlock, int min_write_lock_level, |
2209 | int *write_lock_level) | |
925baedd CM |
2210 | { |
2211 | int i; | |
2212 | int skip_level = level; | |
051e1b9f | 2213 | int no_skips = 0; |
925baedd CM |
2214 | struct extent_buffer *t; |
2215 | ||
2216 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2217 | if (!path->nodes[i]) | |
2218 | break; | |
2219 | if (!path->locks[i]) | |
2220 | break; | |
051e1b9f | 2221 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2222 | skip_level = i + 1; |
2223 | continue; | |
2224 | } | |
051e1b9f | 2225 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2226 | u32 nritems; |
2227 | t = path->nodes[i]; | |
2228 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2229 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2230 | skip_level = i + 1; |
2231 | continue; | |
2232 | } | |
2233 | } | |
051e1b9f CM |
2234 | if (skip_level < i && i >= lowest_unlock) |
2235 | no_skips = 1; | |
2236 | ||
925baedd CM |
2237 | t = path->nodes[i]; |
2238 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2239 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2240 | path->locks[i] = 0; |
f7c79f30 CM |
2241 | if (write_lock_level && |
2242 | i > min_write_lock_level && | |
2243 | i <= *write_lock_level) { | |
2244 | *write_lock_level = i - 1; | |
2245 | } | |
925baedd CM |
2246 | } |
2247 | } | |
2248 | } | |
2249 | ||
b4ce94de CM |
2250 | /* |
2251 | * This releases any locks held in the path starting at level and | |
2252 | * going all the way up to the root. | |
2253 | * | |
2254 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2255 | * corner cases, such as COW of the block at slot zero in the node. This | |
2256 | * ignores those rules, and it should only be called when there are no | |
2257 | * more updates to be done higher up in the tree. | |
2258 | */ | |
2259 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2260 | { | |
2261 | int i; | |
2262 | ||
09a2a8f9 | 2263 | if (path->keep_locks) |
b4ce94de CM |
2264 | return; |
2265 | ||
2266 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2267 | if (!path->nodes[i]) | |
12f4dacc | 2268 | continue; |
b4ce94de | 2269 | if (!path->locks[i]) |
12f4dacc | 2270 | continue; |
bd681513 | 2271 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2272 | path->locks[i] = 0; |
2273 | } | |
2274 | } | |
2275 | ||
c8c42864 CM |
2276 | /* |
2277 | * helper function for btrfs_search_slot. The goal is to find a block | |
2278 | * in cache without setting the path to blocking. If we find the block | |
2279 | * we return zero and the path is unchanged. | |
2280 | * | |
2281 | * If we can't find the block, we set the path blocking and do some | |
2282 | * reada. -EAGAIN is returned and the search must be repeated. | |
2283 | */ | |
2284 | static int | |
2285 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2286 | struct btrfs_root *root, struct btrfs_path *p, | |
2287 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2288 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2289 | { |
2290 | u64 blocknr; | |
2291 | u64 gen; | |
2292 | u32 blocksize; | |
2293 | struct extent_buffer *b = *eb_ret; | |
2294 | struct extent_buffer *tmp; | |
76a05b35 | 2295 | int ret; |
c8c42864 CM |
2296 | |
2297 | blocknr = btrfs_node_blockptr(b, slot); | |
2298 | gen = btrfs_node_ptr_generation(b, slot); | |
2299 | blocksize = btrfs_level_size(root, level - 1); | |
2300 | ||
2301 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2302 | if (tmp) { |
b9fab919 | 2303 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2304 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2305 | *eb_ret = tmp; | |
2306 | return 0; | |
2307 | } | |
2308 | ||
2309 | /* the pages were up to date, but we failed | |
2310 | * the generation number check. Do a full | |
2311 | * read for the generation number that is correct. | |
2312 | * We must do this without dropping locks so | |
2313 | * we can trust our generation number | |
2314 | */ | |
2315 | btrfs_set_path_blocking(p); | |
2316 | ||
2317 | /* now we're allowed to do a blocking uptodate check */ | |
2318 | ret = btrfs_read_buffer(tmp, gen); | |
2319 | if (!ret) { | |
2320 | *eb_ret = tmp; | |
2321 | return 0; | |
cb44921a | 2322 | } |
bdf7c00e JB |
2323 | free_extent_buffer(tmp); |
2324 | btrfs_release_path(p); | |
2325 | return -EIO; | |
c8c42864 CM |
2326 | } |
2327 | ||
2328 | /* | |
2329 | * reduce lock contention at high levels | |
2330 | * of the btree by dropping locks before | |
76a05b35 CM |
2331 | * we read. Don't release the lock on the current |
2332 | * level because we need to walk this node to figure | |
2333 | * out which blocks to read. | |
c8c42864 | 2334 | */ |
8c594ea8 CM |
2335 | btrfs_unlock_up_safe(p, level + 1); |
2336 | btrfs_set_path_blocking(p); | |
2337 | ||
cb44921a | 2338 | free_extent_buffer(tmp); |
c8c42864 CM |
2339 | if (p->reada) |
2340 | reada_for_search(root, p, level, slot, key->objectid); | |
2341 | ||
b3b4aa74 | 2342 | btrfs_release_path(p); |
76a05b35 CM |
2343 | |
2344 | ret = -EAGAIN; | |
5bdd3536 | 2345 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2346 | if (tmp) { |
2347 | /* | |
2348 | * If the read above didn't mark this buffer up to date, | |
2349 | * it will never end up being up to date. Set ret to EIO now | |
2350 | * and give up so that our caller doesn't loop forever | |
2351 | * on our EAGAINs. | |
2352 | */ | |
b9fab919 | 2353 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2354 | ret = -EIO; |
c8c42864 | 2355 | free_extent_buffer(tmp); |
76a05b35 CM |
2356 | } |
2357 | return ret; | |
c8c42864 CM |
2358 | } |
2359 | ||
2360 | /* | |
2361 | * helper function for btrfs_search_slot. This does all of the checks | |
2362 | * for node-level blocks and does any balancing required based on | |
2363 | * the ins_len. | |
2364 | * | |
2365 | * If no extra work was required, zero is returned. If we had to | |
2366 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2367 | * start over | |
2368 | */ | |
2369 | static int | |
2370 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2371 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2372 | struct extent_buffer *b, int level, int ins_len, |
2373 | int *write_lock_level) | |
c8c42864 CM |
2374 | { |
2375 | int ret; | |
2376 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2377 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2378 | int sret; | |
2379 | ||
bd681513 CM |
2380 | if (*write_lock_level < level + 1) { |
2381 | *write_lock_level = level + 1; | |
2382 | btrfs_release_path(p); | |
2383 | goto again; | |
2384 | } | |
2385 | ||
c8c42864 | 2386 | btrfs_set_path_blocking(p); |
0b08851f | 2387 | reada_for_balance(root, p, level); |
c8c42864 | 2388 | sret = split_node(trans, root, p, level); |
bd681513 | 2389 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2390 | |
2391 | BUG_ON(sret > 0); | |
2392 | if (sret) { | |
2393 | ret = sret; | |
2394 | goto done; | |
2395 | } | |
2396 | b = p->nodes[level]; | |
2397 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2398 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2399 | int sret; |
2400 | ||
bd681513 CM |
2401 | if (*write_lock_level < level + 1) { |
2402 | *write_lock_level = level + 1; | |
2403 | btrfs_release_path(p); | |
2404 | goto again; | |
2405 | } | |
2406 | ||
c8c42864 | 2407 | btrfs_set_path_blocking(p); |
0b08851f | 2408 | reada_for_balance(root, p, level); |
c8c42864 | 2409 | sret = balance_level(trans, root, p, level); |
bd681513 | 2410 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2411 | |
2412 | if (sret) { | |
2413 | ret = sret; | |
2414 | goto done; | |
2415 | } | |
2416 | b = p->nodes[level]; | |
2417 | if (!b) { | |
b3b4aa74 | 2418 | btrfs_release_path(p); |
c8c42864 CM |
2419 | goto again; |
2420 | } | |
2421 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2422 | } | |
2423 | return 0; | |
2424 | ||
2425 | again: | |
2426 | ret = -EAGAIN; | |
2427 | done: | |
2428 | return ret; | |
2429 | } | |
2430 | ||
74123bd7 CM |
2431 | /* |
2432 | * look for key in the tree. path is filled in with nodes along the way | |
2433 | * if key is found, we return zero and you can find the item in the leaf | |
2434 | * level of the path (level 0) | |
2435 | * | |
2436 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2437 | * be inserted, and 1 is returned. If there are other errors during the |
2438 | * search a negative error number is returned. | |
97571fd0 CM |
2439 | * |
2440 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2441 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2442 | * possible) | |
74123bd7 | 2443 | */ |
e089f05c CM |
2444 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2445 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2446 | ins_len, int cow) | |
be0e5c09 | 2447 | { |
5f39d397 | 2448 | struct extent_buffer *b; |
be0e5c09 CM |
2449 | int slot; |
2450 | int ret; | |
33c66f43 | 2451 | int err; |
be0e5c09 | 2452 | int level; |
925baedd | 2453 | int lowest_unlock = 1; |
bd681513 CM |
2454 | int root_lock; |
2455 | /* everything at write_lock_level or lower must be write locked */ | |
2456 | int write_lock_level = 0; | |
9f3a7427 | 2457 | u8 lowest_level = 0; |
f7c79f30 | 2458 | int min_write_lock_level; |
9f3a7427 | 2459 | |
6702ed49 | 2460 | lowest_level = p->lowest_level; |
323ac95b | 2461 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2462 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2463 | |
bd681513 | 2464 | if (ins_len < 0) { |
925baedd | 2465 | lowest_unlock = 2; |
65b51a00 | 2466 | |
bd681513 CM |
2467 | /* when we are removing items, we might have to go up to level |
2468 | * two as we update tree pointers Make sure we keep write | |
2469 | * for those levels as well | |
2470 | */ | |
2471 | write_lock_level = 2; | |
2472 | } else if (ins_len > 0) { | |
2473 | /* | |
2474 | * for inserting items, make sure we have a write lock on | |
2475 | * level 1 so we can update keys | |
2476 | */ | |
2477 | write_lock_level = 1; | |
2478 | } | |
2479 | ||
2480 | if (!cow) | |
2481 | write_lock_level = -1; | |
2482 | ||
09a2a8f9 | 2483 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2484 | write_lock_level = BTRFS_MAX_LEVEL; |
2485 | ||
f7c79f30 CM |
2486 | min_write_lock_level = write_lock_level; |
2487 | ||
bb803951 | 2488 | again: |
bd681513 CM |
2489 | /* |
2490 | * we try very hard to do read locks on the root | |
2491 | */ | |
2492 | root_lock = BTRFS_READ_LOCK; | |
2493 | level = 0; | |
5d4f98a2 | 2494 | if (p->search_commit_root) { |
bd681513 CM |
2495 | /* |
2496 | * the commit roots are read only | |
2497 | * so we always do read locks | |
2498 | */ | |
5d4f98a2 YZ |
2499 | b = root->commit_root; |
2500 | extent_buffer_get(b); | |
bd681513 | 2501 | level = btrfs_header_level(b); |
5d4f98a2 | 2502 | if (!p->skip_locking) |
bd681513 | 2503 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2504 | } else { |
bd681513 | 2505 | if (p->skip_locking) { |
5d4f98a2 | 2506 | b = btrfs_root_node(root); |
bd681513 CM |
2507 | level = btrfs_header_level(b); |
2508 | } else { | |
2509 | /* we don't know the level of the root node | |
2510 | * until we actually have it read locked | |
2511 | */ | |
2512 | b = btrfs_read_lock_root_node(root); | |
2513 | level = btrfs_header_level(b); | |
2514 | if (level <= write_lock_level) { | |
2515 | /* whoops, must trade for write lock */ | |
2516 | btrfs_tree_read_unlock(b); | |
2517 | free_extent_buffer(b); | |
2518 | b = btrfs_lock_root_node(root); | |
2519 | root_lock = BTRFS_WRITE_LOCK; | |
2520 | ||
2521 | /* the level might have changed, check again */ | |
2522 | level = btrfs_header_level(b); | |
2523 | } | |
2524 | } | |
5d4f98a2 | 2525 | } |
bd681513 CM |
2526 | p->nodes[level] = b; |
2527 | if (!p->skip_locking) | |
2528 | p->locks[level] = root_lock; | |
925baedd | 2529 | |
eb60ceac | 2530 | while (b) { |
5f39d397 | 2531 | level = btrfs_header_level(b); |
65b51a00 CM |
2532 | |
2533 | /* | |
2534 | * setup the path here so we can release it under lock | |
2535 | * contention with the cow code | |
2536 | */ | |
02217ed2 | 2537 | if (cow) { |
c8c42864 CM |
2538 | /* |
2539 | * if we don't really need to cow this block | |
2540 | * then we don't want to set the path blocking, | |
2541 | * so we test it here | |
2542 | */ | |
5d4f98a2 | 2543 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2544 | goto cow_done; |
5d4f98a2 | 2545 | |
b4ce94de CM |
2546 | btrfs_set_path_blocking(p); |
2547 | ||
bd681513 CM |
2548 | /* |
2549 | * must have write locks on this node and the | |
2550 | * parent | |
2551 | */ | |
5124e00e JB |
2552 | if (level > write_lock_level || |
2553 | (level + 1 > write_lock_level && | |
2554 | level + 1 < BTRFS_MAX_LEVEL && | |
2555 | p->nodes[level + 1])) { | |
bd681513 CM |
2556 | write_lock_level = level + 1; |
2557 | btrfs_release_path(p); | |
2558 | goto again; | |
2559 | } | |
2560 | ||
33c66f43 YZ |
2561 | err = btrfs_cow_block(trans, root, b, |
2562 | p->nodes[level + 1], | |
2563 | p->slots[level + 1], &b); | |
2564 | if (err) { | |
33c66f43 | 2565 | ret = err; |
65b51a00 | 2566 | goto done; |
54aa1f4d | 2567 | } |
02217ed2 | 2568 | } |
65b51a00 | 2569 | cow_done: |
02217ed2 | 2570 | BUG_ON(!cow && ins_len); |
65b51a00 | 2571 | |
eb60ceac | 2572 | p->nodes[level] = b; |
bd681513 | 2573 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2574 | |
2575 | /* | |
2576 | * we have a lock on b and as long as we aren't changing | |
2577 | * the tree, there is no way to for the items in b to change. | |
2578 | * It is safe to drop the lock on our parent before we | |
2579 | * go through the expensive btree search on b. | |
2580 | * | |
2581 | * If cow is true, then we might be changing slot zero, | |
2582 | * which may require changing the parent. So, we can't | |
2583 | * drop the lock until after we know which slot we're | |
2584 | * operating on. | |
2585 | */ | |
2586 | if (!cow) | |
2587 | btrfs_unlock_up_safe(p, level + 1); | |
2588 | ||
5f39d397 | 2589 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2590 | |
5f39d397 | 2591 | if (level != 0) { |
33c66f43 YZ |
2592 | int dec = 0; |
2593 | if (ret && slot > 0) { | |
2594 | dec = 1; | |
be0e5c09 | 2595 | slot -= 1; |
33c66f43 | 2596 | } |
be0e5c09 | 2597 | p->slots[level] = slot; |
33c66f43 | 2598 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2599 | ins_len, &write_lock_level); |
33c66f43 | 2600 | if (err == -EAGAIN) |
c8c42864 | 2601 | goto again; |
33c66f43 YZ |
2602 | if (err) { |
2603 | ret = err; | |
c8c42864 | 2604 | goto done; |
33c66f43 | 2605 | } |
c8c42864 CM |
2606 | b = p->nodes[level]; |
2607 | slot = p->slots[level]; | |
b4ce94de | 2608 | |
bd681513 CM |
2609 | /* |
2610 | * slot 0 is special, if we change the key | |
2611 | * we have to update the parent pointer | |
2612 | * which means we must have a write lock | |
2613 | * on the parent | |
2614 | */ | |
2615 | if (slot == 0 && cow && | |
2616 | write_lock_level < level + 1) { | |
2617 | write_lock_level = level + 1; | |
2618 | btrfs_release_path(p); | |
2619 | goto again; | |
2620 | } | |
2621 | ||
f7c79f30 CM |
2622 | unlock_up(p, level, lowest_unlock, |
2623 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2624 | |
925baedd | 2625 | if (level == lowest_level) { |
33c66f43 YZ |
2626 | if (dec) |
2627 | p->slots[level]++; | |
5b21f2ed | 2628 | goto done; |
925baedd | 2629 | } |
ca7a79ad | 2630 | |
33c66f43 | 2631 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2632 | &b, level, slot, key, 0); |
33c66f43 | 2633 | if (err == -EAGAIN) |
c8c42864 | 2634 | goto again; |
33c66f43 YZ |
2635 | if (err) { |
2636 | ret = err; | |
76a05b35 | 2637 | goto done; |
33c66f43 | 2638 | } |
76a05b35 | 2639 | |
b4ce94de | 2640 | if (!p->skip_locking) { |
bd681513 CM |
2641 | level = btrfs_header_level(b); |
2642 | if (level <= write_lock_level) { | |
2643 | err = btrfs_try_tree_write_lock(b); | |
2644 | if (!err) { | |
2645 | btrfs_set_path_blocking(p); | |
2646 | btrfs_tree_lock(b); | |
2647 | btrfs_clear_path_blocking(p, b, | |
2648 | BTRFS_WRITE_LOCK); | |
2649 | } | |
