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 | |
fba6aa75 | 277 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(cow), |
2b82032c YZ |
278 | BTRFS_FSID_SIZE); |
279 | ||
be20aa9d | 280 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 281 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 282 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 283 | else |
66d7e7f0 | 284 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 285 | |
be20aa9d CM |
286 | if (ret) |
287 | return ret; | |
288 | ||
289 | btrfs_mark_buffer_dirty(cow); | |
290 | *cow_ret = cow; | |
291 | return 0; | |
292 | } | |
293 | ||
bd989ba3 JS |
294 | enum mod_log_op { |
295 | MOD_LOG_KEY_REPLACE, | |
296 | MOD_LOG_KEY_ADD, | |
297 | MOD_LOG_KEY_REMOVE, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
299 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
300 | MOD_LOG_MOVE_KEYS, | |
301 | MOD_LOG_ROOT_REPLACE, | |
302 | }; | |
303 | ||
304 | struct tree_mod_move { | |
305 | int dst_slot; | |
306 | int nr_items; | |
307 | }; | |
308 | ||
309 | struct tree_mod_root { | |
310 | u64 logical; | |
311 | u8 level; | |
312 | }; | |
313 | ||
314 | struct tree_mod_elem { | |
315 | struct rb_node node; | |
316 | u64 index; /* shifted logical */ | |
097b8a7c | 317 | u64 seq; |
bd989ba3 JS |
318 | enum mod_log_op op; |
319 | ||
320 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
321 | int slot; | |
322 | ||
323 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
324 | u64 generation; | |
325 | ||
326 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
327 | struct btrfs_disk_key key; | |
328 | u64 blockptr; | |
329 | ||
330 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
331 | struct tree_mod_move move; | |
332 | ||
333 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
334 | struct tree_mod_root old_root; | |
335 | }; | |
336 | ||
097b8a7c | 337 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 338 | { |
097b8a7c | 339 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
340 | } |
341 | ||
097b8a7c JS |
342 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
343 | { | |
344 | read_unlock(&fs_info->tree_mod_log_lock); | |
345 | } | |
346 | ||
347 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
348 | { | |
349 | write_lock(&fs_info->tree_mod_log_lock); | |
350 | } | |
351 | ||
352 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
353 | { | |
354 | write_unlock(&fs_info->tree_mod_log_lock); | |
355 | } | |
356 | ||
fc36ed7e JS |
357 | /* |
358 | * Increment the upper half of tree_mod_seq, set lower half zero. | |
359 | * | |
360 | * Must be called with fs_info->tree_mod_seq_lock held. | |
361 | */ | |
362 | static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info) | |
363 | { | |
364 | u64 seq = atomic64_read(&fs_info->tree_mod_seq); | |
365 | seq &= 0xffffffff00000000ull; | |
366 | seq += 1ull << 32; | |
367 | atomic64_set(&fs_info->tree_mod_seq, seq); | |
368 | return seq; | |
369 | } | |
370 | ||
371 | /* | |
372 | * Increment the lower half of tree_mod_seq. | |
373 | * | |
374 | * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers | |
375 | * are generated should not technically require a spin lock here. (Rationale: | |
376 | * incrementing the minor while incrementing the major seq number is between its | |
377 | * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it | |
378 | * just returns a unique sequence number as usual.) We have decided to leave | |
379 | * that requirement in here and rethink it once we notice it really imposes a | |
380 | * problem on some workload. | |
381 | */ | |
382 | static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info) | |
383 | { | |
384 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
385 | } | |
386 | ||
387 | /* | |
388 | * return the last minor in the previous major tree_mod_seq number | |
389 | */ | |
390 | u64 btrfs_tree_mod_seq_prev(u64 seq) | |
391 | { | |
392 | return (seq & 0xffffffff00000000ull) - 1ull; | |
393 | } | |
394 | ||
097b8a7c JS |
395 | /* |
396 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
397 | * passed does not already have a sequence number set. So when a caller expects | |
398 | * to record tree modifications, it should ensure to set elem->seq to zero | |
399 | * before calling btrfs_get_tree_mod_seq. | |
400 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
401 | * blocker was added. | |
402 | */ | |
403 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
404 | struct seq_list *elem) | |
bd989ba3 | 405 | { |
097b8a7c JS |
406 | u64 seq; |
407 | ||
408 | tree_mod_log_write_lock(fs_info); | |
bd989ba3 | 409 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 410 | if (!elem->seq) { |
fc36ed7e | 411 | elem->seq = btrfs_inc_tree_mod_seq_major(fs_info); |
097b8a7c JS |
412 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
413 | } | |
fc36ed7e | 414 | seq = btrfs_inc_tree_mod_seq_minor(fs_info); |
bd989ba3 | 415 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
416 | tree_mod_log_write_unlock(fs_info); |
417 | ||
418 | return seq; | |
bd989ba3 JS |
419 | } |
420 | ||
421 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
422 | struct seq_list *elem) | |
423 | { | |
424 | struct rb_root *tm_root; | |
425 | struct rb_node *node; | |
426 | struct rb_node *next; | |
427 | struct seq_list *cur_elem; | |
428 | struct tree_mod_elem *tm; | |
429 | u64 min_seq = (u64)-1; | |
430 | u64 seq_putting = elem->seq; | |
431 | ||
432 | if (!seq_putting) | |
433 | return; | |
434 | ||
bd989ba3 JS |
435 | spin_lock(&fs_info->tree_mod_seq_lock); |
436 | list_del(&elem->list); | |
097b8a7c | 437 | elem->seq = 0; |
bd989ba3 JS |
438 | |
439 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 440 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
441 | if (seq_putting > cur_elem->seq) { |
442 | /* | |
443 | * blocker with lower sequence number exists, we | |
444 | * cannot remove anything from the log | |
445 | */ | |
097b8a7c JS |
446 | spin_unlock(&fs_info->tree_mod_seq_lock); |
447 | return; | |
bd989ba3 JS |
448 | } |
449 | min_seq = cur_elem->seq; | |
450 | } | |
451 | } | |
097b8a7c JS |
452 | spin_unlock(&fs_info->tree_mod_seq_lock); |
453 | ||
bd989ba3 JS |
454 | /* |
455 | * anything that's lower than the lowest existing (read: blocked) | |
456 | * sequence number can be removed from the tree. | |
457 | */ | |
097b8a7c | 458 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
459 | tm_root = &fs_info->tree_mod_log; |
460 | for (node = rb_first(tm_root); node; node = next) { | |
461 | next = rb_next(node); | |
462 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 463 | if (tm->seq > min_seq) |
bd989ba3 JS |
464 | continue; |
465 | rb_erase(node, tm_root); | |
bd989ba3 JS |
466 | kfree(tm); |
467 | } | |
097b8a7c | 468 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
469 | } |
470 | ||
471 | /* | |
472 | * key order of the log: | |
473 | * index -> sequence | |
474 | * | |
475 | * the index is the shifted logical of the *new* root node for root replace | |
476 | * operations, or the shifted logical of the affected block for all other | |
477 | * operations. | |
478 | */ | |
479 | static noinline int | |
480 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
481 | { | |
482 | struct rb_root *tm_root; | |
483 | struct rb_node **new; | |
484 | struct rb_node *parent = NULL; | |
485 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
486 | int ret = 0; |
487 | ||
488 | BUG_ON(!tm); | |
489 | ||
490 | tree_mod_log_write_lock(fs_info); | |
491 | if (list_empty(&fs_info->tree_mod_seq_list)) { | |
492 | tree_mod_log_write_unlock(fs_info); | |
493 | /* | |
494 | * Ok we no longer care about logging modifications, free up tm | |
495 | * and return 0. Any callers shouldn't be using tm after | |
496 | * calling tree_mod_log_insert, but if they do we can just | |
497 | * change this to return a special error code to let the callers | |
498 | * do their own thing. | |
499 | */ | |
500 | kfree(tm); | |
501 | return 0; | |
502 | } | |
bd989ba3 | 503 | |
c8cc6341 JB |
504 | spin_lock(&fs_info->tree_mod_seq_lock); |
505 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
506 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
bd989ba3 | 507 | |
bd989ba3 JS |
508 | tm_root = &fs_info->tree_mod_log; |
509 | new = &tm_root->rb_node; | |
510 | while (*new) { | |
511 | cur = container_of(*new, struct tree_mod_elem, node); | |
512 | parent = *new; | |
513 | if (cur->index < tm->index) | |
514 | new = &((*new)->rb_left); | |
515 | else if (cur->index > tm->index) | |
516 | new = &((*new)->rb_right); | |
097b8a7c | 517 | else if (cur->seq < tm->seq) |
bd989ba3 | 518 | new = &((*new)->rb_left); |
097b8a7c | 519 | else if (cur->seq > tm->seq) |
bd989ba3 JS |
520 | new = &((*new)->rb_right); |
521 | else { | |
c8cc6341 | 522 | ret = -EEXIST; |
bd989ba3 | 523 | kfree(tm); |
c8cc6341 | 524 | goto out; |
bd989ba3 JS |
525 | } |
526 | } | |
527 | ||
528 | rb_link_node(&tm->node, parent, new); | |
529 | rb_insert_color(&tm->node, tm_root); | |
c8cc6341 JB |
530 | out: |
531 | tree_mod_log_write_unlock(fs_info); | |
532 | return ret; | |
bd989ba3 JS |
533 | } |
534 | ||
097b8a7c JS |
535 | /* |
536 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
537 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
538 | * this until all tree mod log insertions are recorded in the rb tree and then | |
539 | * call tree_mod_log_write_unlock() to release. | |
540 | */ | |
e9b7fd4d JS |
541 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
542 | struct extent_buffer *eb) { | |
543 | smp_mb(); | |
544 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
545 | return 1; | |
097b8a7c JS |
546 | if (eb && btrfs_header_level(eb) == 0) |
547 | return 1; | |
e9b7fd4d JS |
548 | return 0; |
549 | } | |
550 | ||
097b8a7c JS |
551 | static inline int |
552 | __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, | |
553 | struct extent_buffer *eb, int slot, | |
554 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 555 | { |
097b8a7c | 556 | struct tree_mod_elem *tm; |
bd989ba3 | 557 | |
c8cc6341 JB |
558 | tm = kzalloc(sizeof(*tm), flags); |
559 | if (!tm) | |
560 | return -ENOMEM; | |
bd989ba3 JS |
561 | |
562 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
563 | if (op != MOD_LOG_KEY_ADD) { | |
564 | btrfs_node_key(eb, &tm->key, slot); | |
565 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
566 | } | |
567 | tm->op = op; | |
568 | tm->slot = slot; | |
569 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
570 | ||
097b8a7c JS |
571 | return __tree_mod_log_insert(fs_info, tm); |
572 | } | |
573 | ||
574 | static noinline int | |
c8cc6341 JB |
575 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
576 | struct extent_buffer *eb, int slot, | |
577 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 578 | { |
097b8a7c JS |
579 | if (tree_mod_dont_log(fs_info, eb)) |
580 | return 0; | |
581 | ||
c8cc6341 | 582 | return __tree_mod_log_insert_key(fs_info, eb, slot, op, flags); |
097b8a7c JS |
583 | } |
584 | ||
bd989ba3 JS |
585 | static noinline int |
586 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
587 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
588 | int nr_items, gfp_t flags) | |
589 | { | |
590 | struct tree_mod_elem *tm; | |
591 | int ret; | |
592 | int i; | |
593 | ||
f395694c JS |
594 | if (tree_mod_dont_log(fs_info, eb)) |
595 | return 0; | |
bd989ba3 | 596 | |
01763a2e JS |
597 | /* |
598 | * When we override something during the move, we log these removals. | |
599 | * This can only happen when we move towards the beginning of the | |
600 | * buffer, i.e. dst_slot < src_slot. | |
601 | */ | |
bd989ba3 | 602 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
c8cc6341 JB |
603 | ret = __tree_mod_log_insert_key(fs_info, eb, i + dst_slot, |
604 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); | |
bd989ba3 JS |
605 | BUG_ON(ret < 0); |
606 | } | |
607 | ||
c8cc6341 JB |
608 | tm = kzalloc(sizeof(*tm), flags); |
609 | if (!tm) | |
610 | return -ENOMEM; | |
f395694c | 611 | |
bd989ba3 JS |
612 | tm->index = eb->start >> PAGE_CACHE_SHIFT; |
613 | tm->slot = src_slot; | |
614 | tm->move.dst_slot = dst_slot; | |
615 | tm->move.nr_items = nr_items; | |
616 | tm->op = MOD_LOG_MOVE_KEYS; | |
617 | ||
c8cc6341 | 618 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
619 | } |
620 | ||
097b8a7c JS |
621 | static inline void |
622 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
623 | { | |
624 | int i; | |
625 | u32 nritems; | |
626 | int ret; | |
627 | ||
b12a3b1e CM |
628 | if (btrfs_header_level(eb) == 0) |
629 | return; | |
630 | ||
097b8a7c JS |
631 | nritems = btrfs_header_nritems(eb); |
632 | for (i = nritems - 1; i >= 0; i--) { | |
c8cc6341 JB |
633 | ret = __tree_mod_log_insert_key(fs_info, eb, i, |
634 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
097b8a7c JS |
635 | BUG_ON(ret < 0); |
636 | } | |
637 | } | |
638 | ||
bd989ba3 JS |
639 | static noinline int |
640 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
641 | struct extent_buffer *old_root, | |
90f8d62e JS |
642 | struct extent_buffer *new_root, gfp_t flags, |
643 | int log_removal) | |
bd989ba3 JS |
644 | { |
645 | struct tree_mod_elem *tm; | |
bd989ba3 | 646 | |
097b8a7c JS |
647 | if (tree_mod_dont_log(fs_info, NULL)) |
648 | return 0; | |
649 | ||
90f8d62e JS |
650 | if (log_removal) |
651 | __tree_mod_log_free_eb(fs_info, old_root); | |
d9abbf1c | 652 | |
c8cc6341 JB |
653 | tm = kzalloc(sizeof(*tm), flags); |
654 | if (!tm) | |
655 | return -ENOMEM; | |
bd989ba3 JS |
656 | |
657 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
658 | tm->old_root.logical = old_root->start; | |
659 | tm->old_root.level = btrfs_header_level(old_root); | |
660 | tm->generation = btrfs_header_generation(old_root); | |
661 | tm->op = MOD_LOG_ROOT_REPLACE; | |
662 | ||
c8cc6341 | 663 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
664 | } |
665 | ||
666 | static struct tree_mod_elem * | |
667 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
668 | int smallest) | |
669 | { | |
670 | struct rb_root *tm_root; | |
671 | struct rb_node *node; | |
672 | struct tree_mod_elem *cur = NULL; | |
673 | struct tree_mod_elem *found = NULL; | |
674 | u64 index = start >> PAGE_CACHE_SHIFT; | |
675 | ||
097b8a7c | 676 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
677 | tm_root = &fs_info->tree_mod_log; |
678 | node = tm_root->rb_node; | |
679 | while (node) { | |
680 | cur = container_of(node, struct tree_mod_elem, node); | |
681 | if (cur->index < index) { | |
682 | node = node->rb_left; | |
683 | } else if (cur->index > index) { | |
684 | node = node->rb_right; | |
097b8a7c | 685 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
686 | node = node->rb_left; |
687 | } else if (!smallest) { | |
688 | /* we want the node with the highest seq */ | |
689 | if (found) | |
097b8a7c | 690 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
691 | found = cur; |
692 | node = node->rb_left; | |
097b8a7c | 693 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
694 | /* we want the node with the smallest seq */ |
695 | if (found) | |
097b8a7c | 696 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
697 | found = cur; |
698 | node = node->rb_right; | |
699 | } else { | |
700 | found = cur; | |
701 | break; | |
702 | } | |
703 | } | |
097b8a7c | 704 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
705 | |
706 | return found; | |
707 | } | |
708 | ||
709 | /* | |
710 | * this returns the element from the log with the smallest time sequence | |
711 | * value that's in the log (the oldest log item). any element with a time | |
712 | * sequence lower than min_seq will be ignored. | |
713 | */ | |
714 | static struct tree_mod_elem * | |
715 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
716 | u64 min_seq) | |
717 | { | |
718 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
719 | } | |
720 | ||
721 | /* | |
722 | * this returns the element from the log with the largest time sequence | |
723 | * value that's in the log (the most recent log item). any element with | |
724 | * a time sequence lower than min_seq will be ignored. | |
725 | */ | |
726 | static struct tree_mod_elem * | |
727 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
728 | { | |
729 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
730 | } | |
731 | ||
097b8a7c | 732 | static noinline void |
bd989ba3 JS |
733 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
734 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 735 | unsigned long src_offset, int nr_items) |
bd989ba3 JS |
736 | { |
737 | int ret; | |
738 | int i; | |
739 | ||
e9b7fd4d | 740 | if (tree_mod_dont_log(fs_info, NULL)) |
bd989ba3 JS |
741 | return; |
742 | ||
c8cc6341 | 743 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
bd989ba3 JS |
744 | return; |
745 | ||
bd989ba3 | 746 | for (i = 0; i < nr_items; i++) { |
c8cc6341 | 747 | ret = __tree_mod_log_insert_key(fs_info, src, |
90f8d62e | 748 | i + src_offset, |
c8cc6341 | 749 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
90f8d62e | 750 | BUG_ON(ret < 0); |
c8cc6341 | 751 | ret = __tree_mod_log_insert_key(fs_info, dst, |
097b8a7c | 752 | i + dst_offset, |
c8cc6341 JB |
753 | MOD_LOG_KEY_ADD, |
754 | GFP_NOFS); | |
bd989ba3 JS |
755 | BUG_ON(ret < 0); |
756 | } | |
757 | } | |
758 | ||
759 | static inline void | |
760 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
761 | int dst_offset, int src_offset, int nr_items) | |
762 | { | |
763 | int ret; | |
764 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
765 | nr_items, GFP_NOFS); | |
766 | BUG_ON(ret < 0); | |
767 | } | |
768 | ||
097b8a7c | 769 | static noinline void |
bd989ba3 | 770 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 771 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
772 | { |
773 | int ret; | |
774 | ||
c8cc6341 JB |
775 | ret = __tree_mod_log_insert_key(fs_info, eb, slot, |
776 | MOD_LOG_KEY_REPLACE, | |
777 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
778 | BUG_ON(ret < 0); |
779 | } | |
780 | ||
097b8a7c JS |
781 | static noinline void |
782 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
bd989ba3 | 783 | { |
e9b7fd4d | 784 | if (tree_mod_dont_log(fs_info, eb)) |
bd989ba3 | 785 | return; |
097b8a7c | 786 | __tree_mod_log_free_eb(fs_info, eb); |
bd989ba3 JS |
787 | } |
788 | ||
097b8a7c | 789 | static noinline void |
bd989ba3 | 790 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
791 | struct extent_buffer *new_root_node, |
792 | int log_removal) | |
bd989ba3 JS |
793 | { |
794 | int ret; | |
bd989ba3 | 795 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 796 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
797 | BUG_ON(ret < 0); |
798 | } | |
799 | ||
5d4f98a2 YZ |
800 | /* |
801 | * check if the tree block can be shared by multiple trees | |
802 | */ | |
803 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
804 | struct extent_buffer *buf) | |
805 | { | |
806 | /* | |
807 | * Tree blocks not in refernece counted trees and tree roots | |
808 | * are never shared. If a block was allocated after the last | |
809 | * snapshot and the block was not allocated by tree relocation, | |
810 | * we know the block is not shared. | |
811 | */ | |
812 | if (root->ref_cows && | |
813 | buf != root->node && buf != root->commit_root && | |
814 | (btrfs_header_generation(buf) <= | |
815 | btrfs_root_last_snapshot(&root->root_item) || | |
816 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
817 | return 1; | |
818 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
819 | if (root->ref_cows && | |
820 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
821 | return 1; | |
822 | #endif | |
823 | return 0; | |
824 | } | |
825 | ||
826 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
827 | struct btrfs_root *root, | |
828 | struct extent_buffer *buf, | |
f0486c68 YZ |
829 | struct extent_buffer *cow, |
830 | int *last_ref) | |
5d4f98a2 YZ |
831 | { |
832 | u64 refs; | |
833 | u64 owner; | |
834 | u64 flags; | |
835 | u64 new_flags = 0; | |
836 | int ret; | |
837 | ||
838 | /* | |
839 | * Backrefs update rules: | |
840 | * | |
841 | * Always use full backrefs for extent pointers in tree block | |
842 | * allocated by tree relocation. | |
843 | * | |
844 | * If a shared tree block is no longer referenced by its owner | |
845 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
846 | * use full backrefs for extent pointers in tree block. | |
847 | * | |
848 | * If a tree block is been relocating | |
849 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
850 | * use full backrefs for extent pointers in tree block. | |
851 | * The reason for this is some operations (such as drop tree) | |
852 | * are only allowed for blocks use full backrefs. | |
853 | */ | |
854 | ||
855 | if (btrfs_block_can_be_shared(root, buf)) { | |
856 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
857 | btrfs_header_level(buf), 1, |
858 | &refs, &flags); | |
be1a5564 MF |
859 | if (ret) |
860 | return ret; | |
e5df9573 MF |
861 | if (refs == 0) { |
862 | ret = -EROFS; | |
863 | btrfs_std_error(root->fs_info, ret); | |
864 | return ret; | |
865 | } | |
5d4f98a2 YZ |
866 | } else { |
867 | refs = 1; | |
868 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
869 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
870 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
871 | else | |
872 | flags = 0; | |
873 | } | |
874 | ||
875 | owner = btrfs_header_owner(buf); | |
876 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
877 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
878 | ||
879 | if (refs > 1) { | |
880 | if ((owner == root->root_key.objectid || | |
881 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
882 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 883 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 884 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
885 | |
886 | if (root->root_key.objectid == | |
887 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 888 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 889 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 890 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 891 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
892 | } |
893 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
894 | } else { | |
895 | ||
896 | if (root->root_key.objectid == | |
897 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 898 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 899 | else |
66d7e7f0 | 900 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 901 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
902 | } |
903 | if (new_flags != 0) { | |
b1c79e09 JB |
904 | int level = btrfs_header_level(buf); |
905 | ||
5d4f98a2 YZ |
906 | ret = btrfs_set_disk_extent_flags(trans, root, |
907 | buf->start, | |
908 | buf->len, | |
b1c79e09 | 909 | new_flags, level, 0); |
be1a5564 MF |
910 | if (ret) |
911 | return ret; | |
5d4f98a2 YZ |
912 | } |
913 | } else { | |
914 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
915 | if (root->root_key.objectid == | |
916 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 917 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 918 | else |
66d7e7f0 | 919 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 920 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 921 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 922 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
923 | } |
924 | clean_tree_block(trans, root, buf); | |
f0486c68 | 925 | *last_ref = 1; |
5d4f98a2 YZ |
926 | } |
927 | return 0; | |
928 | } | |
929 | ||
d352ac68 | 930 | /* |
d397712b CM |
931 | * does the dirty work in cow of a single block. The parent block (if |
932 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
933 | * dirty and returned locked. If you modify the block it needs to be marked | |
934 | * dirty again. | |
d352ac68 CM |
935 | * |
936 | * search_start -- an allocation hint for the new block | |
937 | * | |
d397712b CM |
938 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
939 | * bytes the allocator should try to find free next to the block it returns. | |
940 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 941 | */ |
d397712b | 942 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
943 | struct btrfs_root *root, |
944 | struct extent_buffer *buf, | |
945 | struct extent_buffer *parent, int parent_slot, | |
946 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 947 | u64 search_start, u64 empty_size) |
02217ed2 | 948 | { |
5d4f98a2 | 949 | struct btrfs_disk_key disk_key; |
5f39d397 | 950 | struct extent_buffer *cow; |
be1a5564 | 951 | int level, ret; |
f0486c68 | 952 | int last_ref = 0; |
925baedd | 953 | int unlock_orig = 0; |
5d4f98a2 | 954 | u64 parent_start; |
7bb86316 | 955 | |
925baedd CM |
956 | if (*cow_ret == buf) |
957 | unlock_orig = 1; | |
958 | ||
b9447ef8 | 959 | btrfs_assert_tree_locked(buf); |
925baedd | 960 | |
7bb86316 CM |
961 | WARN_ON(root->ref_cows && trans->transid != |
962 | root->fs_info->running_transaction->transid); | |
6702ed49 | 963 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 964 | |
