]>
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); | |
5de865ee | 42 | static int tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
f230475e | 43 | struct extent_buffer *eb); |
d97e63b6 | 44 | |
df24a2b9 | 45 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 46 | { |
df24a2b9 | 47 | struct btrfs_path *path; |
e00f7308 | 48 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 49 | return path; |
2c90e5d6 CM |
50 | } |
51 | ||
b4ce94de CM |
52 | /* |
53 | * set all locked nodes in the path to blocking locks. This should | |
54 | * be done before scheduling | |
55 | */ | |
56 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
57 | { | |
58 | int i; | |
59 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
60 | if (!p->nodes[i] || !p->locks[i]) |
61 | continue; | |
62 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
63 | if (p->locks[i] == BTRFS_READ_LOCK) | |
64 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
65 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
66 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
67 | } |
68 | } | |
69 | ||
70 | /* | |
71 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
72 | * |
73 | * held is used to keep lockdep happy, when lockdep is enabled | |
74 | * we set held to a blocking lock before we go around and | |
75 | * retake all the spinlocks in the path. You can safely use NULL | |
76 | * for held | |
b4ce94de | 77 | */ |
4008c04a | 78 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 79 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
80 | { |
81 | int i; | |
4008c04a CM |
82 | |
83 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
84 | /* lockdep really cares that we take all of these spinlocks | |
85 | * in the right order. If any of the locks in the path are not | |
86 | * currently blocking, it is going to complain. So, make really | |
87 | * really sure by forcing the path to blocking before we clear | |
88 | * the path blocking. | |
89 | */ | |
bd681513 CM |
90 | if (held) { |
91 | btrfs_set_lock_blocking_rw(held, held_rw); | |
92 | if (held_rw == BTRFS_WRITE_LOCK) | |
93 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
94 | else if (held_rw == BTRFS_READ_LOCK) | |
95 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
96 | } | |
4008c04a CM |
97 | btrfs_set_path_blocking(p); |
98 | #endif | |
99 | ||
100 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
101 | if (p->nodes[i] && p->locks[i]) { |
102 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
103 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
104 | p->locks[i] = BTRFS_WRITE_LOCK; | |
105 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
106 | p->locks[i] = BTRFS_READ_LOCK; | |
107 | } | |
b4ce94de | 108 | } |
4008c04a CM |
109 | |
110 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
111 | if (held) | |
bd681513 | 112 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 113 | #endif |
b4ce94de CM |
114 | } |
115 | ||
d352ac68 | 116 | /* this also releases the path */ |
df24a2b9 | 117 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 118 | { |
ff175d57 JJ |
119 | if (!p) |
120 | return; | |
b3b4aa74 | 121 | btrfs_release_path(p); |
df24a2b9 | 122 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
123 | } |
124 | ||
d352ac68 CM |
125 | /* |
126 | * path release drops references on the extent buffers in the path | |
127 | * and it drops any locks held by this path | |
128 | * | |
129 | * It is safe to call this on paths that no locks or extent buffers held. | |
130 | */ | |
b3b4aa74 | 131 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
132 | { |
133 | int i; | |
a2135011 | 134 | |
234b63a0 | 135 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 136 | p->slots[i] = 0; |
eb60ceac | 137 | if (!p->nodes[i]) |
925baedd CM |
138 | continue; |
139 | if (p->locks[i]) { | |
bd681513 | 140 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
141 | p->locks[i] = 0; |
142 | } | |
5f39d397 | 143 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 144 | p->nodes[i] = NULL; |
eb60ceac CM |
145 | } |
146 | } | |
147 | ||
d352ac68 CM |
148 | /* |
149 | * safely gets a reference on the root node of a tree. A lock | |
150 | * is not taken, so a concurrent writer may put a different node | |
151 | * at the root of the tree. See btrfs_lock_root_node for the | |
152 | * looping required. | |
153 | * | |
154 | * The extent buffer returned by this has a reference taken, so | |
155 | * it won't disappear. It may stop being the root of the tree | |
156 | * at any time because there are no locks held. | |
157 | */ | |
925baedd CM |
158 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
159 | { | |
160 | struct extent_buffer *eb; | |
240f62c8 | 161 | |
3083ee2e JB |
162 | while (1) { |
163 | rcu_read_lock(); | |
164 | eb = rcu_dereference(root->node); | |
165 | ||
166 | /* | |
167 | * RCU really hurts here, we could free up the root node because | |
168 | * it was cow'ed but we may not get the new root node yet so do | |
169 | * the inc_not_zero dance and if it doesn't work then | |
170 | * synchronize_rcu and try again. | |
171 | */ | |
172 | if (atomic_inc_not_zero(&eb->refs)) { | |
173 | rcu_read_unlock(); | |
174 | break; | |
175 | } | |
176 | rcu_read_unlock(); | |
177 | synchronize_rcu(); | |
178 | } | |
925baedd CM |
179 | return eb; |
180 | } | |
181 | ||
d352ac68 CM |
182 | /* loop around taking references on and locking the root node of the |
183 | * tree until you end up with a lock on the root. A locked buffer | |
184 | * is returned, with a reference held. | |
185 | */ | |
925baedd CM |
186 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
187 | { | |
188 | struct extent_buffer *eb; | |
189 | ||
d397712b | 190 | while (1) { |
925baedd CM |
191 | eb = btrfs_root_node(root); |
192 | btrfs_tree_lock(eb); | |
240f62c8 | 193 | if (eb == root->node) |
925baedd | 194 | break; |
925baedd CM |
195 | btrfs_tree_unlock(eb); |
196 | free_extent_buffer(eb); | |
197 | } | |
198 | return eb; | |
199 | } | |
200 | ||
bd681513 CM |
201 | /* loop around taking references on and locking the root node of the |
202 | * tree until you end up with a lock on the root. A locked buffer | |
203 | * is returned, with a reference held. | |
204 | */ | |
48a3b636 | 205 | static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) |
bd681513 CM |
206 | { |
207 | struct extent_buffer *eb; | |
208 | ||
209 | while (1) { | |