2650 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2651 | } else { | |
2652 | err = btrfs_try_tree_read_lock(b); | |
2653 | if (!err) { | |
2654 | btrfs_set_path_blocking(p); | |
2655 | btrfs_tree_read_lock(b); | |
2656 | btrfs_clear_path_blocking(p, b, | |
2657 | BTRFS_READ_LOCK); | |
2658 | } | |
2659 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2660 | } |
bd681513 | 2661 | p->nodes[level] = b; |
b4ce94de | 2662 | } |
be0e5c09 CM |
2663 | } else { |
2664 | p->slots[level] = slot; | |
87b29b20 YZ |
2665 | if (ins_len > 0 && |
2666 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2667 | if (write_lock_level < 1) { |
2668 | write_lock_level = 1; | |
2669 | btrfs_release_path(p); | |
2670 | goto again; | |
2671 | } | |
2672 | ||
b4ce94de | 2673 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2674 | err = split_leaf(trans, root, key, |
2675 | p, ins_len, ret == 0); | |
bd681513 | 2676 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2677 | |
33c66f43 YZ |
2678 | BUG_ON(err > 0); |
2679 | if (err) { | |
2680 | ret = err; | |
65b51a00 CM |
2681 | goto done; |
2682 | } | |
5c680ed6 | 2683 | } |
459931ec | 2684 | if (!p->search_for_split) |
f7c79f30 CM |
2685 | unlock_up(p, level, lowest_unlock, |
2686 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2687 | goto done; |
be0e5c09 CM |
2688 | } |
2689 | } | |
65b51a00 CM |
2690 | ret = 1; |
2691 | done: | |
b4ce94de CM |
2692 | /* |
2693 | * we don't really know what they plan on doing with the path | |
2694 | * from here on, so for now just mark it as blocking | |
2695 | */ | |
b9473439 CM |
2696 | if (!p->leave_spinning) |
2697 | btrfs_set_path_blocking(p); | |
76a05b35 | 2698 | if (ret < 0) |
b3b4aa74 | 2699 | btrfs_release_path(p); |
65b51a00 | 2700 | return ret; |
be0e5c09 CM |
2701 | } |
2702 | ||
5d9e75c4 JS |
2703 | /* |
2704 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2705 | * current state of the tree together with the operations recorded in the tree | |
2706 | * modification log to search for the key in a previous version of this tree, as | |
2707 | * denoted by the time_seq parameter. | |
2708 | * | |
2709 | * Naturally, there is no support for insert, delete or cow operations. | |
2710 | * | |
2711 | * The resulting path and return value will be set up as if we called | |
2712 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2713 | */ | |
2714 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2715 | struct btrfs_path *p, u64 time_seq) | |
2716 | { | |
2717 | struct extent_buffer *b; | |
2718 | int slot; | |
2719 | int ret; | |
2720 | int err; | |
2721 | int level; | |
2722 | int lowest_unlock = 1; | |
2723 | u8 lowest_level = 0; | |
2724 | ||
2725 | lowest_level = p->lowest_level; | |
2726 | WARN_ON(p->nodes[0] != NULL); | |
2727 | ||
2728 | if (p->search_commit_root) { | |
2729 | BUG_ON(time_seq); | |
2730 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2731 | } | |
2732 | ||
2733 | again: | |
5d9e75c4 | 2734 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2735 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2736 | p->locks[level] = BTRFS_READ_LOCK; |
2737 | ||
2738 | while (b) { | |
2739 | level = btrfs_header_level(b); | |
2740 | p->nodes[level] = b; | |
2741 | btrfs_clear_path_blocking(p, NULL, 0); | |
2742 | ||
2743 | /* | |
2744 | * we have a lock on b and as long as we aren't changing | |
2745 | * the tree, there is no way to for the items in b to change. | |
2746 | * It is safe to drop the lock on our parent before we | |
2747 | * go through the expensive btree search on b. | |
2748 | */ | |
2749 | btrfs_unlock_up_safe(p, level + 1); | |
2750 | ||
2751 | ret = bin_search(b, key, level, &slot); | |
2752 | ||
2753 | if (level != 0) { | |
2754 | int dec = 0; | |
2755 | if (ret && slot > 0) { | |
2756 | dec = 1; | |
2757 | slot -= 1; | |
2758 | } | |
2759 | p->slots[level] = slot; | |
2760 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2761 | ||
2762 | if (level == lowest_level) { | |
2763 | if (dec) | |
2764 | p->slots[level]++; | |
2765 | goto done; | |
2766 | } | |
2767 | ||
2768 | err = read_block_for_search(NULL, root, p, &b, level, | |
2769 | slot, key, time_seq); | |
2770 | if (err == -EAGAIN) | |
2771 | goto again; | |
2772 | if (err) { | |
2773 | ret = err; | |
2774 | goto done; | |
2775 | } | |
2776 | ||
2777 | level = btrfs_header_level(b); | |
2778 | err = btrfs_try_tree_read_lock(b); | |
2779 | if (!err) { | |
2780 | btrfs_set_path_blocking(p); | |
2781 | btrfs_tree_read_lock(b); | |
2782 | btrfs_clear_path_blocking(p, b, | |
2783 | BTRFS_READ_LOCK); | |
2784 | } | |
9ec72677 | 2785 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
2786 | if (!b) { |
2787 | ret = -ENOMEM; | |
2788 | goto done; | |
2789 | } | |
5d9e75c4 JS |
2790 | p->locks[level] = BTRFS_READ_LOCK; |
2791 | p->nodes[level] = b; | |
5d9e75c4 JS |
2792 | } else { |
2793 | p->slots[level] = slot; | |
2794 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2795 | goto done; | |
2796 | } | |
2797 | } | |
2798 | ret = 1; | |
2799 | done: | |
2800 | if (!p->leave_spinning) | |
2801 | btrfs_set_path_blocking(p); | |
2802 | if (ret < 0) | |
2803 | btrfs_release_path(p); | |
2804 | ||
2805 | return ret; | |
2806 | } | |
2807 | ||
2f38b3e1 AJ |
2808 | /* |
2809 | * helper to use instead of search slot if no exact match is needed but | |
2810 | * instead the next or previous item should be returned. | |
2811 | * When find_higher is true, the next higher item is returned, the next lower | |
2812 | * otherwise. | |
2813 | * When return_any and find_higher are both true, and no higher item is found, | |
2814 | * return the next lower instead. | |
2815 | * When return_any is true and find_higher is false, and no lower item is found, | |
2816 | * return the next higher instead. | |
2817 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2818 | * < 0 on error | |
2819 | */ | |
2820 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
2821 | struct btrfs_key *key, struct btrfs_path *p, | |
2822 | int find_higher, int return_any) | |
2823 | { | |
2824 | int ret; | |
2825 | struct extent_buffer *leaf; | |
2826 | ||
2827 | again: | |
2828 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2829 | if (ret <= 0) | |
2830 | return ret; | |
2831 | /* | |
2832 | * a return value of 1 means the path is at the position where the | |
2833 | * item should be inserted. Normally this is the next bigger item, | |
2834 | * but in case the previous item is the last in a leaf, path points | |
2835 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2836 | * item. | |
2837 | */ | |
2838 | leaf = p->nodes[0]; | |
2839 | ||
2840 | if (find_higher) { | |
2841 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2842 | ret = btrfs_next_leaf(root, p); | |
2843 | if (ret <= 0) | |
2844 | return ret; | |
2845 | if (!return_any) | |
2846 | return 1; | |
2847 | /* | |
2848 | * no higher item found, return the next | |
2849 | * lower instead | |
2850 | */ | |
2851 | return_any = 0; | |
2852 | find_higher = 0; | |
2853 | btrfs_release_path(p); | |
2854 | goto again; | |
2855 | } | |
2856 | } else { | |
e6793769 AJ |
2857 | if (p->slots[0] == 0) { |
2858 | ret = btrfs_prev_leaf(root, p); | |
2859 | if (ret < 0) | |
2860 | return ret; | |
2861 | if (!ret) { | |
2862 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
2863 | return 0; | |
2f38b3e1 | 2864 | } |
e6793769 AJ |
2865 | if (!return_any) |
2866 | return 1; | |
2867 | /* | |
2868 | * no lower item found, return the next | |
2869 | * higher instead | |
2870 | */ | |
2871 | return_any = 0; | |
2872 | find_higher = 1; | |
2873 | btrfs_release_path(p); | |
2874 | goto again; | |
2875 | } else { | |
2f38b3e1 AJ |
2876 | --p->slots[0]; |
2877 | } | |
2878 | } | |
2879 | return 0; | |
2880 | } | |
2881 | ||
74123bd7 CM |
2882 | /* |
2883 | * adjust the pointers going up the tree, starting at level | |
2884 | * making sure the right key of each node is points to 'key'. | |
2885 | * This is used after shifting pointers to the left, so it stops | |
2886 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2887 | * higher levels | |
aa5d6bed | 2888 | * |
74123bd7 | 2889 | */ |
d6a0a126 | 2890 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2891 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2892 | { |
2893 | int i; | |
5f39d397 CM |
2894 | struct extent_buffer *t; |
2895 | ||
234b63a0 | 2896 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2897 | int tslot = path->slots[i]; |
eb60ceac | 2898 | if (!path->nodes[i]) |
be0e5c09 | 2899 | break; |
5f39d397 | 2900 | t = path->nodes[i]; |
32adf090 | 2901 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 2902 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2903 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2904 | if (tslot != 0) |
2905 | break; | |
2906 | } | |
2907 | } | |
2908 | ||
31840ae1 ZY |
2909 | /* |
2910 | * update item key. | |
2911 | * | |
2912 | * This function isn't completely safe. It's the caller's responsibility | |
2913 | * that the new key won't break the order | |
2914 | */ | |
afe5fea7 | 2915 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2916 | struct btrfs_key *new_key) |
31840ae1 ZY |
2917 | { |
2918 | struct btrfs_disk_key disk_key; | |
2919 | struct extent_buffer *eb; | |
2920 | int slot; | |
2921 | ||
2922 | eb = path->nodes[0]; | |
2923 | slot = path->slots[0]; | |
2924 | if (slot > 0) { | |
2925 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2926 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2927 | } |
2928 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2929 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2930 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2931 | } |
2932 | ||
2933 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2934 | btrfs_set_item_key(eb, &disk_key, slot); | |
2935 | btrfs_mark_buffer_dirty(eb); | |
2936 | if (slot == 0) | |
d6a0a126 | 2937 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
2938 | } |
2939 | ||
74123bd7 CM |
2940 | /* |
2941 | * try to push data from one node into the next node left in the | |
79f95c82 | 2942 | * tree. |
aa5d6bed CM |
2943 | * |
2944 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2945 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2946 | */ |
98ed5174 CM |
2947 | static int push_node_left(struct btrfs_trans_handle *trans, |
2948 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 2949 | struct extent_buffer *src, int empty) |
be0e5c09 | 2950 | { |
be0e5c09 | 2951 | int push_items = 0; |
bb803951 CM |
2952 | int src_nritems; |
2953 | int dst_nritems; | |
aa5d6bed | 2954 | int ret = 0; |
be0e5c09 | 2955 | |
5f39d397 CM |
2956 | src_nritems = btrfs_header_nritems(src); |
2957 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2958 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2959 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2960 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2961 | |
bce4eae9 | 2962 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2963 | return 1; |
2964 | ||
d397712b | 2965 | if (push_items <= 0) |
be0e5c09 CM |
2966 | return 1; |
2967 | ||
bce4eae9 | 2968 | if (empty) { |
971a1f66 | 2969 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2970 | if (push_items < src_nritems) { |
2971 | /* leave at least 8 pointers in the node if | |
2972 | * we aren't going to empty it | |
2973 | */ | |
2974 | if (src_nritems - push_items < 8) { | |
2975 | if (push_items <= 8) | |
2976 | return 1; | |
2977 | push_items -= 8; | |
2978 | } | |
2979 | } | |
2980 | } else | |
2981 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2982 | |
f230475e | 2983 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
90f8d62e | 2984 | push_items); |
5f39d397 CM |
2985 | copy_extent_buffer(dst, src, |
2986 | btrfs_node_key_ptr_offset(dst_nritems), | |
2987 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2988 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2989 | |
bb803951 | 2990 | if (push_items < src_nritems) { |
57911b8b JS |
2991 | /* |
2992 | * don't call tree_mod_log_eb_move here, key removal was already | |
2993 | * fully logged by tree_mod_log_eb_copy above. | |
2994 | */ | |
5f39d397 CM |
2995 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2996 | btrfs_node_key_ptr_offset(push_items), | |
2997 | (src_nritems - push_items) * | |
2998 | sizeof(struct btrfs_key_ptr)); | |
2999 | } | |
3000 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3001 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3002 | btrfs_mark_buffer_dirty(src); | |
3003 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3004 | |
79f95c82 CM |
3005 | return ret; |
3006 | } | |
3007 | ||
3008 | /* | |
3009 | * try to push data from one node into the next node right in the | |
3010 | * tree. | |
3011 | * | |
3012 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3013 | * error, and > 0 if there was no room in the right hand block. | |
3014 | * | |
3015 | * this will only push up to 1/2 the contents of the left node over | |
3016 | */ | |
5f39d397 CM |
3017 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3018 | struct btrfs_root *root, | |
3019 | struct extent_buffer *dst, | |
3020 | struct extent_buffer *src) | |
79f95c82 | 3021 | { |
79f95c82 CM |
3022 | int push_items = 0; |
3023 | int max_push; | |
3024 | int src_nritems; | |
3025 | int dst_nritems; | |
3026 | int ret = 0; | |
79f95c82 | 3027 | |
7bb86316 CM |
3028 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3029 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3030 | ||
5f39d397 CM |
3031 | src_nritems = btrfs_header_nritems(src); |
3032 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3033 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3034 | if (push_items <= 0) |
79f95c82 | 3035 | return 1; |
bce4eae9 | 3036 | |
d397712b | 3037 | if (src_nritems < 4) |
bce4eae9 | 3038 | return 1; |
79f95c82 CM |
3039 | |
3040 | max_push = src_nritems / 2 + 1; | |
3041 | /* don't try to empty the node */ | |
d397712b | 3042 | if (max_push >= src_nritems) |
79f95c82 | 3043 | return 1; |
252c38f0 | 3044 | |
79f95c82 CM |
3045 | if (max_push < push_items) |
3046 | push_items = max_push; | |
3047 | ||
f230475e | 3048 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3049 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3050 | btrfs_node_key_ptr_offset(0), | |
3051 | (dst_nritems) * | |
3052 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3053 | |
f230475e | 3054 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
90f8d62e | 3055 | src_nritems - push_items, push_items); |
5f39d397 CM |
3056 | copy_extent_buffer(dst, src, |
3057 | btrfs_node_key_ptr_offset(0), | |
3058 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3059 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3060 | |
5f39d397 CM |
3061 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3062 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3063 | |
5f39d397 CM |
3064 | btrfs_mark_buffer_dirty(src); |
3065 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3066 | |
aa5d6bed | 3067 | return ret; |
be0e5c09 CM |
3068 | } |
3069 | ||
97571fd0 CM |
3070 | /* |
3071 | * helper function to insert a new root level in the tree. | |
3072 | * A new node is allocated, and a single item is inserted to | |
3073 | * point to the existing root | |
aa5d6bed CM |
3074 | * |
3075 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3076 | */ |
d397712b | 3077 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3078 | struct btrfs_root *root, |
fdd99c72 | 3079 | struct btrfs_path *path, int level) |
5c680ed6 | 3080 | { |
7bb86316 | 3081 | u64 lower_gen; |
5f39d397 CM |
3082 | struct extent_buffer *lower; |
3083 | struct extent_buffer *c; | |
925baedd | 3084 | struct extent_buffer *old; |
5f39d397 | 3085 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3086 | |
3087 | BUG_ON(path->nodes[level]); | |
3088 | BUG_ON(path->nodes[level-1] != root->node); | |
3089 | ||
7bb86316 CM |
3090 | lower = path->nodes[level-1]; |
3091 | if (level == 1) | |