7bb86316 | 965 | level = btrfs_header_level(buf); |
31840ae1 | 966 | |
5d4f98a2 YZ |
967 | if (level == 0) |
968 | btrfs_item_key(buf, &disk_key, 0); | |
969 | else | |
970 | btrfs_node_key(buf, &disk_key, 0); | |
971 | ||
972 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
973 | if (parent) | |
974 | parent_start = parent->start; | |
975 | else | |
976 | parent_start = 0; | |
977 | } else | |
978 | parent_start = 0; | |
979 | ||
980 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
981 | root->root_key.objectid, &disk_key, | |
5581a51a | 982 | level, search_start, empty_size); |
54aa1f4d CM |
983 | if (IS_ERR(cow)) |
984 | return PTR_ERR(cow); | |
6702ed49 | 985 | |
b4ce94de CM |
986 | /* cow is set to blocking by btrfs_init_new_buffer */ |
987 | ||
5f39d397 | 988 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 989 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 990 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
991 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
992 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
993 | BTRFS_HEADER_FLAG_RELOC); | |
994 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
995 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
996 | else | |
997 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 998 | |
fba6aa75 | 999 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(cow), |
2b82032c YZ |
1000 | BTRFS_FSID_SIZE); |
1001 | ||
be1a5564 | 1002 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1003 | if (ret) { |
79787eaa | 1004 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1005 | return ret; |
1006 | } | |
1a40e23b | 1007 | |
3fd0a558 YZ |
1008 | if (root->ref_cows) |
1009 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
1010 | ||
02217ed2 | 1011 | if (buf == root->node) { |
925baedd | 1012 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1013 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1014 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1015 | parent_start = buf->start; | |
1016 | else | |
1017 | parent_start = 0; | |
925baedd | 1018 | |
5f39d397 | 1019 | extent_buffer_get(cow); |
90f8d62e | 1020 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1021 | rcu_assign_pointer(root->node, cow); |
925baedd | 1022 | |
f0486c68 | 1023 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1024 | last_ref); |
5f39d397 | 1025 | free_extent_buffer(buf); |
0b86a832 | 1026 | add_root_to_dirty_list(root); |
02217ed2 | 1027 | } else { |
5d4f98a2 YZ |
1028 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1029 | parent_start = parent->start; | |
1030 | else | |
1031 | parent_start = 0; | |
1032 | ||
1033 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1034 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1035 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1036 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1037 | cow->start); |
74493f7a CM |
1038 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1039 | trans->transid); | |
d6025579 | 1040 | btrfs_mark_buffer_dirty(parent); |
7fb7d76f JB |
1041 | if (last_ref) |
1042 | tree_mod_log_free_eb(root->fs_info, buf); | |
f0486c68 | 1043 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1044 | last_ref); |
02217ed2 | 1045 | } |
925baedd CM |
1046 | if (unlock_orig) |
1047 | btrfs_tree_unlock(buf); | |
3083ee2e | 1048 | free_extent_buffer_stale(buf); |
ccd467d6 | 1049 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1050 | *cow_ret = cow; |
02217ed2 CM |
1051 | return 0; |
1052 | } | |
1053 | ||
5d9e75c4 JS |
1054 | /* |
1055 | * returns the logical address of the oldest predecessor of the given root. | |
1056 | * entries older than time_seq are ignored. | |
1057 | */ | |
1058 | static struct tree_mod_elem * | |
1059 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1060 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1061 | { |
1062 | struct tree_mod_elem *tm; | |
1063 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1064 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1065 | int looped = 0; |
1066 | ||
1067 | if (!time_seq) | |
35a3621b | 1068 | return NULL; |
5d9e75c4 JS |
1069 | |
1070 | /* | |
1071 | * the very last operation that's logged for a root is the replacement | |
1072 | * operation (if it is replaced at all). this has the index of the *new* | |
1073 | * root, making it the very first operation that's logged for this root. | |
1074 | */ | |
1075 | while (1) { | |
1076 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1077 | time_seq); | |
1078 | if (!looped && !tm) | |
35a3621b | 1079 | return NULL; |
5d9e75c4 | 1080 | /* |
28da9fb4 JS |
1081 | * if there are no tree operation for the oldest root, we simply |
1082 | * return it. this should only happen if that (old) root is at | |
1083 | * level 0. | |
5d9e75c4 | 1084 | */ |
28da9fb4 JS |
1085 | if (!tm) |
1086 | break; | |
5d9e75c4 | 1087 | |
28da9fb4 JS |
1088 | /* |
1089 | * if there's an operation that's not a root replacement, we | |
1090 | * found the oldest version of our root. normally, we'll find a | |
1091 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1092 | */ | |
5d9e75c4 JS |
1093 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1094 | break; | |
1095 | ||
1096 | found = tm; | |
1097 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1098 | looped = 1; |
1099 | } | |
1100 | ||
a95236d9 JS |
1101 | /* if there's no old root to return, return what we found instead */ |
1102 | if (!found) | |
1103 | found = tm; | |
1104 | ||
5d9e75c4 JS |
1105 | return found; |
1106 | } | |
1107 | ||
1108 | /* | |
1109 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1110 | * previous operations will be rewinded (until we reach something older than | |
1111 | * time_seq). | |
1112 | */ | |
1113 | static void | |
f1ca7e98 JB |
1114 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1115 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1116 | { |
1117 | u32 n; | |
1118 | struct rb_node *next; | |
1119 | struct tree_mod_elem *tm = first_tm; | |
1120 | unsigned long o_dst; | |
1121 | unsigned long o_src; | |
1122 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1123 | ||
1124 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1125 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1126 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1127 | /* |
1128 | * all the operations are recorded with the operator used for | |
1129 | * the modification. as we're going backwards, we do the | |
1130 | * opposite of each operation here. | |
1131 | */ | |
1132 | switch (tm->op) { | |
1133 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1134 | BUG_ON(tm->slot < n); | |
1c697d4a | 1135 | /* Fallthrough */ |
95c80bb1 | 1136 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1137 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1138 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1139 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1140 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1141 | tm->generation); | |
4c3e6969 | 1142 | n++; |
5d9e75c4 JS |
1143 | break; |
1144 | case MOD_LOG_KEY_REPLACE: | |
1145 | BUG_ON(tm->slot >= n); | |
1146 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1147 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1148 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1149 | tm->generation); | |
1150 | break; | |
1151 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1152 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1153 | n--; |
1154 | break; | |
1155 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1156 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1157 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1158 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1159 | tm->move.nr_items * p_size); |
1160 | break; | |
1161 | case MOD_LOG_ROOT_REPLACE: | |
1162 | /* | |
1163 | * this operation is special. for roots, this must be | |
1164 | * handled explicitly before rewinding. | |
1165 | * for non-roots, this operation may exist if the node | |
1166 | * was a root: root A -> child B; then A gets empty and | |
1167 | * B is promoted to the new root. in the mod log, we'll | |
1168 | * have a root-replace operation for B, a tree block | |
1169 | * that is no root. we simply ignore that operation. | |
1170 | */ | |
1171 | break; | |
1172 | } | |
1173 | next = rb_next(&tm->node); | |
1174 | if (!next) | |
1175 | break; | |
1176 | tm = container_of(next, struct tree_mod_elem, node); | |
1177 | if (tm->index != first_tm->index) | |
1178 | break; | |
1179 | } | |
f1ca7e98 | 1180 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1181 | btrfs_set_header_nritems(eb, n); |
1182 | } | |
1183 | ||
47fb091f JS |
1184 | /* |
1185 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1186 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1187 | * returned buffer is always read-locked. If the returned buffer is not the | |
1188 | * input buffer, the lock on the input buffer is released and the input buffer | |
1189 | * is freed (its refcount is decremented). | |
1190 | */ | |
5d9e75c4 | 1191 | static struct extent_buffer * |
9ec72677 JB |
1192 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1193 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1194 | { |
1195 | struct extent_buffer *eb_rewin; | |
1196 | struct tree_mod_elem *tm; | |
1197 | ||
1198 | if (!time_seq) | |
1199 | return eb; | |
1200 | ||
1201 | if (btrfs_header_level(eb) == 0) | |
1202 | return eb; | |
1203 | ||
1204 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1205 | if (!tm) | |
1206 | return eb; | |
1207 | ||
9ec72677 JB |
1208 | btrfs_set_path_blocking(path); |
1209 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1210 | ||
5d9e75c4 JS |
1211 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1212 | BUG_ON(tm->slot != 0); | |
1213 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1214 | fs_info->tree_root->nodesize); | |
db7f3436 | 1215 | if (!eb_rewin) { |
9ec72677 | 1216 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1217 | free_extent_buffer(eb); |
1218 | return NULL; | |
1219 | } | |
5d9e75c4 JS |
1220 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1221 | btrfs_set_header_backref_rev(eb_rewin, | |
1222 | btrfs_header_backref_rev(eb)); | |
1223 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1224 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1225 | } else { |
1226 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1227 | if (!eb_rewin) { |
9ec72677 | 1228 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1229 | free_extent_buffer(eb); |
1230 | return NULL; | |
1231 | } | |
5d9e75c4 JS |
1232 | } |
1233 | ||
9ec72677 JB |
1234 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1235 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1236 | free_extent_buffer(eb); |
1237 | ||
47fb091f JS |
1238 | extent_buffer_get(eb_rewin); |
1239 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1240 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1241 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1242 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1243 | |
1244 | return eb_rewin; | |
1245 | } | |
1246 | ||
8ba97a15 JS |
1247 | /* |
1248 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1249 | * value. If there are no changes, the current root->root_node is returned. If | |
1250 | * anything changed in between, there's a fresh buffer allocated on which the | |
1251 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1252 | * Returns NULL on error (with no locks held). | |
1253 | */ | |
5d9e75c4 JS |
1254 | static inline struct extent_buffer * |
1255 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1256 | { | |
1257 | struct tree_mod_elem *tm; | |
30b0463a JS |
1258 | struct extent_buffer *eb = NULL; |
1259 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1260 | struct extent_buffer *old; |
a95236d9 | 1261 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1262 | u64 old_generation = 0; |
a95236d9 | 1263 | u64 logical; |
834328a8 | 1264 | u32 blocksize; |
5d9e75c4 | 1265 | |
30b0463a JS |
1266 | eb_root = btrfs_read_lock_root_node(root); |
1267 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1268 | if (!tm) |
30b0463a | 1269 | return eb_root; |
5d9e75c4 | 1270 | |
a95236d9 JS |
1271 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1272 | old_root = &tm->old_root; | |
1273 | old_generation = tm->generation; | |
1274 | logical = old_root->logical; | |
1275 | } else { | |
30b0463a | 1276 | logical = eb_root->start; |
a95236d9 | 1277 | } |
5d9e75c4 | 1278 | |
a95236d9 | 1279 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1280 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1281 | btrfs_tree_read_unlock(eb_root); |
1282 | free_extent_buffer(eb_root); | |
834328a8 | 1283 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1284 | old = read_tree_block(root, logical, blocksize, 0); |
416bc658 JB |
1285 | if (!old || !extent_buffer_uptodate(old)) { |
1286 | free_extent_buffer(old); | |
834328a8 JS |
1287 | pr_warn("btrfs: failed to read tree block %llu from get_old_root\n", |
1288 | logical); | |
1289 | WARN_ON(1); | |
1290 | } else { | |
7bfdcf7f LB |
1291 | eb = btrfs_clone_extent_buffer(old); |
1292 | free_extent_buffer(old); | |
834328a8 JS |
1293 | } |
1294 | } else if (old_root) { | |
30b0463a JS |
1295 | btrfs_tree_read_unlock(eb_root); |
1296 | free_extent_buffer(eb_root); | |
28da9fb4 | 1297 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1298 | } else { |
9ec72677 | 1299 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1300 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1301 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1302 | free_extent_buffer(eb_root); |
834328a8 JS |
1303 | } |
1304 | ||
8ba97a15 JS |
1305 | if (!eb) |
1306 | return NULL; | |
d6381084 | 1307 | extent_buffer_get(eb); |
8ba97a15 | 1308 | btrfs_tree_read_lock(eb); |
a95236d9 | 1309 | if (old_root) { |
5d9e75c4 JS |
1310 | btrfs_set_header_bytenr(eb, eb->start); |
1311 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1312 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1313 | btrfs_set_header_level(eb, old_root->level); |
1314 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1315 | } |
28da9fb4 | 1316 | if (tm) |
f1ca7e98 | 1317 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1318 | else |
1319 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1320 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1321 | |
1322 | return eb; | |
1323 | } | |
1324 | ||
5b6602e7 JS |
1325 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1326 | { | |
1327 | struct tree_mod_elem *tm; | |
1328 | int level; | |
30b0463a | 1329 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1330 | |
30b0463a | 1331 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1332 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1333 | level = tm->old_root.level; | |
1334 | } else { | |
30b0463a | 1335 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1336 | } |
30b0463a | 1337 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1338 | |
1339 | return level; | |
1340 | } | |
1341 | ||
5d4f98a2 YZ |
1342 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1343 | struct btrfs_root *root, | |
1344 | struct extent_buffer *buf) | |
1345 | { | |
f1ebcc74 LB |
1346 | /* ensure we can see the force_cow */ |
1347 | smp_rmb(); | |
1348 | ||
1349 | /* | |
1350 | * We do not need to cow a block if | |
1351 | * 1) this block is not created or changed in this transaction; | |
1352 | * 2) this block does not belong to TREE_RELOC tree; | |
1353 | * 3) the root is not forced COW. | |
1354 | * | |
1355 | * What is forced COW: | |
1356 | * when we create snapshot during commiting the transaction, | |
1357 | * after we've finished coping src root, we must COW the shared | |
1358 | * block to ensure the metadata consistency. | |
1359 | */ | |
5d4f98a2 YZ |
1360 | if (btrfs_header_generation(buf) == trans->transid && |
1361 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1362 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1363 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1364 | !root->force_cow) | |
5d4f98a2 YZ |
1365 | return 0; |
1366 | return 1; | |
1367 | } | |
1368 | ||
d352ac68 CM |
1369 | /* |
1370 | * cows a single block, see __btrfs_cow_block for the real work. | |
1371 | * This version of it has extra checks so that a block isn't cow'd more than | |
1372 | * once per transaction, as long as it hasn't been written yet | |
1373 | */ | |
d397712b | 1374 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1375 | struct btrfs_root *root, struct extent_buffer *buf, |
1376 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1377 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1378 | { |
1379 | u64 search_start; | |
f510cfec | 1380 | int ret; |
dc17ff8f | 1381 | |
31b1a2bd JL |
1382 | if (trans->transaction != root->fs_info->running_transaction) |
1383 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1384 | trans->transid, |
6702ed49 | 1385 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1386 | |
1387 | if (trans->transid != root->fs_info->generation) | |
1388 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1389 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1390 | |
5d4f98a2 | 1391 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1392 | *cow_ret = buf; |
1393 | return 0; | |
1394 | } | |
c487685d | 1395 | |
0b86a832 | 1396 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1397 | |
1398 | if (parent) | |
1399 | btrfs_set_lock_blocking(parent); | |
1400 | btrfs_set_lock_blocking(buf); | |
1401 | ||
f510cfec | 1402 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1403 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1404 | |
1405 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1406 | ||
f510cfec | 1407 | return ret; |
6702ed49 CM |
1408 | } |
1409 | ||
d352ac68 CM |
1410 | /* |
1411 | * helper function for defrag to decide if two blocks pointed to by a | |
1412 | * node are actually close by | |
1413 | */ | |
6b80053d | 1414 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1415 | { |
6b80053d | 1416 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1417 | return 1; |
6b80053d | 1418 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1419 | return 1; |
1420 | return 0; | |
1421 | } | |
1422 | ||
081e9573 CM |
1423 | /* |
1424 | * compare two keys in a memcmp fashion | |
1425 | */ | |
1426 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1427 | { | |
1428 | struct btrfs_key k1; | |
1429 | ||
1430 | btrfs_disk_key_to_cpu(&k1, disk); | |
1431 | ||
20736aba | 1432 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1433 | } |
1434 | ||
f3465ca4 JB |
1435 | /* |
1436 | * same as comp_keys only with two btrfs_key's | |
1437 | */ | |
5d4f98a2 | 1438 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1439 | { |
1440 | if (k1->objectid > k2->objectid) | |
1441 | return 1; | |
1442 | if (k1->objectid < k2->objectid) | |
1443 | return -1; | |
1444 | if (k1->type > k2->type) | |
1445 | return 1; | |
1446 | if (k1->type < k2->type) | |
1447 | return -1; | |
1448 | if (k1->offset > k2->offset) | |
1449 | return 1; | |
1450 | if (k1->offset < k2->offset) | |
1451 | return -1; | |
1452 | return 0; | |
1453 | } | |
081e9573 | 1454 | |
d352ac68 CM |
1455 | /* |
1456 | * this is used by the defrag code to go through all the | |
1457 | * leaves pointed to by a node and reallocate them so that | |
1458 | * disk order is close to key order | |
1459 | */ | |
6702ed49 | 1460 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1461 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1462 | int start_slot, u64 *last_ret, |
a6b6e75e | 1463 | struct btrfs_key *progress) |
6702ed49 | 1464 | { |
6b80053d | 1465 | struct extent_buffer *cur; |
6702ed49 | 1466 | u64 blocknr; |
ca7a79ad | 1467 | u64 gen; |
e9d0b13b CM |
1468 | u64 search_start = *last_ret; |
1469 | u64 last_block = 0; | |
6702ed49 CM |
1470 | u64 other; |
1471 | u32 parent_nritems; | |
6702ed49 CM |
1472 | int end_slot; |
1473 | int i; | |
1474 | int err = 0; | |
f2183bde | 1475 | int parent_level; |
6b80053d CM |
1476 | int uptodate; |
1477 | u32 blocksize; | |
081e9573 CM |
1478 | int progress_passed = 0; |
1479 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1480 | |
5708b959 | 1481 | parent_level = btrfs_header_level(parent); |
5708b959 | 1482 | |
6c1500f2 JL |
1483 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1484 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1485 | |
6b80053d | 1486 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1487 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1488 | end_slot = parent_nritems; |
1489 | ||
1490 | if (parent_nritems == 1) | |
1491 | return 0; | |
1492 | ||
b4ce94de CM |
1493 | btrfs_set_lock_blocking(parent); |
1494 | ||
6702ed49 CM |
1495 | for (i = start_slot; i < end_slot; i++) { |
1496 | int close = 1; | |
a6b6e75e | 1497 | |
081e9573 CM |
1498 | btrfs_node_key(parent, &disk_key, i); |
1499 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1500 | continue; | |
1501 | ||
1502 | progress_passed = 1; | |
6b80053d | 1503 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1504 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1505 | if (last_block == 0) |
1506 | last_block = blocknr; | |
5708b959 | 1507 | |
6702ed49 | 1508 | if (i > 0) { |
6b80053d CM |
1509 | other = btrfs_node_blockptr(parent, i - 1); |
1510 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1511 | } |
0ef3e66b | 1512 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1513 | other = btrfs_node_blockptr(parent, i + 1); |
1514 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1515 | } |
e9d0b13b CM |
1516 | if (close) { |
1517 | last_block = blocknr; | |
6702ed49 | 1518 | continue; |
e9d0b13b | 1519 | } |
6702ed49 | 1520 | |
6b80053d CM |
1521 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1522 | if (cur) | |
b9fab919 | 1523 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1524 | else |
1525 | uptodate = 0; | |
5708b959 | 1526 | if (!cur || !uptodate) { |
6b80053d CM |
1527 | if (!cur) { |
1528 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1529 | blocksize, gen); |
416bc658 JB |
1530 | if (!cur || !extent_buffer_uptodate(cur)) { |
1531 | free_extent_buffer(cur); | |
97d9a8a4 | 1532 | return -EIO; |
416bc658 | 1533 | } |
6b80053d | 1534 | } else if (!uptodate) { |
018642a1 TI |
1535 | err = btrfs_read_buffer(cur, gen); |
1536 | if (err) { | |
1537 | free_extent_buffer(cur); | |
1538 | return err; | |
1539 | } | |
f2183bde | 1540 | } |
6702ed49 | 1541 | } |
e9d0b13b | 1542 | if (search_start == 0) |
6b80053d | 1543 | search_start = last_block; |
e9d0b13b | 1544 | |
e7a84565 | 1545 | btrfs_tree_lock(cur); |
b4ce94de | 1546 | btrfs_set_lock_blocking(cur); |
6b80053d | 1547 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1548 | &cur, search_start, |
6b80053d | 1549 | min(16 * blocksize, |
9fa8cfe7 | 1550 | (end_slot - i) * blocksize)); |
252c38f0 | 1551 | if (err) { |
e7a84565 | 1552 | btrfs_tree_unlock(cur); |
6b80053d | 1553 | free_extent_buffer(cur); |
6702ed49 | 1554 | break; |
252c38f0 | 1555 | } |
e7a84565 CM |
1556 | search_start = cur->start; |
1557 | last_block = cur->start; | |
f2183bde | 1558 | *last_ret = search_start; |
e7a84565 CM |
1559 | btrfs_tree_unlock(cur); |
1560 | free_extent_buffer(cur); | |
6702ed49 CM |
1561 | } |
1562 | return err; | |
1563 | } | |
1564 | ||
74123bd7 CM |
1565 | /* |
1566 | * The leaf data grows from end-to-front in the node. | |
1567 | * this returns the address of the start of the last item, | |
1568 | * which is the stop of the leaf data stack | |
1569 | */ | |
123abc88 | 1570 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1571 | struct extent_buffer *leaf) |
be0e5c09 | 1572 | { |
5f39d397 | 1573 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1574 | if (nr == 0) |
123abc88 | 1575 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1576 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1577 | } |
1578 | ||
aa5d6bed | 1579 | |
74123bd7 | 1580 | /* |
5f39d397 CM |
1581 | * search for key in the extent_buffer. The items start at offset p, |
1582 | * and they are item_size apart. There are 'max' items in p. | |
1583 | * | |
74123bd7 CM |
1584 | * the slot in the array is returned via slot, and it points to |
1585 | * the place where you would insert key if it is not found in | |
1586 | * the array. | |
1587 | * | |
1588 | * slot may point to max if the key is bigger than all of the keys | |
1589 | */ | |
e02119d5 CM |
1590 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1591 | unsigned long p, | |
1592 | int item_size, struct btrfs_key *key, | |
1593 | int max, int *slot) | |
be0e5c09 CM |
1594 | { |
1595 | int low = 0; | |
1596 | int high = max; | |
1597 | int mid; | |
1598 | int ret; | |
479965d6 | 1599 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1600 | struct btrfs_disk_key unaligned; |
1601 | unsigned long offset; | |
5f39d397 CM |
1602 | char *kaddr = NULL; |
1603 | unsigned long map_start = 0; | |
1604 | unsigned long map_len = 0; | |