210 | eb = btrfs_root_node(root); | |
211 | btrfs_tree_read_lock(eb); | |
212 | if (eb == root->node) | |
213 | break; | |
214 | btrfs_tree_read_unlock(eb); | |
215 | free_extent_buffer(eb); | |
216 | } | |
217 | return eb; | |
218 | } | |
219 | ||
d352ac68 CM |
220 | /* cowonly root (everything not a reference counted cow subvolume), just get |
221 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
222 | * get properly updated on disk. | |
223 | */ | |
0b86a832 CM |
224 | static void add_root_to_dirty_list(struct btrfs_root *root) |
225 | { | |
e5846fc6 | 226 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
227 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
228 | list_add(&root->dirty_list, | |
229 | &root->fs_info->dirty_cowonly_roots); | |
230 | } | |
e5846fc6 | 231 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
232 | } |
233 | ||
d352ac68 CM |
234 | /* |
235 | * used by snapshot creation to make a copy of a root for a tree with | |
236 | * a given objectid. The buffer with the new root node is returned in | |
237 | * cow_ret, and this func returns zero on success or a negative error code. | |
238 | */ | |
be20aa9d CM |
239 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
240 | struct btrfs_root *root, | |
241 | struct extent_buffer *buf, | |
242 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
243 | { | |
244 | struct extent_buffer *cow; | |
be20aa9d CM |
245 | int ret = 0; |
246 | int level; | |
5d4f98a2 | 247 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
248 | |
249 | WARN_ON(root->ref_cows && trans->transid != | |
250 | root->fs_info->running_transaction->transid); | |
251 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
252 | ||
253 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
254 | if (level == 0) |
255 | btrfs_item_key(buf, &disk_key, 0); | |
256 | else | |
257 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 258 | |
5d4f98a2 YZ |
259 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
260 | new_root_objectid, &disk_key, level, | |
5581a51a | 261 | buf->start, 0); |
5d4f98a2 | 262 | if (IS_ERR(cow)) |
be20aa9d CM |
263 | return PTR_ERR(cow); |
264 | ||
265 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
266 | btrfs_set_header_bytenr(cow, cow->start); | |
267 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
268 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
269 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
270 | BTRFS_HEADER_FLAG_RELOC); | |
271 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
272 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
273 | else | |
274 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 275 | |
0a4e5586 | 276 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
277 | BTRFS_FSID_SIZE); |
278 | ||
be20aa9d | 279 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 280 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 281 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 282 | else |
66d7e7f0 | 283 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 284 | |
be20aa9d CM |
285 | if (ret) |
286 | return ret; | |
287 | ||
288 | btrfs_mark_buffer_dirty(cow); | |
289 | *cow_ret = cow; | |
290 | return 0; | |
291 | } | |
292 | ||
bd989ba3 JS |
293 | enum mod_log_op { |
294 | MOD_LOG_KEY_REPLACE, | |
295 | MOD_LOG_KEY_ADD, | |
296 | MOD_LOG_KEY_REMOVE, | |
297 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
298 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
299 | MOD_LOG_MOVE_KEYS, | |
300 | MOD_LOG_ROOT_REPLACE, | |
301 | }; | |
302 | ||
303 | struct tree_mod_move { | |
304 | int dst_slot; | |
305 | int nr_items; | |
306 | }; | |
307 | ||
308 | struct tree_mod_root { | |
309 | u64 logical; | |
310 | u8 level; | |
311 | }; | |
312 | ||
313 | struct tree_mod_elem { | |
314 | struct rb_node node; | |
315 | u64 index; /* shifted logical */ | |
097b8a7c | 316 | u64 seq; |
bd989ba3 JS |
317 | enum mod_log_op op; |
318 | ||
319 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
320 | int slot; | |
321 | ||
322 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
323 | u64 generation; | |
324 | ||
325 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
326 | struct btrfs_disk_key key; | |
327 | u64 blockptr; | |
328 | ||
329 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
330 | struct tree_mod_move move; | |
331 | ||
332 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
333 | struct tree_mod_root old_root; | |
334 | }; | |
335 | ||
097b8a7c | 336 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 337 | { |
097b8a7c | 338 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
339 | } |
340 | ||
097b8a7c JS |
341 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
342 | { | |
343 | read_unlock(&fs_info->tree_mod_log_lock); | |
344 | } | |
345 | ||
346 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
347 | { | |
348 | write_lock(&fs_info->tree_mod_log_lock); | |
349 | } | |
350 | ||
351 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
352 | { | |
353 | write_unlock(&fs_info->tree_mod_log_lock); | |
354 | } | |
355 | ||
fc36ed7e JS |
356 | /* |
357 | * Increment the upper half of tree_mod_seq, set lower half zero. | |
358 | * | |
359 | * Must be called with fs_info->tree_mod_seq_lock held. | |
360 | */ | |
361 | static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info) | |
362 | { | |
363 | u64 seq = atomic64_read(&fs_info->tree_mod_seq); | |
364 | seq &= 0xffffffff00000000ull; | |
365 | seq += 1ull << 32; | |
366 | atomic64_set(&fs_info->tree_mod_seq, seq); | |
367 | return seq; | |
368 | } | |
369 | ||
370 | /* | |
371 | * Increment the lower half of tree_mod_seq. | |
372 | * | |
373 | * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers | |
374 | * are generated should not technically require a spin lock here. (Rationale: | |
375 | * incrementing the minor while incrementing the major seq number is between its | |
376 | * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it | |
377 | * just returns a unique sequence number as usual.) We have decided to leave | |
378 | * that requirement in here and rethink it once we notice it really imposes a | |
379 | * problem on some workload. | |
380 | */ | |
381 | static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info) | |
382 | { | |
383 | return atomic64_inc_return(&fs_info->tree_mod_seq); | |
384 | } | |
385 | ||
386 | /* | |