3092 | btrfs_item_key(lower, &lower_key, 0); | |
3093 | else | |
3094 | btrfs_node_key(lower, &lower_key, 0); | |
3095 | ||
31840ae1 | 3096 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3097 | root->root_key.objectid, &lower_key, |
5581a51a | 3098 | level, root->node->start, 0); |
5f39d397 CM |
3099 | if (IS_ERR(c)) |
3100 | return PTR_ERR(c); | |
925baedd | 3101 | |
f0486c68 YZ |
3102 | root_add_used(root, root->nodesize); |
3103 | ||
5d4f98a2 | 3104 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3105 | btrfs_set_header_nritems(c, 1); |
3106 | btrfs_set_header_level(c, level); | |
db94535d | 3107 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3108 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3109 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3110 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
3111 | |
3112 | write_extent_buffer(c, root->fs_info->fsid, | |
3113 | (unsigned long)btrfs_header_fsid(c), | |
3114 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3115 | |
3116 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
3117 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
3118 | BTRFS_UUID_SIZE); | |
3119 | ||
5f39d397 | 3120 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3121 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3122 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3123 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3124 | |
3125 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3126 | |
5f39d397 | 3127 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3128 | |
925baedd | 3129 | old = root->node; |
fdd99c72 | 3130 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3131 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3132 | |
3133 | /* the super has an extra ref to root->node */ | |
3134 | free_extent_buffer(old); | |
3135 | ||
0b86a832 | 3136 | add_root_to_dirty_list(root); |
5f39d397 CM |
3137 | extent_buffer_get(c); |
3138 | path->nodes[level] = c; | |
bd681513 | 3139 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3140 | path->slots[level] = 0; |
3141 | return 0; | |
3142 | } | |
3143 | ||
74123bd7 CM |
3144 | /* |
3145 | * worker function to insert a single pointer in a node. | |
3146 | * the node should have enough room for the pointer already | |
97571fd0 | 3147 | * |
74123bd7 CM |
3148 | * slot and level indicate where you want the key to go, and |
3149 | * blocknr is the block the key points to. | |
3150 | */ | |
143bede5 JM |
3151 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3152 | struct btrfs_root *root, struct btrfs_path *path, | |
3153 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3154 | int slot, int level) |
74123bd7 | 3155 | { |
5f39d397 | 3156 | struct extent_buffer *lower; |
74123bd7 | 3157 | int nritems; |
f3ea38da | 3158 | int ret; |
5c680ed6 CM |
3159 | |
3160 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3161 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3162 | lower = path->nodes[level]; |
3163 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3164 | BUG_ON(slot > nritems); |
143bede5 | 3165 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3166 | if (slot != nritems) { |
c3e06965 | 3167 | if (level) |
f3ea38da JS |
3168 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3169 | slot, nritems - slot); | |
5f39d397 CM |
3170 | memmove_extent_buffer(lower, |
3171 | btrfs_node_key_ptr_offset(slot + 1), | |
3172 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3173 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3174 | } |
c3e06965 | 3175 | if (level) { |
f3ea38da | 3176 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3177 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3178 | BUG_ON(ret < 0); |
3179 | } | |
5f39d397 | 3180 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3181 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3182 | WARN_ON(trans->transid == 0); |
3183 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3184 | btrfs_set_header_nritems(lower, nritems + 1); |
3185 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3186 | } |
3187 | ||
97571fd0 CM |
3188 | /* |
3189 | * split the node at the specified level in path in two. | |
3190 | * The path is corrected to point to the appropriate node after the split | |
3191 | * | |
3192 | * Before splitting this tries to make some room in the node by pushing | |
3193 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3194 | * |
3195 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3196 | */ |
e02119d5 CM |
3197 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3198 | struct btrfs_root *root, | |
3199 | struct btrfs_path *path, int level) | |
be0e5c09 | 3200 | { |
5f39d397 CM |
3201 | struct extent_buffer *c; |
3202 | struct extent_buffer *split; | |
3203 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3204 | int mid; |
5c680ed6 | 3205 | int ret; |
7518a238 | 3206 | u32 c_nritems; |
eb60ceac | 3207 | |
5f39d397 | 3208 | c = path->nodes[level]; |
7bb86316 | 3209 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3210 | if (c == root->node) { |
d9abbf1c | 3211 | /* |
90f8d62e JS |
3212 | * trying to split the root, lets make a new one |
3213 | * | |
fdd99c72 | 3214 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3215 | * insert_new_root, because that root buffer will be kept as a |
3216 | * normal node. We are going to log removal of half of the | |
3217 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3218 | * tree lock on the buffer, which is why we cannot race with | |
3219 | * other tree_mod_log users. | |
d9abbf1c | 3220 | */ |
fdd99c72 | 3221 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3222 | if (ret) |
3223 | return ret; | |
b3612421 | 3224 | } else { |
e66f709b | 3225 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3226 | c = path->nodes[level]; |
3227 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3228 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3229 | return 0; |
54aa1f4d CM |
3230 | if (ret < 0) |
3231 | return ret; | |
be0e5c09 | 3232 | } |
e66f709b | 3233 | |
5f39d397 | 3234 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3235 | mid = (c_nritems + 1) / 2; |
3236 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3237 | |
5d4f98a2 | 3238 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3239 | root->root_key.objectid, |
5581a51a | 3240 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3241 | if (IS_ERR(split)) |
3242 | return PTR_ERR(split); | |
3243 | ||
f0486c68 YZ |
3244 | root_add_used(root, root->nodesize); |
3245 | ||
5d4f98a2 | 3246 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3247 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3248 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3249 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3250 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3251 | btrfs_set_header_owner(split, root->root_key.objectid); |
3252 | write_extent_buffer(split, root->fs_info->fsid, | |
3253 | (unsigned long)btrfs_header_fsid(split), | |
3254 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3255 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
3256 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
3257 | BTRFS_UUID_SIZE); | |
54aa1f4d | 3258 | |
90f8d62e | 3259 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
3260 | copy_extent_buffer(split, c, |
3261 | btrfs_node_key_ptr_offset(0), | |
3262 | btrfs_node_key_ptr_offset(mid), | |
3263 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3264 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3265 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3266 | ret = 0; |
3267 | ||
5f39d397 CM |
3268 | btrfs_mark_buffer_dirty(c); |
3269 | btrfs_mark_buffer_dirty(split); | |
3270 | ||
143bede5 | 3271 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3272 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3273 | |
5de08d7d | 3274 | if (path->slots[level] >= mid) { |
5c680ed6 | 3275 | path->slots[level] -= mid; |
925baedd | 3276 | btrfs_tree_unlock(c); |
5f39d397 CM |
3277 | free_extent_buffer(c); |
3278 | path->nodes[level] = split; | |
5c680ed6 CM |
3279 | path->slots[level + 1] += 1; |
3280 | } else { | |
925baedd | 3281 | btrfs_tree_unlock(split); |
5f39d397 | 3282 | free_extent_buffer(split); |
be0e5c09 | 3283 | } |
aa5d6bed | 3284 | return ret; |
be0e5c09 CM |
3285 | } |
3286 | ||
74123bd7 CM |
3287 | /* |
3288 | * how many bytes are required to store the items in a leaf. start | |
3289 | * and nr indicate which items in the leaf to check. This totals up the | |
3290 | * space used both by the item structs and the item data | |
3291 | */ | |
5f39d397 | 3292 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3293 | { |
41be1f3b JB |
3294 | struct btrfs_item *start_item; |
3295 | struct btrfs_item *end_item; | |
3296 | struct btrfs_map_token token; | |
be0e5c09 | 3297 | int data_len; |
5f39d397 | 3298 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3299 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3300 | |
3301 | if (!nr) | |
3302 | return 0; | |
41be1f3b JB |
3303 | btrfs_init_map_token(&token); |
3304 | start_item = btrfs_item_nr(l, start); | |
3305 | end_item = btrfs_item_nr(l, end); | |
3306 | data_len = btrfs_token_item_offset(l, start_item, &token) + | |
3307 | btrfs_token_item_size(l, start_item, &token); | |
3308 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3309 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3310 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3311 | return data_len; |
3312 | } | |
3313 | ||
d4dbff95 CM |
3314 | /* |
3315 | * The space between the end of the leaf items and | |
3316 | * the start of the leaf data. IOW, how much room | |
3317 | * the leaf has left for both items and data | |
3318 | */ | |
d397712b | 3319 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3320 | struct extent_buffer *leaf) |
d4dbff95 | 3321 | { |
5f39d397 CM |
3322 | int nritems = btrfs_header_nritems(leaf); |
3323 | int ret; | |
3324 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3325 | if (ret < 0) { | |
d397712b CM |
3326 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
3327 | "used %d nritems %d\n", | |
ae2f5411 | 3328 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3329 | leaf_space_used(leaf, 0, nritems), nritems); |
3330 | } | |
3331 | return ret; | |
d4dbff95 CM |
3332 | } |
3333 | ||
99d8f83c CM |
3334 | /* |
3335 | * min slot controls the lowest index we're willing to push to the | |
3336 | * right. We'll push up to and including min_slot, but no lower | |
3337 | */ | |
44871b1b CM |
3338 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3339 | struct btrfs_root *root, | |
3340 | struct btrfs_path *path, | |
3341 | int data_size, int empty, | |
3342 | struct extent_buffer *right, | |
99d8f83c CM |
3343 | int free_space, u32 left_nritems, |
3344 | u32 min_slot) | |
00ec4c51 | 3345 | { |
5f39d397 | 3346 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3347 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3348 | struct btrfs_map_token token; |
5f39d397 | 3349 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3350 | int slot; |
34a38218 | 3351 | u32 i; |
00ec4c51 CM |
3352 | int push_space = 0; |
3353 | int push_items = 0; | |
0783fcfc | 3354 | struct btrfs_item *item; |
34a38218 | 3355 | u32 nr; |
7518a238 | 3356 | u32 right_nritems; |
5f39d397 | 3357 | u32 data_end; |
db94535d | 3358 | u32 this_item_size; |
00ec4c51 | 3359 | |
cfed81a0 CM |
3360 | btrfs_init_map_token(&token); |
3361 | ||
34a38218 CM |
3362 | if (empty) |
3363 | nr = 0; | |
3364 | else | |
99d8f83c | 3365 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3366 | |
31840ae1 | 3367 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3368 | push_space += data_size; |
31840ae1 | 3369 | |
44871b1b | 3370 | slot = path->slots[1]; |
34a38218 CM |
3371 | i = left_nritems - 1; |
3372 | while (i >= nr) { | |
5f39d397 | 3373 | item = btrfs_item_nr(left, i); |
db94535d | 3374 | |
31840ae1 ZY |
3375 | if (!empty && push_items > 0) { |
3376 | if (path->slots[0] > i) | |
3377 | break; | |
3378 | if (path->slots[0] == i) { | |
3379 | int space = btrfs_leaf_free_space(root, left); | |
3380 | if (space + push_space * 2 > free_space) | |
3381 | break; | |
3382 | } | |
3383 | } | |
3384 | ||
00ec4c51 | 3385 | if (path->slots[0] == i) |
87b29b20 | 3386 | push_space += data_size; |
db94535d | 3387 | |
db94535d CM |
3388 | this_item_size = btrfs_item_size(left, item); |
3389 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3390 | break; |
31840ae1 | 3391 | |
00ec4c51 | 3392 | push_items++; |
db94535d | 3393 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3394 | if (i == 0) |
3395 | break; | |
3396 | i--; | |
db94535d | 3397 | } |
5f39d397 | 3398 | |
925baedd CM |
3399 | if (push_items == 0) |
3400 | goto out_unlock; | |
5f39d397 | 3401 | |
6c1500f2 | 3402 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3403 | |
00ec4c51 | 3404 | /* push left to right */ |
5f39d397 | 3405 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3406 | |
5f39d397 | 3407 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3408 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3409 | |
00ec4c51 | 3410 | /* make room in the right data area */ |
5f39d397 CM |
3411 | data_end = leaf_data_end(root, right); |
3412 | memmove_extent_buffer(right, | |
3413 | btrfs_leaf_data(right) + data_end - push_space, | |
3414 | btrfs_leaf_data(right) + data_end, | |
3415 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3416 | ||
00ec4c51 | 3417 | /* copy from the left data area */ |
5f39d397 | 3418 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3419 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3420 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3421 | push_space); | |
5f39d397 CM |
3422 | |
3423 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3424 | btrfs_item_nr_offset(0), | |
3425 | right_nritems * sizeof(struct btrfs_item)); | |
3426 | ||
00ec4c51 | 3427 | /* copy the items from left to right */ |
5f39d397 CM |
3428 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3429 | btrfs_item_nr_offset(left_nritems - push_items), | |
3430 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3431 | |
3432 | /* update the item pointers */ | |
7518a238 | 3433 | right_nritems += push_items; |
5f39d397 | 3434 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3435 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3436 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 3437 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
3438 | push_space -= btrfs_token_item_size(right, item, &token); |
3439 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3440 | } |
3441 | ||
7518a238 | 3442 | left_nritems -= push_items; |
5f39d397 | 3443 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3444 | |
34a38218 CM |
3445 | if (left_nritems) |
3446 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3447 | else |
3448 | clean_tree_block(trans, root, left); | |
3449 | ||
5f39d397 | 3450 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3451 | |
5f39d397 CM |
3452 | btrfs_item_key(right, &disk_key, 0); |
3453 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3454 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3455 | |
00ec4c51 | 3456 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3457 | if (path->slots[0] >= left_nritems) { |
3458 | path->slots[0] -= left_nritems; | |
925baedd CM |
3459 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3460 | clean_tree_block(trans, root, path->nodes[0]); | |
3461 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3462 | free_extent_buffer(path->nodes[0]); |
3463 | path->nodes[0] = right; | |
00ec4c51 CM |
3464 | path->slots[1] += 1; |
3465 | } else { | |
925baedd | 3466 | btrfs_tree_unlock(right); |
5f39d397 | 3467 | free_extent_buffer(right); |
00ec4c51 CM |
3468 | } |
3469 | return 0; | |
925baedd CM |
3470 | |
3471 | out_unlock: | |
3472 | btrfs_tree_unlock(right); | |
3473 | free_extent_buffer(right); | |
3474 | return 1; | |