479965d6 | 1605 | int err; |
be0e5c09 | 1606 | |
d397712b | 1607 | while (low < high) { |
be0e5c09 | 1608 | mid = (low + high) / 2; |
5f39d397 CM |
1609 | offset = p + mid * item_size; |
1610 | ||
a6591715 | 1611 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1612 | (offset + sizeof(struct btrfs_disk_key)) > |
1613 | map_start + map_len) { | |
934d375b CM |
1614 | |
1615 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1616 | sizeof(struct btrfs_disk_key), |
a6591715 | 1617 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1618 | |
1619 | if (!err) { | |
1620 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1621 | map_start); | |
1622 | } else { | |
1623 | read_extent_buffer(eb, &unaligned, | |
1624 | offset, sizeof(unaligned)); | |
1625 | tmp = &unaligned; | |
1626 | } | |
5f39d397 | 1627 | |
5f39d397 CM |
1628 | } else { |
1629 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1630 | map_start); | |
1631 | } | |
be0e5c09 CM |
1632 | ret = comp_keys(tmp, key); |
1633 | ||
1634 | if (ret < 0) | |
1635 | low = mid + 1; | |
1636 | else if (ret > 0) | |
1637 | high = mid; | |
1638 | else { | |
1639 | *slot = mid; | |
1640 | return 0; | |
1641 | } | |
1642 | } | |
1643 | *slot = low; | |
1644 | return 1; | |
1645 | } | |
1646 | ||
97571fd0 CM |
1647 | /* |
1648 | * simple bin_search frontend that does the right thing for | |
1649 | * leaves vs nodes | |
1650 | */ | |
5f39d397 CM |
1651 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1652 | int level, int *slot) | |
be0e5c09 | 1653 | { |
f775738f | 1654 | if (level == 0) |
5f39d397 CM |
1655 | return generic_bin_search(eb, |
1656 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1657 | sizeof(struct btrfs_item), |
5f39d397 | 1658 | key, btrfs_header_nritems(eb), |
7518a238 | 1659 | slot); |
f775738f | 1660 | else |
5f39d397 CM |
1661 | return generic_bin_search(eb, |
1662 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1663 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1664 | key, btrfs_header_nritems(eb), |
7518a238 | 1665 | slot); |
be0e5c09 CM |
1666 | } |
1667 | ||
5d4f98a2 YZ |
1668 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1669 | int level, int *slot) | |
1670 | { | |
1671 | return bin_search(eb, key, level, slot); | |
1672 | } | |
1673 | ||
f0486c68 YZ |
1674 | static void root_add_used(struct btrfs_root *root, u32 size) |
1675 | { | |
1676 | spin_lock(&root->accounting_lock); | |
1677 | btrfs_set_root_used(&root->root_item, | |
1678 | btrfs_root_used(&root->root_item) + size); | |
1679 | spin_unlock(&root->accounting_lock); | |
1680 | } | |
1681 | ||
1682 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1683 | { | |
1684 | spin_lock(&root->accounting_lock); | |
1685 | btrfs_set_root_used(&root->root_item, | |
1686 | btrfs_root_used(&root->root_item) - size); | |
1687 | spin_unlock(&root->accounting_lock); | |
1688 | } | |
1689 | ||
d352ac68 CM |
1690 | /* given a node and slot number, this reads the blocks it points to. The |
1691 | * extent buffer is returned with a reference taken (but unlocked). | |
1692 | * NULL is returned on error. | |
1693 | */ | |
e02119d5 | 1694 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1695 | struct extent_buffer *parent, int slot) |
bb803951 | 1696 | { |
ca7a79ad | 1697 | int level = btrfs_header_level(parent); |
416bc658 JB |
1698 | struct extent_buffer *eb; |
1699 | ||
bb803951 CM |
1700 | if (slot < 0) |
1701 | return NULL; | |
5f39d397 | 1702 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1703 | return NULL; |
ca7a79ad CM |
1704 | |
1705 | BUG_ON(level == 0); | |
1706 | ||
416bc658 JB |
1707 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1708 | btrfs_level_size(root, level - 1), | |
1709 | btrfs_node_ptr_generation(parent, slot)); | |
1710 | if (eb && !extent_buffer_uptodate(eb)) { | |
1711 | free_extent_buffer(eb); | |
1712 | eb = NULL; | |
1713 | } | |
1714 | ||
1715 | return eb; | |
bb803951 CM |
1716 | } |
1717 | ||
d352ac68 CM |
1718 | /* |
1719 | * node level balancing, used to make sure nodes are in proper order for | |
1720 | * item deletion. We balance from the top down, so we have to make sure | |
1721 | * that a deletion won't leave an node completely empty later on. | |
1722 | */ | |
e02119d5 | 1723 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1724 | struct btrfs_root *root, |
1725 | struct btrfs_path *path, int level) | |
bb803951 | 1726 | { |
5f39d397 CM |
1727 | struct extent_buffer *right = NULL; |
1728 | struct extent_buffer *mid; | |
1729 | struct extent_buffer *left = NULL; | |
1730 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1731 | int ret = 0; |
1732 | int wret; | |
1733 | int pslot; | |
bb803951 | 1734 | int orig_slot = path->slots[level]; |
79f95c82 | 1735 | u64 orig_ptr; |
bb803951 CM |
1736 | |
1737 | if (level == 0) | |
1738 | return 0; | |
1739 | ||
5f39d397 | 1740 | mid = path->nodes[level]; |
b4ce94de | 1741 | |
bd681513 CM |
1742 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1743 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1744 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1745 | ||
1d4f8a0c | 1746 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1747 | |
a05a9bb1 | 1748 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1749 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1750 | pslot = path->slots[level + 1]; |
1751 | } | |
bb803951 | 1752 | |
40689478 CM |
1753 | /* |
1754 | * deal with the case where there is only one pointer in the root | |
1755 | * by promoting the node below to a root | |
1756 | */ | |
5f39d397 CM |
1757 | if (!parent) { |
1758 | struct extent_buffer *child; | |
bb803951 | 1759 | |
5f39d397 | 1760 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1761 | return 0; |
1762 | ||
1763 | /* promote the child to a root */ | |
5f39d397 | 1764 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1765 | if (!child) { |
1766 | ret = -EROFS; | |
1767 | btrfs_std_error(root->fs_info, ret); | |
1768 | goto enospc; | |
1769 | } | |
1770 | ||
925baedd | 1771 | btrfs_tree_lock(child); |
b4ce94de | 1772 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1773 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1774 | if (ret) { |
1775 | btrfs_tree_unlock(child); | |
1776 | free_extent_buffer(child); | |
1777 | goto enospc; | |
1778 | } | |
2f375ab9 | 1779 | |
90f8d62e | 1780 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1781 | rcu_assign_pointer(root->node, child); |
925baedd | 1782 | |
0b86a832 | 1783 | add_root_to_dirty_list(root); |
925baedd | 1784 | btrfs_tree_unlock(child); |
b4ce94de | 1785 | |
925baedd | 1786 | path->locks[level] = 0; |
bb803951 | 1787 | path->nodes[level] = NULL; |
5f39d397 | 1788 | clean_tree_block(trans, root, mid); |
925baedd | 1789 | btrfs_tree_unlock(mid); |
bb803951 | 1790 | /* once for the path */ |
5f39d397 | 1791 | free_extent_buffer(mid); |
f0486c68 YZ |
1792 | |
1793 | root_sub_used(root, mid->len); | |
5581a51a | 1794 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1795 | /* once for the root ptr */ |
3083ee2e | 1796 | free_extent_buffer_stale(mid); |
f0486c68 | 1797 | return 0; |
bb803951 | 1798 | } |
5f39d397 | 1799 | if (btrfs_header_nritems(mid) > |
123abc88 | 1800 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1801 | return 0; |
1802 | ||
5f39d397 CM |
1803 | left = read_node_slot(root, parent, pslot - 1); |
1804 | if (left) { | |
925baedd | 1805 | btrfs_tree_lock(left); |
b4ce94de | 1806 | btrfs_set_lock_blocking(left); |
5f39d397 | 1807 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1808 | parent, pslot - 1, &left); |
54aa1f4d CM |
1809 | if (wret) { |
1810 | ret = wret; | |
1811 | goto enospc; | |
1812 | } | |
2cc58cf2 | 1813 | } |
5f39d397 CM |
1814 | right = read_node_slot(root, parent, pslot + 1); |
1815 | if (right) { | |
925baedd | 1816 | btrfs_tree_lock(right); |
b4ce94de | 1817 | btrfs_set_lock_blocking(right); |
5f39d397 | 1818 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1819 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1820 | if (wret) { |
1821 | ret = wret; | |
1822 | goto enospc; | |
1823 | } | |
1824 | } | |
1825 | ||
1826 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1827 | if (left) { |
1828 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1829 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1830 | if (wret < 0) |
1831 | ret = wret; | |
bb803951 | 1832 | } |
79f95c82 CM |
1833 | |
1834 | /* | |
1835 | * then try to empty the right most buffer into the middle | |
1836 | */ | |
5f39d397 | 1837 | if (right) { |
971a1f66 | 1838 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1839 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1840 | ret = wret; |
5f39d397 | 1841 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1842 | clean_tree_block(trans, root, right); |
925baedd | 1843 | btrfs_tree_unlock(right); |
afe5fea7 | 1844 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1845 | root_sub_used(root, right->len); |
5581a51a | 1846 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1847 | free_extent_buffer_stale(right); |
f0486c68 | 1848 | right = NULL; |
bb803951 | 1849 | } else { |
5f39d397 CM |
1850 | struct btrfs_disk_key right_key; |
1851 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 1852 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1853 | pslot + 1, 0); |
5f39d397 CM |
1854 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1855 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1856 | } |
1857 | } | |
5f39d397 | 1858 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1859 | /* |
1860 | * we're not allowed to leave a node with one item in the | |
1861 | * tree during a delete. A deletion from lower in the tree | |
1862 | * could try to delete the only pointer in this node. | |
1863 | * So, pull some keys from the left. | |
1864 | * There has to be a left pointer at this point because | |
1865 | * otherwise we would have pulled some pointers from the | |
1866 | * right | |
1867 | */ | |
305a26af MF |
1868 | if (!left) { |
1869 | ret = -EROFS; | |
1870 | btrfs_std_error(root->fs_info, ret); | |
1871 | goto enospc; | |
1872 | } | |
5f39d397 | 1873 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1874 | if (wret < 0) { |
79f95c82 | 1875 | ret = wret; |
54aa1f4d CM |
1876 | goto enospc; |
1877 | } | |
bce4eae9 CM |
1878 | if (wret == 1) { |
1879 | wret = push_node_left(trans, root, left, mid, 1); | |
1880 | if (wret < 0) | |
1881 | ret = wret; | |
1882 | } | |
79f95c82 CM |
1883 | BUG_ON(wret == 1); |
1884 | } | |
5f39d397 | 1885 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1886 | clean_tree_block(trans, root, mid); |
925baedd | 1887 | btrfs_tree_unlock(mid); |
afe5fea7 | 1888 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1889 | root_sub_used(root, mid->len); |
5581a51a | 1890 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1891 | free_extent_buffer_stale(mid); |
f0486c68 | 1892 | mid = NULL; |
79f95c82 CM |
1893 | } else { |
1894 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1895 | struct btrfs_disk_key mid_key; |
1896 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 1897 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 1898 | pslot, 0); |
5f39d397 CM |
1899 | btrfs_set_node_key(parent, &mid_key, pslot); |
1900 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1901 | } |
bb803951 | 1902 | |
79f95c82 | 1903 | /* update the path */ |
5f39d397 CM |
1904 | if (left) { |
1905 | if (btrfs_header_nritems(left) > orig_slot) { | |
1906 | extent_buffer_get(left); | |
925baedd | 1907 | /* left was locked after cow */ |
5f39d397 | 1908 | path->nodes[level] = left; |
bb803951 CM |
1909 | path->slots[level + 1] -= 1; |
1910 | path->slots[level] = orig_slot; | |
925baedd CM |
1911 | if (mid) { |
1912 | btrfs_tree_unlock(mid); | |
5f39d397 | 1913 | free_extent_buffer(mid); |
925baedd | 1914 | } |
bb803951 | 1915 | } else { |
5f39d397 | 1916 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1917 | path->slots[level] = orig_slot; |
1918 | } | |
1919 | } | |
79f95c82 | 1920 | /* double check we haven't messed things up */ |
e20d96d6 | 1921 | if (orig_ptr != |
5f39d397 | 1922 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1923 | BUG(); |
54aa1f4d | 1924 | enospc: |
925baedd CM |
1925 | if (right) { |
1926 | btrfs_tree_unlock(right); | |
5f39d397 | 1927 | free_extent_buffer(right); |
925baedd CM |
1928 | } |
1929 | if (left) { | |
1930 | if (path->nodes[level] != left) | |
1931 | btrfs_tree_unlock(left); | |
5f39d397 | 1932 | free_extent_buffer(left); |
925baedd | 1933 | } |
bb803951 CM |
1934 | return ret; |
1935 | } | |
1936 | ||
d352ac68 CM |
1937 | /* Node balancing for insertion. Here we only split or push nodes around |
1938 | * when they are completely full. This is also done top down, so we | |
1939 | * have to be pessimistic. | |
1940 | */ | |
d397712b | 1941 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1942 | struct btrfs_root *root, |
1943 | struct btrfs_path *path, int level) | |
e66f709b | 1944 | { |
5f39d397 CM |
1945 | struct extent_buffer *right = NULL; |
1946 | struct extent_buffer *mid; | |
1947 | struct extent_buffer *left = NULL; | |
1948 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1949 | int ret = 0; |
1950 | int wret; | |
1951 | int pslot; | |
1952 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1953 | |
1954 | if (level == 0) | |
1955 | return 1; | |
1956 | ||
5f39d397 | 1957 | mid = path->nodes[level]; |
7bb86316 | 1958 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1959 | |
a05a9bb1 | 1960 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1961 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1962 | pslot = path->slots[level + 1]; |
1963 | } | |
e66f709b | 1964 | |
5f39d397 | 1965 | if (!parent) |
e66f709b | 1966 | return 1; |
e66f709b | 1967 | |
5f39d397 | 1968 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1969 | |
1970 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1971 | if (left) { |
e66f709b | 1972 | u32 left_nr; |
925baedd CM |
1973 | |
1974 | btrfs_tree_lock(left); | |
b4ce94de CM |
1975 | btrfs_set_lock_blocking(left); |
1976 | ||
5f39d397 | 1977 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1978 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1979 | wret = 1; | |
1980 | } else { | |
5f39d397 | 1981 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1982 | pslot - 1, &left); |
54aa1f4d CM |
1983 | if (ret) |
1984 | wret = 1; | |
1985 | else { | |
54aa1f4d | 1986 | wret = push_node_left(trans, root, |
971a1f66 | 1987 | left, mid, 0); |
54aa1f4d | 1988 | } |
33ade1f8 | 1989 | } |
e66f709b CM |
1990 | if (wret < 0) |
1991 | ret = wret; | |
1992 | if (wret == 0) { | |
5f39d397 | 1993 | struct btrfs_disk_key disk_key; |
e66f709b | 1994 | orig_slot += left_nr; |
5f39d397 | 1995 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 1996 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1997 | pslot, 0); |
5f39d397 CM |
1998 | btrfs_set_node_key(parent, &disk_key, pslot); |
1999 | btrfs_mark_buffer_dirty(parent); | |
2000 | if (btrfs_header_nritems(left) > orig_slot) { | |
2001 | path->nodes[level] = left; | |
e66f709b CM |
2002 | path->slots[level + 1] -= 1; |
2003 | path->slots[level] = orig_slot; | |
925baedd | 2004 | btrfs_tree_unlock(mid); |
5f39d397 | 2005 | free_extent_buffer(mid); |
e66f709b CM |
2006 | } else { |
2007 | orig_slot -= | |
5f39d397 | 2008 | btrfs_header_nritems(left); |
e66f709b | 2009 | path->slots[level] = orig_slot; |
925baedd | 2010 | btrfs_tree_unlock(left); |
5f39d397 | 2011 | free_extent_buffer(left); |
e66f709b | 2012 | } |
e66f709b CM |
2013 | return 0; |
2014 | } | |
925baedd | 2015 | btrfs_tree_unlock(left); |
5f39d397 | 2016 | free_extent_buffer(left); |
e66f709b | 2017 | } |
925baedd | 2018 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2019 | |
2020 | /* | |
2021 | * then try to empty the right most buffer into the middle | |
2022 | */ | |
5f39d397 | 2023 | if (right) { |
33ade1f8 | 2024 | u32 right_nr; |
b4ce94de | 2025 | |
925baedd | 2026 | btrfs_tree_lock(right); |
b4ce94de CM |
2027 | btrfs_set_lock_blocking(right); |
2028 | ||
5f39d397 | 2029 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2030 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2031 | wret = 1; | |
2032 | } else { | |
5f39d397 CM |
2033 | ret = btrfs_cow_block(trans, root, right, |
2034 | parent, pslot + 1, | |
9fa8cfe7 | 2035 | &right); |
54aa1f4d CM |
2036 | if (ret) |
2037 | wret = 1; | |
2038 | else { | |
54aa1f4d | 2039 | wret = balance_node_right(trans, root, |
5f39d397 | 2040 | right, mid); |
54aa1f4d | 2041 | } |
33ade1f8 | 2042 | } |
e66f709b CM |
2043 | if (wret < 0) |
2044 | ret = wret; | |
2045 | if (wret == 0) { | |
5f39d397 CM |
2046 | struct btrfs_disk_key disk_key; |
2047 | ||
2048 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2049 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2050 | pslot + 1, 0); |
5f39d397 CM |
2051 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2052 | btrfs_mark_buffer_dirty(parent); | |
2053 | ||
2054 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2055 | path->nodes[level] = right; | |
e66f709b CM |
2056 | path->slots[level + 1] += 1; |
2057 | path->slots[level] = orig_slot - | |
5f39d397 | 2058 | btrfs_header_nritems(mid); |
925baedd | 2059 | btrfs_tree_unlock(mid); |
5f39d397 | 2060 | free_extent_buffer(mid); |
e66f709b | 2061 | } else { |
925baedd | 2062 | btrfs_tree_unlock(right); |
5f39d397 | 2063 | free_extent_buffer(right); |
e66f709b | 2064 | } |
e66f709b CM |
2065 | return 0; |
2066 | } | |
925baedd | 2067 | btrfs_tree_unlock(right); |
5f39d397 | 2068 | free_extent_buffer(right); |
e66f709b | 2069 | } |
e66f709b CM |
2070 | return 1; |
2071 | } | |
2072 | ||
3c69faec | 2073 | /* |
d352ac68 CM |
2074 | * readahead one full node of leaves, finding things that are close |
2075 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2076 | */ |
c8c42864 CM |
2077 | static void reada_for_search(struct btrfs_root *root, |
2078 | struct btrfs_path *path, | |
2079 | int level, int slot, u64 objectid) | |
3c69faec | 2080 | { |
5f39d397 | 2081 | struct extent_buffer *node; |
01f46658 | 2082 | struct btrfs_disk_key disk_key; |
3c69faec | 2083 | u32 nritems; |
3c69faec | 2084 | u64 search; |
a7175319 | 2085 | u64 target; |
6b80053d | 2086 | u64 nread = 0; |
cb25c2ea | 2087 | u64 gen; |
3c69faec | 2088 | int direction = path->reada; |
5f39d397 | 2089 | struct extent_buffer *eb; |
6b80053d CM |
2090 | u32 nr; |
2091 | u32 blocksize; | |
2092 | u32 nscan = 0; | |
db94535d | 2093 | |
a6b6e75e | 2094 | if (level != 1) |
6702ed49 CM |
2095 | return; |
2096 | ||
2097 | if (!path->nodes[level]) | |
3c69faec CM |
2098 | return; |
2099 | ||
5f39d397 | 2100 | node = path->nodes[level]; |
925baedd | 2101 | |
3c69faec | 2102 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2103 | blocksize = btrfs_level_size(root, level - 1); |
2104 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2105 | if (eb) { |
2106 | free_extent_buffer(eb); | |
3c69faec CM |
2107 | return; |
2108 | } | |
2109 | ||
a7175319 | 2110 | target = search; |
6b80053d | 2111 | |
5f39d397 | 2112 | nritems = btrfs_header_nritems(node); |
6b80053d | 2113 | nr = slot; |
25b8b936 | 2114 | |
d397712b | 2115 | while (1) { |
6b80053d CM |
2116 | if (direction < 0) { |
2117 | if (nr == 0) | |
2118 | break; | |
2119 | nr--; | |
2120 | } else if (direction > 0) { | |
2121 | nr++; | |
2122 | if (nr >= nritems) | |
2123 | break; | |
3c69faec | 2124 | } |
01f46658 CM |
2125 | if (path->reada < 0 && objectid) { |
2126 | btrfs_node_key(node, &disk_key, nr); | |
2127 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2128 | break; | |
2129 | } | |
6b80053d | 2130 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2131 | if ((search <= target && target - search <= 65536) || |
2132 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2133 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2134 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2135 | nread += blocksize; |
2136 | } | |
2137 | nscan++; | |
a7175319 | 2138 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2139 | break; |
3c69faec CM |
2140 | } |
2141 | } | |
925baedd | 2142 | |
0b08851f JB |
2143 | static noinline void reada_for_balance(struct btrfs_root *root, |
2144 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2145 | { |
2146 | int slot; | |
2147 | int nritems; | |
2148 | struct extent_buffer *parent; | |
2149 | struct extent_buffer *eb; | |
2150 | u64 gen; | |
2151 | u64 block1 = 0; | |
2152 | u64 block2 = 0; | |
b4ce94de CM |
2153 | int blocksize; |
2154 | ||
8c594ea8 | 2155 | parent = path->nodes[level + 1]; |
b4ce94de | 2156 | if (!parent) |
0b08851f | 2157 | return; |
b4ce94de CM |
2158 | |
2159 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2160 | slot = path->slots[level + 1]; |
b4ce94de CM |
2161 | blocksize = btrfs_level_size(root, level); |
2162 | ||
2163 | if (slot > 0) { | |
2164 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2165 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2166 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2167 | /* |
2168 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2169 | * don't want to return eagain here. That will loop | |
2170 | * forever | |
2171 | */ | |
2172 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2173 | block1 = 0; |
2174 | free_extent_buffer(eb); | |
2175 | } | |
8c594ea8 | 2176 | if (slot + 1 < nritems) { |
b4ce94de CM |
2177 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2178 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2179 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2180 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2181 | block2 = 0; |
2182 | free_extent_buffer(eb); | |
2183 | } | |
8c594ea8 | 2184 | |
0b08851f JB |
2185 | if (block1) |
2186 | readahead_tree_block(root, block1, blocksize, 0); | |
2187 | if (block2) | |
2188 | readahead_tree_block(root, block2, blocksize, 0); | |
b4ce94de CM |
2189 | } |
2190 | ||
2191 | ||
d352ac68 | 2192 | /* |
d397712b CM |
2193 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2194 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2195 | * operations on the tree might require changing key pointers higher up in the | |
2196 | * tree. | |
d352ac68 | 2197 | * |
d397712b CM |
2198 | * callers might also have set path->keep_locks, which tells this code to keep |
2199 | * the lock if the path points to the last slot in the block. This is part of | |
2200 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2201 | * |
d397712b CM |
2202 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2203 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2204 | */ |
e02119d5 | 2205 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2206 | int lowest_unlock, int min_write_lock_level, |
2207 | int *write_lock_level) | |
925baedd CM |
2208 | { |
2209 | int i; | |
2210 | int skip_level = level; | |
051e1b9f | 2211 | int no_skips = 0; |
925baedd CM |
2212 | struct extent_buffer *t; |
2213 | ||
2214 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2215 | if (!path->nodes[i]) | |
2216 | break; | |
2217 | if (!path->locks[i]) | |
2218 | break; | |
051e1b9f | 2219 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2220 | skip_level = i + 1; |
2221 | continue; | |
2222 | } | |
051e1b9f | 2223 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2224 | u32 nritems; |
2225 | t = path->nodes[i]; | |
2226 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2227 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2228 | skip_level = i + 1; |
2229 | continue; | |
2230 | } | |
2231 | } | |
051e1b9f CM |
2232 | if (skip_level < i && i >= lowest_unlock) |
2233 | no_skips = 1; | |
2234 | ||
925baedd CM |
2235 | t = path->nodes[i]; |
2236 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2237 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2238 | path->locks[i] = 0; |
f7c79f30 CM |
2239 | if (write_lock_level && |
2240 | i > min_write_lock_level && | |
2241 | i <= *write_lock_level) { | |
2242 | *write_lock_level = i - 1; | |
2243 | } | |
925baedd CM |
2244 | } |
2245 | } | |
2246 | } | |
2247 | ||
b4ce94de CM |
2248 | /* |
2249 | * This releases any locks held in the path starting at level and | |
2250 | * going all the way up to the root. | |
2251 | * | |
2252 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2253 | * corner cases, such as COW of the block at slot zero in the node. This | |
2254 | * ignores those rules, and it should only be called when there are no | |
2255 | * more updates to be done higher up in the tree. | |
2256 | */ | |
2257 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2258 | { | |
2259 | int i; | |
2260 | ||
09a2a8f9 | 2261 | if (path->keep_locks) |
b4ce94de CM |
2262 | return; |
2263 | ||
2264 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2265 | if (!path->nodes[i]) | |
12f4dacc | 2266 | continue; |
b4ce94de | 2267 | if (!path->locks[i]) |
12f4dacc | 2268 | continue; |
bd681513 | 2269 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2270 | path->locks[i] = 0; |
2271 | } | |
2272 | } | |
2273 | ||
c8c42864 CM |
2274 | /* |