387 | * return the last minor in the previous major tree_mod_seq number | |
388 | */ | |
389 | u64 btrfs_tree_mod_seq_prev(u64 seq) | |
390 | { | |
391 | return (seq & 0xffffffff00000000ull) - 1ull; | |
392 | } | |
393 | ||
097b8a7c JS |
394 | /* |
395 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
396 | * passed does not already have a sequence number set. So when a caller expects | |
397 | * to record tree modifications, it should ensure to set elem->seq to zero | |
398 | * before calling btrfs_get_tree_mod_seq. | |
399 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
400 | * blocker was added. | |
401 | */ | |
402 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
403 | struct seq_list *elem) | |
bd989ba3 | 404 | { |
097b8a7c JS |
405 | u64 seq; |
406 | ||
407 | tree_mod_log_write_lock(fs_info); | |
bd989ba3 | 408 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c | 409 | if (!elem->seq) { |
fc36ed7e | 410 | elem->seq = btrfs_inc_tree_mod_seq_major(fs_info); |
097b8a7c JS |
411 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); |
412 | } | |
fc36ed7e | 413 | seq = btrfs_inc_tree_mod_seq_minor(fs_info); |
bd989ba3 | 414 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
415 | tree_mod_log_write_unlock(fs_info); |
416 | ||
417 | return seq; | |
bd989ba3 JS |
418 | } |
419 | ||
420 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
421 | struct seq_list *elem) | |
422 | { | |
423 | struct rb_root *tm_root; | |
424 | struct rb_node *node; | |
425 | struct rb_node *next; | |
426 | struct seq_list *cur_elem; | |
427 | struct tree_mod_elem *tm; | |
428 | u64 min_seq = (u64)-1; | |
429 | u64 seq_putting = elem->seq; | |
430 | ||
431 | if (!seq_putting) | |
432 | return; | |
433 | ||
bd989ba3 JS |
434 | spin_lock(&fs_info->tree_mod_seq_lock); |
435 | list_del(&elem->list); | |
097b8a7c | 436 | elem->seq = 0; |
bd989ba3 JS |
437 | |
438 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 439 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
440 | if (seq_putting > cur_elem->seq) { |
441 | /* | |
442 | * blocker with lower sequence number exists, we | |
443 | * cannot remove anything from the log | |
444 | */ | |
097b8a7c JS |
445 | spin_unlock(&fs_info->tree_mod_seq_lock); |
446 | return; | |
bd989ba3 JS |
447 | } |
448 | min_seq = cur_elem->seq; | |
449 | } | |
450 | } | |
097b8a7c JS |
451 | spin_unlock(&fs_info->tree_mod_seq_lock); |
452 | ||
bd989ba3 JS |
453 | /* |
454 | * anything that's lower than the lowest existing (read: blocked) | |
455 | * sequence number can be removed from the tree. | |
456 | */ | |
097b8a7c | 457 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
458 | tm_root = &fs_info->tree_mod_log; |
459 | for (node = rb_first(tm_root); node; node = next) { | |
460 | next = rb_next(node); | |
461 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 462 | if (tm->seq > min_seq) |
bd989ba3 JS |
463 | continue; |
464 | rb_erase(node, tm_root); | |
bd989ba3 JS |
465 | kfree(tm); |
466 | } | |
097b8a7c | 467 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
468 | } |
469 | ||
470 | /* | |
471 | * key order of the log: | |
472 | * index -> sequence | |
473 | * | |
474 | * the index is the shifted logical of the *new* root node for root replace | |
475 | * operations, or the shifted logical of the affected block for all other | |
476 | * operations. | |
5de865ee FDBM |
477 | * |
478 | * Note: must be called with write lock (tree_mod_log_write_lock). | |
bd989ba3 JS |
479 | */ |
480 | static noinline int | |
481 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
482 | { | |
483 | struct rb_root *tm_root; | |
484 | struct rb_node **new; | |
485 | struct rb_node *parent = NULL; | |
486 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
487 | |
488 | BUG_ON(!tm); | |
489 | ||
c8cc6341 JB |
490 | spin_lock(&fs_info->tree_mod_seq_lock); |
491 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
492 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
bd989ba3 | 493 | |
bd989ba3 JS |
494 | tm_root = &fs_info->tree_mod_log; |
495 | new = &tm_root->rb_node; | |
496 | while (*new) { | |
497 | cur = container_of(*new, struct tree_mod_elem, node); | |
498 | parent = *new; | |
499 | if (cur->index < tm->index) | |
500 | new = &((*new)->rb_left); | |
501 | else if (cur->index > tm->index) | |
502 | new = &((*new)->rb_right); | |
097b8a7c | 503 | else if (cur->seq < tm->seq) |
bd989ba3 | 504 | new = &((*new)->rb_left); |
097b8a7c | 505 | else if (cur->seq > tm->seq) |
bd989ba3 | 506 | new = &((*new)->rb_right); |
5de865ee FDBM |
507 | else |
508 | return -EEXIST; | |
bd989ba3 JS |
509 | } |
510 | ||
511 | rb_link_node(&tm->node, parent, new); | |
512 | rb_insert_color(&tm->node, tm_root); | |
5de865ee | 513 | return 0; |
bd989ba3 JS |
514 | } |
515 | ||
097b8a7c JS |
516 | /* |
517 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
518 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
519 | * this until all tree mod log insertions are recorded in the rb tree and then | |
520 | * call tree_mod_log_write_unlock() to release. | |
521 | */ | |
e9b7fd4d JS |
522 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
523 | struct extent_buffer *eb) { | |
524 | smp_mb(); | |
525 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
526 | return 1; | |
097b8a7c JS |
527 | if (eb && btrfs_header_level(eb) == 0) |
528 | return 1; | |
5de865ee FDBM |
529 | |
530 | tree_mod_log_write_lock(fs_info); | |
531 | if (list_empty(&(fs_info)->tree_mod_seq_list)) { | |
532 | tree_mod_log_write_unlock(fs_info); | |
533 | return 1; | |
534 | } | |
535 | ||
e9b7fd4d JS |
536 | return 0; |
537 | } | |
538 | ||
5de865ee FDBM |
539 | /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */ |
540 | static inline int tree_mod_need_log(const struct btrfs_fs_info *fs_info, | |
541 | struct extent_buffer *eb) | |
542 | { | |
543 | smp_mb(); | |
544 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
545 | return 0; | |
546 | if (eb && btrfs_header_level(eb) == 0) | |
547 | return 0; | |
548 | ||
549 | return 1; | |
550 | } | |
551 | ||
552 | static struct tree_mod_elem * | |
553 | alloc_tree_mod_elem(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) | |
5de865ee | 560 | return NULL; |
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); | |
5de865ee | 570 | RB_CLEAR_NODE(&tm->node); |
bd989ba3 | 571 | |
5de865ee | 572 | return tm; |
097b8a7c JS |
573 | } |
574 | ||