00ec4c51 | 3475 | } |
925baedd | 3476 | |
44871b1b CM |
3477 | /* |
3478 | * push some data in the path leaf to the right, trying to free up at | |
3479 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3480 | * | |
3481 | * returns 1 if the push failed because the other node didn't have enough | |
3482 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3483 | * |
3484 | * this will push starting from min_slot to the end of the leaf. It won't | |
3485 | * push any slot lower than min_slot | |
44871b1b CM |
3486 | */ |
3487 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3488 | *root, struct btrfs_path *path, |
3489 | int min_data_size, int data_size, | |
3490 | int empty, u32 min_slot) | |
44871b1b CM |
3491 | { |
3492 | struct extent_buffer *left = path->nodes[0]; | |
3493 | struct extent_buffer *right; | |
3494 | struct extent_buffer *upper; | |
3495 | int slot; | |
3496 | int free_space; | |
3497 | u32 left_nritems; | |
3498 | int ret; | |
3499 | ||
3500 | if (!path->nodes[1]) | |
3501 | return 1; | |
3502 | ||
3503 | slot = path->slots[1]; | |
3504 | upper = path->nodes[1]; | |
3505 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3506 | return 1; | |
3507 | ||
3508 | btrfs_assert_tree_locked(path->nodes[1]); | |
3509 | ||
3510 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3511 | if (right == NULL) |
3512 | return 1; | |
3513 | ||
44871b1b CM |
3514 | btrfs_tree_lock(right); |
3515 | btrfs_set_lock_blocking(right); | |
3516 | ||
3517 | free_space = btrfs_leaf_free_space(root, right); | |
3518 | if (free_space < data_size) | |
3519 | goto out_unlock; | |
3520 | ||
3521 | /* cow and double check */ | |
3522 | ret = btrfs_cow_block(trans, root, right, upper, | |
3523 | slot + 1, &right); | |
3524 | if (ret) | |
3525 | goto out_unlock; | |
3526 | ||
3527 | free_space = btrfs_leaf_free_space(root, right); | |
3528 | if (free_space < data_size) | |
3529 | goto out_unlock; | |
3530 | ||
3531 | left_nritems = btrfs_header_nritems(left); | |
3532 | if (left_nritems == 0) | |
3533 | goto out_unlock; | |
3534 | ||
99d8f83c CM |
3535 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3536 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3537 | out_unlock: |
3538 | btrfs_tree_unlock(right); | |
3539 | free_extent_buffer(right); | |
3540 | return 1; | |
3541 | } | |
3542 | ||
74123bd7 CM |
3543 | /* |
3544 | * push some data in the path leaf to the left, trying to free up at | |
3545 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3546 | * |
3547 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3548 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3549 | * items | |
74123bd7 | 3550 | */ |
44871b1b CM |
3551 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3552 | struct btrfs_root *root, | |
3553 | struct btrfs_path *path, int data_size, | |
3554 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3555 | int free_space, u32 right_nritems, |
3556 | u32 max_slot) | |
be0e5c09 | 3557 | { |
5f39d397 CM |
3558 | struct btrfs_disk_key disk_key; |
3559 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3560 | int i; |
be0e5c09 CM |
3561 | int push_space = 0; |
3562 | int push_items = 0; | |
0783fcfc | 3563 | struct btrfs_item *item; |
7518a238 | 3564 | u32 old_left_nritems; |
34a38218 | 3565 | u32 nr; |
aa5d6bed | 3566 | int ret = 0; |
db94535d CM |
3567 | u32 this_item_size; |
3568 | u32 old_left_item_size; | |
cfed81a0 CM |
3569 | struct btrfs_map_token token; |
3570 | ||
3571 | btrfs_init_map_token(&token); | |
be0e5c09 | 3572 | |
34a38218 | 3573 | if (empty) |
99d8f83c | 3574 | nr = min(right_nritems, max_slot); |
34a38218 | 3575 | else |
99d8f83c | 3576 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3577 | |
3578 | for (i = 0; i < nr; i++) { | |
5f39d397 | 3579 | item = btrfs_item_nr(right, i); |
db94535d | 3580 | |
31840ae1 ZY |
3581 | if (!empty && push_items > 0) { |
3582 | if (path->slots[0] < i) | |
3583 | break; | |
3584 | if (path->slots[0] == i) { | |
3585 | int space = btrfs_leaf_free_space(root, right); | |
3586 | if (space + push_space * 2 > free_space) | |
3587 | break; | |
3588 | } | |
3589 | } | |
3590 | ||
be0e5c09 | 3591 | if (path->slots[0] == i) |
87b29b20 | 3592 | push_space += data_size; |
db94535d CM |
3593 | |
3594 | this_item_size = btrfs_item_size(right, item); | |
3595 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3596 | break; |
db94535d | 3597 | |
be0e5c09 | 3598 | push_items++; |
db94535d CM |
3599 | push_space += this_item_size + sizeof(*item); |
3600 | } | |
3601 | ||
be0e5c09 | 3602 | if (push_items == 0) { |
925baedd CM |
3603 | ret = 1; |
3604 | goto out; | |
be0e5c09 | 3605 | } |
34a38218 | 3606 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3607 | WARN_ON(1); |
5f39d397 | 3608 | |
be0e5c09 | 3609 | /* push data from right to left */ |
5f39d397 CM |
3610 | copy_extent_buffer(left, right, |
3611 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3612 | btrfs_item_nr_offset(0), | |
3613 | push_items * sizeof(struct btrfs_item)); | |
3614 | ||
123abc88 | 3615 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3616 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3617 | |
3618 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3619 | leaf_data_end(root, left) - push_space, |
3620 | btrfs_leaf_data(right) + | |
5f39d397 | 3621 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3622 | push_space); |
5f39d397 | 3623 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3624 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3625 | |
db94535d | 3626 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3627 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3628 | u32 ioff; |
db94535d | 3629 | |
5f39d397 | 3630 | item = btrfs_item_nr(left, i); |
db94535d | 3631 | |
cfed81a0 CM |
3632 | ioff = btrfs_token_item_offset(left, item, &token); |
3633 | btrfs_set_token_item_offset(left, item, | |
3634 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3635 | &token); | |
be0e5c09 | 3636 | } |
5f39d397 | 3637 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3638 | |
3639 | /* fixup right node */ | |
31b1a2bd JL |
3640 | if (push_items > right_nritems) |
3641 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3642 | right_nritems); |
34a38218 CM |
3643 | |
3644 | if (push_items < right_nritems) { | |
3645 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3646 | leaf_data_end(root, right); | |
3647 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3648 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3649 | btrfs_leaf_data(right) + | |
3650 | leaf_data_end(root, right), push_space); | |
3651 | ||
3652 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3653 | btrfs_item_nr_offset(push_items), |
3654 | (btrfs_header_nritems(right) - push_items) * | |
3655 | sizeof(struct btrfs_item)); | |
34a38218 | 3656 | } |
eef1c494 Y |
3657 | right_nritems -= push_items; |
3658 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3659 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3660 | for (i = 0; i < right_nritems; i++) { |
3661 | item = btrfs_item_nr(right, i); | |
db94535d | 3662 | |
cfed81a0 CM |
3663 | push_space = push_space - btrfs_token_item_size(right, |
3664 | item, &token); | |
3665 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3666 | } |
eb60ceac | 3667 | |
5f39d397 | 3668 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3669 | if (right_nritems) |
3670 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3671 | else |
3672 | clean_tree_block(trans, root, right); | |
098f59c2 | 3673 | |
5f39d397 | 3674 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3675 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3676 | |
3677 | /* then fixup the leaf pointer in the path */ | |
3678 | if (path->slots[0] < push_items) { | |
3679 | path->slots[0] += old_left_nritems; | |
925baedd | 3680 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3681 | free_extent_buffer(path->nodes[0]); |
3682 | path->nodes[0] = left; | |
be0e5c09 CM |
3683 | path->slots[1] -= 1; |
3684 | } else { | |
925baedd | 3685 | btrfs_tree_unlock(left); |
5f39d397 | 3686 | free_extent_buffer(left); |
be0e5c09 CM |
3687 | path->slots[0] -= push_items; |
3688 | } | |
eb60ceac | 3689 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3690 | return ret; |
925baedd CM |
3691 | out: |
3692 | btrfs_tree_unlock(left); | |
3693 | free_extent_buffer(left); | |
3694 | return ret; | |
be0e5c09 CM |
3695 | } |
3696 | ||
44871b1b CM |
3697 | /* |
3698 | * push some data in the path leaf to the left, trying to free up at | |
3699 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3700 | * |
3701 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3702 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3703 | * items | |
44871b1b CM |
3704 | */ |
3705 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3706 | *root, struct btrfs_path *path, int min_data_size, |
3707 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3708 | { |
3709 | struct extent_buffer *right = path->nodes[0]; | |
3710 | struct extent_buffer *left; | |
3711 | int slot; | |
3712 | int free_space; | |
3713 | u32 right_nritems; | |
3714 | int ret = 0; | |
3715 | ||
3716 | slot = path->slots[1]; | |
3717 | if (slot == 0) | |
3718 | return 1; | |
3719 | if (!path->nodes[1]) | |
3720 | return 1; | |
3721 | ||
3722 | right_nritems = btrfs_header_nritems(right); | |
3723 | if (right_nritems == 0) | |
3724 | return 1; | |
3725 | ||
3726 | btrfs_assert_tree_locked(path->nodes[1]); | |
3727 | ||
3728 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3729 | if (left == NULL) |
3730 | return 1; | |
3731 | ||
44871b1b CM |
3732 | btrfs_tree_lock(left); |
3733 | btrfs_set_lock_blocking(left); | |
3734 | ||
3735 | free_space = btrfs_leaf_free_space(root, left); | |
3736 | if (free_space < data_size) { | |
3737 | ret = 1; | |
3738 | goto out; | |
3739 | } | |
3740 | ||
3741 | /* cow and double check */ | |
3742 | ret = btrfs_cow_block(trans, root, left, | |
3743 | path->nodes[1], slot - 1, &left); | |
3744 | if (ret) { | |
3745 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3746 | if (ret == -ENOSPC) |
3747 | ret = 1; | |
44871b1b CM |
3748 | goto out; |
3749 | } | |
3750 | ||
3751 | free_space = btrfs_leaf_free_space(root, left); | |
3752 | if (free_space < data_size) { | |
3753 | ret = 1; | |
3754 | goto out; | |
3755 | } | |
3756 | ||
99d8f83c CM |
3757 | return __push_leaf_left(trans, root, path, min_data_size, |
3758 | empty, left, free_space, right_nritems, | |
3759 | max_slot); | |
44871b1b CM |
3760 | out: |
3761 | btrfs_tree_unlock(left); | |
3762 | free_extent_buffer(left); | |
3763 | return ret; | |
3764 | } | |
3765 | ||
3766 | /* | |
3767 | * split the path's leaf in two, making sure there is at least data_size | |
3768 | * available for the resulting leaf level of the path. | |
44871b1b | 3769 | */ |
143bede5 JM |
3770 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3771 | struct btrfs_root *root, | |
3772 | struct btrfs_path *path, | |
3773 | struct extent_buffer *l, | |
3774 | struct extent_buffer *right, | |
3775 | int slot, int mid, int nritems) | |
44871b1b CM |
3776 | { |
3777 | int data_copy_size; | |
3778 | int rt_data_off; | |
3779 | int i; | |
44871b1b | 3780 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3781 | struct btrfs_map_token token; |
3782 | ||
3783 | btrfs_init_map_token(&token); | |
44871b1b CM |
3784 | |
3785 | nritems = nritems - mid; | |
3786 | btrfs_set_header_nritems(right, nritems); | |
3787 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3788 | ||
3789 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3790 | btrfs_item_nr_offset(mid), | |
3791 | nritems * sizeof(struct btrfs_item)); | |
3792 | ||
3793 | copy_extent_buffer(right, l, | |
3794 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3795 | data_copy_size, btrfs_leaf_data(l) + | |
3796 | leaf_data_end(root, l), data_copy_size); | |
3797 | ||
3798 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3799 | btrfs_item_end_nr(l, mid); | |
3800 | ||
3801 | for (i = 0; i < nritems; i++) { | |
3802 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3803 | u32 ioff; | |
3804 | ||
cfed81a0 CM |
3805 | ioff = btrfs_token_item_offset(right, item, &token); |
3806 | btrfs_set_token_item_offset(right, item, | |
3807 | ioff + rt_data_off, &token); | |
44871b1b CM |
3808 | } |
3809 | ||
44871b1b | 3810 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3811 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3812 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 3813 | path->slots[1] + 1, 1); |
44871b1b CM |
3814 | |
3815 | btrfs_mark_buffer_dirty(right); | |
3816 | btrfs_mark_buffer_dirty(l); | |
3817 | BUG_ON(path->slots[0] != slot); | |
3818 | ||
44871b1b CM |
3819 | if (mid <= slot) { |
3820 | btrfs_tree_unlock(path->nodes[0]); | |
3821 | free_extent_buffer(path->nodes[0]); | |
3822 | path->nodes[0] = right; | |
3823 | path->slots[0] -= mid; | |
3824 | path->slots[1] += 1; | |
3825 | } else { | |
3826 | btrfs_tree_unlock(right); | |
3827 | free_extent_buffer(right); | |
3828 | } | |
3829 | ||
3830 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3831 | } |
3832 | ||
99d8f83c CM |
3833 | /* |
3834 | * double splits happen when we need to insert a big item in the middle | |
3835 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3836 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3837 | * A B C | |
3838 | * | |
3839 | * We avoid this by trying to push the items on either side of our target | |
3840 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3841 | * completely. | |
3842 | */ | |
3843 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3844 | struct btrfs_root *root, | |
3845 | struct btrfs_path *path, | |
3846 | int data_size) | |
3847 | { | |
3848 | int ret; | |
3849 | int progress = 0; | |
3850 | int slot; | |
3851 | u32 nritems; | |
3852 | ||
3853 | slot = path->slots[0]; | |
3854 | ||
3855 | /* | |
3856 | * try to push all the items after our slot into the | |
3857 | * right leaf | |
3858 | */ | |
3859 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3860 | if (ret < 0) | |
3861 | return ret; | |
3862 | ||
3863 | if (ret == 0) | |
3864 | progress++; | |
3865 | ||
3866 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3867 | /* | |
3868 | * our goal is to get our slot at the start or end of a leaf. If | |
3869 | * we've done so we're done | |
3870 | */ | |
3871 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3872 | return 0; | |
3873 | ||
3874 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3875 | return 0; | |
3876 | ||
3877 | /* try to push all the items before our slot into the next leaf */ | |
3878 | slot = path->slots[0]; | |
3879 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3880 | if (ret < 0) | |
3881 | return ret; | |
3882 | ||
3883 | if (ret == 0) | |
3884 | progress++; | |
3885 | ||
3886 | if (progress) | |
3887 | return 0; | |
3888 | return 1; | |
3889 | } | |
3890 | ||
74123bd7 CM |
3891 | /* |
3892 | * split the path's leaf in two, making sure there is at least data_size | |
3893 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3894 | * |
3895 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3896 | */ |
e02119d5 CM |
3897 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3898 | struct btrfs_root *root, | |
3899 | struct btrfs_key *ins_key, | |
3900 | struct btrfs_path *path, int data_size, | |
3901 | int extend) | |
be0e5c09 | 3902 | { |
5d4f98a2 | 3903 | struct btrfs_disk_key disk_key; |
5f39d397 | 3904 | struct extent_buffer *l; |
7518a238 | 3905 | u32 nritems; |
eb60ceac CM |
3906 | int mid; |
3907 | int slot; | |
5f39d397 | 3908 | struct extent_buffer *right; |
d4dbff95 | 3909 | int ret = 0; |
aa5d6bed | 3910 | int wret; |
5d4f98a2 | 3911 | int split; |
cc0c5538 | 3912 | int num_doubles = 0; |
99d8f83c | 3913 | int tried_avoid_double = 0; |
aa5d6bed | 3914 | |
a5719521 YZ |
3915 | l = path->nodes[0]; |
3916 | slot = path->slots[0]; | |