2275 | * helper function for btrfs_search_slot. The goal is to find a block | |
2276 | * in cache without setting the path to blocking. If we find the block | |
2277 | * we return zero and the path is unchanged. | |
2278 | * | |
2279 | * If we can't find the block, we set the path blocking and do some | |
2280 | * reada. -EAGAIN is returned and the search must be repeated. | |
2281 | */ | |
2282 | static int | |
2283 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2284 | struct btrfs_root *root, struct btrfs_path *p, | |
2285 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2286 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2287 | { |
2288 | u64 blocknr; | |
2289 | u64 gen; | |
2290 | u32 blocksize; | |
2291 | struct extent_buffer *b = *eb_ret; | |
2292 | struct extent_buffer *tmp; | |
76a05b35 | 2293 | int ret; |
c8c42864 CM |
2294 | |
2295 | blocknr = btrfs_node_blockptr(b, slot); | |
2296 | gen = btrfs_node_ptr_generation(b, slot); | |
2297 | blocksize = btrfs_level_size(root, level - 1); | |
2298 | ||
2299 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2300 | if (tmp) { |
b9fab919 | 2301 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2302 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2303 | *eb_ret = tmp; | |
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | /* the pages were up to date, but we failed | |
2308 | * the generation number check. Do a full | |
2309 | * read for the generation number that is correct. | |
2310 | * We must do this without dropping locks so | |
2311 | * we can trust our generation number | |
2312 | */ | |
2313 | btrfs_set_path_blocking(p); | |
2314 | ||
2315 | /* now we're allowed to do a blocking uptodate check */ | |
2316 | ret = btrfs_read_buffer(tmp, gen); | |
2317 | if (!ret) { | |
2318 | *eb_ret = tmp; | |
2319 | return 0; | |
cb44921a | 2320 | } |
bdf7c00e JB |
2321 | free_extent_buffer(tmp); |
2322 | btrfs_release_path(p); | |
2323 | return -EIO; | |
c8c42864 CM |
2324 | } |
2325 | ||
2326 | /* | |
2327 | * reduce lock contention at high levels | |
2328 | * of the btree by dropping locks before | |
76a05b35 CM |
2329 | * we read. Don't release the lock on the current |
2330 | * level because we need to walk this node to figure | |
2331 | * out which blocks to read. | |
c8c42864 | 2332 | */ |
8c594ea8 CM |
2333 | btrfs_unlock_up_safe(p, level + 1); |
2334 | btrfs_set_path_blocking(p); | |
2335 | ||
cb44921a | 2336 | free_extent_buffer(tmp); |
c8c42864 CM |
2337 | if (p->reada) |
2338 | reada_for_search(root, p, level, slot, key->objectid); | |
2339 | ||
b3b4aa74 | 2340 | btrfs_release_path(p); |
76a05b35 CM |
2341 | |
2342 | ret = -EAGAIN; | |
5bdd3536 | 2343 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2344 | if (tmp) { |
2345 | /* | |
2346 | * If the read above didn't mark this buffer up to date, | |
2347 | * it will never end up being up to date. Set ret to EIO now | |
2348 | * and give up so that our caller doesn't loop forever | |
2349 | * on our EAGAINs. | |
2350 | */ | |
b9fab919 | 2351 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2352 | ret = -EIO; |
c8c42864 | 2353 | free_extent_buffer(tmp); |
76a05b35 CM |
2354 | } |
2355 | return ret; | |
c8c42864 CM |
2356 | } |
2357 | ||
2358 | /* | |
2359 | * helper function for btrfs_search_slot. This does all of the checks | |
2360 | * for node-level blocks and does any balancing required based on | |
2361 | * the ins_len. | |
2362 | * | |
2363 | * If no extra work was required, zero is returned. If we had to | |
2364 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2365 | * start over | |
2366 | */ | |
2367 | static int | |
2368 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2369 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2370 | struct extent_buffer *b, int level, int ins_len, |
2371 | int *write_lock_level) | |
c8c42864 CM |
2372 | { |
2373 | int ret; | |
2374 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2375 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2376 | int sret; | |
2377 | ||
bd681513 CM |
2378 | if (*write_lock_level < level + 1) { |
2379 | *write_lock_level = level + 1; | |
2380 | btrfs_release_path(p); | |
2381 | goto again; | |
2382 | } | |
2383 | ||
c8c42864 | 2384 | btrfs_set_path_blocking(p); |
0b08851f | 2385 | reada_for_balance(root, p, level); |
c8c42864 | 2386 | sret = split_node(trans, root, p, level); |
bd681513 | 2387 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2388 | |
2389 | BUG_ON(sret > 0); | |
2390 | if (sret) { | |
2391 | ret = sret; | |
2392 | goto done; | |
2393 | } | |
2394 | b = p->nodes[level]; | |
2395 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2396 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2397 | int sret; |
2398 | ||
bd681513 CM |
2399 | if (*write_lock_level < level + 1) { |
2400 | *write_lock_level = level + 1; | |
2401 | btrfs_release_path(p); | |
2402 | goto again; | |
2403 | } | |
2404 | ||
c8c42864 | 2405 | btrfs_set_path_blocking(p); |
0b08851f | 2406 | reada_for_balance(root, p, level); |
c8c42864 | 2407 | sret = balance_level(trans, root, p, level); |
bd681513 | 2408 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2409 | |
2410 | if (sret) { | |
2411 | ret = sret; | |
2412 | goto done; | |
2413 | } | |
2414 | b = p->nodes[level]; | |
2415 | if (!b) { | |
b3b4aa74 | 2416 | btrfs_release_path(p); |
c8c42864 CM |
2417 | goto again; |
2418 | } | |
2419 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2420 | } | |
2421 | return 0; | |
2422 | ||
2423 | again: | |
2424 | ret = -EAGAIN; | |
2425 | done: | |
2426 | return ret; | |
2427 | } | |
2428 | ||
74123bd7 CM |
2429 | /* |
2430 | * look for key in the tree. path is filled in with nodes along the way | |
2431 | * if key is found, we return zero and you can find the item in the leaf | |
2432 | * level of the path (level 0) | |
2433 | * | |
2434 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2435 | * be inserted, and 1 is returned. If there are other errors during the |
2436 | * search a negative error number is returned. | |
97571fd0 CM |
2437 | * |
2438 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2439 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2440 | * possible) | |
74123bd7 | 2441 | */ |
e089f05c CM |
2442 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2443 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2444 | ins_len, int cow) | |
be0e5c09 | 2445 | { |
5f39d397 | 2446 | struct extent_buffer *b; |
be0e5c09 CM |
2447 | int slot; |
2448 | int ret; | |
33c66f43 | 2449 | int err; |
be0e5c09 | 2450 | int level; |
925baedd | 2451 | int lowest_unlock = 1; |
bd681513 CM |
2452 | int root_lock; |
2453 | /* everything at write_lock_level or lower must be write locked */ | |
2454 | int write_lock_level = 0; | |
9f3a7427 | 2455 | u8 lowest_level = 0; |
f7c79f30 | 2456 | int min_write_lock_level; |
9f3a7427 | 2457 | |
6702ed49 | 2458 | lowest_level = p->lowest_level; |
323ac95b | 2459 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2460 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2461 | |
bd681513 | 2462 | if (ins_len < 0) { |
925baedd | 2463 | lowest_unlock = 2; |
65b51a00 | 2464 | |
bd681513 CM |
2465 | /* when we are removing items, we might have to go up to level |
2466 | * two as we update tree pointers Make sure we keep write | |
2467 | * for those levels as well | |
2468 | */ | |
2469 | write_lock_level = 2; | |
2470 | } else if (ins_len > 0) { | |
2471 | /* | |
2472 | * for inserting items, make sure we have a write lock on | |
2473 | * level 1 so we can update keys | |
2474 | */ | |
2475 | write_lock_level = 1; | |
2476 | } | |
2477 | ||
2478 | if (!cow) | |
2479 | write_lock_level = -1; | |
2480 | ||
09a2a8f9 | 2481 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2482 | write_lock_level = BTRFS_MAX_LEVEL; |
2483 | ||
f7c79f30 CM |
2484 | min_write_lock_level = write_lock_level; |
2485 | ||
bb803951 | 2486 | again: |
bd681513 CM |
2487 | /* |
2488 | * we try very hard to do read locks on the root | |
2489 | */ | |
2490 | root_lock = BTRFS_READ_LOCK; | |
2491 | level = 0; | |
5d4f98a2 | 2492 | if (p->search_commit_root) { |
bd681513 CM |
2493 | /* |
2494 | * the commit roots are read only | |
2495 | * so we always do read locks | |
2496 | */ | |
5d4f98a2 YZ |
2497 | b = root->commit_root; |
2498 | extent_buffer_get(b); | |
bd681513 | 2499 | level = btrfs_header_level(b); |
5d4f98a2 | 2500 | if (!p->skip_locking) |
bd681513 | 2501 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2502 | } else { |
bd681513 | 2503 | if (p->skip_locking) { |
5d4f98a2 | 2504 | b = btrfs_root_node(root); |
bd681513 CM |
2505 | level = btrfs_header_level(b); |
2506 | } else { | |
2507 | /* we don't know the level of the root node | |
2508 | * until we actually have it read locked | |
2509 | */ | |
2510 | b = btrfs_read_lock_root_node(root); | |
2511 | level = btrfs_header_level(b); | |
2512 | if (level <= write_lock_level) { | |
2513 | /* whoops, must trade for write lock */ | |
2514 | btrfs_tree_read_unlock(b); | |
2515 | free_extent_buffer(b); | |
2516 | b = btrfs_lock_root_node(root); | |
2517 | root_lock = BTRFS_WRITE_LOCK; | |
2518 | ||
2519 | /* the level might have changed, check again */ | |
2520 | level = btrfs_header_level(b); | |
2521 | } | |
2522 | } | |
5d4f98a2 | 2523 | } |
bd681513 CM |
2524 | p->nodes[level] = b; |
2525 | if (!p->skip_locking) | |
2526 | p->locks[level] = root_lock; | |
925baedd | 2527 | |
eb60ceac | 2528 | while (b) { |
5f39d397 | 2529 | level = btrfs_header_level(b); |
65b51a00 CM |
2530 | |
2531 | /* | |
2532 | * setup the path here so we can release it under lock | |
2533 | * contention with the cow code | |
2534 | */ | |
02217ed2 | 2535 | if (cow) { |
c8c42864 CM |
2536 | /* |
2537 | * if we don't really need to cow this block | |
2538 | * then we don't want to set the path blocking, | |
2539 | * so we test it here | |
2540 | */ | |
5d4f98a2 | 2541 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2542 | goto cow_done; |
5d4f98a2 | 2543 | |
b4ce94de CM |
2544 | btrfs_set_path_blocking(p); |
2545 | ||
bd681513 CM |
2546 | /* |
2547 | * must have write locks on this node and the | |
2548 | * parent | |
2549 | */ | |
5124e00e JB |
2550 | if (level > write_lock_level || |
2551 | (level + 1 > write_lock_level && | |
2552 | level + 1 < BTRFS_MAX_LEVEL && | |
2553 | p->nodes[level + 1])) { | |
bd681513 CM |
2554 | write_lock_level = level + 1; |
2555 | btrfs_release_path(p); | |
2556 | goto again; | |
2557 | } | |
2558 | ||
33c66f43 YZ |
2559 | err = btrfs_cow_block(trans, root, b, |
2560 | p->nodes[level + 1], | |
2561 | p->slots[level + 1], &b); | |
2562 | if (err) { | |
33c66f43 | 2563 | ret = err; |
65b51a00 | 2564 | goto done; |
54aa1f4d | 2565 | } |
02217ed2 | 2566 | } |
65b51a00 | 2567 | cow_done: |
02217ed2 | 2568 | BUG_ON(!cow && ins_len); |
65b51a00 | 2569 | |
eb60ceac | 2570 | p->nodes[level] = b; |
bd681513 | 2571 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2572 | |
2573 | /* | |
2574 | * we have a lock on b and as long as we aren't changing | |
2575 | * the tree, there is no way to for the items in b to change. | |
2576 | * It is safe to drop the lock on our parent before we | |
2577 | * go through the expensive btree search on b. | |
2578 | * | |
2579 | * If cow is true, then we might be changing slot zero, | |
2580 | * which may require changing the parent. So, we can't | |
2581 | * drop the lock until after we know which slot we're | |
2582 | * operating on. | |
2583 | */ | |
2584 | if (!cow) | |
2585 | btrfs_unlock_up_safe(p, level + 1); | |
2586 | ||
5f39d397 | 2587 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2588 | |
5f39d397 | 2589 | if (level != 0) { |
33c66f43 YZ |
2590 | int dec = 0; |
2591 | if (ret && slot > 0) { | |
2592 | dec = 1; | |
be0e5c09 | 2593 | slot -= 1; |
33c66f43 | 2594 | } |
be0e5c09 | 2595 | p->slots[level] = slot; |
33c66f43 | 2596 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2597 | ins_len, &write_lock_level); |
33c66f43 | 2598 | if (err == -EAGAIN) |
c8c42864 | 2599 | goto again; |
33c66f43 YZ |
2600 | if (err) { |
2601 | ret = err; | |
c8c42864 | 2602 | goto done; |
33c66f43 | 2603 | } |
c8c42864 CM |
2604 | b = p->nodes[level]; |
2605 | slot = p->slots[level]; | |
b4ce94de | 2606 | |
bd681513 CM |
2607 | /* |
2608 | * slot 0 is special, if we change the key | |
2609 | * we have to update the parent pointer | |
2610 | * which means we must have a write lock | |
2611 | * on the parent | |
2612 | */ | |
2613 | if (slot == 0 && cow && | |
2614 | write_lock_level < level + 1) { | |
2615 | write_lock_level = level + 1; | |
2616 | btrfs_release_path(p); | |
2617 | goto again; | |
2618 | } | |
2619 | ||
f7c79f30 CM |
2620 | unlock_up(p, level, lowest_unlock, |
2621 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2622 | |
925baedd | 2623 | if (level == lowest_level) { |
33c66f43 YZ |
2624 | if (dec) |
2625 | p->slots[level]++; | |
5b21f2ed | 2626 | goto done; |
925baedd | 2627 | } |
ca7a79ad | 2628 | |
33c66f43 | 2629 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2630 | &b, level, slot, key, 0); |
33c66f43 | 2631 | if (err == -EAGAIN) |
c8c42864 | 2632 | goto again; |
33c66f43 YZ |
2633 | if (err) { |
2634 | ret = err; | |
76a05b35 | 2635 | goto done; |
33c66f43 | 2636 | } |
76a05b35 | 2637 | |
b4ce94de | 2638 | if (!p->skip_locking) { |
bd681513 CM |
2639 | level = btrfs_header_level(b); |
2640 | if (level <= write_lock_level) { | |
2641 | err = btrfs_try_tree_write_lock(b); | |
2642 | if (!err) { | |
2643 | btrfs_set_path_blocking(p); | |
2644 | btrfs_tree_lock(b); | |
2645 | btrfs_clear_path_blocking(p, b, | |
2646 | BTRFS_WRITE_LOCK); | |
2647 | } | |
2648 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2649 | } else { | |
2650 | err = btrfs_try_tree_read_lock(b); | |
2651 | if (!err) { | |
2652 | btrfs_set_path_blocking(p); | |
2653 | btrfs_tree_read_lock(b); | |
2654 | btrfs_clear_path_blocking(p, b, | |
2655 | BTRFS_READ_LOCK); | |
2656 | } | |
2657 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2658 | } |
bd681513 | 2659 | p->nodes[level] = b; |
b4ce94de | 2660 | } |
be0e5c09 CM |
2661 | } else { |
2662 | p->slots[level] = slot; | |
87b29b20 YZ |
2663 | if (ins_len > 0 && |
2664 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2665 | if (write_lock_level < 1) { |
2666 | write_lock_level = 1; | |
2667 | btrfs_release_path(p); | |
2668 | goto again; | |
2669 | } | |
2670 | ||
b4ce94de | 2671 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2672 | err = split_leaf(trans, root, key, |
2673 | p, ins_len, ret == 0); | |
bd681513 | 2674 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2675 | |
33c66f43 YZ |
2676 | BUG_ON(err > 0); |
2677 | if (err) { | |
2678 | ret = err; | |
65b51a00 CM |
2679 | goto done; |
2680 | } | |
5c680ed6 | 2681 | } |
459931ec | 2682 | if (!p->search_for_split) |
f7c79f30 CM |
2683 | unlock_up(p, level, lowest_unlock, |
2684 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2685 | goto done; |
be0e5c09 CM |
2686 | } |
2687 | } | |
65b51a00 CM |
2688 | ret = 1; |
2689 | done: | |
b4ce94de CM |
2690 | /* |
2691 | * we don't really know what they plan on doing with the path | |
2692 | * from here on, so for now just mark it as blocking | |
2693 | */ | |
b9473439 CM |
2694 | if (!p->leave_spinning) |
2695 | btrfs_set_path_blocking(p); | |
76a05b35 | 2696 | if (ret < 0) |
b3b4aa74 | 2697 | btrfs_release_path(p); |
65b51a00 | 2698 | return ret; |
be0e5c09 CM |
2699 | } |
2700 | ||
5d9e75c4 JS |
2701 | /* |
2702 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2703 | * current state of the tree together with the operations recorded in the tree | |
2704 | * modification log to search for the key in a previous version of this tree, as | |
2705 | * denoted by the time_seq parameter. | |
2706 | * | |
2707 | * Naturally, there is no support for insert, delete or cow operations. | |
2708 | * | |
2709 | * The resulting path and return value will be set up as if we called | |
2710 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2711 | */ | |
2712 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2713 | struct btrfs_path *p, u64 time_seq) | |
2714 | { | |
2715 | struct extent_buffer *b; | |
2716 | int slot; | |
2717 | int ret; | |
2718 | int err; | |
2719 | int level; | |
2720 | int lowest_unlock = 1; | |
2721 | u8 lowest_level = 0; | |
2722 | ||
2723 | lowest_level = p->lowest_level; | |
2724 | WARN_ON(p->nodes[0] != NULL); | |
2725 | ||
2726 | if (p->search_commit_root) { | |
2727 | BUG_ON(time_seq); | |
2728 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2729 | } | |
2730 | ||
2731 | again: | |
5d9e75c4 | 2732 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2733 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2734 | p->locks[level] = BTRFS_READ_LOCK; |
2735 | ||
2736 | while (b) { | |
2737 | level = btrfs_header_level(b); | |
2738 | p->nodes[level] = b; | |
2739 | btrfs_clear_path_blocking(p, NULL, 0); | |
2740 | ||
2741 | /* | |
2742 | * we have a lock on b and as long as we aren't changing | |
2743 | * the tree, there is no way to for the items in b to change. | |
2744 | * It is safe to drop the lock on our parent before we | |
2745 | * go through the expensive btree search on b. | |
2746 | */ | |
2747 | btrfs_unlock_up_safe(p, level + 1); | |
2748 | ||
2749 | ret = bin_search(b, key, level, &slot); | |
2750 | ||
2751 | if (level != 0) { | |
2752 | int dec = 0; | |
2753 | if (ret && slot > 0) { | |
2754 | dec = 1; | |
2755 | slot -= 1; | |
2756 | } | |
2757 | p->slots[level] = slot; | |
2758 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2759 | ||
2760 | if (level == lowest_level) { | |
2761 | if (dec) | |
2762 | p->slots[level]++; | |
2763 | goto done; | |
2764 | } | |
2765 | ||
2766 | err = read_block_for_search(NULL, root, p, &b, level, | |
2767 | slot, key, time_seq); | |
2768 | if (err == -EAGAIN) | |
2769 | goto again; | |
2770 | if (err) { | |
2771 | ret = err; | |
2772 | goto done; | |
2773 | } | |
2774 | ||
2775 | level = btrfs_header_level(b); | |
2776 | err = btrfs_try_tree_read_lock(b); | |
2777 | if (!err) { | |
2778 | btrfs_set_path_blocking(p); | |
2779 | btrfs_tree_read_lock(b); | |
2780 | btrfs_clear_path_blocking(p, b, | |
2781 | BTRFS_READ_LOCK); | |
2782 | } | |
9ec72677 | 2783 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
2784 | if (!b) { |
2785 | ret = -ENOMEM; | |
2786 | goto done; | |
2787 | } | |
5d9e75c4 JS |
2788 | p->locks[level] = BTRFS_READ_LOCK; |
2789 | p->nodes[level] = b; | |
5d9e75c4 JS |
2790 | } else { |
2791 | p->slots[level] = slot; | |
2792 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2793 | goto done; | |
2794 | } | |
2795 | } | |
2796 | ret = 1; | |
2797 | done: | |
2798 | if (!p->leave_spinning) | |
2799 | btrfs_set_path_blocking(p); | |
2800 | if (ret < 0) | |
2801 | btrfs_release_path(p); | |
2802 | ||
2803 | return ret; | |
2804 | } | |
2805 | ||
2f38b3e1 AJ |
2806 | /* |
2807 | * helper to use instead of search slot if no exact match is needed but | |
2808 | * instead the next or previous item should be returned. | |
2809 | * When find_higher is true, the next higher item is returned, the next lower | |
2810 | * otherwise. | |
2811 | * When return_any and find_higher are both true, and no higher item is found, | |
2812 | * return the next lower instead. | |
2813 | * When return_any is true and find_higher is false, and no lower item is found, | |
2814 | * return the next higher instead. | |
2815 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2816 | * < 0 on error | |
2817 | */ | |
2818 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
2819 | struct btrfs_key *key, struct btrfs_path *p, | |
2820 | int find_higher, int return_any) | |
2821 | { | |
2822 | int ret; | |
2823 | struct extent_buffer *leaf; | |
2824 | ||
2825 | again: | |
2826 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2827 | if (ret <= 0) | |
2828 | return ret; | |
2829 | /* | |
2830 | * a return value of 1 means the path is at the position where the | |
2831 | * item should be inserted. Normally this is the next bigger item, | |
2832 | * but in case the previous item is the last in a leaf, path points | |
2833 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2834 | * item. | |
2835 | */ | |
2836 | leaf = p->nodes[0]; | |
2837 | ||
2838 | if (find_higher) { | |
2839 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2840 | ret = btrfs_next_leaf(root, p); | |
2841 | if (ret <= 0) | |
2842 | return ret; | |
2843 | if (!return_any) | |
2844 | return 1; | |
2845 | /* | |
2846 | * no higher item found, return the next | |
2847 | * lower instead | |
2848 | */ | |
2849 | return_any = 0; | |
2850 | find_higher = 0; | |
2851 | btrfs_release_path(p); | |
2852 | goto again; | |
2853 | } | |
2854 | } else { | |
e6793769 AJ |
2855 | if (p->slots[0] == 0) { |
2856 | ret = btrfs_prev_leaf(root, p); | |
2857 | if (ret < 0) | |
2858 | return ret; | |
2859 | if (!ret) { | |
2860 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
2861 | return 0; | |
2f38b3e1 | 2862 | } |
e6793769 AJ |
2863 | if (!return_any) |
2864 | return 1; | |
2865 | /* | |
2866 | * no lower item found, return the next | |
2867 | * higher instead | |
2868 | */ | |
2869 | return_any = 0; | |
2870 | find_higher = 1; | |
2871 | btrfs_release_path(p); | |
2872 | goto again; | |
2873 | } else { | |
2f38b3e1 AJ |
2874 | --p->slots[0]; |
2875 | } | |
2876 | } | |
2877 | return 0; | |
2878 | } | |
2879 | ||
74123bd7 CM |
2880 | /* |
2881 | * adjust the pointers going up the tree, starting at level | |
2882 | * making sure the right key of each node is points to 'key'. | |
2883 | * This is used after shifting pointers to the left, so it stops | |
2884 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2885 | * higher levels | |
aa5d6bed | 2886 | * |
74123bd7 | 2887 | */ |
d6a0a126 | 2888 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2889 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2890 | { |
2891 | int i; | |
5f39d397 CM |
2892 | struct extent_buffer *t; |
2893 | ||
234b63a0 | 2894 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2895 | int tslot = path->slots[i]; |
eb60ceac | 2896 | if (!path->nodes[i]) |
be0e5c09 | 2897 | break; |
5f39d397 | 2898 | t = path->nodes[i]; |
32adf090 | 2899 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 2900 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2901 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2902 | if (tslot != 0) |
2903 | break; | |
2904 | } | |
2905 | } | |
2906 | ||
31840ae1 ZY |
2907 | /* |
2908 | * update item key. | |
2909 | * | |
2910 | * This function isn't completely safe. It's the caller's responsibility | |
2911 | * that the new key won't break the order | |
2912 | */ | |
afe5fea7 | 2913 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2914 | struct btrfs_key *new_key) |
31840ae1 ZY |
2915 | { |
2916 | struct btrfs_disk_key disk_key; | |
2917 | struct extent_buffer *eb; | |
2918 | int slot; | |
2919 | ||
2920 | eb = path->nodes[0]; | |
2921 | slot = path->slots[0]; | |
2922 | if (slot > 0) { | |
2923 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2924 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2925 | } |
2926 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2927 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2928 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2929 | } |
2930 | ||
2931 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2932 | btrfs_set_item_key(eb, &disk_key, slot); | |
2933 | btrfs_mark_buffer_dirty(eb); | |
2934 | if (slot == 0) | |
d6a0a126 | 2935 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
2936 | } |
2937 | ||
74123bd7 CM |
2938 | /* |
2939 | * try to push data from one node into the next node left in the | |
79f95c82 | 2940 | * tree. |
aa5d6bed CM |
2941 | * |
2942 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2943 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2944 | */ |
98ed5174 CM |
2945 | static int push_node_left(struct btrfs_trans_handle *trans, |
2946 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 2947 | struct extent_buffer *src, int empty) |
be0e5c09 | 2948 | { |
be0e5c09 | 2949 | int push_items = 0; |
bb803951 CM |
2950 | int src_nritems; |
2951 | int dst_nritems; | |
aa5d6bed | 2952 | int ret = 0; |
be0e5c09 | 2953 | |
5f39d397 CM |
2954 | src_nritems = btrfs_header_nritems(src); |
2955 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2956 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2957 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2958 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2959 | |
bce4eae9 | 2960 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2961 | return 1; |
2962 | ||
d397712b | 2963 | if (push_items <= 0) |
be0e5c09 CM |
2964 | return 1; |
2965 | ||
bce4eae9 | 2966 | if (empty) { |
971a1f66 | 2967 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2968 | if (push_items < src_nritems) { |
2969 | /* leave at least 8 pointers in the node if | |
2970 | * we aren't going to empty it | |
2971 | */ | |
2972 | if (src_nritems - push_items < 8) { | |
2973 | if (push_items <= 8) | |
2974 | return 1; | |
2975 | push_items -= 8; | |
2976 | } | |