575 | static noinline int | |
c8cc6341 JB |
576 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
577 | struct extent_buffer *eb, int slot, | |
578 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 579 | { |
5de865ee FDBM |
580 | struct tree_mod_elem *tm; |
581 | int ret; | |
582 | ||
583 | if (!tree_mod_need_log(fs_info, eb)) | |
584 | return 0; | |
585 | ||
586 | tm = alloc_tree_mod_elem(eb, slot, op, flags); | |
587 | if (!tm) | |
588 | return -ENOMEM; | |
589 | ||
590 | if (tree_mod_dont_log(fs_info, eb)) { | |
591 | kfree(tm); | |
097b8a7c | 592 | return 0; |
5de865ee FDBM |
593 | } |
594 | ||
595 | ret = __tree_mod_log_insert(fs_info, tm); | |
596 | tree_mod_log_write_unlock(fs_info); | |
597 | if (ret) | |
598 | kfree(tm); | |
097b8a7c | 599 | |
5de865ee | 600 | return ret; |
097b8a7c JS |
601 | } |
602 | ||
bd989ba3 JS |
603 | static noinline int |
604 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
605 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
606 | int nr_items, gfp_t flags) | |
607 | { | |
5de865ee FDBM |
608 | struct tree_mod_elem *tm = NULL; |
609 | struct tree_mod_elem **tm_list = NULL; | |
610 | int ret = 0; | |
bd989ba3 | 611 | int i; |
5de865ee | 612 | int locked = 0; |
bd989ba3 | 613 | |
5de865ee | 614 | if (!tree_mod_need_log(fs_info, eb)) |
f395694c | 615 | return 0; |
bd989ba3 | 616 | |
5de865ee FDBM |
617 | tm_list = kzalloc(nr_items * sizeof(struct tree_mod_elem *), flags); |
618 | if (!tm_list) | |
619 | return -ENOMEM; | |
620 | ||
621 | tm = kzalloc(sizeof(*tm), flags); | |
622 | if (!tm) { | |
623 | ret = -ENOMEM; | |
624 | goto free_tms; | |
625 | } | |
626 | ||
627 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
628 | tm->slot = src_slot; | |
629 | tm->move.dst_slot = dst_slot; | |
630 | tm->move.nr_items = nr_items; | |
631 | tm->op = MOD_LOG_MOVE_KEYS; | |
632 | ||
633 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { | |
634 | tm_list[i] = alloc_tree_mod_elem(eb, i + dst_slot, | |
635 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, flags); | |
636 | if (!tm_list[i]) { | |
637 | ret = -ENOMEM; | |
638 | goto free_tms; | |
639 | } | |
640 | } | |
641 | ||
642 | if (tree_mod_dont_log(fs_info, eb)) | |
643 | goto free_tms; | |
644 | locked = 1; | |
645 | ||
01763a2e JS |
646 | /* |
647 | * When we override something during the move, we log these removals. | |
648 | * This can only happen when we move towards the beginning of the | |
649 | * buffer, i.e. dst_slot < src_slot. | |
650 | */ | |
bd989ba3 | 651 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
5de865ee FDBM |
652 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
653 | if (ret) | |
654 | goto free_tms; | |
bd989ba3 JS |
655 | } |
656 | ||
5de865ee FDBM |
657 | ret = __tree_mod_log_insert(fs_info, tm); |
658 | if (ret) | |
659 | goto free_tms; | |
660 | tree_mod_log_write_unlock(fs_info); | |
661 | kfree(tm_list); | |
f395694c | 662 | |
5de865ee FDBM |
663 | return 0; |
664 | free_tms: | |
665 | for (i = 0; i < nr_items; i++) { | |
666 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
667 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
668 | kfree(tm_list[i]); | |
669 | } | |
670 | if (locked) | |
671 | tree_mod_log_write_unlock(fs_info); | |
672 | kfree(tm_list); | |
673 | kfree(tm); | |
bd989ba3 | 674 | |
5de865ee | 675 | return ret; |
bd989ba3 JS |
676 | } |
677 | ||
5de865ee FDBM |
678 | static inline int |
679 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, | |
680 | struct tree_mod_elem **tm_list, | |
681 | int nritems) | |
097b8a7c | 682 | { |
5de865ee | 683 | int i, j; |
097b8a7c JS |
684 | int ret; |
685 | ||
097b8a7c | 686 | for (i = nritems - 1; i >= 0; i--) { |
5de865ee FDBM |
687 | ret = __tree_mod_log_insert(fs_info, tm_list[i]); |
688 | if (ret) { | |
689 | for (j = nritems - 1; j > i; j--) | |
690 | rb_erase(&tm_list[j]->node, | |
691 | &fs_info->tree_mod_log); | |
692 | return ret; | |
693 | } | |
097b8a7c | 694 | } |
5de865ee FDBM |
695 | |
696 | return 0; | |
097b8a7c JS |
697 | } |
698 | ||
bd989ba3 JS |
699 | static noinline int |
700 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
701 | struct extent_buffer *old_root, | |
90f8d62e JS |
702 | struct extent_buffer *new_root, gfp_t flags, |
703 | int log_removal) | |
bd989ba3 | 704 | { |
5de865ee FDBM |
705 | struct tree_mod_elem *tm = NULL; |
706 | struct tree_mod_elem **tm_list = NULL; | |
707 | int nritems = 0; | |
708 | int ret = 0; | |
709 | int i; | |
bd989ba3 | 710 | |
5de865ee | 711 | if (!tree_mod_need_log(fs_info, NULL)) |
097b8a7c JS |
712 | return 0; |
713 | ||
5de865ee FDBM |
714 | if (log_removal && btrfs_header_level(old_root) > 0) { |
715 | nritems = btrfs_header_nritems(old_root); | |
716 | tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *), | |
717 | flags); | |
718 | if (!tm_list) { | |
719 | ret = -ENOMEM; | |
720 | goto free_tms; | |
721 | } | |
722 | for (i = 0; i < nritems; i++) { | |
723 | tm_list[i] = alloc_tree_mod_elem(old_root, i, | |
724 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, flags); | |
725 | if (!tm_list[i]) { | |
726 | ret = -ENOMEM; | |
727 | goto free_tms; | |
728 | } | |
729 | } | |
730 | } | |
d9abbf1c | 731 | |
c8cc6341 | 732 | tm = kzalloc(sizeof(*tm), flags); |
5de865ee FDBM |
733 | if (!tm) { |
734 | ret = -ENOMEM; | |
735 | goto free_tms; | |
736 | } | |
bd989ba3 JS |
737 | |
738 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
739 | tm->old_root.logical = old_root->start; | |
740 | tm->old_root.level = btrfs_header_level(old_root); | |
741 | tm->generation = btrfs_header_generation(old_root); | |
742 | tm->op = MOD_LOG_ROOT_REPLACE; | |
743 | ||
5de865ee FDBM |
744 | if (tree_mod_dont_log(fs_info, NULL)) |
745 | goto free_tms; | |
746 | ||
747 | if (tm_list) | |
748 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
749 | if (!ret) | |
750 | ret = __tree_mod_log_insert(fs_info, tm); | |
751 | ||
752 | tree_mod_log_write_unlock(fs_info); | |
753 | if (ret) | |
754 | goto free_tms; | |
755 | kfree(tm_list); | |
756 | ||
757 | return ret; | |
758 | ||
759 | free_tms: | |
760 | if (tm_list) { | |
761 | for (i = 0; i < nritems; i++) | |
762 | kfree(tm_list[i]); | |
763 | kfree(tm_list); | |
764 | } | |
765 | kfree(tm); | |
766 | ||
767 | return ret; | |
bd989ba3 JS |
768 | } |
769 | ||
770 | static struct tree_mod_elem * | |
771 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