3917 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3918 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3919 | return -EOVERFLOW; | |
3920 | ||
40689478 | 3921 | /* first try to make some room by pushing left and right */ |
33157e05 | 3922 | if (data_size && path->nodes[1]) { |
99d8f83c CM |
3923 | wret = push_leaf_right(trans, root, path, data_size, |
3924 | data_size, 0, 0); | |
d397712b | 3925 | if (wret < 0) |
eaee50e8 | 3926 | return wret; |
3685f791 | 3927 | if (wret) { |
99d8f83c CM |
3928 | wret = push_leaf_left(trans, root, path, data_size, |
3929 | data_size, 0, (u32)-1); | |
3685f791 CM |
3930 | if (wret < 0) |
3931 | return wret; | |
3932 | } | |
3933 | l = path->nodes[0]; | |
aa5d6bed | 3934 | |
3685f791 | 3935 | /* did the pushes work? */ |
87b29b20 | 3936 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3937 | return 0; |
3326d1b0 | 3938 | } |
aa5d6bed | 3939 | |
5c680ed6 | 3940 | if (!path->nodes[1]) { |
fdd99c72 | 3941 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3942 | if (ret) |
3943 | return ret; | |
3944 | } | |
cc0c5538 | 3945 | again: |
5d4f98a2 | 3946 | split = 1; |
cc0c5538 | 3947 | l = path->nodes[0]; |
eb60ceac | 3948 | slot = path->slots[0]; |
5f39d397 | 3949 | nritems = btrfs_header_nritems(l); |
d397712b | 3950 | mid = (nritems + 1) / 2; |
54aa1f4d | 3951 | |
5d4f98a2 YZ |
3952 | if (mid <= slot) { |
3953 | if (nritems == 1 || | |
3954 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
3955 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3956 | if (slot >= nritems) { | |
3957 | split = 0; | |
3958 | } else { | |
3959 | mid = slot; | |
3960 | if (mid != nritems && | |
3961 | leaf_space_used(l, mid, nritems - mid) + | |
3962 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3963 | if (data_size && !tried_avoid_double) |
3964 | goto push_for_double; | |
5d4f98a2 YZ |
3965 | split = 2; |
3966 | } | |
3967 | } | |
3968 | } | |
3969 | } else { | |
3970 | if (leaf_space_used(l, 0, mid) + data_size > | |
3971 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3972 | if (!extend && data_size && slot == 0) { | |
3973 | split = 0; | |
3974 | } else if ((extend || !data_size) && slot == 0) { | |
3975 | mid = 1; | |
3976 | } else { | |
3977 | mid = slot; | |
3978 | if (mid != nritems && | |
3979 | leaf_space_used(l, mid, nritems - mid) + | |
3980 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3981 | if (data_size && !tried_avoid_double) |
3982 | goto push_for_double; | |
5d4f98a2 YZ |
3983 | split = 2 ; |
3984 | } | |
3985 | } | |
3986 | } | |
3987 | } | |
3988 | ||
3989 | if (split == 0) | |
3990 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3991 | else | |
3992 | btrfs_item_key(l, &disk_key, mid); | |
3993 | ||
3994 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 3995 | root->root_key.objectid, |
5581a51a | 3996 | &disk_key, 0, l->start, 0); |
f0486c68 | 3997 | if (IS_ERR(right)) |
5f39d397 | 3998 | return PTR_ERR(right); |
f0486c68 YZ |
3999 | |
4000 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4001 | |
4002 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4003 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4004 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4005 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4006 | btrfs_set_header_owner(right, root->root_key.objectid); |
4007 | btrfs_set_header_level(right, 0); | |
4008 | write_extent_buffer(right, root->fs_info->fsid, | |
4009 | (unsigned long)btrfs_header_fsid(right), | |
4010 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
4011 | |
4012 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
4013 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
4014 | BTRFS_UUID_SIZE); | |
44871b1b | 4015 | |
5d4f98a2 YZ |
4016 | if (split == 0) { |
4017 | if (mid <= slot) { | |
4018 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4019 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4020 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4021 | btrfs_tree_unlock(path->nodes[0]); |
4022 | free_extent_buffer(path->nodes[0]); | |
4023 | path->nodes[0] = right; | |
4024 | path->slots[0] = 0; | |
4025 | path->slots[1] += 1; | |
4026 | } else { | |
4027 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4028 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4029 | path->slots[1], 1); |
5d4f98a2 YZ |
4030 | btrfs_tree_unlock(path->nodes[0]); |
4031 | free_extent_buffer(path->nodes[0]); | |
4032 | path->nodes[0] = right; | |
4033 | path->slots[0] = 0; | |
143bede5 | 4034 | if (path->slots[1] == 0) |
d6a0a126 | 4035 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4036 | } |
5d4f98a2 YZ |
4037 | btrfs_mark_buffer_dirty(right); |
4038 | return ret; | |
d4dbff95 | 4039 | } |
74123bd7 | 4040 | |
143bede5 | 4041 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4042 | |
5d4f98a2 | 4043 | if (split == 2) { |
cc0c5538 CM |
4044 | BUG_ON(num_doubles != 0); |
4045 | num_doubles++; | |
4046 | goto again; | |
a429e513 | 4047 | } |
44871b1b | 4048 | |
143bede5 | 4049 | return 0; |
99d8f83c CM |
4050 | |
4051 | push_for_double: | |
4052 | push_for_double_split(trans, root, path, data_size); | |
4053 | tried_avoid_double = 1; | |
4054 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4055 | return 0; | |
4056 | goto again; | |
be0e5c09 CM |
4057 | } |
4058 | ||
ad48fd75 YZ |
4059 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4060 | struct btrfs_root *root, | |
4061 | struct btrfs_path *path, int ins_len) | |
459931ec | 4062 | { |
ad48fd75 | 4063 | struct btrfs_key key; |
459931ec | 4064 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4065 | struct btrfs_file_extent_item *fi; |
4066 | u64 extent_len = 0; | |
4067 | u32 item_size; | |
4068 | int ret; | |
459931ec CM |
4069 | |
4070 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4071 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4072 | ||
4073 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4074 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4075 | ||
4076 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4077 | return 0; | |
459931ec CM |
4078 | |
4079 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4080 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4081 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4082 | struct btrfs_file_extent_item); | |
4083 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4084 | } | |
b3b4aa74 | 4085 | btrfs_release_path(path); |
459931ec | 4086 | |
459931ec | 4087 | path->keep_locks = 1; |
ad48fd75 YZ |
4088 | path->search_for_split = 1; |
4089 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4090 | path->search_for_split = 0; |
ad48fd75 YZ |
4091 | if (ret < 0) |
4092 | goto err; | |
459931ec | 4093 | |
ad48fd75 YZ |
4094 | ret = -EAGAIN; |
4095 | leaf = path->nodes[0]; | |
459931ec | 4096 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4097 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4098 | goto err; | |
4099 | ||
109f6aef CM |
4100 | /* the leaf has changed, it now has room. return now */ |
4101 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4102 | goto err; | |
4103 | ||
ad48fd75 YZ |
4104 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4105 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4106 | struct btrfs_file_extent_item); | |
4107 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4108 | goto err; | |
459931ec CM |
4109 | } |
4110 | ||
b9473439 | 4111 | btrfs_set_path_blocking(path); |
ad48fd75 | 4112 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4113 | if (ret) |
4114 | goto err; | |
459931ec | 4115 | |
ad48fd75 | 4116 | path->keep_locks = 0; |
b9473439 | 4117 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4118 | return 0; |
4119 | err: | |
4120 | path->keep_locks = 0; | |
4121 | return ret; | |
4122 | } | |
4123 | ||
4124 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4125 | struct btrfs_root *root, | |
4126 | struct btrfs_path *path, | |
4127 | struct btrfs_key *new_key, | |
4128 | unsigned long split_offset) | |
4129 | { | |
4130 | struct extent_buffer *leaf; | |
4131 | struct btrfs_item *item; | |
4132 | struct btrfs_item *new_item; | |
4133 | int slot; | |
4134 | char *buf; | |
4135 | u32 nritems; | |
4136 | u32 item_size; | |
4137 | u32 orig_offset; | |
4138 | struct btrfs_disk_key disk_key; | |
4139 | ||
b9473439 CM |
4140 | leaf = path->nodes[0]; |
4141 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4142 | ||
b4ce94de CM |
4143 | btrfs_set_path_blocking(path); |
4144 | ||
459931ec CM |
4145 | item = btrfs_item_nr(leaf, path->slots[0]); |
4146 | orig_offset = btrfs_item_offset(leaf, item); | |
4147 | item_size = btrfs_item_size(leaf, item); | |
4148 | ||
459931ec | 4149 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4150 | if (!buf) |
4151 | return -ENOMEM; | |
4152 | ||
459931ec CM |
4153 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4154 | path->slots[0]), item_size); | |
459931ec | 4155 | |
ad48fd75 | 4156 | slot = path->slots[0] + 1; |
459931ec | 4157 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4158 | if (slot != nritems) { |
4159 | /* shift the items */ | |
4160 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4161 | btrfs_item_nr_offset(slot), |
4162 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4163 | } |
4164 | ||
4165 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4166 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4167 | ||
4168 | new_item = btrfs_item_nr(leaf, slot); | |
4169 | ||
4170 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4171 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4172 | ||
4173 | btrfs_set_item_offset(leaf, item, | |
4174 | orig_offset + item_size - split_offset); | |
4175 | btrfs_set_item_size(leaf, item, split_offset); | |
4176 | ||
4177 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4178 | ||
4179 | /* write the data for the start of the original item */ | |
4180 | write_extent_buffer(leaf, buf, | |
4181 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4182 | split_offset); | |
4183 | ||
4184 | /* write the data for the new item */ | |
4185 | write_extent_buffer(leaf, buf + split_offset, | |
4186 | btrfs_item_ptr_offset(leaf, slot), | |
4187 | item_size - split_offset); | |
4188 | btrfs_mark_buffer_dirty(leaf); | |
4189 | ||
ad48fd75 | 4190 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4191 | kfree(buf); |
ad48fd75 YZ |
4192 | return 0; |
4193 | } | |
4194 | ||
4195 | /* | |
4196 | * This function splits a single item into two items, | |
4197 | * giving 'new_key' to the new item and splitting the | |
4198 | * old one at split_offset (from the start of the item). | |
4199 | * | |
4200 | * The path may be released by this operation. After | |
4201 | * the split, the path is pointing to the old item. The | |
4202 | * new item is going to be in the same node as the old one. | |
4203 | * | |
4204 | * Note, the item being split must be smaller enough to live alone on | |
4205 | * a tree block with room for one extra struct btrfs_item | |
4206 | * | |
4207 | * This allows us to split the item in place, keeping a lock on the | |
4208 | * leaf the entire time. | |
4209 | */ | |
4210 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4211 | struct btrfs_root *root, | |
4212 | struct btrfs_path *path, | |
4213 | struct btrfs_key *new_key, | |
4214 | unsigned long split_offset) | |
4215 | { | |
4216 | int ret; | |
4217 | ret = setup_leaf_for_split(trans, root, path, | |
4218 | sizeof(struct btrfs_item)); | |
4219 | if (ret) | |
4220 | return ret; | |
4221 | ||
4222 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4223 | return ret; |
4224 | } | |
4225 | ||
ad48fd75 YZ |
4226 | /* |
4227 | * This function duplicate a item, giving 'new_key' to the new item. | |
4228 | * It guarantees both items live in the same tree leaf and the new item | |
4229 | * is contiguous with the original item. | |
4230 | * | |
4231 | * This allows us to split file extent in place, keeping a lock on the | |
4232 | * leaf the entire time. | |
4233 | */ | |
4234 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4235 | struct btrfs_root *root, | |
4236 | struct btrfs_path *path, | |
4237 | struct btrfs_key *new_key) | |
4238 | { | |
4239 | struct extent_buffer *leaf; | |
4240 | int ret; | |
4241 | u32 item_size; | |
4242 | ||
4243 | leaf = path->nodes[0]; | |
4244 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4245 | ret = setup_leaf_for_split(trans, root, path, | |
4246 | item_size + sizeof(struct btrfs_item)); | |
4247 | if (ret) | |
4248 | return ret; | |
4249 | ||
4250 | path->slots[0]++; | |
afe5fea7 | 4251 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4252 | item_size, item_size + |
4253 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4254 | leaf = path->nodes[0]; |
4255 | memcpy_extent_buffer(leaf, | |
4256 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4257 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4258 | item_size); | |
4259 | return 0; | |
4260 | } | |
4261 | ||
d352ac68 CM |
4262 | /* |
4263 | * make the item pointed to by the path smaller. new_size indicates | |
4264 | * how small to make it, and from_end tells us if we just chop bytes | |
4265 | * off the end of the item or if we shift the item to chop bytes off | |
4266 | * the front. | |
4267 | */ | |
afe5fea7 | 4268 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4269 | u32 new_size, int from_end) |
b18c6685 | 4270 | { |
b18c6685 | 4271 | int slot; |
5f39d397 CM |
4272 | struct extent_buffer *leaf; |
4273 | struct btrfs_item *item; | |
b18c6685 CM |
4274 | u32 nritems; |
4275 | unsigned int data_end; | |
4276 | unsigned int old_data_start; | |
4277 | unsigned int old_size; | |
4278 | unsigned int size_diff; | |
4279 | int i; | |
cfed81a0 CM |
4280 | struct btrfs_map_token token; |
4281 | ||
4282 | btrfs_init_map_token(&token); | |
b18c6685 | 4283 | |
5f39d397 | 4284 | leaf = path->nodes[0]; |
179e29e4 CM |
4285 | slot = path->slots[0]; |
4286 | ||
4287 | old_size = btrfs_item_size_nr(leaf, slot); | |
4288 | if (old_size == new_size) | |
143bede5 | 4289 | return; |
b18c6685 | 4290 | |
5f39d397 | 4291 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4292 | data_end = leaf_data_end(root, leaf); |
4293 | ||
5f39d397 | 4294 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4295 | |
b18c6685 CM |
4296 | size_diff = old_size - new_size; |
4297 | ||
4298 | BUG_ON(slot < 0); | |
4299 | BUG_ON(slot >= nritems); | |
4300 | ||
4301 | /* | |
4302 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4303 | */ | |
4304 | /* first correct the data pointers */ | |
4305 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4306 | u32 ioff; |
4307 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4308 | |
cfed81a0 CM |
4309 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4310 | btrfs_set_token_item_offset(leaf, item, | |
4311 | ioff + size_diff, &token); | |
b18c6685 | 4312 | } |
db94535d | 4313 | |
b18c6685 | 4314 | /* shift the data */ |
179e29e4 CM |
4315 | if (from_end) { |
4316 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4317 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4318 | data_end, old_data_start + new_size - data_end); | |
4319 | } else { | |
4320 | struct btrfs_disk_key disk_key; | |
4321 | u64 offset; | |
4322 | ||
4323 | btrfs_item_key(leaf, &disk_key, slot); | |
4324 | ||
4325 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4326 | unsigned long ptr; | |
4327 | struct btrfs_file_extent_item *fi; | |
4328 | ||
4329 | fi = btrfs_item_ptr(leaf, slot, | |
4330 | struct btrfs_file_extent_item); | |
4331 | fi = (struct btrfs_file_extent_item *)( | |
4332 | (unsigned long)fi - size_diff); | |
4333 | ||
4334 | if (btrfs_file_extent_type(leaf, fi) == | |
4335 | BTRFS_FILE_EXTENT_INLINE) { | |
4336 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4337 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4338 | (unsigned long)fi, |
4339 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4340 | disk_bytenr)); |
4341 | } | |
4342 | } | |
4343 | ||