2977 | } | |
2978 | } else | |
2979 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2980 | |
f230475e | 2981 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
90f8d62e | 2982 | push_items); |
5f39d397 CM |
2983 | copy_extent_buffer(dst, src, |
2984 | btrfs_node_key_ptr_offset(dst_nritems), | |
2985 | btrfs_node_key_ptr_offset(0), | |
d397712b | 2986 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 2987 | |
bb803951 | 2988 | if (push_items < src_nritems) { |
57911b8b JS |
2989 | /* |
2990 | * don't call tree_mod_log_eb_move here, key removal was already | |
2991 | * fully logged by tree_mod_log_eb_copy above. | |
2992 | */ | |
5f39d397 CM |
2993 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
2994 | btrfs_node_key_ptr_offset(push_items), | |
2995 | (src_nritems - push_items) * | |
2996 | sizeof(struct btrfs_key_ptr)); | |
2997 | } | |
2998 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
2999 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3000 | btrfs_mark_buffer_dirty(src); | |
3001 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3002 | |
79f95c82 CM |
3003 | return ret; |
3004 | } | |
3005 | ||
3006 | /* | |
3007 | * try to push data from one node into the next node right in the | |
3008 | * tree. | |
3009 | * | |
3010 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3011 | * error, and > 0 if there was no room in the right hand block. | |
3012 | * | |
3013 | * this will only push up to 1/2 the contents of the left node over | |
3014 | */ | |
5f39d397 CM |
3015 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3016 | struct btrfs_root *root, | |
3017 | struct extent_buffer *dst, | |
3018 | struct extent_buffer *src) | |
79f95c82 | 3019 | { |
79f95c82 CM |
3020 | int push_items = 0; |
3021 | int max_push; | |
3022 | int src_nritems; | |
3023 | int dst_nritems; | |
3024 | int ret = 0; | |
79f95c82 | 3025 | |
7bb86316 CM |
3026 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3027 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3028 | ||
5f39d397 CM |
3029 | src_nritems = btrfs_header_nritems(src); |
3030 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3031 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3032 | if (push_items <= 0) |
79f95c82 | 3033 | return 1; |
bce4eae9 | 3034 | |
d397712b | 3035 | if (src_nritems < 4) |
bce4eae9 | 3036 | return 1; |
79f95c82 CM |
3037 | |
3038 | max_push = src_nritems / 2 + 1; | |
3039 | /* don't try to empty the node */ | |
d397712b | 3040 | if (max_push >= src_nritems) |
79f95c82 | 3041 | return 1; |
252c38f0 | 3042 | |
79f95c82 CM |
3043 | if (max_push < push_items) |
3044 | push_items = max_push; | |
3045 | ||
f230475e | 3046 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3047 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3048 | btrfs_node_key_ptr_offset(0), | |
3049 | (dst_nritems) * | |
3050 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3051 | |
f230475e | 3052 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
90f8d62e | 3053 | src_nritems - push_items, push_items); |
5f39d397 CM |
3054 | copy_extent_buffer(dst, src, |
3055 | btrfs_node_key_ptr_offset(0), | |
3056 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3057 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3058 | |
5f39d397 CM |
3059 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3060 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3061 | |
5f39d397 CM |
3062 | btrfs_mark_buffer_dirty(src); |
3063 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3064 | |
aa5d6bed | 3065 | return ret; |
be0e5c09 CM |
3066 | } |
3067 | ||
97571fd0 CM |
3068 | /* |
3069 | * helper function to insert a new root level in the tree. | |
3070 | * A new node is allocated, and a single item is inserted to | |
3071 | * point to the existing root | |
aa5d6bed CM |
3072 | * |
3073 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3074 | */ |
d397712b | 3075 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3076 | struct btrfs_root *root, |
fdd99c72 | 3077 | struct btrfs_path *path, int level) |
5c680ed6 | 3078 | { |
7bb86316 | 3079 | u64 lower_gen; |
5f39d397 CM |
3080 | struct extent_buffer *lower; |
3081 | struct extent_buffer *c; | |
925baedd | 3082 | struct extent_buffer *old; |
5f39d397 | 3083 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3084 | |
3085 | BUG_ON(path->nodes[level]); | |
3086 | BUG_ON(path->nodes[level-1] != root->node); | |
3087 | ||
7bb86316 CM |
3088 | lower = path->nodes[level-1]; |
3089 | if (level == 1) | |
3090 | btrfs_item_key(lower, &lower_key, 0); | |
3091 | else | |
3092 | btrfs_node_key(lower, &lower_key, 0); | |
3093 | ||
31840ae1 | 3094 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3095 | root->root_key.objectid, &lower_key, |
5581a51a | 3096 | level, root->node->start, 0); |
5f39d397 CM |
3097 | if (IS_ERR(c)) |
3098 | return PTR_ERR(c); | |
925baedd | 3099 | |
f0486c68 YZ |
3100 | root_add_used(root, root->nodesize); |
3101 | ||
5d4f98a2 | 3102 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3103 | btrfs_set_header_nritems(c, 1); |
3104 | btrfs_set_header_level(c, level); | |
db94535d | 3105 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3106 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3107 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3108 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3109 | |
fba6aa75 | 3110 | write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(c), |
5f39d397 | 3111 | BTRFS_FSID_SIZE); |
e17cade2 CM |
3112 | |
3113 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 3114 | btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE); |
e17cade2 | 3115 | |
5f39d397 | 3116 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3117 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3118 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3119 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3120 | |
3121 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3122 | |
5f39d397 | 3123 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3124 | |
925baedd | 3125 | old = root->node; |
fdd99c72 | 3126 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3127 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3128 | |
3129 | /* the super has an extra ref to root->node */ | |
3130 | free_extent_buffer(old); | |
3131 | ||
0b86a832 | 3132 | add_root_to_dirty_list(root); |
5f39d397 CM |
3133 | extent_buffer_get(c); |
3134 | path->nodes[level] = c; | |
bd681513 | 3135 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3136 | path->slots[level] = 0; |
3137 | return 0; | |
3138 | } | |
3139 | ||
74123bd7 CM |
3140 | /* |
3141 | * worker function to insert a single pointer in a node. | |
3142 | * the node should have enough room for the pointer already | |
97571fd0 | 3143 | * |
74123bd7 CM |
3144 | * slot and level indicate where you want the key to go, and |
3145 | * blocknr is the block the key points to. | |
3146 | */ | |
143bede5 JM |
3147 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3148 | struct btrfs_root *root, struct btrfs_path *path, | |
3149 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3150 | int slot, int level) |
74123bd7 | 3151 | { |
5f39d397 | 3152 | struct extent_buffer *lower; |
74123bd7 | 3153 | int nritems; |
f3ea38da | 3154 | int ret; |
5c680ed6 CM |
3155 | |
3156 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3157 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3158 | lower = path->nodes[level]; |
3159 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3160 | BUG_ON(slot > nritems); |
143bede5 | 3161 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3162 | if (slot != nritems) { |
c3e06965 | 3163 | if (level) |
f3ea38da JS |
3164 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3165 | slot, nritems - slot); | |
5f39d397 CM |
3166 | memmove_extent_buffer(lower, |
3167 | btrfs_node_key_ptr_offset(slot + 1), | |
3168 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3169 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3170 | } |
c3e06965 | 3171 | if (level) { |
f3ea38da | 3172 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3173 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3174 | BUG_ON(ret < 0); |
3175 | } | |
5f39d397 | 3176 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3177 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3178 | WARN_ON(trans->transid == 0); |
3179 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3180 | btrfs_set_header_nritems(lower, nritems + 1); |
3181 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3182 | } |
3183 | ||
97571fd0 CM |
3184 | /* |
3185 | * split the node at the specified level in path in two. | |
3186 | * The path is corrected to point to the appropriate node after the split | |
3187 | * | |
3188 | * Before splitting this tries to make some room in the node by pushing | |
3189 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3190 | * |
3191 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3192 | */ |
e02119d5 CM |
3193 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3194 | struct btrfs_root *root, | |
3195 | struct btrfs_path *path, int level) | |
be0e5c09 | 3196 | { |
5f39d397 CM |
3197 | struct extent_buffer *c; |
3198 | struct extent_buffer *split; | |
3199 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3200 | int mid; |
5c680ed6 | 3201 | int ret; |
7518a238 | 3202 | u32 c_nritems; |
eb60ceac | 3203 | |
5f39d397 | 3204 | c = path->nodes[level]; |
7bb86316 | 3205 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3206 | if (c == root->node) { |
d9abbf1c | 3207 | /* |
90f8d62e JS |
3208 | * trying to split the root, lets make a new one |
3209 | * | |
fdd99c72 | 3210 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3211 | * insert_new_root, because that root buffer will be kept as a |
3212 | * normal node. We are going to log removal of half of the | |
3213 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3214 | * tree lock on the buffer, which is why we cannot race with | |
3215 | * other tree_mod_log users. | |
d9abbf1c | 3216 | */ |
fdd99c72 | 3217 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3218 | if (ret) |
3219 | return ret; | |
b3612421 | 3220 | } else { |
e66f709b | 3221 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3222 | c = path->nodes[level]; |
3223 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3224 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3225 | return 0; |
54aa1f4d CM |
3226 | if (ret < 0) |
3227 | return ret; | |
be0e5c09 | 3228 | } |
e66f709b | 3229 | |
5f39d397 | 3230 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3231 | mid = (c_nritems + 1) / 2; |
3232 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3233 | |
5d4f98a2 | 3234 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3235 | root->root_key.objectid, |
5581a51a | 3236 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3237 | if (IS_ERR(split)) |
3238 | return PTR_ERR(split); | |
3239 | ||
f0486c68 YZ |
3240 | root_add_used(root, root->nodesize); |
3241 | ||
5d4f98a2 | 3242 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3243 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3244 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3245 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3246 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3247 | btrfs_set_header_owner(split, root->root_key.objectid); |
3248 | write_extent_buffer(split, root->fs_info->fsid, | |
fba6aa75 | 3249 | btrfs_header_fsid(split), BTRFS_FSID_SIZE); |
e17cade2 | 3250 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
b308bc2f | 3251 | btrfs_header_chunk_tree_uuid(split), |
e17cade2 | 3252 | BTRFS_UUID_SIZE); |
54aa1f4d | 3253 | |
90f8d62e | 3254 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
3255 | copy_extent_buffer(split, c, |
3256 | btrfs_node_key_ptr_offset(0), | |
3257 | btrfs_node_key_ptr_offset(mid), | |
3258 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3259 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3260 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3261 | ret = 0; |
3262 | ||
5f39d397 CM |
3263 | btrfs_mark_buffer_dirty(c); |
3264 | btrfs_mark_buffer_dirty(split); | |
3265 | ||
143bede5 | 3266 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3267 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3268 | |
5de08d7d | 3269 | if (path->slots[level] >= mid) { |
5c680ed6 | 3270 | path->slots[level] -= mid; |
925baedd | 3271 | btrfs_tree_unlock(c); |
5f39d397 CM |
3272 | free_extent_buffer(c); |
3273 | path->nodes[level] = split; | |
5c680ed6 CM |
3274 | path->slots[level + 1] += 1; |
3275 | } else { | |
925baedd | 3276 | btrfs_tree_unlock(split); |
5f39d397 | 3277 | free_extent_buffer(split); |
be0e5c09 | 3278 | } |
aa5d6bed | 3279 | return ret; |
be0e5c09 CM |
3280 | } |
3281 | ||
74123bd7 CM |
3282 | /* |
3283 | * how many bytes are required to store the items in a leaf. start | |
3284 | * and nr indicate which items in the leaf to check. This totals up the | |
3285 | * space used both by the item structs and the item data | |
3286 | */ | |
5f39d397 | 3287 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3288 | { |
41be1f3b JB |
3289 | struct btrfs_item *start_item; |
3290 | struct btrfs_item *end_item; | |
3291 | struct btrfs_map_token token; | |
be0e5c09 | 3292 | int data_len; |
5f39d397 | 3293 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3294 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3295 | |
3296 | if (!nr) | |
3297 | return 0; | |
41be1f3b JB |
3298 | btrfs_init_map_token(&token); |
3299 | start_item = btrfs_item_nr(l, start); | |
3300 | end_item = btrfs_item_nr(l, end); | |
3301 | data_len = btrfs_token_item_offset(l, start_item, &token) + | |
3302 | btrfs_token_item_size(l, start_item, &token); | |
3303 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3304 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3305 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3306 | return data_len; |
3307 | } | |
3308 | ||
d4dbff95 CM |
3309 | /* |
3310 | * The space between the end of the leaf items and | |
3311 | * the start of the leaf data. IOW, how much room | |
3312 | * the leaf has left for both items and data | |
3313 | */ | |
d397712b | 3314 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3315 | struct extent_buffer *leaf) |
d4dbff95 | 3316 | { |
5f39d397 CM |
3317 | int nritems = btrfs_header_nritems(leaf); |
3318 | int ret; | |
3319 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3320 | if (ret < 0) { | |
d397712b CM |
3321 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
3322 | "used %d nritems %d\n", | |
ae2f5411 | 3323 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3324 | leaf_space_used(leaf, 0, nritems), nritems); |
3325 | } | |
3326 | return ret; | |
d4dbff95 CM |
3327 | } |
3328 | ||
99d8f83c CM |
3329 | /* |
3330 | * min slot controls the lowest index we're willing to push to the | |
3331 | * right. We'll push up to and including min_slot, but no lower | |
3332 | */ | |
44871b1b CM |
3333 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3334 | struct btrfs_root *root, | |
3335 | struct btrfs_path *path, | |
3336 | int data_size, int empty, | |
3337 | struct extent_buffer *right, | |
99d8f83c CM |
3338 | int free_space, u32 left_nritems, |
3339 | u32 min_slot) | |
00ec4c51 | 3340 | { |
5f39d397 | 3341 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3342 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3343 | struct btrfs_map_token token; |
5f39d397 | 3344 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3345 | int slot; |
34a38218 | 3346 | u32 i; |
00ec4c51 CM |
3347 | int push_space = 0; |
3348 | int push_items = 0; | |
0783fcfc | 3349 | struct btrfs_item *item; |
34a38218 | 3350 | u32 nr; |
7518a238 | 3351 | u32 right_nritems; |
5f39d397 | 3352 | u32 data_end; |
db94535d | 3353 | u32 this_item_size; |
00ec4c51 | 3354 | |
cfed81a0 CM |
3355 | btrfs_init_map_token(&token); |
3356 | ||
34a38218 CM |
3357 | if (empty) |
3358 | nr = 0; | |
3359 | else | |
99d8f83c | 3360 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3361 | |
31840ae1 | 3362 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3363 | push_space += data_size; |
31840ae1 | 3364 | |
44871b1b | 3365 | slot = path->slots[1]; |
34a38218 CM |
3366 | i = left_nritems - 1; |
3367 | while (i >= nr) { | |
5f39d397 | 3368 | item = btrfs_item_nr(left, i); |
db94535d | 3369 | |
31840ae1 ZY |
3370 | if (!empty && push_items > 0) { |
3371 | if (path->slots[0] > i) | |
3372 | break; | |
3373 | if (path->slots[0] == i) { | |
3374 | int space = btrfs_leaf_free_space(root, left); | |
3375 | if (space + push_space * 2 > free_space) | |
3376 | break; | |
3377 | } | |
3378 | } | |
3379 | ||
00ec4c51 | 3380 | if (path->slots[0] == i) |
87b29b20 | 3381 | push_space += data_size; |
db94535d | 3382 | |
db94535d CM |
3383 | this_item_size = btrfs_item_size(left, item); |
3384 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3385 | break; |
31840ae1 | 3386 | |
00ec4c51 | 3387 | push_items++; |
db94535d | 3388 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3389 | if (i == 0) |
3390 | break; | |
3391 | i--; | |
db94535d | 3392 | } |
5f39d397 | 3393 | |
925baedd CM |
3394 | if (push_items == 0) |
3395 | goto out_unlock; | |
5f39d397 | 3396 | |
6c1500f2 | 3397 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3398 | |
00ec4c51 | 3399 | /* push left to right */ |
5f39d397 | 3400 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3401 | |
5f39d397 | 3402 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3403 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3404 | |
00ec4c51 | 3405 | /* make room in the right data area */ |
5f39d397 CM |
3406 | data_end = leaf_data_end(root, right); |
3407 | memmove_extent_buffer(right, | |
3408 | btrfs_leaf_data(right) + data_end - push_space, | |
3409 | btrfs_leaf_data(right) + data_end, | |
3410 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3411 | ||
00ec4c51 | 3412 | /* copy from the left data area */ |
5f39d397 | 3413 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3414 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3415 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3416 | push_space); | |
5f39d397 CM |
3417 | |
3418 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3419 | btrfs_item_nr_offset(0), | |
3420 | right_nritems * sizeof(struct btrfs_item)); | |
3421 | ||
00ec4c51 | 3422 | /* copy the items from left to right */ |
5f39d397 CM |
3423 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3424 | btrfs_item_nr_offset(left_nritems - push_items), | |
3425 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3426 | |
3427 | /* update the item pointers */ | |
7518a238 | 3428 | right_nritems += push_items; |
5f39d397 | 3429 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3430 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3431 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 3432 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
3433 | push_space -= btrfs_token_item_size(right, item, &token); |
3434 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3435 | } |
3436 | ||
7518a238 | 3437 | left_nritems -= push_items; |
5f39d397 | 3438 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3439 | |
34a38218 CM |
3440 | if (left_nritems) |
3441 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3442 | else |
3443 | clean_tree_block(trans, root, left); | |
3444 | ||
5f39d397 | 3445 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3446 | |
5f39d397 CM |
3447 | btrfs_item_key(right, &disk_key, 0); |
3448 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3449 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3450 | |
00ec4c51 | 3451 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3452 | if (path->slots[0] >= left_nritems) { |
3453 | path->slots[0] -= left_nritems; | |
925baedd CM |
3454 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3455 | clean_tree_block(trans, root, path->nodes[0]); | |
3456 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3457 | free_extent_buffer(path->nodes[0]); |
3458 | path->nodes[0] = right; | |
00ec4c51 CM |
3459 | path->slots[1] += 1; |
3460 | } else { | |
925baedd | 3461 | btrfs_tree_unlock(right); |
5f39d397 | 3462 | free_extent_buffer(right); |
00ec4c51 CM |
3463 | } |
3464 | return 0; | |
925baedd CM |
3465 | |
3466 | out_unlock: | |
3467 | btrfs_tree_unlock(right); | |
3468 | free_extent_buffer(right); | |
3469 | return 1; | |
00ec4c51 | 3470 | } |
925baedd | 3471 | |
44871b1b CM |
3472 | /* |
3473 | * push some data in the path leaf to the right, trying to free up at | |
3474 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3475 | * | |
3476 | * returns 1 if the push failed because the other node didn't have enough | |
3477 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3478 | * |
3479 | * this will push starting from min_slot to the end of the leaf. It won't | |
3480 | * push any slot lower than min_slot | |
44871b1b CM |
3481 | */ |
3482 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3483 | *root, struct btrfs_path *path, |
3484 | int min_data_size, int data_size, | |
3485 | int empty, u32 min_slot) | |
44871b1b CM |
3486 | { |
3487 | struct extent_buffer *left = path->nodes[0]; | |
3488 | struct extent_buffer *right; | |
3489 | struct extent_buffer *upper; | |
3490 | int slot; | |
3491 | int free_space; | |
3492 | u32 left_nritems; | |
3493 | int ret; | |
3494 | ||
3495 | if (!path->nodes[1]) | |
3496 | return 1; | |
3497 | ||
3498 | slot = path->slots[1]; | |
3499 | upper = path->nodes[1]; | |
3500 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3501 | return 1; | |
3502 | ||
3503 | btrfs_assert_tree_locked(path->nodes[1]); | |
3504 | ||
3505 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3506 | if (right == NULL) |
3507 | return 1; | |
3508 | ||
44871b1b CM |
3509 | btrfs_tree_lock(right); |
3510 | btrfs_set_lock_blocking(right); | |
3511 | ||
3512 | free_space = btrfs_leaf_free_space(root, right); | |
3513 | if (free_space < data_size) | |
3514 | goto out_unlock; | |
3515 | ||
3516 | /* cow and double check */ | |
3517 | ret = btrfs_cow_block(trans, root, right, upper, | |
3518 | slot + 1, &right); | |
3519 | if (ret) | |
3520 | goto out_unlock; | |
3521 | ||
3522 | free_space = btrfs_leaf_free_space(root, right); | |
3523 | if (free_space < data_size) | |
3524 | goto out_unlock; | |
3525 | ||
3526 | left_nritems = btrfs_header_nritems(left); | |
3527 | if (left_nritems == 0) | |
3528 | goto out_unlock; | |
3529 | ||
99d8f83c CM |
3530 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3531 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3532 | out_unlock: |
3533 | btrfs_tree_unlock(right); | |
3534 | free_extent_buffer(right); | |
3535 | return 1; | |
3536 | } | |
3537 | ||
74123bd7 CM |
3538 | /* |
3539 | * push some data in the path leaf to the left, trying to free up at | |
3540 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3541 | * |
3542 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3543 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3544 | * items | |
74123bd7 | 3545 | */ |
44871b1b CM |
3546 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3547 | struct btrfs_root *root, | |
3548 | struct btrfs_path *path, int data_size, | |
3549 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3550 | int free_space, u32 right_nritems, |
3551 | u32 max_slot) | |
be0e5c09 | 3552 | { |
5f39d397 CM |
3553 | struct btrfs_disk_key disk_key; |
3554 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3555 | int i; |
be0e5c09 CM |
3556 | int push_space = 0; |
3557 | int push_items = 0; | |
0783fcfc | 3558 | struct btrfs_item *item; |
7518a238 | 3559 | u32 old_left_nritems; |
34a38218 | 3560 | u32 nr; |
aa5d6bed | 3561 | int ret = 0; |
db94535d CM |
3562 | u32 this_item_size; |
3563 | u32 old_left_item_size; | |
cfed81a0 CM |
3564 | struct btrfs_map_token token; |
3565 | ||
3566 | btrfs_init_map_token(&token); | |
be0e5c09 | 3567 | |
34a38218 | 3568 | if (empty) |
99d8f83c | 3569 | nr = min(right_nritems, max_slot); |
34a38218 | 3570 | else |
99d8f83c | 3571 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3572 | |
3573 | for (i = 0; i < nr; i++) { | |
5f39d397 | 3574 | item = btrfs_item_nr(right, i); |
db94535d | 3575 | |
31840ae1 ZY |
3576 | if (!empty && push_items > 0) { |
3577 | if (path->slots[0] < i) | |
3578 | break; | |
3579 | if (path->slots[0] == i) { | |
3580 | int space = btrfs_leaf_free_space(root, right); | |
3581 | if (space + push_space * 2 > free_space) | |
3582 | break; | |
3583 | } | |
3584 | } | |
3585 | ||
be0e5c09 | 3586 | if (path->slots[0] == i) |
87b29b20 | 3587 | push_space += data_size; |
db94535d CM |
3588 | |
3589 | this_item_size = btrfs_item_size(right, item); | |
3590 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3591 | break; |
db94535d | 3592 | |
be0e5c09 | 3593 | push_items++; |
db94535d CM |
3594 | push_space += this_item_size + sizeof(*item); |
3595 | } | |
3596 | ||
be0e5c09 | 3597 | if (push_items == 0) { |
925baedd CM |
3598 | ret = 1; |
3599 | goto out; | |
be0e5c09 | 3600 | } |