772 | int smallest) | |
773 | { | |
774 | struct rb_root *tm_root; | |
775 | struct rb_node *node; | |
776 | struct tree_mod_elem *cur = NULL; | |
777 | struct tree_mod_elem *found = NULL; | |
778 | u64 index = start >> PAGE_CACHE_SHIFT; | |
779 | ||
097b8a7c | 780 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
781 | tm_root = &fs_info->tree_mod_log; |
782 | node = tm_root->rb_node; | |
783 | while (node) { | |
784 | cur = container_of(node, struct tree_mod_elem, node); | |
785 | if (cur->index < index) { | |
786 | node = node->rb_left; | |
787 | } else if (cur->index > index) { | |
788 | node = node->rb_right; | |
097b8a7c | 789 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
790 | node = node->rb_left; |
791 | } else if (!smallest) { | |
792 | /* we want the node with the highest seq */ | |
793 | if (found) | |
097b8a7c | 794 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
795 | found = cur; |
796 | node = node->rb_left; | |
097b8a7c | 797 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
798 | /* we want the node with the smallest seq */ |
799 | if (found) | |
097b8a7c | 800 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
801 | found = cur; |
802 | node = node->rb_right; | |
803 | } else { | |
804 | found = cur; | |
805 | break; | |
806 | } | |
807 | } | |
097b8a7c | 808 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
809 | |
810 | return found; | |
811 | } | |
812 | ||
813 | /* | |
814 | * this returns the element from the log with the smallest time sequence | |
815 | * value that's in the log (the oldest log item). any element with a time | |
816 | * sequence lower than min_seq will be ignored. | |
817 | */ | |
818 | static struct tree_mod_elem * | |
819 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
820 | u64 min_seq) | |
821 | { | |
822 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
823 | } | |
824 | ||
825 | /* | |
826 | * this returns the element from the log with the largest time sequence | |
827 | * value that's in the log (the most recent log item). any element with | |
828 | * a time sequence lower than min_seq will be ignored. | |
829 | */ | |
830 | static struct tree_mod_elem * | |
831 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
832 | { | |
833 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
834 | } | |
835 | ||
5de865ee | 836 | static noinline int |
bd989ba3 JS |
837 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
838 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 839 | unsigned long src_offset, int nr_items) |
bd989ba3 | 840 | { |
5de865ee FDBM |
841 | int ret = 0; |
842 | struct tree_mod_elem **tm_list = NULL; | |
843 | struct tree_mod_elem **tm_list_add, **tm_list_rem; | |
bd989ba3 | 844 | int i; |
5de865ee | 845 | int locked = 0; |
bd989ba3 | 846 | |
5de865ee FDBM |
847 | if (!tree_mod_need_log(fs_info, NULL)) |
848 | return 0; | |
bd989ba3 | 849 | |
c8cc6341 | 850 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
5de865ee FDBM |
851 | return 0; |
852 | ||
853 | tm_list = kzalloc(nr_items * 2 * sizeof(struct tree_mod_elem *), | |
854 | GFP_NOFS); | |
855 | if (!tm_list) | |
856 | return -ENOMEM; | |
bd989ba3 | 857 | |
5de865ee FDBM |
858 | tm_list_add = tm_list; |
859 | tm_list_rem = tm_list + nr_items; | |
bd989ba3 | 860 | for (i = 0; i < nr_items; i++) { |
5de865ee FDBM |
861 | tm_list_rem[i] = alloc_tree_mod_elem(src, i + src_offset, |
862 | MOD_LOG_KEY_REMOVE, GFP_NOFS); | |
863 | if (!tm_list_rem[i]) { | |
864 | ret = -ENOMEM; | |
865 | goto free_tms; | |
866 | } | |
867 | ||
868 | tm_list_add[i] = alloc_tree_mod_elem(dst, i + dst_offset, | |
869 | MOD_LOG_KEY_ADD, GFP_NOFS); | |
870 | if (!tm_list_add[i]) { | |
871 | ret = -ENOMEM; | |
872 | goto free_tms; | |
873 | } | |
874 | } | |
875 | ||
876 | if (tree_mod_dont_log(fs_info, NULL)) | |
877 | goto free_tms; | |
878 | locked = 1; | |
879 | ||
880 | for (i = 0; i < nr_items; i++) { | |
881 | ret = __tree_mod_log_insert(fs_info, tm_list_rem[i]); | |
882 | if (ret) | |
883 | goto free_tms; | |
884 | ret = __tree_mod_log_insert(fs_info, tm_list_add[i]); | |
885 | if (ret) | |
886 | goto free_tms; | |
bd989ba3 | 887 | } |
5de865ee FDBM |
888 | |
889 | tree_mod_log_write_unlock(fs_info); | |
890 | kfree(tm_list); | |
891 | ||
892 | return 0; | |
893 | ||
894 | free_tms: | |
895 | for (i = 0; i < nr_items * 2; i++) { | |
896 | if (tm_list[i] && !RB_EMPTY_NODE(&tm_list[i]->node)) | |
897 | rb_erase(&tm_list[i]->node, &fs_info->tree_mod_log); | |
898 | kfree(tm_list[i]); | |
899 | } | |
900 | if (locked) | |
901 | tree_mod_log_write_unlock(fs_info); | |
902 | kfree(tm_list); | |
903 | ||
904 | return ret; | |
bd989ba3 JS |
905 | } |
906 | ||
907 | static inline void | |
908 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
909 | int dst_offset, int src_offset, int nr_items) | |
910 | { | |
911 | int ret; | |
912 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
913 | nr_items, GFP_NOFS); | |
914 | BUG_ON(ret < 0); | |
915 | } | |
916 | ||
097b8a7c | 917 | static noinline void |
bd989ba3 | 918 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 919 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
920 | { |
921 | int ret; | |
922 | ||
78357766 | 923 | ret = tree_mod_log_insert_key(fs_info, eb, slot, |
c8cc6341 JB |
924 | MOD_LOG_KEY_REPLACE, |
925 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
926 | BUG_ON(ret < 0); |
927 | } | |
928 | ||
5de865ee | 929 | static noinline int |
097b8a7c | 930 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) |
bd989ba3 | 931 | { |
5de865ee FDBM |
932 | struct tree_mod_elem **tm_list = NULL; |
933 | int nritems = 0; | |
934 | int i; | |
935 | int ret = 0; | |
936 | ||
937 | if (btrfs_header_level(eb) == 0) | |
938 | return 0; | |
939 | ||
940 | if (!tree_mod_need_log(fs_info, NULL)) | |
941 | return 0; | |
942 | ||
943 | nritems = btrfs_header_nritems(eb); | |
944 | tm_list = kzalloc(nritems * sizeof(struct tree_mod_elem *), | |
945 | GFP_NOFS); | |
946 | if (!tm_list) | |
947 | return -ENOMEM; | |
948 | ||
949 | for (i = 0; i < nritems; i++) { | |
950 | tm_list[i] = alloc_tree_mod_elem(eb, i, | |
951 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
952 | if (!tm_list[i]) { | |
953 | ret = -ENOMEM; | |
954 | goto free_tms; | |
955 | } | |
956 | } | |
957 | ||
e9b7fd4d | 958 | if (tree_mod_dont_log(fs_info, eb)) |
5de865ee FDBM |
959 | goto free_tms; |
960 | ||