4344 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4345 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4346 | data_end, old_data_start - data_end); | |
4347 | ||
4348 | offset = btrfs_disk_key_offset(&disk_key); | |
4349 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4350 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4351 | if (slot == 0) | |
d6a0a126 | 4352 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4353 | } |
5f39d397 CM |
4354 | |
4355 | item = btrfs_item_nr(leaf, slot); | |
4356 | btrfs_set_item_size(leaf, item, new_size); | |
4357 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4358 | |
5f39d397 CM |
4359 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4360 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4361 | BUG(); |
5f39d397 | 4362 | } |
b18c6685 CM |
4363 | } |
4364 | ||
d352ac68 | 4365 | /* |
8f69dbd2 | 4366 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4367 | */ |
4b90c680 | 4368 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4369 | u32 data_size) |
6567e837 | 4370 | { |
6567e837 | 4371 | int slot; |
5f39d397 CM |
4372 | struct extent_buffer *leaf; |
4373 | struct btrfs_item *item; | |
6567e837 CM |
4374 | u32 nritems; |
4375 | unsigned int data_end; | |
4376 | unsigned int old_data; | |
4377 | unsigned int old_size; | |
4378 | int i; | |
cfed81a0 CM |
4379 | struct btrfs_map_token token; |
4380 | ||
4381 | btrfs_init_map_token(&token); | |
6567e837 | 4382 | |
5f39d397 | 4383 | leaf = path->nodes[0]; |
6567e837 | 4384 | |
5f39d397 | 4385 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4386 | data_end = leaf_data_end(root, leaf); |
4387 | ||
5f39d397 CM |
4388 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4389 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4390 | BUG(); |
5f39d397 | 4391 | } |
6567e837 | 4392 | slot = path->slots[0]; |
5f39d397 | 4393 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4394 | |
4395 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4396 | if (slot >= nritems) { |
4397 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
4398 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
4399 | slot, nritems); | |
3326d1b0 CM |
4400 | BUG_ON(1); |
4401 | } | |
6567e837 CM |
4402 | |
4403 | /* | |
4404 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4405 | */ | |
4406 | /* first correct the data pointers */ | |
4407 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4408 | u32 ioff; |
4409 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4410 | |
cfed81a0 CM |
4411 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4412 | btrfs_set_token_item_offset(leaf, item, | |
4413 | ioff - data_size, &token); | |
6567e837 | 4414 | } |
5f39d397 | 4415 | |
6567e837 | 4416 | /* shift the data */ |
5f39d397 | 4417 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4418 | data_end - data_size, btrfs_leaf_data(leaf) + |
4419 | data_end, old_data - data_end); | |
5f39d397 | 4420 | |
6567e837 | 4421 | data_end = old_data; |
5f39d397 CM |
4422 | old_size = btrfs_item_size_nr(leaf, slot); |
4423 | item = btrfs_item_nr(leaf, slot); | |
4424 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
4425 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4426 | |
5f39d397 CM |
4427 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4428 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4429 | BUG(); |
5f39d397 | 4430 | } |
6567e837 CM |
4431 | } |
4432 | ||
74123bd7 | 4433 | /* |
44871b1b CM |
4434 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4435 | * to save stack depth by doing the bulk of the work in a function | |
4436 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4437 | */ |
afe5fea7 | 4438 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4439 | struct btrfs_key *cpu_key, u32 *data_size, |
4440 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4441 | { |
5f39d397 | 4442 | struct btrfs_item *item; |
9c58309d | 4443 | int i; |
7518a238 | 4444 | u32 nritems; |
be0e5c09 | 4445 | unsigned int data_end; |
e2fa7227 | 4446 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4447 | struct extent_buffer *leaf; |
4448 | int slot; | |
cfed81a0 CM |
4449 | struct btrfs_map_token token; |
4450 | ||
4451 | btrfs_init_map_token(&token); | |
e2fa7227 | 4452 | |
5f39d397 | 4453 | leaf = path->nodes[0]; |
44871b1b | 4454 | slot = path->slots[0]; |
74123bd7 | 4455 | |
5f39d397 | 4456 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4457 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4458 | |
f25956cc | 4459 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4460 | btrfs_print_leaf(root, leaf); |
d397712b | 4461 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4462 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4463 | BUG(); |
d4dbff95 | 4464 | } |
5f39d397 | 4465 | |
be0e5c09 | 4466 | if (slot != nritems) { |
5f39d397 | 4467 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4468 | |
5f39d397 CM |
4469 | if (old_data < data_end) { |
4470 | btrfs_print_leaf(root, leaf); | |
d397712b | 4471 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4472 | slot, old_data, data_end); |
4473 | BUG_ON(1); | |
4474 | } | |
be0e5c09 CM |
4475 | /* |
4476 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4477 | */ | |
4478 | /* first correct the data pointers */ | |
0783fcfc | 4479 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4480 | u32 ioff; |
db94535d | 4481 | |
5f39d397 | 4482 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4483 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4484 | btrfs_set_token_item_offset(leaf, item, | |
4485 | ioff - total_data, &token); | |
0783fcfc | 4486 | } |
be0e5c09 | 4487 | /* shift the items */ |
9c58309d | 4488 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4489 | btrfs_item_nr_offset(slot), |
d6025579 | 4490 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4491 | |
4492 | /* shift the data */ | |
5f39d397 | 4493 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4494 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4495 | data_end, old_data - data_end); |
be0e5c09 CM |
4496 | data_end = old_data; |
4497 | } | |
5f39d397 | 4498 | |
62e2749e | 4499 | /* setup the item for the new data */ |
9c58309d CM |
4500 | for (i = 0; i < nr; i++) { |
4501 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4502 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4503 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4504 | btrfs_set_token_item_offset(leaf, item, |
4505 | data_end - data_size[i], &token); | |
9c58309d | 4506 | data_end -= data_size[i]; |
cfed81a0 | 4507 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4508 | } |
44871b1b | 4509 | |
9c58309d | 4510 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4511 | |
5a01a2e3 CM |
4512 | if (slot == 0) { |
4513 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4514 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4515 | } |
b9473439 CM |
4516 | btrfs_unlock_up_safe(path, 1); |
4517 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4518 | |
5f39d397 CM |
4519 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4520 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4521 | BUG(); |
5f39d397 | 4522 | } |
44871b1b CM |
4523 | } |
4524 | ||
4525 | /* | |
4526 | * Given a key and some data, insert items into the tree. | |
4527 | * This does all the path init required, making room in the tree if needed. | |
4528 | */ | |
4529 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4530 | struct btrfs_root *root, | |
4531 | struct btrfs_path *path, | |
4532 | struct btrfs_key *cpu_key, u32 *data_size, | |
4533 | int nr) | |
4534 | { | |
44871b1b CM |
4535 | int ret = 0; |
4536 | int slot; | |
4537 | int i; | |
4538 | u32 total_size = 0; | |
4539 | u32 total_data = 0; | |
4540 | ||
4541 | for (i = 0; i < nr; i++) | |
4542 | total_data += data_size[i]; | |
4543 | ||
4544 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4545 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4546 | if (ret == 0) | |
4547 | return -EEXIST; | |
4548 | if (ret < 0) | |
143bede5 | 4549 | return ret; |
44871b1b | 4550 | |
44871b1b CM |
4551 | slot = path->slots[0]; |
4552 | BUG_ON(slot < 0); | |
4553 | ||
afe5fea7 | 4554 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4555 | total_data, total_size, nr); |
143bede5 | 4556 | return 0; |
62e2749e CM |
4557 | } |
4558 | ||
4559 | /* | |
4560 | * Given a key and some data, insert an item into the tree. | |
4561 | * This does all the path init required, making room in the tree if needed. | |
4562 | */ | |
e089f05c CM |
4563 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4564 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4565 | data_size) | |
62e2749e CM |
4566 | { |
4567 | int ret = 0; | |
2c90e5d6 | 4568 | struct btrfs_path *path; |
5f39d397 CM |
4569 | struct extent_buffer *leaf; |
4570 | unsigned long ptr; | |
62e2749e | 4571 | |
2c90e5d6 | 4572 | path = btrfs_alloc_path(); |
db5b493a TI |
4573 | if (!path) |
4574 | return -ENOMEM; | |
2c90e5d6 | 4575 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4576 | if (!ret) { |
5f39d397 CM |
4577 | leaf = path->nodes[0]; |
4578 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4579 | write_extent_buffer(leaf, data, ptr, data_size); | |
4580 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4581 | } |
2c90e5d6 | 4582 | btrfs_free_path(path); |
aa5d6bed | 4583 | return ret; |
be0e5c09 CM |
4584 | } |
4585 | ||
74123bd7 | 4586 | /* |
5de08d7d | 4587 | * delete the pointer from a given node. |
74123bd7 | 4588 | * |
d352ac68 CM |
4589 | * the tree should have been previously balanced so the deletion does not |
4590 | * empty a node. | |
74123bd7 | 4591 | */ |
afe5fea7 TI |
4592 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4593 | int level, int slot) | |
be0e5c09 | 4594 | { |
5f39d397 | 4595 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4596 | u32 nritems; |
f3ea38da | 4597 | int ret; |
be0e5c09 | 4598 | |
5f39d397 | 4599 | nritems = btrfs_header_nritems(parent); |
d397712b | 4600 | if (slot != nritems - 1) { |
0e411ece | 4601 | if (level) |
f3ea38da JS |
4602 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4603 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4604 | memmove_extent_buffer(parent, |
4605 | btrfs_node_key_ptr_offset(slot), | |
4606 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4607 | sizeof(struct btrfs_key_ptr) * |
4608 | (nritems - slot - 1)); | |
57ba86c0 CM |
4609 | } else if (level) { |
4610 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4611 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4612 | BUG_ON(ret < 0); |
bb803951 | 4613 | } |
f3ea38da | 4614 | |
7518a238 | 4615 | nritems--; |
5f39d397 | 4616 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4617 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4618 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4619 | /* just turn the root into a leaf and break */ |
5f39d397 | 4620 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4621 | } else if (slot == 0) { |
5f39d397 CM |
4622 | struct btrfs_disk_key disk_key; |
4623 | ||
4624 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4625 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4626 | } |
d6025579 | 4627 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4628 | } |
4629 | ||
323ac95b CM |
4630 | /* |
4631 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4632 | * path->nodes[1]. |
323ac95b CM |
4633 | * |
4634 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4635 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4636 | * | |
4637 | * The path must have already been setup for deleting the leaf, including | |
4638 | * all the proper balancing. path->nodes[1] must be locked. | |
4639 | */ | |
143bede5 JM |
4640 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4641 | struct btrfs_root *root, | |
4642 | struct btrfs_path *path, | |
4643 | struct extent_buffer *leaf) | |
323ac95b | 4644 | { |
5d4f98a2 | 4645 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4646 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4647 | |
4d081c41 CM |
4648 | /* |
4649 | * btrfs_free_extent is expensive, we want to make sure we | |
4650 | * aren't holding any locks when we call it | |
4651 | */ | |
4652 | btrfs_unlock_up_safe(path, 0); | |
4653 | ||
f0486c68 YZ |
4654 | root_sub_used(root, leaf->len); |
4655 | ||
3083ee2e | 4656 | extent_buffer_get(leaf); |
5581a51a | 4657 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4658 | free_extent_buffer_stale(leaf); |
323ac95b | 4659 | } |
74123bd7 CM |
4660 | /* |
4661 | * delete the item at the leaf level in path. If that empties | |
4662 | * the leaf, remove it from the tree | |
4663 | */ | |
85e21bac CM |
4664 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4665 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4666 | { |
5f39d397 CM |
4667 | struct extent_buffer *leaf; |
4668 | struct btrfs_item *item; | |
85e21bac CM |
4669 | int last_off; |
4670 | int dsize = 0; | |
aa5d6bed CM |
4671 | int ret = 0; |
4672 | int wret; | |
85e21bac | 4673 | int i; |
7518a238 | 4674 | u32 nritems; |
cfed81a0 CM |
4675 | struct btrfs_map_token token; |
4676 | ||
4677 | btrfs_init_map_token(&token); | |
be0e5c09 | 4678 | |
5f39d397 | 4679 | leaf = path->nodes[0]; |
85e21bac CM |
4680 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4681 | ||
4682 | for (i = 0; i < nr; i++) | |
4683 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4684 | ||
5f39d397 | 4685 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4686 | |
85e21bac | 4687 | if (slot + nr != nritems) { |
123abc88 | 4688 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4689 | |
4690 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4691 | data_end + dsize, |
4692 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4693 | last_off - data_end); |
5f39d397 | 4694 | |
85e21bac | 4695 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4696 | u32 ioff; |
db94535d | 4697 | |
5f39d397 | 4698 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4699 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4700 | btrfs_set_token_item_offset(leaf, item, | |
4701 | ioff + dsize, &token); | |
0783fcfc | 4702 | } |
db94535d | 4703 | |
5f39d397 | 4704 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4705 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4706 | sizeof(struct btrfs_item) * |
85e21bac | 4707 | (nritems - slot - nr)); |
be0e5c09 | 4708 | } |
85e21bac CM |
4709 | btrfs_set_header_nritems(leaf, nritems - nr); |
4710 | nritems -= nr; | |
5f39d397 | 4711 | |
74123bd7 | 4712 | /* delete the leaf if we've emptied it */ |
7518a238 | 4713 | if (nritems == 0) { |
5f39d397 CM |
4714 | if (leaf == root->node) { |
4715 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4716 | } else { |
f0486c68 YZ |
4717 | btrfs_set_path_blocking(path); |
4718 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4719 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4720 | } |
be0e5c09 | 4721 | } else { |
7518a238 | 4722 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4723 | if (slot == 0) { |
5f39d397 CM |
4724 | struct btrfs_disk_key disk_key; |
4725 | ||
4726 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 4727 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 4728 | } |
aa5d6bed | 4729 | |
74123bd7 | 4730 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4731 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4732 | /* push_leaf_left fixes the path. |
4733 | * make sure the path still points to our leaf | |
4734 | * for possible call to del_ptr below | |
4735 | */ | |
4920c9ac | 4736 | slot = path->slots[1]; |
5f39d397 CM |
4737 | extent_buffer_get(leaf); |
4738 | ||
b9473439 | 4739 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4740 | wret = push_leaf_left(trans, root, path, 1, 1, |
4741 | 1, (u32)-1); | |
54aa1f4d | 4742 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4743 | ret = wret; |
5f39d397 CM |
4744 | |
4745 | if (path->nodes[0] == leaf && | |
4746 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4747 | wret = push_leaf_right(trans, root, path, 1, |