34a38218 | 3601 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3602 | WARN_ON(1); |
5f39d397 | 3603 | |
be0e5c09 | 3604 | /* push data from right to left */ |
5f39d397 CM |
3605 | copy_extent_buffer(left, right, |
3606 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3607 | btrfs_item_nr_offset(0), | |
3608 | push_items * sizeof(struct btrfs_item)); | |
3609 | ||
123abc88 | 3610 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3611 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3612 | |
3613 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3614 | leaf_data_end(root, left) - push_space, |
3615 | btrfs_leaf_data(right) + | |
5f39d397 | 3616 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3617 | push_space); |
5f39d397 | 3618 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3619 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3620 | |
db94535d | 3621 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3622 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3623 | u32 ioff; |
db94535d | 3624 | |
5f39d397 | 3625 | item = btrfs_item_nr(left, i); |
db94535d | 3626 | |
cfed81a0 CM |
3627 | ioff = btrfs_token_item_offset(left, item, &token); |
3628 | btrfs_set_token_item_offset(left, item, | |
3629 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3630 | &token); | |
be0e5c09 | 3631 | } |
5f39d397 | 3632 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3633 | |
3634 | /* fixup right node */ | |
31b1a2bd JL |
3635 | if (push_items > right_nritems) |
3636 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3637 | right_nritems); |
34a38218 CM |
3638 | |
3639 | if (push_items < right_nritems) { | |
3640 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3641 | leaf_data_end(root, right); | |
3642 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3643 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3644 | btrfs_leaf_data(right) + | |
3645 | leaf_data_end(root, right), push_space); | |
3646 | ||
3647 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3648 | btrfs_item_nr_offset(push_items), |
3649 | (btrfs_header_nritems(right) - push_items) * | |
3650 | sizeof(struct btrfs_item)); | |
34a38218 | 3651 | } |
eef1c494 Y |
3652 | right_nritems -= push_items; |
3653 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3654 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3655 | for (i = 0; i < right_nritems; i++) { |
3656 | item = btrfs_item_nr(right, i); | |
db94535d | 3657 | |
cfed81a0 CM |
3658 | push_space = push_space - btrfs_token_item_size(right, |
3659 | item, &token); | |
3660 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3661 | } |
eb60ceac | 3662 | |
5f39d397 | 3663 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3664 | if (right_nritems) |
3665 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3666 | else |
3667 | clean_tree_block(trans, root, right); | |
098f59c2 | 3668 | |
5f39d397 | 3669 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3670 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3671 | |
3672 | /* then fixup the leaf pointer in the path */ | |
3673 | if (path->slots[0] < push_items) { | |
3674 | path->slots[0] += old_left_nritems; | |
925baedd | 3675 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3676 | free_extent_buffer(path->nodes[0]); |
3677 | path->nodes[0] = left; | |
be0e5c09 CM |
3678 | path->slots[1] -= 1; |
3679 | } else { | |
925baedd | 3680 | btrfs_tree_unlock(left); |
5f39d397 | 3681 | free_extent_buffer(left); |
be0e5c09 CM |
3682 | path->slots[0] -= push_items; |
3683 | } | |
eb60ceac | 3684 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3685 | return ret; |
925baedd CM |
3686 | out: |
3687 | btrfs_tree_unlock(left); | |
3688 | free_extent_buffer(left); | |
3689 | return ret; | |
be0e5c09 CM |
3690 | } |
3691 | ||
44871b1b CM |
3692 | /* |
3693 | * push some data in the path leaf to the left, trying to free up at | |
3694 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3695 | * |
3696 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3697 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3698 | * items | |
44871b1b CM |
3699 | */ |
3700 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3701 | *root, struct btrfs_path *path, int min_data_size, |
3702 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3703 | { |
3704 | struct extent_buffer *right = path->nodes[0]; | |
3705 | struct extent_buffer *left; | |
3706 | int slot; | |
3707 | int free_space; | |
3708 | u32 right_nritems; | |
3709 | int ret = 0; | |
3710 | ||
3711 | slot = path->slots[1]; | |
3712 | if (slot == 0) | |
3713 | return 1; | |
3714 | if (!path->nodes[1]) | |
3715 | return 1; | |
3716 | ||
3717 | right_nritems = btrfs_header_nritems(right); | |
3718 | if (right_nritems == 0) | |
3719 | return 1; | |
3720 | ||
3721 | btrfs_assert_tree_locked(path->nodes[1]); | |
3722 | ||
3723 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3724 | if (left == NULL) |
3725 | return 1; | |
3726 | ||
44871b1b CM |
3727 | btrfs_tree_lock(left); |
3728 | btrfs_set_lock_blocking(left); | |
3729 | ||
3730 | free_space = btrfs_leaf_free_space(root, left); | |
3731 | if (free_space < data_size) { | |
3732 | ret = 1; | |
3733 | goto out; | |
3734 | } | |
3735 | ||
3736 | /* cow and double check */ | |
3737 | ret = btrfs_cow_block(trans, root, left, | |
3738 | path->nodes[1], slot - 1, &left); | |
3739 | if (ret) { | |
3740 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3741 | if (ret == -ENOSPC) |
3742 | ret = 1; | |
44871b1b CM |
3743 | goto out; |
3744 | } | |
3745 | ||
3746 | free_space = btrfs_leaf_free_space(root, left); | |
3747 | if (free_space < data_size) { | |
3748 | ret = 1; | |
3749 | goto out; | |
3750 | } | |
3751 | ||
99d8f83c CM |
3752 | return __push_leaf_left(trans, root, path, min_data_size, |
3753 | empty, left, free_space, right_nritems, | |
3754 | max_slot); | |
44871b1b CM |
3755 | out: |
3756 | btrfs_tree_unlock(left); | |
3757 | free_extent_buffer(left); | |
3758 | return ret; | |
3759 | } | |
3760 | ||
3761 | /* | |
3762 | * split the path's leaf in two, making sure there is at least data_size | |
3763 | * available for the resulting leaf level of the path. | |
44871b1b | 3764 | */ |
143bede5 JM |
3765 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3766 | struct btrfs_root *root, | |
3767 | struct btrfs_path *path, | |
3768 | struct extent_buffer *l, | |
3769 | struct extent_buffer *right, | |
3770 | int slot, int mid, int nritems) | |
44871b1b CM |
3771 | { |
3772 | int data_copy_size; | |
3773 | int rt_data_off; | |
3774 | int i; | |
44871b1b | 3775 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3776 | struct btrfs_map_token token; |
3777 | ||
3778 | btrfs_init_map_token(&token); | |
44871b1b CM |
3779 | |
3780 | nritems = nritems - mid; | |
3781 | btrfs_set_header_nritems(right, nritems); | |
3782 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3783 | ||
3784 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3785 | btrfs_item_nr_offset(mid), | |
3786 | nritems * sizeof(struct btrfs_item)); | |
3787 | ||
3788 | copy_extent_buffer(right, l, | |
3789 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3790 | data_copy_size, btrfs_leaf_data(l) + | |
3791 | leaf_data_end(root, l), data_copy_size); | |
3792 | ||
3793 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3794 | btrfs_item_end_nr(l, mid); | |
3795 | ||
3796 | for (i = 0; i < nritems; i++) { | |
3797 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3798 | u32 ioff; | |
3799 | ||
cfed81a0 CM |
3800 | ioff = btrfs_token_item_offset(right, item, &token); |
3801 | btrfs_set_token_item_offset(right, item, | |
3802 | ioff + rt_data_off, &token); | |
44871b1b CM |
3803 | } |
3804 | ||
44871b1b | 3805 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3806 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3807 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 3808 | path->slots[1] + 1, 1); |
44871b1b CM |
3809 | |
3810 | btrfs_mark_buffer_dirty(right); | |
3811 | btrfs_mark_buffer_dirty(l); | |
3812 | BUG_ON(path->slots[0] != slot); | |
3813 | ||
44871b1b CM |
3814 | if (mid <= slot) { |
3815 | btrfs_tree_unlock(path->nodes[0]); | |
3816 | free_extent_buffer(path->nodes[0]); | |
3817 | path->nodes[0] = right; | |
3818 | path->slots[0] -= mid; | |
3819 | path->slots[1] += 1; | |
3820 | } else { | |
3821 | btrfs_tree_unlock(right); | |
3822 | free_extent_buffer(right); | |
3823 | } | |
3824 | ||
3825 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3826 | } |
3827 | ||
99d8f83c CM |
3828 | /* |
3829 | * double splits happen when we need to insert a big item in the middle | |
3830 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3831 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3832 | * A B C | |
3833 | * | |
3834 | * We avoid this by trying to push the items on either side of our target | |
3835 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3836 | * completely. | |
3837 | */ | |
3838 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3839 | struct btrfs_root *root, | |
3840 | struct btrfs_path *path, | |
3841 | int data_size) | |
3842 | { | |
3843 | int ret; | |
3844 | int progress = 0; | |
3845 | int slot; | |
3846 | u32 nritems; | |
3847 | ||
3848 | slot = path->slots[0]; | |
3849 | ||
3850 | /* | |
3851 | * try to push all the items after our slot into the | |
3852 | * right leaf | |
3853 | */ | |
3854 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3855 | if (ret < 0) | |
3856 | return ret; | |
3857 | ||
3858 | if (ret == 0) | |
3859 | progress++; | |
3860 | ||
3861 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3862 | /* | |
3863 | * our goal is to get our slot at the start or end of a leaf. If | |
3864 | * we've done so we're done | |
3865 | */ | |
3866 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3867 | return 0; | |
3868 | ||
3869 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3870 | return 0; | |
3871 | ||
3872 | /* try to push all the items before our slot into the next leaf */ | |
3873 | slot = path->slots[0]; | |
3874 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3875 | if (ret < 0) | |
3876 | return ret; | |
3877 | ||
3878 | if (ret == 0) | |
3879 | progress++; | |
3880 | ||
3881 | if (progress) | |
3882 | return 0; | |
3883 | return 1; | |
3884 | } | |
3885 | ||
74123bd7 CM |
3886 | /* |
3887 | * split the path's leaf in two, making sure there is at least data_size | |
3888 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3889 | * |
3890 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3891 | */ |
e02119d5 CM |
3892 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3893 | struct btrfs_root *root, | |
3894 | struct btrfs_key *ins_key, | |
3895 | struct btrfs_path *path, int data_size, | |
3896 | int extend) | |
be0e5c09 | 3897 | { |
5d4f98a2 | 3898 | struct btrfs_disk_key disk_key; |
5f39d397 | 3899 | struct extent_buffer *l; |
7518a238 | 3900 | u32 nritems; |
eb60ceac CM |
3901 | int mid; |
3902 | int slot; | |
5f39d397 | 3903 | struct extent_buffer *right; |
d4dbff95 | 3904 | int ret = 0; |
aa5d6bed | 3905 | int wret; |
5d4f98a2 | 3906 | int split; |
cc0c5538 | 3907 | int num_doubles = 0; |
99d8f83c | 3908 | int tried_avoid_double = 0; |
aa5d6bed | 3909 | |
a5719521 YZ |
3910 | l = path->nodes[0]; |
3911 | slot = path->slots[0]; | |
3912 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3913 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3914 | return -EOVERFLOW; | |
3915 | ||
40689478 | 3916 | /* first try to make some room by pushing left and right */ |
33157e05 | 3917 | if (data_size && path->nodes[1]) { |
99d8f83c CM |
3918 | wret = push_leaf_right(trans, root, path, data_size, |
3919 | data_size, 0, 0); | |
d397712b | 3920 | if (wret < 0) |
eaee50e8 | 3921 | return wret; |
3685f791 | 3922 | if (wret) { |
99d8f83c CM |
3923 | wret = push_leaf_left(trans, root, path, data_size, |
3924 | data_size, 0, (u32)-1); | |
3685f791 CM |
3925 | if (wret < 0) |
3926 | return wret; | |
3927 | } | |
3928 | l = path->nodes[0]; | |
aa5d6bed | 3929 | |
3685f791 | 3930 | /* did the pushes work? */ |
87b29b20 | 3931 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3932 | return 0; |
3326d1b0 | 3933 | } |
aa5d6bed | 3934 | |
5c680ed6 | 3935 | if (!path->nodes[1]) { |
fdd99c72 | 3936 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3937 | if (ret) |
3938 | return ret; | |
3939 | } | |
cc0c5538 | 3940 | again: |
5d4f98a2 | 3941 | split = 1; |
cc0c5538 | 3942 | l = path->nodes[0]; |
eb60ceac | 3943 | slot = path->slots[0]; |
5f39d397 | 3944 | nritems = btrfs_header_nritems(l); |
d397712b | 3945 | mid = (nritems + 1) / 2; |
54aa1f4d | 3946 | |
5d4f98a2 YZ |
3947 | if (mid <= slot) { |
3948 | if (nritems == 1 || | |
3949 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
3950 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3951 | if (slot >= nritems) { | |
3952 | split = 0; | |
3953 | } else { | |
3954 | mid = slot; | |
3955 | if (mid != nritems && | |
3956 | leaf_space_used(l, mid, nritems - mid) + | |
3957 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3958 | if (data_size && !tried_avoid_double) |
3959 | goto push_for_double; | |
5d4f98a2 YZ |
3960 | split = 2; |
3961 | } | |
3962 | } | |
3963 | } | |
3964 | } else { | |
3965 | if (leaf_space_used(l, 0, mid) + data_size > | |
3966 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3967 | if (!extend && data_size && slot == 0) { | |
3968 | split = 0; | |
3969 | } else if ((extend || !data_size) && slot == 0) { | |
3970 | mid = 1; | |
3971 | } else { | |
3972 | mid = slot; | |
3973 | if (mid != nritems && | |
3974 | leaf_space_used(l, mid, nritems - mid) + | |
3975 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3976 | if (data_size && !tried_avoid_double) |
3977 | goto push_for_double; | |
5d4f98a2 YZ |
3978 | split = 2 ; |
3979 | } | |
3980 | } | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | if (split == 0) | |
3985 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3986 | else | |
3987 | btrfs_item_key(l, &disk_key, mid); | |
3988 | ||
3989 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 3990 | root->root_key.objectid, |
5581a51a | 3991 | &disk_key, 0, l->start, 0); |
f0486c68 | 3992 | if (IS_ERR(right)) |
5f39d397 | 3993 | return PTR_ERR(right); |
f0486c68 YZ |
3994 | |
3995 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
3996 | |
3997 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 3998 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 3999 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4000 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4001 | btrfs_set_header_owner(right, root->root_key.objectid); |
4002 | btrfs_set_header_level(right, 0); | |
4003 | write_extent_buffer(right, root->fs_info->fsid, | |
fba6aa75 | 4004 | btrfs_header_fsid(right), BTRFS_FSID_SIZE); |
e17cade2 CM |
4005 | |
4006 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 4007 | btrfs_header_chunk_tree_uuid(right), |
e17cade2 | 4008 | BTRFS_UUID_SIZE); |
44871b1b | 4009 | |
5d4f98a2 YZ |
4010 | if (split == 0) { |
4011 | if (mid <= slot) { | |
4012 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4013 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4014 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4015 | btrfs_tree_unlock(path->nodes[0]); |
4016 | free_extent_buffer(path->nodes[0]); | |
4017 | path->nodes[0] = right; | |
4018 | path->slots[0] = 0; | |
4019 | path->slots[1] += 1; | |
4020 | } else { | |
4021 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4022 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4023 | path->slots[1], 1); |
5d4f98a2 YZ |
4024 | btrfs_tree_unlock(path->nodes[0]); |
4025 | free_extent_buffer(path->nodes[0]); | |
4026 | path->nodes[0] = right; | |
4027 | path->slots[0] = 0; | |
143bede5 | 4028 | if (path->slots[1] == 0) |
d6a0a126 | 4029 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4030 | } |
5d4f98a2 YZ |
4031 | btrfs_mark_buffer_dirty(right); |
4032 | return ret; | |
d4dbff95 | 4033 | } |
74123bd7 | 4034 | |
143bede5 | 4035 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4036 | |
5d4f98a2 | 4037 | if (split == 2) { |
cc0c5538 CM |
4038 | BUG_ON(num_doubles != 0); |
4039 | num_doubles++; | |
4040 | goto again; | |
a429e513 | 4041 | } |
44871b1b | 4042 | |
143bede5 | 4043 | return 0; |
99d8f83c CM |
4044 | |
4045 | push_for_double: | |
4046 | push_for_double_split(trans, root, path, data_size); | |
4047 | tried_avoid_double = 1; | |
4048 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4049 | return 0; | |
4050 | goto again; | |
be0e5c09 CM |
4051 | } |
4052 | ||
ad48fd75 YZ |
4053 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4054 | struct btrfs_root *root, | |
4055 | struct btrfs_path *path, int ins_len) | |
459931ec | 4056 | { |
ad48fd75 | 4057 | struct btrfs_key key; |
459931ec | 4058 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4059 | struct btrfs_file_extent_item *fi; |
4060 | u64 extent_len = 0; | |
4061 | u32 item_size; | |
4062 | int ret; | |
459931ec CM |
4063 | |
4064 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4065 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4066 | ||
4067 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4068 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4069 | ||
4070 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4071 | return 0; | |
459931ec CM |
4072 | |
4073 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4074 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4075 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4076 | struct btrfs_file_extent_item); | |
4077 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4078 | } | |
b3b4aa74 | 4079 | btrfs_release_path(path); |
459931ec | 4080 | |
459931ec | 4081 | path->keep_locks = 1; |
ad48fd75 YZ |
4082 | path->search_for_split = 1; |
4083 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4084 | path->search_for_split = 0; |
ad48fd75 YZ |
4085 | if (ret < 0) |
4086 | goto err; | |
459931ec | 4087 | |
ad48fd75 YZ |
4088 | ret = -EAGAIN; |
4089 | leaf = path->nodes[0]; | |
459931ec | 4090 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4091 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4092 | goto err; | |
4093 | ||
109f6aef CM |
4094 | /* the leaf has changed, it now has room. return now */ |
4095 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4096 | goto err; | |
4097 | ||
ad48fd75 YZ |
4098 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4099 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4100 | struct btrfs_file_extent_item); | |
4101 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4102 | goto err; | |
459931ec CM |
4103 | } |
4104 | ||
b9473439 | 4105 | btrfs_set_path_blocking(path); |
ad48fd75 | 4106 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4107 | if (ret) |
4108 | goto err; | |
459931ec | 4109 | |
ad48fd75 | 4110 | path->keep_locks = 0; |
b9473439 | 4111 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4112 | return 0; |
4113 | err: | |
4114 | path->keep_locks = 0; | |
4115 | return ret; | |
4116 | } | |
4117 | ||
4118 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4119 | struct btrfs_root *root, | |
4120 | struct btrfs_path *path, | |
4121 | struct btrfs_key *new_key, | |
4122 | unsigned long split_offset) | |
4123 | { | |
4124 | struct extent_buffer *leaf; | |
4125 | struct btrfs_item *item; | |
4126 | struct btrfs_item *new_item; | |
4127 | int slot; | |
4128 | char *buf; | |
4129 | u32 nritems; | |
4130 | u32 item_size; | |
4131 | u32 orig_offset; | |
4132 | struct btrfs_disk_key disk_key; | |
4133 | ||
b9473439 CM |
4134 | leaf = path->nodes[0]; |
4135 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4136 | ||
b4ce94de CM |
4137 | btrfs_set_path_blocking(path); |
4138 | ||
459931ec CM |
4139 | item = btrfs_item_nr(leaf, path->slots[0]); |
4140 | orig_offset = btrfs_item_offset(leaf, item); | |
4141 | item_size = btrfs_item_size(leaf, item); | |
4142 | ||
459931ec | 4143 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4144 | if (!buf) |
4145 | return -ENOMEM; | |
4146 | ||
459931ec CM |
4147 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4148 | path->slots[0]), item_size); | |
459931ec | 4149 | |
ad48fd75 | 4150 | slot = path->slots[0] + 1; |
459931ec | 4151 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4152 | if (slot != nritems) { |
4153 | /* shift the items */ | |
4154 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4155 | btrfs_item_nr_offset(slot), |
4156 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4157 | } |
4158 | ||
4159 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4160 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4161 | ||
4162 | new_item = btrfs_item_nr(leaf, slot); | |
4163 | ||
4164 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4165 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4166 | ||
4167 | btrfs_set_item_offset(leaf, item, | |
4168 | orig_offset + item_size - split_offset); | |
4169 | btrfs_set_item_size(leaf, item, split_offset); | |
4170 | ||
4171 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4172 | ||
4173 | /* write the data for the start of the original item */ | |
4174 | write_extent_buffer(leaf, buf, | |
4175 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4176 | split_offset); | |
4177 | ||
4178 | /* write the data for the new item */ | |
4179 | write_extent_buffer(leaf, buf + split_offset, | |
4180 | btrfs_item_ptr_offset(leaf, slot), | |
4181 | item_size - split_offset); | |
4182 | btrfs_mark_buffer_dirty(leaf); | |
4183 | ||
ad48fd75 | 4184 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4185 | kfree(buf); |
ad48fd75 YZ |
4186 | return 0; |
4187 | } | |
4188 | ||
4189 | /* | |
4190 | * This function splits a single item into two items, | |
4191 | * giving 'new_key' to the new item and splitting the | |
4192 | * old one at split_offset (from the start of the item). | |
4193 | * | |
4194 | * The path may be released by this operation. After | |
4195 | * the split, the path is pointing to the old item. The | |
4196 | * new item is going to be in the same node as the old one. | |
4197 | * | |
4198 | * Note, the item being split must be smaller enough to live alone on | |
4199 | * a tree block with room for one extra struct btrfs_item | |
4200 | * | |
4201 | * This allows us to split the item in place, keeping a lock on the | |
4202 | * leaf the entire time. | |
4203 | */ | |
4204 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4205 | struct btrfs_root *root, | |
4206 | struct btrfs_path *path, | |
4207 | struct btrfs_key *new_key, | |
4208 | unsigned long split_offset) | |
4209 | { | |
4210 | int ret; | |
4211 | ret = setup_leaf_for_split(trans, root, path, | |
4212 | sizeof(struct btrfs_item)); | |
4213 | if (ret) | |
4214 | return ret; | |
4215 | ||
4216 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4217 | return ret; |
4218 | } | |
4219 | ||
ad48fd75 YZ |
4220 | /* |
4221 | * This function duplicate a item, giving 'new_key' to the new item. | |
4222 | * It guarantees both items live in the same tree leaf and the new item | |
4223 | * is contiguous with the original item. | |
4224 | * | |
4225 | * This allows us to split file extent in place, keeping a lock on the | |
4226 | * leaf the entire time. | |
4227 | */ | |
4228 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4229 | struct btrfs_root *root, | |
4230 | struct btrfs_path *path, | |
4231 | struct btrfs_key *new_key) | |
4232 | { | |
4233 | struct extent_buffer *leaf; | |
4234 | int ret; | |
4235 | u32 item_size; | |
4236 | ||
4237 | leaf = path->nodes[0]; | |
4238 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4239 | ret = setup_leaf_for_split(trans, root, path, | |
4240 | item_size + sizeof(struct btrfs_item)); | |
4241 | if (ret) | |
4242 | return ret; | |
4243 | ||
4244 | path->slots[0]++; | |
afe5fea7 | 4245 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4246 | item_size, item_size + |
4247 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4248 | leaf = path->nodes[0]; |
4249 | memcpy_extent_buffer(leaf, | |
4250 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4251 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4252 | item_size); | |
4253 | return 0; | |
4254 | } | |
4255 | ||
d352ac68 CM |
4256 | /* |
4257 | * make the item pointed to by the path smaller. new_size indicates | |
4258 | * how small to make it, and from_end tells us if we just chop bytes | |
4259 | * off the end of the item or if we shift the item to chop bytes off | |
4260 | * the front. | |
4261 | */ | |
afe5fea7 | 4262 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4263 | u32 new_size, int from_end) |
b18c6685 | 4264 | { |
b18c6685 | 4265 | int slot; |
5f39d397 CM |
4266 | struct extent_buffer *leaf; |
4267 | struct btrfs_item *item; | |
b18c6685 CM |
4268 | u32 nritems; |
4269 | unsigned int data_end; | |
4270 | unsigned int old_data_start; | |
4271 | unsigned int old_size; | |
4272 | unsigned int size_diff; | |
4273 | int i; | |
cfed81a0 CM |
4274 | struct btrfs_map_token token; |
4275 | ||
4276 | btrfs_init_map_token(&token); | |
b18c6685 | 4277 | |
5f39d397 | 4278 | leaf = path->nodes[0]; |