961 | ret = __tree_mod_log_free_eb(fs_info, tm_list, nritems); | |
962 | tree_mod_log_write_unlock(fs_info); | |
963 | if (ret) | |
964 | goto free_tms; | |
965 | kfree(tm_list); | |
966 | ||
967 | return 0; | |
968 | ||
969 | free_tms: | |
970 | for (i = 0; i < nritems; i++) | |
971 | kfree(tm_list[i]); | |
972 | kfree(tm_list); | |
973 | ||
974 | return ret; | |
bd989ba3 JS |
975 | } |
976 | ||
097b8a7c | 977 | static noinline void |
bd989ba3 | 978 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
979 | struct extent_buffer *new_root_node, |
980 | int log_removal) | |
bd989ba3 JS |
981 | { |
982 | int ret; | |
bd989ba3 | 983 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 984 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
985 | BUG_ON(ret < 0); |
986 | } | |
987 | ||
5d4f98a2 YZ |
988 | /* |
989 | * check if the tree block can be shared by multiple trees | |
990 | */ | |
991 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
992 | struct extent_buffer *buf) | |
993 | { | |
994 | /* | |
995 | * Tree blocks not in refernece counted trees and tree roots | |
996 | * are never shared. If a block was allocated after the last | |
997 | * snapshot and the block was not allocated by tree relocation, | |
998 | * we know the block is not shared. | |
999 | */ | |
1000 | if (root->ref_cows && | |
1001 | buf != root->node && buf != root->commit_root && | |
1002 | (btrfs_header_generation(buf) <= | |
1003 | btrfs_root_last_snapshot(&root->root_item) || | |
1004 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
1005 | return 1; | |
1006 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1007 | if (root->ref_cows && | |
1008 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1009 | return 1; | |
1010 | #endif | |
1011 | return 0; | |
1012 | } | |
1013 | ||
1014 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
1015 | struct btrfs_root *root, | |
1016 | struct extent_buffer *buf, | |
f0486c68 YZ |
1017 | struct extent_buffer *cow, |
1018 | int *last_ref) | |
5d4f98a2 YZ |
1019 | { |
1020 | u64 refs; | |
1021 | u64 owner; | |
1022 | u64 flags; | |
1023 | u64 new_flags = 0; | |
1024 | int ret; | |
1025 | ||
1026 | /* | |
1027 | * Backrefs update rules: | |
1028 | * | |
1029 | * Always use full backrefs for extent pointers in tree block | |
1030 | * allocated by tree relocation. | |
1031 | * | |
1032 | * If a shared tree block is no longer referenced by its owner | |
1033 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
1034 | * use full backrefs for extent pointers in tree block. | |
1035 | * | |
1036 | * If a tree block is been relocating | |
1037 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
1038 | * use full backrefs for extent pointers in tree block. | |
1039 | * The reason for this is some operations (such as drop tree) | |
1040 | * are only allowed for blocks use full backrefs. | |
1041 | */ | |
1042 | ||
1043 | if (btrfs_block_can_be_shared(root, buf)) { | |
1044 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
1045 | btrfs_header_level(buf), 1, |
1046 | &refs, &flags); | |
be1a5564 MF |
1047 | if (ret) |
1048 | return ret; | |
e5df9573 MF |
1049 | if (refs == 0) { |
1050 | ret = -EROFS; | |
1051 | btrfs_std_error(root->fs_info, ret); | |
1052 | return ret; | |
1053 | } | |
5d4f98a2 YZ |
1054 | } else { |
1055 | refs = 1; | |
1056 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
1057 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1058 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1059 | else | |
1060 | flags = 0; | |
1061 | } | |
1062 | ||
1063 | owner = btrfs_header_owner(buf); | |
1064 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
1065 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
1066 | ||
1067 | if (refs > 1) { | |
1068 | if ((owner == root->root_key.objectid || | |
1069 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
1070 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 1071 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 1072 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1073 | |
1074 | if (root->root_key.objectid == | |
1075 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 1076 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 1077 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 1078 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 1079 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1080 | } |
1081 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
1082 | } else { | |
1083 | ||
1084 | if (root->root_key.objectid == | |
1085 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 1086 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 1087 | else |
66d7e7f0 | 1088 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 1089 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1090 | } |
1091 | if (new_flags != 0) { | |
b1c79e09 JB |
1092 | int level = btrfs_header_level(buf); |
1093 | ||
5d4f98a2 YZ |
1094 | ret = btrfs_set_disk_extent_flags(trans, root, |
1095 | buf->start, | |
1096 | buf->len, | |
b1c79e09 | 1097 | new_flags, level, 0); |
be1a5564 MF |
1098 | if (ret) |
1099 | return ret; | |
5d4f98a2 YZ |
1100 | } |
1101 | } else { | |
1102 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
1103 | if (root->root_key.objectid == | |
1104 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 1105 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 1106 | else |
66d7e7f0 | 1107 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 1108 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 1109 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 1110 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
1111 | } |
1112 | clean_tree_block(trans, root, buf); | |
f0486c68 | 1113 | *last_ref = 1; |
5d4f98a2 YZ |
1114 | } |
1115 | return 0; | |
1116 | } | |
1117 | ||
d352ac68 | 1118 | /* |
d397712b CM |
1119 | * does the dirty work in cow of a single block. The parent block (if |
1120 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
1121 | * dirty and returned locked. If you modify the block it needs to be marked | |
1122 | * dirty again. | |
d352ac68 CM |
1123 | * |
1124 | * search_start -- an allocation hint for the new block | |
1125 | * | |
d397712b CM |
1126 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
1127 | * bytes the allocator should try to find free next to the block it returns. | |
1128 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 1129 | */ |