4748 | 1, 1, 0); | |
54aa1f4d | 4749 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4750 | ret = wret; |
4751 | } | |
5f39d397 CM |
4752 | |
4753 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4754 | path->slots[1] = slot; |
143bede5 | 4755 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4756 | free_extent_buffer(leaf); |
143bede5 | 4757 | ret = 0; |
5de08d7d | 4758 | } else { |
925baedd CM |
4759 | /* if we're still in the path, make sure |
4760 | * we're dirty. Otherwise, one of the | |
4761 | * push_leaf functions must have already | |
4762 | * dirtied this buffer | |
4763 | */ | |
4764 | if (path->nodes[0] == leaf) | |
4765 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4766 | free_extent_buffer(leaf); |
be0e5c09 | 4767 | } |
d5719762 | 4768 | } else { |
5f39d397 | 4769 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4770 | } |
4771 | } | |
aa5d6bed | 4772 | return ret; |
be0e5c09 CM |
4773 | } |
4774 | ||
7bb86316 | 4775 | /* |
925baedd | 4776 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4777 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4778 | * returns < 0 on io errors. | |
d352ac68 CM |
4779 | * |
4780 | * This may release the path, and so you may lose any locks held at the | |
4781 | * time you call it. | |
7bb86316 | 4782 | */ |
35a3621b | 4783 | static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4784 | { |
925baedd CM |
4785 | struct btrfs_key key; |
4786 | struct btrfs_disk_key found_key; | |
4787 | int ret; | |
7bb86316 | 4788 | |
925baedd | 4789 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4790 | |
925baedd CM |
4791 | if (key.offset > 0) |
4792 | key.offset--; | |
4793 | else if (key.type > 0) | |
4794 | key.type--; | |
4795 | else if (key.objectid > 0) | |
4796 | key.objectid--; | |
4797 | else | |
4798 | return 1; | |
7bb86316 | 4799 | |
b3b4aa74 | 4800 | btrfs_release_path(path); |
925baedd CM |
4801 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4802 | if (ret < 0) | |
4803 | return ret; | |
4804 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4805 | ret = comp_keys(&found_key, &key); | |
4806 | if (ret < 0) | |
4807 | return 0; | |
4808 | return 1; | |
7bb86316 CM |
4809 | } |
4810 | ||
3f157a2f CM |
4811 | /* |
4812 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4813 | * for nodes or leaves that are have a minimum transaction id. |
4814 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4815 | * |
4816 | * This does not cow, but it does stuff the starting key it finds back | |
4817 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4818 | * key and get a writable path. | |
4819 | * | |
4820 | * This does lock as it descends, and path->keep_locks should be set | |
4821 | * to 1 by the caller. | |
4822 | * | |
4823 | * This honors path->lowest_level to prevent descent past a given level | |
4824 | * of the tree. | |
4825 | * | |
d352ac68 CM |
4826 | * min_trans indicates the oldest transaction that you are interested |
4827 | * in walking through. Any nodes or leaves older than min_trans are | |
4828 | * skipped over (without reading them). | |
4829 | * | |
3f157a2f CM |
4830 | * returns zero if something useful was found, < 0 on error and 1 if there |
4831 | * was nothing in the tree that matched the search criteria. | |
4832 | */ | |
4833 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4834 | struct btrfs_key *max_key, |
de78b51a | 4835 | struct btrfs_path *path, |
3f157a2f CM |
4836 | u64 min_trans) |
4837 | { | |
4838 | struct extent_buffer *cur; | |
4839 | struct btrfs_key found_key; | |
4840 | int slot; | |
9652480b | 4841 | int sret; |
3f157a2f CM |
4842 | u32 nritems; |
4843 | int level; | |
4844 | int ret = 1; | |
4845 | ||
934d375b | 4846 | WARN_ON(!path->keep_locks); |
3f157a2f | 4847 | again: |
bd681513 | 4848 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4849 | level = btrfs_header_level(cur); |
e02119d5 | 4850 | WARN_ON(path->nodes[level]); |
3f157a2f | 4851 | path->nodes[level] = cur; |
bd681513 | 4852 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4853 | |
4854 | if (btrfs_header_generation(cur) < min_trans) { | |
4855 | ret = 1; | |
4856 | goto out; | |
4857 | } | |
d397712b | 4858 | while (1) { |
3f157a2f CM |
4859 | nritems = btrfs_header_nritems(cur); |
4860 | level = btrfs_header_level(cur); | |
9652480b | 4861 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4862 | |
323ac95b CM |
4863 | /* at the lowest level, we're done, setup the path and exit */ |
4864 | if (level == path->lowest_level) { | |
e02119d5 CM |
4865 | if (slot >= nritems) |
4866 | goto find_next_key; | |
3f157a2f CM |
4867 | ret = 0; |
4868 | path->slots[level] = slot; | |
4869 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4870 | goto out; | |
4871 | } | |
9652480b Y |
4872 | if (sret && slot > 0) |
4873 | slot--; | |
3f157a2f | 4874 | /* |
de78b51a ES |
4875 | * check this node pointer against the min_trans parameters. |
4876 | * If it is too old, old, skip to the next one. | |
3f157a2f | 4877 | */ |
d397712b | 4878 | while (slot < nritems) { |
3f157a2f CM |
4879 | u64 blockptr; |
4880 | u64 gen; | |
e02119d5 | 4881 | |
3f157a2f CM |
4882 | blockptr = btrfs_node_blockptr(cur, slot); |
4883 | gen = btrfs_node_ptr_generation(cur, slot); | |
4884 | if (gen < min_trans) { | |
4885 | slot++; | |
4886 | continue; | |
4887 | } | |
de78b51a | 4888 | break; |
3f157a2f | 4889 | } |
e02119d5 | 4890 | find_next_key: |
3f157a2f CM |
4891 | /* |
4892 | * we didn't find a candidate key in this node, walk forward | |
4893 | * and find another one | |
4894 | */ | |
4895 | if (slot >= nritems) { | |
e02119d5 | 4896 | path->slots[level] = slot; |
b4ce94de | 4897 | btrfs_set_path_blocking(path); |
e02119d5 | 4898 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 4899 | min_trans); |
e02119d5 | 4900 | if (sret == 0) { |
b3b4aa74 | 4901 | btrfs_release_path(path); |
3f157a2f CM |
4902 | goto again; |
4903 | } else { | |
4904 | goto out; | |
4905 | } | |
4906 | } | |
4907 | /* save our key for returning back */ | |
4908 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4909 | path->slots[level] = slot; | |
4910 | if (level == path->lowest_level) { | |
4911 | ret = 0; | |
f7c79f30 | 4912 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4913 | goto out; |
4914 | } | |
b4ce94de | 4915 | btrfs_set_path_blocking(path); |
3f157a2f | 4916 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4917 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4918 | |
bd681513 | 4919 | btrfs_tree_read_lock(cur); |
b4ce94de | 4920 | |
bd681513 | 4921 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4922 | path->nodes[level - 1] = cur; |
f7c79f30 | 4923 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4924 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4925 | } |
4926 | out: | |
4927 | if (ret == 0) | |
4928 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4929 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4930 | return ret; |
4931 | } | |
4932 | ||
7069830a AB |
4933 | static void tree_move_down(struct btrfs_root *root, |
4934 | struct btrfs_path *path, | |
4935 | int *level, int root_level) | |
4936 | { | |
74dd17fb | 4937 | BUG_ON(*level == 0); |
7069830a AB |
4938 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
4939 | path->slots[*level]); | |
4940 | path->slots[*level - 1] = 0; | |
4941 | (*level)--; | |
4942 | } | |
4943 | ||
4944 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
4945 | struct btrfs_path *path, | |
4946 | int *level, int root_level) | |
4947 | { | |
4948 | int ret = 0; | |
4949 | int nritems; | |
4950 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4951 | ||
4952 | path->slots[*level]++; | |
4953 | ||
74dd17fb | 4954 | while (path->slots[*level] >= nritems) { |
7069830a AB |
4955 | if (*level == root_level) |
4956 | return -1; | |
4957 | ||
4958 | /* move upnext */ | |
4959 | path->slots[*level] = 0; | |
4960 | free_extent_buffer(path->nodes[*level]); | |
4961 | path->nodes[*level] = NULL; | |
4962 | (*level)++; | |
4963 | path->slots[*level]++; | |
4964 | ||
4965 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4966 | ret = 1; | |
4967 | } | |
4968 | return ret; | |
4969 | } | |
4970 | ||
4971 | /* | |
4972 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
4973 | * or down. | |
4974 | */ | |
4975 | static int tree_advance(struct btrfs_root *root, | |
4976 | struct btrfs_path *path, | |
4977 | int *level, int root_level, | |
4978 | int allow_down, | |
4979 | struct btrfs_key *key) | |
4980 | { | |
4981 | int ret; | |
4982 | ||
4983 | if (*level == 0 || !allow_down) { | |
4984 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
4985 | } else { | |
4986 | tree_move_down(root, path, level, root_level); | |
4987 | ret = 0; | |
4988 | } | |
4989 | if (ret >= 0) { | |
4990 | if (*level == 0) | |
4991 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
4992 | path->slots[*level]); | |
4993 | else | |
4994 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
4995 | path->slots[*level]); | |
4996 | } | |
4997 | return ret; | |
4998 | } | |
4999 | ||
5000 | static int tree_compare_item(struct btrfs_root *left_root, | |
5001 | struct btrfs_path *left_path, | |
5002 | struct btrfs_path *right_path, | |
5003 | char *tmp_buf) | |
5004 | { | |
5005 | int cmp; | |
5006 | int len1, len2; | |
5007 | unsigned long off1, off2; | |
5008 | ||
5009 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5010 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5011 | if (len1 != len2) | |
5012 | return 1; | |
5013 | ||
5014 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5015 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5016 | right_path->slots[0]); | |
5017 | ||
5018 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5019 | ||
5020 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5021 | if (cmp) | |
5022 | return 1; | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | #define ADVANCE 1 | |
5027 | #define ADVANCE_ONLY_NEXT -1 | |
5028 | ||
5029 | /* | |
5030 | * This function compares two trees and calls the provided callback for | |
5031 | * every changed/new/deleted item it finds. | |
5032 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5033 | * the compare pretty fast on snapshotted subvolumes. | |
5034 | * | |
5035 | * This currently works on commit roots only. As commit roots are read only, | |
5036 | * we don't do any locking. The commit roots are protected with transactions. | |
5037 | * Transactions are ended and rejoined when a commit is tried in between. | |
5038 | * | |
5039 | * This function checks for modifications done to the trees while comparing. | |
5040 | * If it detects a change, it aborts immediately. | |
5041 | */ | |
5042 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5043 | struct btrfs_root *right_root, | |
5044 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5045 | { | |
5046 | int ret; | |
5047 | int cmp; | |
5048 | struct btrfs_trans_handle *trans = NULL; | |
5049 | struct btrfs_path *left_path = NULL; | |
5050 | struct btrfs_path *right_path = NULL; | |
5051 | struct btrfs_key left_key; | |
5052 | struct btrfs_key right_key; | |
5053 | char *tmp_buf = NULL; | |
5054 | int left_root_level; | |
5055 | int right_root_level; | |
5056 | int left_level; | |
5057 | int right_level; | |
5058 | int left_end_reached; | |
5059 | int right_end_reached; | |
5060 | int advance_left; | |
5061 | int advance_right; | |
5062 | u64 left_blockptr; | |
5063 | u64 right_blockptr; | |
5064 | u64 left_start_ctransid; | |
5065 | u64 right_start_ctransid; | |
5066 | u64 ctransid; | |
5067 | ||
5068 | left_path = btrfs_alloc_path(); | |
5069 | if (!left_path) { | |
5070 | ret = -ENOMEM; | |
5071 | goto out; | |
5072 | } | |
5073 | right_path = btrfs_alloc_path(); | |
5074 | if (!right_path) { | |
5075 | ret = -ENOMEM; | |
5076 | goto out; | |
5077 | } | |
5078 | ||
5079 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5080 | if (!tmp_buf) { | |
5081 | ret = -ENOMEM; | |
5082 | goto out; | |
5083 | } | |
5084 | ||
5085 | left_path->search_commit_root = 1; | |
5086 | left_path->skip_locking = 1; | |
5087 | right_path->search_commit_root = 1; | |
5088 | right_path->skip_locking = 1; | |
5089 | ||
5f3ab90a | 5090 | spin_lock(&left_root->root_item_lock); |
7069830a | 5091 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5092 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5093 | |
5f3ab90a | 5094 | spin_lock(&right_root->root_item_lock); |
7069830a | 5095 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5096 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5097 | |
5098 | trans = btrfs_join_transaction(left_root); | |
5099 | if (IS_ERR(trans)) { | |
5100 | ret = PTR_ERR(trans); | |
5101 | trans = NULL; | |
5102 | goto out; | |
5103 | } | |
5104 | ||
5105 | /* | |
5106 | * Strategy: Go to the first items of both trees. Then do | |
5107 | * | |
5108 | * If both trees are at level 0 | |
5109 | * Compare keys of current items | |
5110 | * If left < right treat left item as new, advance left tree | |
5111 | * and repeat | |
5112 | * If left > right treat right item as deleted, advance right tree | |
5113 | * and repeat | |
5114 | * If left == right do deep compare of items, treat as changed if | |
5115 | * needed, advance both trees and repeat | |
5116 | * If both trees are at the same level but not at level 0 | |
5117 | * Compare keys of current nodes/leafs | |
5118 | * If left < right advance left tree and repeat | |
5119 | * If left > right advance right tree and repeat | |
5120 | * If left == right compare blockptrs of the next nodes/leafs | |
5121 | * If they match advance both trees but stay at the same level | |
5122 | * and repeat | |
5123 | * If they don't match advance both trees while allowing to go | |
5124 | * deeper and repeat | |
5125 | * If tree levels are different | |
5126 | * Advance the tree that needs it and repeat | |
5127 | * | |
5128 | * Advancing a tree means: | |
5129 | * If we are at level 0, try to go to the next slot. If that's not | |
5130 | * possible, go one level up and repeat. Stop when we found a level | |
5131 | * where we could go to the next slot. We may at this point be on a | |
5132 | * node or a leaf. | |
5133 | * | |
5134 | * If we are not at level 0 and not on shared tree blocks, go one | |
5135 | * level deeper. | |
5136 | * | |
5137 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5138 | * the right if possible or go up and right. | |
5139 | */ | |
5140 | ||
5141 | left_level = btrfs_header_level(left_root->commit_root); | |
5142 | left_root_level = left_level; | |
5143 | left_path->nodes[left_level] = left_root->commit_root; | |
5144 | extent_buffer_get(left_path->nodes[left_level]); | |
5145 | ||
5146 | right_level = btrfs_header_level(right_root->commit_root); | |
5147 | right_root_level = right_level; | |
5148 | right_path->nodes[right_level] = right_root->commit_root; | |
5149 | extent_buffer_get(right_path->nodes[right_level]); | |
5150 | ||
5151 | if (left_level == 0) | |
5152 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5153 | &left_key, left_path->slots[left_level]); | |
5154 | else | |
5155 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5156 | &left_key, left_path->slots[left_level]); | |
5157 | if (right_level == 0) | |
5158 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5159 | &right_key, right_path->slots[right_level]); | |
5160 | else | |
5161 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5162 | &right_key, right_path->slots[right_level]); | |
5163 | ||
5164 | left_end_reached = right_end_reached = 0; | |
5165 | advance_left = advance_right = 0; | |
5166 | ||
5167 | while (1) { | |
5168 | /* | |
5169 | * We need to make sure the transaction does not get committed | |
5170 | * while we do anything on commit roots. This means, we need to | |
5171 | * join and leave transactions for every item that we process. | |
5172 | */ | |
5173 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5174 | btrfs_release_path(left_path); | |