179e29e4 CM |
4279 | slot = path->slots[0]; |
4280 | ||
4281 | old_size = btrfs_item_size_nr(leaf, slot); | |
4282 | if (old_size == new_size) | |
143bede5 | 4283 | return; |
b18c6685 | 4284 | |
5f39d397 | 4285 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4286 | data_end = leaf_data_end(root, leaf); |
4287 | ||
5f39d397 | 4288 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4289 | |
b18c6685 CM |
4290 | size_diff = old_size - new_size; |
4291 | ||
4292 | BUG_ON(slot < 0); | |
4293 | BUG_ON(slot >= nritems); | |
4294 | ||
4295 | /* | |
4296 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4297 | */ | |
4298 | /* first correct the data pointers */ | |
4299 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4300 | u32 ioff; |
4301 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4302 | |
cfed81a0 CM |
4303 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4304 | btrfs_set_token_item_offset(leaf, item, | |
4305 | ioff + size_diff, &token); | |
b18c6685 | 4306 | } |
db94535d | 4307 | |
b18c6685 | 4308 | /* shift the data */ |
179e29e4 CM |
4309 | if (from_end) { |
4310 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4311 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4312 | data_end, old_data_start + new_size - data_end); | |
4313 | } else { | |
4314 | struct btrfs_disk_key disk_key; | |
4315 | u64 offset; | |
4316 | ||
4317 | btrfs_item_key(leaf, &disk_key, slot); | |
4318 | ||
4319 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4320 | unsigned long ptr; | |
4321 | struct btrfs_file_extent_item *fi; | |
4322 | ||
4323 | fi = btrfs_item_ptr(leaf, slot, | |
4324 | struct btrfs_file_extent_item); | |
4325 | fi = (struct btrfs_file_extent_item *)( | |
4326 | (unsigned long)fi - size_diff); | |
4327 | ||
4328 | if (btrfs_file_extent_type(leaf, fi) == | |
4329 | BTRFS_FILE_EXTENT_INLINE) { | |
4330 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4331 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4332 | (unsigned long)fi, |
4333 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4334 | disk_bytenr)); |
4335 | } | |
4336 | } | |
4337 | ||
4338 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4339 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4340 | data_end, old_data_start - data_end); | |
4341 | ||
4342 | offset = btrfs_disk_key_offset(&disk_key); | |
4343 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4344 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4345 | if (slot == 0) | |
d6a0a126 | 4346 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4347 | } |
5f39d397 CM |
4348 | |
4349 | item = btrfs_item_nr(leaf, slot); | |
4350 | btrfs_set_item_size(leaf, item, new_size); | |
4351 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4352 | |
5f39d397 CM |
4353 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4354 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4355 | BUG(); |
5f39d397 | 4356 | } |
b18c6685 CM |
4357 | } |
4358 | ||
d352ac68 | 4359 | /* |
8f69dbd2 | 4360 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4361 | */ |
4b90c680 | 4362 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4363 | u32 data_size) |
6567e837 | 4364 | { |
6567e837 | 4365 | int slot; |
5f39d397 CM |
4366 | struct extent_buffer *leaf; |
4367 | struct btrfs_item *item; | |
6567e837 CM |
4368 | u32 nritems; |
4369 | unsigned int data_end; | |
4370 | unsigned int old_data; | |
4371 | unsigned int old_size; | |
4372 | int i; | |
cfed81a0 CM |
4373 | struct btrfs_map_token token; |
4374 | ||
4375 | btrfs_init_map_token(&token); | |
6567e837 | 4376 | |
5f39d397 | 4377 | leaf = path->nodes[0]; |
6567e837 | 4378 | |
5f39d397 | 4379 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4380 | data_end = leaf_data_end(root, leaf); |
4381 | ||
5f39d397 CM |
4382 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4383 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4384 | BUG(); |
5f39d397 | 4385 | } |
6567e837 | 4386 | slot = path->slots[0]; |
5f39d397 | 4387 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4388 | |
4389 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4390 | if (slot >= nritems) { |
4391 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
4392 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
4393 | slot, nritems); | |
3326d1b0 CM |
4394 | BUG_ON(1); |
4395 | } | |
6567e837 CM |
4396 | |
4397 | /* | |
4398 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4399 | */ | |
4400 | /* first correct the data pointers */ | |
4401 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4402 | u32 ioff; |
4403 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4404 | |
cfed81a0 CM |
4405 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4406 | btrfs_set_token_item_offset(leaf, item, | |
4407 | ioff - data_size, &token); | |
6567e837 | 4408 | } |
5f39d397 | 4409 | |
6567e837 | 4410 | /* shift the data */ |
5f39d397 | 4411 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4412 | data_end - data_size, btrfs_leaf_data(leaf) + |
4413 | data_end, old_data - data_end); | |
5f39d397 | 4414 | |
6567e837 | 4415 | data_end = old_data; |
5f39d397 CM |
4416 | old_size = btrfs_item_size_nr(leaf, slot); |
4417 | item = btrfs_item_nr(leaf, slot); | |
4418 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
4419 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4420 | |
5f39d397 CM |
4421 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4422 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4423 | BUG(); |
5f39d397 | 4424 | } |
6567e837 CM |
4425 | } |
4426 | ||
74123bd7 | 4427 | /* |
44871b1b CM |
4428 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4429 | * to save stack depth by doing the bulk of the work in a function | |
4430 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4431 | */ |
afe5fea7 | 4432 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4433 | struct btrfs_key *cpu_key, u32 *data_size, |
4434 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4435 | { |
5f39d397 | 4436 | struct btrfs_item *item; |
9c58309d | 4437 | int i; |
7518a238 | 4438 | u32 nritems; |
be0e5c09 | 4439 | unsigned int data_end; |
e2fa7227 | 4440 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4441 | struct extent_buffer *leaf; |
4442 | int slot; | |
cfed81a0 CM |
4443 | struct btrfs_map_token token; |
4444 | ||
4445 | btrfs_init_map_token(&token); | |
e2fa7227 | 4446 | |
5f39d397 | 4447 | leaf = path->nodes[0]; |
44871b1b | 4448 | slot = path->slots[0]; |
74123bd7 | 4449 | |
5f39d397 | 4450 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4451 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4452 | |
f25956cc | 4453 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4454 | btrfs_print_leaf(root, leaf); |
d397712b | 4455 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4456 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4457 | BUG(); |
d4dbff95 | 4458 | } |
5f39d397 | 4459 | |
be0e5c09 | 4460 | if (slot != nritems) { |
5f39d397 | 4461 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4462 | |
5f39d397 CM |
4463 | if (old_data < data_end) { |
4464 | btrfs_print_leaf(root, leaf); | |
d397712b | 4465 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4466 | slot, old_data, data_end); |
4467 | BUG_ON(1); | |
4468 | } | |
be0e5c09 CM |
4469 | /* |
4470 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4471 | */ | |
4472 | /* first correct the data pointers */ | |
0783fcfc | 4473 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4474 | u32 ioff; |
db94535d | 4475 | |
5f39d397 | 4476 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4477 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4478 | btrfs_set_token_item_offset(leaf, item, | |
4479 | ioff - total_data, &token); | |
0783fcfc | 4480 | } |
be0e5c09 | 4481 | /* shift the items */ |
9c58309d | 4482 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4483 | btrfs_item_nr_offset(slot), |
d6025579 | 4484 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4485 | |
4486 | /* shift the data */ | |
5f39d397 | 4487 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4488 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4489 | data_end, old_data - data_end); |
be0e5c09 CM |
4490 | data_end = old_data; |
4491 | } | |
5f39d397 | 4492 | |
62e2749e | 4493 | /* setup the item for the new data */ |
9c58309d CM |
4494 | for (i = 0; i < nr; i++) { |
4495 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4496 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4497 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4498 | btrfs_set_token_item_offset(leaf, item, |
4499 | data_end - data_size[i], &token); | |
9c58309d | 4500 | data_end -= data_size[i]; |
cfed81a0 | 4501 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4502 | } |
44871b1b | 4503 | |
9c58309d | 4504 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4505 | |
5a01a2e3 CM |
4506 | if (slot == 0) { |
4507 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4508 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4509 | } |
b9473439 CM |
4510 | btrfs_unlock_up_safe(path, 1); |
4511 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4512 | |
5f39d397 CM |
4513 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4514 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4515 | BUG(); |
5f39d397 | 4516 | } |
44871b1b CM |
4517 | } |
4518 | ||
4519 | /* | |
4520 | * Given a key and some data, insert items into the tree. | |
4521 | * This does all the path init required, making room in the tree if needed. | |
4522 | */ | |
4523 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4524 | struct btrfs_root *root, | |
4525 | struct btrfs_path *path, | |
4526 | struct btrfs_key *cpu_key, u32 *data_size, | |
4527 | int nr) | |
4528 | { | |
44871b1b CM |
4529 | int ret = 0; |
4530 | int slot; | |
4531 | int i; | |
4532 | u32 total_size = 0; | |
4533 | u32 total_data = 0; | |
4534 | ||
4535 | for (i = 0; i < nr; i++) | |
4536 | total_data += data_size[i]; | |
4537 | ||
4538 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4539 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4540 | if (ret == 0) | |
4541 | return -EEXIST; | |
4542 | if (ret < 0) | |
143bede5 | 4543 | return ret; |
44871b1b | 4544 | |
44871b1b CM |
4545 | slot = path->slots[0]; |
4546 | BUG_ON(slot < 0); | |
4547 | ||
afe5fea7 | 4548 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4549 | total_data, total_size, nr); |
143bede5 | 4550 | return 0; |
62e2749e CM |
4551 | } |
4552 | ||
4553 | /* | |
4554 | * Given a key and some data, insert an item into the tree. | |
4555 | * This does all the path init required, making room in the tree if needed. | |
4556 | */ | |
e089f05c CM |
4557 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4558 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4559 | data_size) | |
62e2749e CM |
4560 | { |
4561 | int ret = 0; | |
2c90e5d6 | 4562 | struct btrfs_path *path; |
5f39d397 CM |
4563 | struct extent_buffer *leaf; |
4564 | unsigned long ptr; | |
62e2749e | 4565 | |
2c90e5d6 | 4566 | path = btrfs_alloc_path(); |
db5b493a TI |
4567 | if (!path) |
4568 | return -ENOMEM; | |
2c90e5d6 | 4569 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4570 | if (!ret) { |
5f39d397 CM |
4571 | leaf = path->nodes[0]; |
4572 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4573 | write_extent_buffer(leaf, data, ptr, data_size); | |
4574 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4575 | } |
2c90e5d6 | 4576 | btrfs_free_path(path); |
aa5d6bed | 4577 | return ret; |
be0e5c09 CM |
4578 | } |
4579 | ||
74123bd7 | 4580 | /* |
5de08d7d | 4581 | * delete the pointer from a given node. |
74123bd7 | 4582 | * |
d352ac68 CM |
4583 | * the tree should have been previously balanced so the deletion does not |
4584 | * empty a node. | |
74123bd7 | 4585 | */ |
afe5fea7 TI |
4586 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4587 | int level, int slot) | |
be0e5c09 | 4588 | { |
5f39d397 | 4589 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4590 | u32 nritems; |
f3ea38da | 4591 | int ret; |
be0e5c09 | 4592 | |
5f39d397 | 4593 | nritems = btrfs_header_nritems(parent); |
d397712b | 4594 | if (slot != nritems - 1) { |
0e411ece | 4595 | if (level) |
f3ea38da JS |
4596 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4597 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4598 | memmove_extent_buffer(parent, |
4599 | btrfs_node_key_ptr_offset(slot), | |
4600 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4601 | sizeof(struct btrfs_key_ptr) * |
4602 | (nritems - slot - 1)); | |
57ba86c0 CM |
4603 | } else if (level) { |
4604 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4605 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4606 | BUG_ON(ret < 0); |
bb803951 | 4607 | } |
f3ea38da | 4608 | |
7518a238 | 4609 | nritems--; |
5f39d397 | 4610 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4611 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4612 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4613 | /* just turn the root into a leaf and break */ |
5f39d397 | 4614 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4615 | } else if (slot == 0) { |
5f39d397 CM |
4616 | struct btrfs_disk_key disk_key; |
4617 | ||
4618 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4619 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4620 | } |
d6025579 | 4621 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4622 | } |
4623 | ||
323ac95b CM |
4624 | /* |
4625 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4626 | * path->nodes[1]. |
323ac95b CM |
4627 | * |
4628 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4629 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4630 | * | |
4631 | * The path must have already been setup for deleting the leaf, including | |
4632 | * all the proper balancing. path->nodes[1] must be locked. | |
4633 | */ | |
143bede5 JM |
4634 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4635 | struct btrfs_root *root, | |
4636 | struct btrfs_path *path, | |
4637 | struct extent_buffer *leaf) | |
323ac95b | 4638 | { |
5d4f98a2 | 4639 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4640 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4641 | |
4d081c41 CM |
4642 | /* |
4643 | * btrfs_free_extent is expensive, we want to make sure we | |
4644 | * aren't holding any locks when we call it | |
4645 | */ | |
4646 | btrfs_unlock_up_safe(path, 0); | |
4647 | ||
f0486c68 YZ |
4648 | root_sub_used(root, leaf->len); |
4649 | ||
3083ee2e | 4650 | extent_buffer_get(leaf); |
5581a51a | 4651 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4652 | free_extent_buffer_stale(leaf); |
323ac95b | 4653 | } |
74123bd7 CM |
4654 | /* |
4655 | * delete the item at the leaf level in path. If that empties | |
4656 | * the leaf, remove it from the tree | |
4657 | */ | |
85e21bac CM |
4658 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4659 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4660 | { |
5f39d397 CM |
4661 | struct extent_buffer *leaf; |
4662 | struct btrfs_item *item; | |
85e21bac CM |
4663 | int last_off; |
4664 | int dsize = 0; | |
aa5d6bed CM |
4665 | int ret = 0; |
4666 | int wret; | |
85e21bac | 4667 | int i; |
7518a238 | 4668 | u32 nritems; |
cfed81a0 CM |
4669 | struct btrfs_map_token token; |
4670 | ||
4671 | btrfs_init_map_token(&token); | |
be0e5c09 | 4672 | |
5f39d397 | 4673 | leaf = path->nodes[0]; |
85e21bac CM |
4674 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4675 | ||
4676 | for (i = 0; i < nr; i++) | |
4677 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4678 | ||
5f39d397 | 4679 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4680 | |
85e21bac | 4681 | if (slot + nr != nritems) { |
123abc88 | 4682 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4683 | |
4684 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4685 | data_end + dsize, |
4686 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4687 | last_off - data_end); |
5f39d397 | 4688 | |
85e21bac | 4689 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4690 | u32 ioff; |
db94535d | 4691 | |
5f39d397 | 4692 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4693 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4694 | btrfs_set_token_item_offset(leaf, item, | |
4695 | ioff + dsize, &token); | |
0783fcfc | 4696 | } |
db94535d | 4697 | |
5f39d397 | 4698 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4699 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4700 | sizeof(struct btrfs_item) * |
85e21bac | 4701 | (nritems - slot - nr)); |
be0e5c09 | 4702 | } |
85e21bac CM |
4703 | btrfs_set_header_nritems(leaf, nritems - nr); |
4704 | nritems -= nr; | |
5f39d397 | 4705 | |
74123bd7 | 4706 | /* delete the leaf if we've emptied it */ |
7518a238 | 4707 | if (nritems == 0) { |
5f39d397 CM |
4708 | if (leaf == root->node) { |
4709 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4710 | } else { |
f0486c68 YZ |
4711 | btrfs_set_path_blocking(path); |
4712 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4713 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4714 | } |
be0e5c09 | 4715 | } else { |
7518a238 | 4716 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4717 | if (slot == 0) { |
5f39d397 CM |
4718 | struct btrfs_disk_key disk_key; |
4719 | ||
4720 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 4721 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 4722 | } |
aa5d6bed | 4723 | |
74123bd7 | 4724 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4725 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4726 | /* push_leaf_left fixes the path. |
4727 | * make sure the path still points to our leaf | |
4728 | * for possible call to del_ptr below | |
4729 | */ | |
4920c9ac | 4730 | slot = path->slots[1]; |
5f39d397 CM |
4731 | extent_buffer_get(leaf); |
4732 | ||
b9473439 | 4733 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4734 | wret = push_leaf_left(trans, root, path, 1, 1, |
4735 | 1, (u32)-1); | |
54aa1f4d | 4736 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4737 | ret = wret; |
5f39d397 CM |
4738 | |
4739 | if (path->nodes[0] == leaf && | |
4740 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4741 | wret = push_leaf_right(trans, root, path, 1, |
4742 | 1, 1, 0); | |
54aa1f4d | 4743 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4744 | ret = wret; |
4745 | } | |
5f39d397 CM |
4746 | |
4747 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4748 | path->slots[1] = slot; |
143bede5 | 4749 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4750 | free_extent_buffer(leaf); |
143bede5 | 4751 | ret = 0; |
5de08d7d | 4752 | } else { |
925baedd CM |
4753 | /* if we're still in the path, make sure |
4754 | * we're dirty. Otherwise, one of the | |
4755 | * push_leaf functions must have already | |
4756 | * dirtied this buffer | |
4757 | */ | |
4758 | if (path->nodes[0] == leaf) | |
4759 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4760 | free_extent_buffer(leaf); |
be0e5c09 | 4761 | } |
d5719762 | 4762 | } else { |
5f39d397 | 4763 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4764 | } |
4765 | } | |
aa5d6bed | 4766 | return ret; |
be0e5c09 CM |
4767 | } |
4768 | ||
7bb86316 | 4769 | /* |
925baedd | 4770 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4771 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4772 | * returns < 0 on io errors. | |
d352ac68 CM |
4773 | * |
4774 | * This may release the path, and so you may lose any locks held at the | |
4775 | * time you call it. | |
7bb86316 | 4776 | */ |
35a3621b | 4777 | static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4778 | { |
925baedd CM |
4779 | struct btrfs_key key; |
4780 | struct btrfs_disk_key found_key; | |
4781 | int ret; | |
7bb86316 | 4782 | |
925baedd | 4783 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4784 | |
925baedd CM |
4785 | if (key.offset > 0) |
4786 | key.offset--; | |
4787 | else if (key.type > 0) | |
4788 | key.type--; | |
4789 | else if (key.objectid > 0) | |
4790 | key.objectid--; | |
4791 | else | |
4792 | return 1; | |
7bb86316 | 4793 | |
b3b4aa74 | 4794 | btrfs_release_path(path); |
925baedd CM |
4795 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4796 | if (ret < 0) | |
4797 | return ret; | |
4798 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4799 | ret = comp_keys(&found_key, &key); | |
4800 | if (ret < 0) | |
4801 | return 0; | |
4802 | return 1; | |
7bb86316 CM |
4803 | } |
4804 | ||
3f157a2f CM |
4805 | /* |
4806 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4807 | * for nodes or leaves that are have a minimum transaction id. |
4808 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4809 | * |
4810 | * This does not cow, but it does stuff the starting key it finds back | |
4811 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4812 | * key and get a writable path. | |
4813 | * | |
4814 | * This does lock as it descends, and path->keep_locks should be set | |
4815 | * to 1 by the caller. | |
4816 | * | |
4817 | * This honors path->lowest_level to prevent descent past a given level | |
4818 | * of the tree. | |
4819 | * | |
d352ac68 CM |
4820 | * min_trans indicates the oldest transaction that you are interested |
4821 | * in walking through. Any nodes or leaves older than min_trans are | |
4822 | * skipped over (without reading them). | |
4823 | * | |
3f157a2f CM |
4824 | * returns zero if something useful was found, < 0 on error and 1 if there |
4825 | * was nothing in the tree that matched the search criteria. | |
4826 | */ | |
4827 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4828 | struct btrfs_key *max_key, |
de78b51a | 4829 | struct btrfs_path *path, |
3f157a2f CM |
4830 | u64 min_trans) |
4831 | { | |
4832 | struct extent_buffer *cur; | |
4833 | struct btrfs_key found_key; | |
4834 | int slot; | |
9652480b | 4835 | int sret; |
3f157a2f CM |
4836 | u32 nritems; |
4837 | int level; | |
4838 | int ret = 1; | |
4839 | ||
934d375b | 4840 | WARN_ON(!path->keep_locks); |
3f157a2f | 4841 | again: |
bd681513 | 4842 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4843 | level = btrfs_header_level(cur); |
e02119d5 | 4844 | WARN_ON(path->nodes[level]); |
3f157a2f | 4845 | path->nodes[level] = cur; |
bd681513 | 4846 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4847 | |
4848 | if (btrfs_header_generation(cur) < min_trans) { | |
4849 | ret = 1; | |
4850 | goto out; | |
4851 | } | |
d397712b | 4852 | while (1) { |
3f157a2f CM |
4853 | nritems = btrfs_header_nritems(cur); |
4854 | level = btrfs_header_level(cur); | |
9652480b | 4855 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4856 | |
323ac95b CM |
4857 | /* at the lowest level, we're done, setup the path and exit */ |
4858 | if (level == path->lowest_level) { | |
e02119d5 CM |
4859 | if (slot >= nritems) |
4860 | goto find_next_key; | |
3f157a2f CM |
4861 | ret = 0; |
4862 | path->slots[level] = slot; | |
4863 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4864 | goto out; | |
4865 | } | |
9652480b Y |
4866 | if (sret && slot > 0) |
4867 | slot--; | |
3f157a2f | 4868 | /* |
de78b51a ES |
4869 | * check this node pointer against the min_trans parameters. |
4870 | * If it is too old, old, skip to the next one. | |
3f157a2f | 4871 | */ |
d397712b | 4872 | while (slot < nritems) { |
3f157a2f CM |
4873 | u64 blockptr; |
4874 | u64 gen; | |
e02119d5 | 4875 | |
3f157a2f CM |
4876 | blockptr = btrfs_node_blockptr(cur, slot); |
4877 | gen = btrfs_node_ptr_generation(cur, slot); | |
4878 | if (gen < min_trans) { | |
4879 | slot++; | |
4880 | continue; | |
4881 | } | |
de78b51a | 4882 | break; |
3f157a2f | 4883 | } |
e02119d5 | 4884 | find_next_key: |
3f157a2f CM |
4885 | /* |
4886 | * we didn't find a candidate key in this node, walk forward | |
4887 | * and find another one | |
4888 | */ | |
4889 | if (slot >= nritems) { | |
e02119d5 | 4890 | path->slots[level] = slot; |
b4ce94de | 4891 | btrfs_set_path_blocking(path); |
e02119d5 | 4892 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 4893 | min_trans); |
e02119d5 | 4894 | if (sret == 0) { |
b3b4aa74 | 4895 | btrfs_release_path(path); |
3f157a2f CM |
4896 | goto again; |
4897 | } else { | |
4898 | goto out; | |
4899 | } | |
4900 | } | |
4901 | /* save our key for returning back */ | |
4902 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4903 | path->slots[level] = slot; | |
4904 | if (level == path->lowest_level) { | |
4905 | ret = 0; | |
f7c79f30 | 4906 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4907 | goto out; |
4908 | } | |
b4ce94de | 4909 | btrfs_set_path_blocking(path); |
3f157a2f | 4910 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4911 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4912 | |
bd681513 | 4913 | btrfs_tree_read_lock(cur); |
b4ce94de | 4914 | |
bd681513 | 4915 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4916 | path->nodes[level - 1] = cur; |
f7c79f30 | 4917 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4918 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4919 | } |
4920 | out: | |
4921 | if (ret == 0) | |