d397712b | 1130 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1131 | struct btrfs_root *root, |
1132 | struct extent_buffer *buf, | |
1133 | struct extent_buffer *parent, int parent_slot, | |
1134 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 1135 | u64 search_start, u64 empty_size) |
02217ed2 | 1136 | { |
5d4f98a2 | 1137 | struct btrfs_disk_key disk_key; |
5f39d397 | 1138 | struct extent_buffer *cow; |
be1a5564 | 1139 | int level, ret; |
f0486c68 | 1140 | int last_ref = 0; |
925baedd | 1141 | int unlock_orig = 0; |
5d4f98a2 | 1142 | u64 parent_start; |
7bb86316 | 1143 | |
925baedd CM |
1144 | if (*cow_ret == buf) |
1145 | unlock_orig = 1; | |
1146 | ||
b9447ef8 | 1147 | btrfs_assert_tree_locked(buf); |
925baedd | 1148 | |
7bb86316 CM |
1149 | WARN_ON(root->ref_cows && trans->transid != |
1150 | root->fs_info->running_transaction->transid); | |
6702ed49 | 1151 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 1152 | |
7bb86316 | 1153 | level = btrfs_header_level(buf); |
31840ae1 | 1154 | |
5d4f98a2 YZ |
1155 | if (level == 0) |
1156 | btrfs_item_key(buf, &disk_key, 0); | |
1157 | else | |
1158 | btrfs_node_key(buf, &disk_key, 0); | |
1159 | ||
1160 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
1161 | if (parent) | |
1162 | parent_start = parent->start; | |
1163 | else | |
1164 | parent_start = 0; | |
1165 | } else | |
1166 | parent_start = 0; | |
1167 | ||
1168 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
1169 | root->root_key.objectid, &disk_key, | |
5581a51a | 1170 | level, search_start, empty_size); |
54aa1f4d CM |
1171 | if (IS_ERR(cow)) |
1172 | return PTR_ERR(cow); | |
6702ed49 | 1173 | |
b4ce94de CM |
1174 | /* cow is set to blocking by btrfs_init_new_buffer */ |
1175 | ||
5f39d397 | 1176 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 1177 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 1178 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
1179 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
1180 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1181 | BTRFS_HEADER_FLAG_RELOC); | |
1182 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1183 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1184 | else | |
1185 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1186 | |
0a4e5586 | 1187 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
1188 | BTRFS_FSID_SIZE); |
1189 | ||
be1a5564 | 1190 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1191 | if (ret) { |
79787eaa | 1192 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1193 | return ret; |
1194 | } | |
1a40e23b | 1195 | |
83d4cfd4 JB |
1196 | if (root->ref_cows) { |
1197 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); | |
1198 | if (ret) | |
1199 | return ret; | |
1200 | } | |
3fd0a558 | 1201 | |
02217ed2 | 1202 | if (buf == root->node) { |
925baedd | 1203 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1204 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1205 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1206 | parent_start = buf->start; | |
1207 | else | |
1208 | parent_start = 0; | |
925baedd | 1209 | |
5f39d397 | 1210 | extent_buffer_get(cow); |
90f8d62e | 1211 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1212 | rcu_assign_pointer(root->node, cow); |
925baedd | 1213 | |
f0486c68 | 1214 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1215 | last_ref); |
5f39d397 | 1216 | free_extent_buffer(buf); |
0b86a832 | 1217 | add_root_to_dirty_list(root); |
02217ed2 | 1218 | } else { |
5d4f98a2 YZ |
1219 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1220 | parent_start = parent->start; | |
1221 | else | |
1222 | parent_start = 0; | |
1223 | ||
1224 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1225 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1226 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1227 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1228 | cow->start); |
74493f7a CM |
1229 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1230 | trans->transid); | |
d6025579 | 1231 | btrfs_mark_buffer_dirty(parent); |
5de865ee FDBM |
1232 | if (last_ref) { |
1233 | ret = tree_mod_log_free_eb(root->fs_info, buf); | |
1234 | if (ret) { | |
1235 | btrfs_abort_transaction(trans, root, ret); | |
1236 | return ret; | |
1237 | } | |
1238 | } | |
f0486c68 | 1239 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1240 | last_ref); |
02217ed2 | 1241 | } |
925baedd CM |
1242 | if (unlock_orig) |
1243 | btrfs_tree_unlock(buf); | |
3083ee2e | 1244 | free_extent_buffer_stale(buf); |
ccd467d6 | 1245 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1246 | *cow_ret = cow; |
02217ed2 CM |
1247 | return 0; |
1248 | } | |
1249 | ||
5d9e75c4 JS |
1250 | /* |
1251 | * returns the logical address of the oldest predecessor of the given root. | |
1252 | * entries older than time_seq are ignored. | |
1253 | */ | |
1254 | static struct tree_mod_elem * | |
1255 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1256 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1257 | { |
1258 | struct tree_mod_elem *tm; | |
1259 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1260 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1261 | int looped = 0; |
1262 | ||
1263 | if (!time_seq) | |
35a3621b | 1264 | return NULL; |
5d9e75c4 JS |
1265 | |
1266 | /* | |
1267 | * the very last operation that's logged for a root is the replacement | |
1268 | * operation (if it is replaced at all). this has the index of the *new* | |
1269 | * root, making it the very first operation that's logged for this root. | |
1270 | */ | |
1271 | while (1) { | |
1272 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1273 | time_seq); | |
1274 | if (!looped && !tm) | |
35a3621b | 1275 | return NULL; |
5d9e75c4 | 1276 | /* |
28da9fb4 JS |
1277 | * if there are no tree operation for the oldest root, we simply |
1278 | * return it. this should only happen if that (old) root is at | |
1279 | * level 0. | |
5d9e75c4 | 1280 | */ |
28da9fb4 JS |
1281 | if (!tm) |
1282 | break; | |
5d9e75c4 | 1283 | |
28da9fb4 JS |
1284 | /* |
1285 | * if there's an operation that's not a root replacement, we | |
1286 | * found the oldest version of our root. normally, we'll find a | |
1287 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1288 | */ | |
5d9e75c4 JS |
1289 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1290 | break; | |