5175 | btrfs_release_path(right_path); | |
5176 | ||
5177 | ret = btrfs_end_transaction(trans, left_root); | |
5178 | trans = NULL; | |
5179 | if (ret < 0) | |
5180 | goto out; | |
5181 | } | |
5182 | /* now rejoin the transaction */ | |
5183 | if (!trans) { | |
5184 | trans = btrfs_join_transaction(left_root); | |
5185 | if (IS_ERR(trans)) { | |
5186 | ret = PTR_ERR(trans); | |
5187 | trans = NULL; | |
5188 | goto out; | |
5189 | } | |
5190 | ||
5f3ab90a | 5191 | spin_lock(&left_root->root_item_lock); |
7069830a | 5192 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5193 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5194 | if (ctransid != left_start_ctransid) |
5195 | left_start_ctransid = 0; | |
5196 | ||
5f3ab90a | 5197 | spin_lock(&right_root->root_item_lock); |
7069830a | 5198 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5199 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5200 | if (ctransid != right_start_ctransid) |
5201 | right_start_ctransid = 0; | |
5202 | ||
5203 | if (!left_start_ctransid || !right_start_ctransid) { | |
5204 | WARN(1, KERN_WARNING | |
5205 | "btrfs: btrfs_compare_tree detected " | |
5206 | "a change in one of the trees while " | |
5207 | "iterating. This is probably a " | |
5208 | "bug.\n"); | |
5209 | ret = -EIO; | |
5210 | goto out; | |
5211 | } | |
5212 | ||
5213 | /* | |
5214 | * the commit root may have changed, so start again | |
5215 | * where we stopped | |
5216 | */ | |
5217 | left_path->lowest_level = left_level; | |
5218 | right_path->lowest_level = right_level; | |
5219 | ret = btrfs_search_slot(NULL, left_root, | |
5220 | &left_key, left_path, 0, 0); | |
5221 | if (ret < 0) | |
5222 | goto out; | |
5223 | ret = btrfs_search_slot(NULL, right_root, | |
5224 | &right_key, right_path, 0, 0); | |
5225 | if (ret < 0) | |
5226 | goto out; | |
5227 | } | |
5228 | ||
5229 | if (advance_left && !left_end_reached) { | |
5230 | ret = tree_advance(left_root, left_path, &left_level, | |
5231 | left_root_level, | |
5232 | advance_left != ADVANCE_ONLY_NEXT, | |
5233 | &left_key); | |
5234 | if (ret < 0) | |
5235 | left_end_reached = ADVANCE; | |
5236 | advance_left = 0; | |
5237 | } | |
5238 | if (advance_right && !right_end_reached) { | |
5239 | ret = tree_advance(right_root, right_path, &right_level, | |
5240 | right_root_level, | |
5241 | advance_right != ADVANCE_ONLY_NEXT, | |
5242 | &right_key); | |
5243 | if (ret < 0) | |
5244 | right_end_reached = ADVANCE; | |
5245 | advance_right = 0; | |
5246 | } | |
5247 | ||
5248 | if (left_end_reached && right_end_reached) { | |
5249 | ret = 0; | |
5250 | goto out; | |
5251 | } else if (left_end_reached) { | |
5252 | if (right_level == 0) { | |
5253 | ret = changed_cb(left_root, right_root, | |
5254 | left_path, right_path, | |
5255 | &right_key, | |
5256 | BTRFS_COMPARE_TREE_DELETED, | |
5257 | ctx); | |
5258 | if (ret < 0) | |
5259 | goto out; | |
5260 | } | |
5261 | advance_right = ADVANCE; | |
5262 | continue; | |
5263 | } else if (right_end_reached) { | |
5264 | if (left_level == 0) { | |
5265 | ret = changed_cb(left_root, right_root, | |
5266 | left_path, right_path, | |
5267 | &left_key, | |
5268 | BTRFS_COMPARE_TREE_NEW, | |
5269 | ctx); | |
5270 | if (ret < 0) | |
5271 | goto out; | |
5272 | } | |
5273 | advance_left = ADVANCE; | |
5274 | continue; | |
5275 | } | |
5276 | ||
5277 | if (left_level == 0 && right_level == 0) { | |
5278 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5279 | if (cmp < 0) { | |
5280 | ret = changed_cb(left_root, right_root, | |
5281 | left_path, right_path, | |
5282 | &left_key, | |
5283 | BTRFS_COMPARE_TREE_NEW, | |
5284 | ctx); | |
5285 | if (ret < 0) | |
5286 | goto out; | |
5287 | advance_left = ADVANCE; | |
5288 | } else if (cmp > 0) { | |
5289 | ret = changed_cb(left_root, right_root, | |
5290 | left_path, right_path, | |
5291 | &right_key, | |
5292 | BTRFS_COMPARE_TREE_DELETED, | |
5293 | ctx); | |
5294 | if (ret < 0) | |
5295 | goto out; | |
5296 | advance_right = ADVANCE; | |
5297 | } else { | |
ba5e8f2e JB |
5298 | enum btrfs_compare_tree_result cmp; |
5299 | ||
74dd17fb | 5300 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5301 | ret = tree_compare_item(left_root, left_path, |
5302 | right_path, tmp_buf); | |
ba5e8f2e JB |
5303 | if (ret) |
5304 | cmp = BTRFS_COMPARE_TREE_CHANGED; | |
5305 | else | |
5306 | cmp = BTRFS_COMPARE_TREE_SAME; | |
5307 | ret = changed_cb(left_root, right_root, | |
5308 | left_path, right_path, | |
5309 | &left_key, cmp, ctx); | |
5310 | if (ret < 0) | |
5311 | goto out; | |
7069830a AB |
5312 | advance_left = ADVANCE; |
5313 | advance_right = ADVANCE; | |
5314 | } | |
5315 | } else if (left_level == right_level) { | |
5316 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5317 | if (cmp < 0) { | |
5318 | advance_left = ADVANCE; | |
5319 | } else if (cmp > 0) { | |
5320 | advance_right = ADVANCE; | |
5321 | } else { | |
5322 | left_blockptr = btrfs_node_blockptr( | |
5323 | left_path->nodes[left_level], | |
5324 | left_path->slots[left_level]); | |
5325 | right_blockptr = btrfs_node_blockptr( | |
5326 | right_path->nodes[right_level], | |
5327 | right_path->slots[right_level]); | |
5328 | if (left_blockptr == right_blockptr) { | |
5329 | /* | |
5330 | * As we're on a shared block, don't | |
5331 | * allow to go deeper. | |
5332 | */ | |
5333 | advance_left = ADVANCE_ONLY_NEXT; | |
5334 | advance_right = ADVANCE_ONLY_NEXT; | |
5335 | } else { | |
5336 | advance_left = ADVANCE; | |
5337 | advance_right = ADVANCE; | |
5338 | } | |
5339 | } | |
5340 | } else if (left_level < right_level) { | |
5341 | advance_right = ADVANCE; | |
5342 | } else { | |
5343 | advance_left = ADVANCE; | |
5344 | } | |
5345 | } | |
5346 | ||
5347 | out: | |
5348 | btrfs_free_path(left_path); | |
5349 | btrfs_free_path(right_path); | |
5350 | kfree(tmp_buf); | |
5351 | ||
5352 | if (trans) { | |
5353 | if (!ret) | |
5354 | ret = btrfs_end_transaction(trans, left_root); | |
5355 | else | |
5356 | btrfs_end_transaction(trans, left_root); | |
5357 | } | |
5358 | ||
5359 | return ret; | |
5360 | } | |
5361 | ||
3f157a2f CM |
5362 | /* |
5363 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5364 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5365 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5366 | * |
5367 | * 0 is returned if another key is found, < 0 if there are any errors | |
5368 | * and 1 is returned if there are no higher keys in the tree | |
5369 | * | |
5370 | * path->keep_locks should be set to 1 on the search made before | |
5371 | * calling this function. | |
5372 | */ | |
e7a84565 | 5373 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5374 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5375 | { |
e7a84565 CM |
5376 | int slot; |
5377 | struct extent_buffer *c; | |
5378 | ||
934d375b | 5379 | WARN_ON(!path->keep_locks); |
d397712b | 5380 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5381 | if (!path->nodes[level]) |
5382 | return 1; | |
5383 | ||
5384 | slot = path->slots[level] + 1; | |
5385 | c = path->nodes[level]; | |
3f157a2f | 5386 | next: |
e7a84565 | 5387 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5388 | int ret; |
5389 | int orig_lowest; | |
5390 | struct btrfs_key cur_key; | |
5391 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5392 | !path->nodes[level + 1]) | |
e7a84565 | 5393 | return 1; |
33c66f43 YZ |
5394 | |
5395 | if (path->locks[level + 1]) { | |
5396 | level++; | |
5397 | continue; | |
5398 | } | |
5399 | ||
5400 | slot = btrfs_header_nritems(c) - 1; | |
5401 | if (level == 0) | |
5402 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5403 | else | |
5404 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5405 | ||
5406 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5407 | btrfs_release_path(path); |
33c66f43 YZ |
5408 | path->lowest_level = level; |
5409 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5410 | 0, 0); | |
5411 | path->lowest_level = orig_lowest; | |
5412 | if (ret < 0) | |
5413 | return ret; | |
5414 | ||
5415 | c = path->nodes[level]; | |
5416 | slot = path->slots[level]; | |
5417 | if (ret == 0) | |
5418 | slot++; | |
5419 | goto next; | |
e7a84565 | 5420 | } |
33c66f43 | 5421 | |
e7a84565 CM |
5422 | if (level == 0) |
5423 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5424 | else { |
3f157a2f CM |
5425 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5426 | ||
3f157a2f CM |
5427 | if (gen < min_trans) { |
5428 | slot++; | |
5429 | goto next; | |
5430 | } | |
e7a84565 | 5431 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5432 | } |
e7a84565 CM |
5433 | return 0; |
5434 | } | |
5435 | return 1; | |
5436 | } | |
5437 | ||
97571fd0 | 5438 | /* |
925baedd | 5439 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5440 | * returns 0 if it found something or 1 if there are no greater leaves. |
5441 | * returns < 0 on io errors. | |
97571fd0 | 5442 | */ |
234b63a0 | 5443 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5444 | { |
5445 | return btrfs_next_old_leaf(root, path, 0); | |
5446 | } | |
5447 | ||
5448 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5449 | u64 time_seq) | |
d97e63b6 CM |
5450 | { |
5451 | int slot; | |
8e73f275 | 5452 | int level; |
5f39d397 | 5453 | struct extent_buffer *c; |
8e73f275 | 5454 | struct extent_buffer *next; |
925baedd CM |
5455 | struct btrfs_key key; |
5456 | u32 nritems; | |
5457 | int ret; | |
8e73f275 | 5458 | int old_spinning = path->leave_spinning; |
bd681513 | 5459 | int next_rw_lock = 0; |
925baedd CM |
5460 | |
5461 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5462 | if (nritems == 0) |
925baedd | 5463 | return 1; |
925baedd | 5464 | |
8e73f275 CM |
5465 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5466 | again: | |
5467 | level = 1; | |
5468 | next = NULL; | |
bd681513 | 5469 | next_rw_lock = 0; |
b3b4aa74 | 5470 | btrfs_release_path(path); |
8e73f275 | 5471 | |
a2135011 | 5472 | path->keep_locks = 1; |
31533fb2 | 5473 | path->leave_spinning = 1; |
8e73f275 | 5474 | |
3d7806ec JS |
5475 | if (time_seq) |
5476 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5477 | else | |
5478 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5479 | path->keep_locks = 0; |
5480 | ||
5481 | if (ret < 0) | |
5482 | return ret; | |
5483 | ||
a2135011 | 5484 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5485 | /* |
5486 | * by releasing the path above we dropped all our locks. A balance | |
5487 | * could have added more items next to the key that used to be | |
5488 | * at the very end of the block. So, check again here and | |
5489 | * advance the path if there are now more items available. | |
5490 | */ | |
a2135011 | 5491 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5492 | if (ret == 0) |
5493 | path->slots[0]++; | |
8e73f275 | 5494 | ret = 0; |
925baedd CM |
5495 | goto done; |
5496 | } | |
d97e63b6 | 5497 | |
d397712b | 5498 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5499 | if (!path->nodes[level]) { |
5500 | ret = 1; | |
5501 | goto done; | |
5502 | } | |
5f39d397 | 5503 | |
d97e63b6 CM |
5504 | slot = path->slots[level] + 1; |
5505 | c = path->nodes[level]; | |
5f39d397 | 5506 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5507 | level++; |
8e73f275 CM |
5508 | if (level == BTRFS_MAX_LEVEL) { |
5509 | ret = 1; | |
5510 | goto done; | |
5511 | } | |
d97e63b6 CM |
5512 | continue; |
5513 | } | |
5f39d397 | 5514 | |
925baedd | 5515 | if (next) { |
bd681513 | 5516 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5517 | free_extent_buffer(next); |
925baedd | 5518 | } |
5f39d397 | 5519 | |
8e73f275 | 5520 | next = c; |
bd681513 | 5521 | next_rw_lock = path->locks[level]; |
8e73f275 | 5522 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5523 | slot, &key, 0); |
8e73f275 CM |
5524 | if (ret == -EAGAIN) |
5525 | goto again; | |
5f39d397 | 5526 | |
76a05b35 | 5527 | if (ret < 0) { |
b3b4aa74 | 5528 | btrfs_release_path(path); |
76a05b35 CM |
5529 | goto done; |
5530 | } | |
5531 | ||
5cd57b2c | 5532 | if (!path->skip_locking) { |
bd681513 | 5533 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5534 | if (!ret && time_seq) { |
5535 | /* | |
5536 | * If we don't get the lock, we may be racing | |
5537 | * with push_leaf_left, holding that lock while | |
5538 | * itself waiting for the leaf we've currently | |
5539 | * locked. To solve this situation, we give up | |
5540 | * on our lock and cycle. | |
5541 | */ | |
cf538830 | 5542 | free_extent_buffer(next); |
d42244a0 JS |
5543 | btrfs_release_path(path); |
5544 | cond_resched(); | |
5545 | goto again; | |
5546 | } | |
8e73f275 CM |
5547 | if (!ret) { |
5548 | btrfs_set_path_blocking(path); | |
bd681513 | 5549 | btrfs_tree_read_lock(next); |
31533fb2 | 5550 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5551 | BTRFS_READ_LOCK); |
8e73f275 | 5552 | } |
31533fb2 | 5553 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5554 | } |
d97e63b6 CM |
5555 | break; |
5556 | } | |
5557 | path->slots[level] = slot; | |
d397712b | 5558 | while (1) { |
d97e63b6 CM |
5559 | level--; |
5560 | c = path->nodes[level]; | |
925baedd | 5561 | if (path->locks[level]) |
bd681513 | 5562 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5563 | |
5f39d397 | 5564 | free_extent_buffer(c); |
d97e63b6 CM |
5565 | path->nodes[level] = next; |
5566 | path->slots[level] = 0; | |
a74a4b97 | 5567 | if (!path->skip_locking) |
bd681513 | 5568 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5569 | if (!level) |
5570 | break; | |
b4ce94de | 5571 | |
8e73f275 | 5572 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5573 | 0, &key, 0); |
8e73f275 CM |
5574 | if (ret == -EAGAIN) |
5575 | goto again; | |
5576 | ||
76a05b35 | 5577 | if (ret < 0) { |
b3b4aa74 | 5578 | btrfs_release_path(path); |
76a05b35 CM |
5579 | goto done; |
5580 | } | |
5581 | ||
5cd57b2c | 5582 | if (!path->skip_locking) { |
bd681513 | 5583 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5584 | if (!ret) { |
5585 | btrfs_set_path_blocking(path); | |
bd681513 | 5586 | btrfs_tree_read_lock(next); |
31533fb2 | 5587 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5588 | BTRFS_READ_LOCK); |
5589 | } | |
31533fb2 | 5590 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5591 | } |
d97e63b6 | 5592 | } |
8e73f275 | 5593 | ret = 0; |
925baedd | 5594 | done: |
f7c79f30 | 5595 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5596 | path->leave_spinning = old_spinning; |
5597 | if (!old_spinning) | |
5598 | btrfs_set_path_blocking(path); | |
5599 | ||
5600 | return ret; | |
d97e63b6 | 5601 | } |
0b86a832 | 5602 | |
3f157a2f CM |
5603 | /* |
5604 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5605 | * searching until it gets past min_objectid or finds an item of 'type' | |
5606 | * | |
5607 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5608 | */ | |
0b86a832 CM |
5609 | int btrfs_previous_item(struct btrfs_root *root, |
5610 | struct btrfs_path *path, u64 min_objectid, | |
5611 | int type) | |
5612 | { | |
5613 | struct btrfs_key found_key; | |
5614 | struct extent_buffer *leaf; | |
e02119d5 | 5615 | u32 nritems; |
0b86a832 CM |
5616 | int ret; |
5617 | ||
d397712b | 5618 | while (1) { |
0b86a832 | 5619 | if (path->slots[0] == 0) { |
b4ce94de | 5620 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5621 | ret = btrfs_prev_leaf(root, path); |
5622 | if (ret != 0) | |
5623 | return ret; | |
5624 | } else { | |
5625 | path->slots[0]--; | |
5626 | } | |
5627 | leaf = path->nodes[0]; | |
e02119d5 CM |
5628 | nritems = btrfs_header_nritems(leaf); |
5629 | if (nritems == 0) | |
5630 | return 1; | |
5631 | if (path->slots[0] == nritems) | |
5632 | path->slots[0]--; | |
5633 | ||
0b86a832 | 5634 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5635 | if (found_key.objectid < min_objectid) |
5636 | break; | |
0a4eefbb YZ |
5637 | if (found_key.type == type) |
5638 | return 0; | |
e02119d5 CM |
5639 | if (found_key.objectid == min_objectid && |
5640 | found_key.type < type) | |
5641 | break; | |
0b86a832 CM |
5642 | } |
5643 | return 1; | |
5644 | } |