4922 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4923 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4924 | return ret; |
4925 | } | |
4926 | ||
7069830a AB |
4927 | static void tree_move_down(struct btrfs_root *root, |
4928 | struct btrfs_path *path, | |
4929 | int *level, int root_level) | |
4930 | { | |
74dd17fb | 4931 | BUG_ON(*level == 0); |
7069830a AB |
4932 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
4933 | path->slots[*level]); | |
4934 | path->slots[*level - 1] = 0; | |
4935 | (*level)--; | |
4936 | } | |
4937 | ||
4938 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
4939 | struct btrfs_path *path, | |
4940 | int *level, int root_level) | |
4941 | { | |
4942 | int ret = 0; | |
4943 | int nritems; | |
4944 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4945 | ||
4946 | path->slots[*level]++; | |
4947 | ||
74dd17fb | 4948 | while (path->slots[*level] >= nritems) { |
7069830a AB |
4949 | if (*level == root_level) |
4950 | return -1; | |
4951 | ||
4952 | /* move upnext */ | |
4953 | path->slots[*level] = 0; | |
4954 | free_extent_buffer(path->nodes[*level]); | |
4955 | path->nodes[*level] = NULL; | |
4956 | (*level)++; | |
4957 | path->slots[*level]++; | |
4958 | ||
4959 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4960 | ret = 1; | |
4961 | } | |
4962 | return ret; | |
4963 | } | |
4964 | ||
4965 | /* | |
4966 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
4967 | * or down. | |
4968 | */ | |
4969 | static int tree_advance(struct btrfs_root *root, | |
4970 | struct btrfs_path *path, | |
4971 | int *level, int root_level, | |
4972 | int allow_down, | |
4973 | struct btrfs_key *key) | |
4974 | { | |
4975 | int ret; | |
4976 | ||
4977 | if (*level == 0 || !allow_down) { | |
4978 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
4979 | } else { | |
4980 | tree_move_down(root, path, level, root_level); | |
4981 | ret = 0; | |
4982 | } | |
4983 | if (ret >= 0) { | |
4984 | if (*level == 0) | |
4985 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
4986 | path->slots[*level]); | |
4987 | else | |
4988 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
4989 | path->slots[*level]); | |
4990 | } | |
4991 | return ret; | |
4992 | } | |
4993 | ||
4994 | static int tree_compare_item(struct btrfs_root *left_root, | |
4995 | struct btrfs_path *left_path, | |
4996 | struct btrfs_path *right_path, | |
4997 | char *tmp_buf) | |
4998 | { | |
4999 | int cmp; | |
5000 | int len1, len2; | |
5001 | unsigned long off1, off2; | |
5002 | ||
5003 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5004 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5005 | if (len1 != len2) | |
5006 | return 1; | |
5007 | ||
5008 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5009 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5010 | right_path->slots[0]); | |
5011 | ||
5012 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5013 | ||
5014 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5015 | if (cmp) | |
5016 | return 1; | |
5017 | return 0; | |
5018 | } | |
5019 | ||
5020 | #define ADVANCE 1 | |
5021 | #define ADVANCE_ONLY_NEXT -1 | |
5022 | ||
5023 | /* | |
5024 | * This function compares two trees and calls the provided callback for | |
5025 | * every changed/new/deleted item it finds. | |
5026 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5027 | * the compare pretty fast on snapshotted subvolumes. | |
5028 | * | |
5029 | * This currently works on commit roots only. As commit roots are read only, | |
5030 | * we don't do any locking. The commit roots are protected with transactions. | |
5031 | * Transactions are ended and rejoined when a commit is tried in between. | |
5032 | * | |
5033 | * This function checks for modifications done to the trees while comparing. | |
5034 | * If it detects a change, it aborts immediately. | |
5035 | */ | |
5036 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5037 | struct btrfs_root *right_root, | |
5038 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5039 | { | |
5040 | int ret; | |
5041 | int cmp; | |
5042 | struct btrfs_trans_handle *trans = NULL; | |
5043 | struct btrfs_path *left_path = NULL; | |
5044 | struct btrfs_path *right_path = NULL; | |
5045 | struct btrfs_key left_key; | |
5046 | struct btrfs_key right_key; | |
5047 | char *tmp_buf = NULL; | |
5048 | int left_root_level; | |
5049 | int right_root_level; | |
5050 | int left_level; | |
5051 | int right_level; | |
5052 | int left_end_reached; | |
5053 | int right_end_reached; | |
5054 | int advance_left; | |
5055 | int advance_right; | |
5056 | u64 left_blockptr; | |
5057 | u64 right_blockptr; | |
5058 | u64 left_start_ctransid; | |
5059 | u64 right_start_ctransid; | |
5060 | u64 ctransid; | |
5061 | ||
5062 | left_path = btrfs_alloc_path(); | |
5063 | if (!left_path) { | |
5064 | ret = -ENOMEM; | |
5065 | goto out; | |
5066 | } | |
5067 | right_path = btrfs_alloc_path(); | |
5068 | if (!right_path) { | |
5069 | ret = -ENOMEM; | |
5070 | goto out; | |
5071 | } | |
5072 | ||
5073 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5074 | if (!tmp_buf) { | |
5075 | ret = -ENOMEM; | |
5076 | goto out; | |
5077 | } | |
5078 | ||
5079 | left_path->search_commit_root = 1; | |
5080 | left_path->skip_locking = 1; | |
5081 | right_path->search_commit_root = 1; | |
5082 | right_path->skip_locking = 1; | |
5083 | ||
5f3ab90a | 5084 | spin_lock(&left_root->root_item_lock); |
7069830a | 5085 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5086 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5087 | |
5f3ab90a | 5088 | spin_lock(&right_root->root_item_lock); |
7069830a | 5089 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5090 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5091 | |
5092 | trans = btrfs_join_transaction(left_root); | |
5093 | if (IS_ERR(trans)) { | |
5094 | ret = PTR_ERR(trans); | |
5095 | trans = NULL; | |
5096 | goto out; | |
5097 | } | |
5098 | ||
5099 | /* | |
5100 | * Strategy: Go to the first items of both trees. Then do | |
5101 | * | |
5102 | * If both trees are at level 0 | |
5103 | * Compare keys of current items | |
5104 | * If left < right treat left item as new, advance left tree | |
5105 | * and repeat | |
5106 | * If left > right treat right item as deleted, advance right tree | |
5107 | * and repeat | |
5108 | * If left == right do deep compare of items, treat as changed if | |
5109 | * needed, advance both trees and repeat | |
5110 | * If both trees are at the same level but not at level 0 | |
5111 | * Compare keys of current nodes/leafs | |
5112 | * If left < right advance left tree and repeat | |
5113 | * If left > right advance right tree and repeat | |
5114 | * If left == right compare blockptrs of the next nodes/leafs | |
5115 | * If they match advance both trees but stay at the same level | |
5116 | * and repeat | |
5117 | * If they don't match advance both trees while allowing to go | |
5118 | * deeper and repeat | |
5119 | * If tree levels are different | |
5120 | * Advance the tree that needs it and repeat | |
5121 | * | |
5122 | * Advancing a tree means: | |
5123 | * If we are at level 0, try to go to the next slot. If that's not | |
5124 | * possible, go one level up and repeat. Stop when we found a level | |
5125 | * where we could go to the next slot. We may at this point be on a | |
5126 | * node or a leaf. | |
5127 | * | |
5128 | * If we are not at level 0 and not on shared tree blocks, go one | |
5129 | * level deeper. | |
5130 | * | |
5131 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5132 | * the right if possible or go up and right. | |
5133 | */ | |
5134 | ||
5135 | left_level = btrfs_header_level(left_root->commit_root); | |
5136 | left_root_level = left_level; | |
5137 | left_path->nodes[left_level] = left_root->commit_root; | |
5138 | extent_buffer_get(left_path->nodes[left_level]); | |
5139 | ||
5140 | right_level = btrfs_header_level(right_root->commit_root); | |
5141 | right_root_level = right_level; | |
5142 | right_path->nodes[right_level] = right_root->commit_root; | |
5143 | extent_buffer_get(right_path->nodes[right_level]); | |
5144 | ||
5145 | if (left_level == 0) | |
5146 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5147 | &left_key, left_path->slots[left_level]); | |
5148 | else | |
5149 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5150 | &left_key, left_path->slots[left_level]); | |
5151 | if (right_level == 0) | |
5152 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5153 | &right_key, right_path->slots[right_level]); | |
5154 | else | |
5155 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5156 | &right_key, right_path->slots[right_level]); | |
5157 | ||
5158 | left_end_reached = right_end_reached = 0; | |
5159 | advance_left = advance_right = 0; | |
5160 | ||
5161 | while (1) { | |
5162 | /* | |
5163 | * We need to make sure the transaction does not get committed | |
5164 | * while we do anything on commit roots. This means, we need to | |
5165 | * join and leave transactions for every item that we process. | |
5166 | */ | |
5167 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5168 | btrfs_release_path(left_path); | |
5169 | btrfs_release_path(right_path); | |
5170 | ||
5171 | ret = btrfs_end_transaction(trans, left_root); | |
5172 | trans = NULL; | |
5173 | if (ret < 0) | |
5174 | goto out; | |
5175 | } | |
5176 | /* now rejoin the transaction */ | |
5177 | if (!trans) { | |
5178 | trans = btrfs_join_transaction(left_root); | |
5179 | if (IS_ERR(trans)) { | |
5180 | ret = PTR_ERR(trans); | |
5181 | trans = NULL; | |
5182 | goto out; | |
5183 | } | |
5184 | ||
5f3ab90a | 5185 | spin_lock(&left_root->root_item_lock); |
7069830a | 5186 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5187 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5188 | if (ctransid != left_start_ctransid) |
5189 | left_start_ctransid = 0; | |
5190 | ||
5f3ab90a | 5191 | spin_lock(&right_root->root_item_lock); |
7069830a | 5192 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5193 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5194 | if (ctransid != right_start_ctransid) |
5195 | right_start_ctransid = 0; | |
5196 | ||
5197 | if (!left_start_ctransid || !right_start_ctransid) { | |
5198 | WARN(1, KERN_WARNING | |
5199 | "btrfs: btrfs_compare_tree detected " | |
5200 | "a change in one of the trees while " | |
5201 | "iterating. This is probably a " | |
5202 | "bug.\n"); | |
5203 | ret = -EIO; | |
5204 | goto out; | |
5205 | } | |
5206 | ||
5207 | /* | |
5208 | * the commit root may have changed, so start again | |
5209 | * where we stopped | |
5210 | */ | |
5211 | left_path->lowest_level = left_level; | |
5212 | right_path->lowest_level = right_level; | |
5213 | ret = btrfs_search_slot(NULL, left_root, | |
5214 | &left_key, left_path, 0, 0); | |
5215 | if (ret < 0) | |
5216 | goto out; | |
5217 | ret = btrfs_search_slot(NULL, right_root, | |
5218 | &right_key, right_path, 0, 0); | |
5219 | if (ret < 0) | |
5220 | goto out; | |
5221 | } | |
5222 | ||
5223 | if (advance_left && !left_end_reached) { | |
5224 | ret = tree_advance(left_root, left_path, &left_level, | |
5225 | left_root_level, | |
5226 | advance_left != ADVANCE_ONLY_NEXT, | |
5227 | &left_key); | |
5228 | if (ret < 0) | |
5229 | left_end_reached = ADVANCE; | |
5230 | advance_left = 0; | |
5231 | } | |
5232 | if (advance_right && !right_end_reached) { | |
5233 | ret = tree_advance(right_root, right_path, &right_level, | |
5234 | right_root_level, | |
5235 | advance_right != ADVANCE_ONLY_NEXT, | |
5236 | &right_key); | |
5237 | if (ret < 0) | |
5238 | right_end_reached = ADVANCE; | |
5239 | advance_right = 0; | |
5240 | } | |
5241 | ||
5242 | if (left_end_reached && right_end_reached) { | |
5243 | ret = 0; | |
5244 | goto out; | |
5245 | } else if (left_end_reached) { | |
5246 | if (right_level == 0) { | |
5247 | ret = changed_cb(left_root, right_root, | |
5248 | left_path, right_path, | |
5249 | &right_key, | |
5250 | BTRFS_COMPARE_TREE_DELETED, | |
5251 | ctx); | |
5252 | if (ret < 0) | |
5253 | goto out; | |
5254 | } | |
5255 | advance_right = ADVANCE; | |
5256 | continue; | |
5257 | } else if (right_end_reached) { | |
5258 | if (left_level == 0) { | |
5259 | ret = changed_cb(left_root, right_root, | |
5260 | left_path, right_path, | |
5261 | &left_key, | |
5262 | BTRFS_COMPARE_TREE_NEW, | |
5263 | ctx); | |
5264 | if (ret < 0) | |
5265 | goto out; | |
5266 | } | |
5267 | advance_left = ADVANCE; | |
5268 | continue; | |
5269 | } | |
5270 | ||
5271 | if (left_level == 0 && right_level == 0) { | |
5272 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5273 | if (cmp < 0) { | |
5274 | ret = changed_cb(left_root, right_root, | |
5275 | left_path, right_path, | |
5276 | &left_key, | |
5277 | BTRFS_COMPARE_TREE_NEW, | |
5278 | ctx); | |
5279 | if (ret < 0) | |
5280 | goto out; | |
5281 | advance_left = ADVANCE; | |
5282 | } else if (cmp > 0) { | |
5283 | ret = changed_cb(left_root, right_root, | |
5284 | left_path, right_path, | |
5285 | &right_key, | |
5286 | BTRFS_COMPARE_TREE_DELETED, | |
5287 | ctx); | |
5288 | if (ret < 0) | |
5289 | goto out; | |
5290 | advance_right = ADVANCE; | |
5291 | } else { | |
ba5e8f2e JB |
5292 | enum btrfs_compare_tree_result cmp; |
5293 | ||
74dd17fb | 5294 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5295 | ret = tree_compare_item(left_root, left_path, |
5296 | right_path, tmp_buf); | |
ba5e8f2e JB |
5297 | if (ret) |
5298 | cmp = BTRFS_COMPARE_TREE_CHANGED; | |
5299 | else | |
5300 | cmp = BTRFS_COMPARE_TREE_SAME; | |
5301 | ret = changed_cb(left_root, right_root, | |
5302 | left_path, right_path, | |
5303 | &left_key, cmp, ctx); | |
5304 | if (ret < 0) | |
5305 | goto out; | |
7069830a AB |
5306 | advance_left = ADVANCE; |
5307 | advance_right = ADVANCE; | |
5308 | } | |
5309 | } else if (left_level == right_level) { | |
5310 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5311 | if (cmp < 0) { | |
5312 | advance_left = ADVANCE; | |
5313 | } else if (cmp > 0) { | |
5314 | advance_right = ADVANCE; | |
5315 | } else { | |
5316 | left_blockptr = btrfs_node_blockptr( | |
5317 | left_path->nodes[left_level], | |
5318 | left_path->slots[left_level]); | |
5319 | right_blockptr = btrfs_node_blockptr( | |
5320 | right_path->nodes[right_level], | |
5321 | right_path->slots[right_level]); | |
5322 | if (left_blockptr == right_blockptr) { | |
5323 | /* | |
5324 | * As we're on a shared block, don't | |
5325 | * allow to go deeper. | |
5326 | */ | |
5327 | advance_left = ADVANCE_ONLY_NEXT; | |
5328 | advance_right = ADVANCE_ONLY_NEXT; | |
5329 | } else { | |
5330 | advance_left = ADVANCE; | |
5331 | advance_right = ADVANCE; | |
5332 | } | |
5333 | } | |
5334 | } else if (left_level < right_level) { | |
5335 | advance_right = ADVANCE; | |
5336 | } else { | |
5337 | advance_left = ADVANCE; | |
5338 | } | |
5339 | } | |
5340 | ||
5341 | out: | |
5342 | btrfs_free_path(left_path); | |
5343 | btrfs_free_path(right_path); | |
5344 | kfree(tmp_buf); | |
5345 | ||
5346 | if (trans) { | |
5347 | if (!ret) | |
5348 | ret = btrfs_end_transaction(trans, left_root); | |
5349 | else | |
5350 | btrfs_end_transaction(trans, left_root); | |
5351 | } | |
5352 | ||
5353 | return ret; | |
5354 | } | |
5355 | ||
3f157a2f CM |
5356 | /* |
5357 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5358 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5359 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5360 | * |
5361 | * 0 is returned if another key is found, < 0 if there are any errors | |
5362 | * and 1 is returned if there are no higher keys in the tree | |
5363 | * | |
5364 | * path->keep_locks should be set to 1 on the search made before | |
5365 | * calling this function. | |
5366 | */ | |
e7a84565 | 5367 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5368 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5369 | { |
e7a84565 CM |
5370 | int slot; |
5371 | struct extent_buffer *c; | |
5372 | ||
934d375b | 5373 | WARN_ON(!path->keep_locks); |
d397712b | 5374 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5375 | if (!path->nodes[level]) |
5376 | return 1; | |
5377 | ||
5378 | slot = path->slots[level] + 1; | |
5379 | c = path->nodes[level]; | |
3f157a2f | 5380 | next: |
e7a84565 | 5381 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5382 | int ret; |
5383 | int orig_lowest; | |
5384 | struct btrfs_key cur_key; | |
5385 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5386 | !path->nodes[level + 1]) | |
e7a84565 | 5387 | return 1; |
33c66f43 YZ |
5388 | |
5389 | if (path->locks[level + 1]) { | |
5390 | level++; | |
5391 | continue; | |
5392 | } | |
5393 | ||
5394 | slot = btrfs_header_nritems(c) - 1; | |
5395 | if (level == 0) | |
5396 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5397 | else | |
5398 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5399 | ||
5400 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5401 | btrfs_release_path(path); |
33c66f43 YZ |
5402 | path->lowest_level = level; |
5403 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5404 | 0, 0); | |
5405 | path->lowest_level = orig_lowest; | |
5406 | if (ret < 0) | |
5407 | return ret; | |
5408 | ||
5409 | c = path->nodes[level]; | |
5410 | slot = path->slots[level]; | |
5411 | if (ret == 0) | |
5412 | slot++; | |
5413 | goto next; | |
e7a84565 | 5414 | } |
33c66f43 | 5415 | |
e7a84565 CM |
5416 | if (level == 0) |
5417 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5418 | else { |
3f157a2f CM |
5419 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5420 | ||
3f157a2f CM |
5421 | if (gen < min_trans) { |
5422 | slot++; | |
5423 | goto next; | |
5424 | } | |
e7a84565 | 5425 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5426 | } |
e7a84565 CM |
5427 | return 0; |
5428 | } | |
5429 | return 1; | |
5430 | } | |
5431 | ||
97571fd0 | 5432 | /* |
925baedd | 5433 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5434 | * returns 0 if it found something or 1 if there are no greater leaves. |
5435 | * returns < 0 on io errors. | |
97571fd0 | 5436 | */ |
234b63a0 | 5437 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5438 | { |
5439 | return btrfs_next_old_leaf(root, path, 0); | |
5440 | } | |
5441 | ||
5442 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5443 | u64 time_seq) | |
d97e63b6 CM |
5444 | { |
5445 | int slot; | |
8e73f275 | 5446 | int level; |
5f39d397 | 5447 | struct extent_buffer *c; |
8e73f275 | 5448 | struct extent_buffer *next; |
925baedd CM |
5449 | struct btrfs_key key; |
5450 | u32 nritems; | |
5451 | int ret; | |
8e73f275 | 5452 | int old_spinning = path->leave_spinning; |
bd681513 | 5453 | int next_rw_lock = 0; |
925baedd CM |
5454 | |
5455 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5456 | if (nritems == 0) |
925baedd | 5457 | return 1; |
925baedd | 5458 | |
8e73f275 CM |
5459 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5460 | again: | |
5461 | level = 1; | |
5462 | next = NULL; | |
bd681513 | 5463 | next_rw_lock = 0; |
b3b4aa74 | 5464 | btrfs_release_path(path); |
8e73f275 | 5465 | |
a2135011 | 5466 | path->keep_locks = 1; |
31533fb2 | 5467 | path->leave_spinning = 1; |
8e73f275 | 5468 | |
3d7806ec JS |
5469 | if (time_seq) |
5470 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5471 | else | |
5472 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5473 | path->keep_locks = 0; |
5474 | ||
5475 | if (ret < 0) | |
5476 | return ret; | |
5477 | ||
a2135011 | 5478 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5479 | /* |
5480 | * by releasing the path above we dropped all our locks. A balance | |
5481 | * could have added more items next to the key that used to be | |
5482 | * at the very end of the block. So, check again here and | |
5483 | * advance the path if there are now more items available. | |
5484 | */ | |
a2135011 | 5485 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5486 | if (ret == 0) |
5487 | path->slots[0]++; | |
8e73f275 | 5488 | ret = 0; |
925baedd CM |
5489 | goto done; |
5490 | } | |
d97e63b6 | 5491 | |
d397712b | 5492 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5493 | if (!path->nodes[level]) { |
5494 | ret = 1; | |
5495 | goto done; | |
5496 | } | |
5f39d397 | 5497 | |
d97e63b6 CM |
5498 | slot = path->slots[level] + 1; |
5499 | c = path->nodes[level]; | |
5f39d397 | 5500 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5501 | level++; |
8e73f275 CM |
5502 | if (level == BTRFS_MAX_LEVEL) { |
5503 | ret = 1; | |
5504 | goto done; | |
5505 | } | |
d97e63b6 CM |
5506 | continue; |
5507 | } | |
5f39d397 | 5508 | |
925baedd | 5509 | if (next) { |
bd681513 | 5510 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5511 | free_extent_buffer(next); |
925baedd | 5512 | } |
5f39d397 | 5513 | |
8e73f275 | 5514 | next = c; |
bd681513 | 5515 | next_rw_lock = path->locks[level]; |
8e73f275 | 5516 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5517 | slot, &key, 0); |
8e73f275 CM |
5518 | if (ret == -EAGAIN) |
5519 | goto again; | |
5f39d397 | 5520 | |
76a05b35 | 5521 | if (ret < 0) { |
b3b4aa74 | 5522 | btrfs_release_path(path); |
76a05b35 CM |
5523 | goto done; |
5524 | } | |
5525 | ||
5cd57b2c | 5526 | if (!path->skip_locking) { |
bd681513 | 5527 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5528 | if (!ret && time_seq) { |
5529 | /* | |
5530 | * If we don't get the lock, we may be racing | |
5531 | * with push_leaf_left, holding that lock while | |
5532 | * itself waiting for the leaf we've currently | |
5533 | * locked. To solve this situation, we give up | |
5534 | * on our lock and cycle. | |
5535 | */ | |
cf538830 | 5536 | free_extent_buffer(next); |
d42244a0 JS |
5537 | btrfs_release_path(path); |
5538 | cond_resched(); | |
5539 | goto again; | |
5540 | } | |
8e73f275 CM |
5541 | if (!ret) { |
5542 | btrfs_set_path_blocking(path); | |
bd681513 | 5543 | btrfs_tree_read_lock(next); |
31533fb2 | 5544 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5545 | BTRFS_READ_LOCK); |
8e73f275 | 5546 | } |
31533fb2 | 5547 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5548 | } |
d97e63b6 CM |
5549 | break; |
5550 | } | |
5551 | path->slots[level] = slot; | |
d397712b | 5552 | while (1) { |
d97e63b6 CM |
5553 | level--; |
5554 | c = path->nodes[level]; | |
925baedd | 5555 | if (path->locks[level]) |
bd681513 | 5556 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5557 | |
5f39d397 | 5558 | free_extent_buffer(c); |
d97e63b6 CM |
5559 | path->nodes[level] = next; |
5560 | path->slots[level] = 0; | |
a74a4b97 | 5561 | if (!path->skip_locking) |
bd681513 | 5562 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5563 | if (!level) |
5564 | break; | |
b4ce94de | 5565 | |
8e73f275 | 5566 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5567 | 0, &key, 0); |
8e73f275 CM |
5568 | if (ret == -EAGAIN) |
5569 | goto again; | |
5570 | ||
76a05b35 | 5571 | if (ret < 0) { |
b3b4aa74 | 5572 | btrfs_release_path(path); |
76a05b35 CM |
5573 | goto done; |
5574 | } | |
5575 | ||
5cd57b2c | 5576 | if (!path->skip_locking) { |
bd681513 | 5577 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5578 | if (!ret) { |
5579 | btrfs_set_path_blocking(path); | |
bd681513 | 5580 | btrfs_tree_read_lock(next); |
31533fb2 | 5581 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5582 | BTRFS_READ_LOCK); |
5583 | } | |
31533fb2 | 5584 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5585 | } |
d97e63b6 | 5586 | } |
8e73f275 | 5587 | ret = 0; |
925baedd | 5588 | done: |
f7c79f30 | 5589 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5590 | path->leave_spinning = old_spinning; |
5591 | if (!old_spinning) | |
5592 | btrfs_set_path_blocking(path); | |
5593 | ||
5594 | return ret; | |
d97e63b6 | 5595 | } |
0b86a832 | 5596 | |
3f157a2f CM |
5597 | /* |
5598 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5599 | * searching until it gets past min_objectid or finds an item of 'type' | |
5600 | * | |
5601 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5602 | */ | |
0b86a832 CM |
5603 | int btrfs_previous_item(struct btrfs_root *root, |
5604 | struct btrfs_path *path, u64 min_objectid, | |
5605 | int type) | |
5606 | { | |
5607 | struct btrfs_key found_key; | |
5608 | struct extent_buffer *leaf; | |
e02119d5 | 5609 | u32 nritems; |
0b86a832 CM |
5610 | int ret; |
5611 | ||
d397712b | 5612 | while (1) { |
0b86a832 | 5613 | if (path->slots[0] == 0) { |
b4ce94de | 5614 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5615 | ret = btrfs_prev_leaf(root, path); |
5616 | if (ret != 0) | |
5617 | return ret; | |
5618 | } else { | |
5619 | path->slots[0]--; | |
5620 | } | |
5621 | leaf = path->nodes[0]; | |
e02119d5 CM |
5622 | nritems = btrfs_header_nritems(leaf); |
5623 | if (nritems == 0) | |
5624 | return 1; | |
5625 | if (path->slots[0] == nritems) | |
5626 | path->slots[0]--; | |
5627 | ||
0b86a832 | 5628 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5629 | if (found_key.objectid < min_objectid) |
5630 | break; | |
0a4eefbb YZ |
5631 | if (found_key.type == type) |
5632 | return 0; | |
e02119d5 CM |
5633 | if (found_key.objectid == min_objectid && |
5634 | found_key.type < type) | |
5635 | break; | |
0b86a832 CM |
5636 | } |
5637 | return 1; | |
5638 | } |