1291 | ||
1292 | found = tm; | |
1293 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1294 | looped = 1; |
1295 | } | |
1296 | ||
a95236d9 JS |
1297 | /* if there's no old root to return, return what we found instead */ |
1298 | if (!found) | |
1299 | found = tm; | |
1300 | ||
5d9e75c4 JS |
1301 | return found; |
1302 | } | |
1303 | ||
1304 | /* | |
1305 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1306 | * previous operations will be rewinded (until we reach something older than | |
1307 | * time_seq). | |
1308 | */ | |
1309 | static void | |
f1ca7e98 JB |
1310 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1311 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1312 | { |
1313 | u32 n; | |
1314 | struct rb_node *next; | |
1315 | struct tree_mod_elem *tm = first_tm; | |
1316 | unsigned long o_dst; | |
1317 | unsigned long o_src; | |
1318 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1319 | ||
1320 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1321 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1322 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1323 | /* |
1324 | * all the operations are recorded with the operator used for | |
1325 | * the modification. as we're going backwards, we do the | |
1326 | * opposite of each operation here. | |
1327 | */ | |
1328 | switch (tm->op) { | |
1329 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1330 | BUG_ON(tm->slot < n); | |
1c697d4a | 1331 | /* Fallthrough */ |
95c80bb1 | 1332 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1333 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1334 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1335 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1336 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1337 | tm->generation); | |
4c3e6969 | 1338 | n++; |
5d9e75c4 JS |
1339 | break; |
1340 | case MOD_LOG_KEY_REPLACE: | |
1341 | BUG_ON(tm->slot >= n); | |
1342 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1343 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1344 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1345 | tm->generation); | |
1346 | break; | |
1347 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1348 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1349 | n--; |
1350 | break; | |
1351 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1352 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1353 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1354 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1355 | tm->move.nr_items * p_size); |
1356 | break; | |
1357 | case MOD_LOG_ROOT_REPLACE: | |
1358 | /* | |
1359 | * this operation is special. for roots, this must be | |
1360 | * handled explicitly before rewinding. | |
1361 | * for non-roots, this operation may exist if the node | |
1362 | * was a root: root A -> child B; then A gets empty and | |
1363 | * B is promoted to the new root. in the mod log, we'll | |
1364 | * have a root-replace operation for B, a tree block | |
1365 | * that is no root. we simply ignore that operation. | |
1366 | */ | |
1367 | break; | |
1368 | } | |
1369 | next = rb_next(&tm->node); | |
1370 | if (!next) | |
1371 | break; | |
1372 | tm = container_of(next, struct tree_mod_elem, node); | |
1373 | if (tm->index != first_tm->index) | |
1374 | break; | |
1375 | } | |
f1ca7e98 | 1376 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1377 | btrfs_set_header_nritems(eb, n); |
1378 | } | |
1379 | ||
47fb091f JS |
1380 | /* |
1381 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1382 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1383 | * returned buffer is always read-locked. If the returned buffer is not the | |
1384 | * input buffer, the lock on the input buffer is released and the input buffer | |
1385 | * is freed (its refcount is decremented). | |
1386 | */ | |
5d9e75c4 | 1387 | static struct extent_buffer * |
9ec72677 JB |
1388 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1389 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1390 | { |
1391 | struct extent_buffer *eb_rewin; | |
1392 | struct tree_mod_elem *tm; | |
1393 | ||
1394 | if (!time_seq) | |
1395 | return eb; | |
1396 | ||
1397 | if (btrfs_header_level(eb) == 0) | |
1398 | return eb; | |
1399 | ||
1400 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1401 | if (!tm) | |
1402 | return eb; | |
1403 | ||
9ec72677 JB |
1404 | btrfs_set_path_blocking(path); |
1405 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1406 | ||
5d9e75c4 JS |
1407 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1408 | BUG_ON(tm->slot != 0); | |
1409 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1410 | fs_info->tree_root->nodesize); | |
db7f3436 | 1411 | if (!eb_rewin) { |
9ec72677 | 1412 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1413 | free_extent_buffer(eb); |
1414 | return NULL; | |
1415 | } | |
5d9e75c4 JS |
1416 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1417 | btrfs_set_header_backref_rev(eb_rewin, | |
1418 | btrfs_header_backref_rev(eb)); | |
1419 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1420 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1421 | } else { |
1422 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1423 | if (!eb_rewin) { |
9ec72677 | 1424 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1425 | free_extent_buffer(eb); |
1426 | return NULL; | |
1427 | } | |
5d9e75c4 JS |
1428 | } |
1429 | ||
9ec72677 JB |
1430 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1431 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1432 | free_extent_buffer(eb); |
1433 | ||
47fb091f JS |
1434 | extent_buffer_get(eb_rewin); |
1435 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1436 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1437 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1438 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1439 | |
1440 | return eb_rewin; | |
1441 | } | |
1442 | ||
8ba97a15 JS |
1443 | /* |
1444 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1445 | * value. If there are no changes, the current root->root_node is returned. If | |
1446 | * anything changed in between, there's a fresh buffer allocated on which the | |
1447 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1448 | * Returns NULL on error (with no locks held). | |
1449 | */ | |
5d9e75c4 JS |
1450 | static inline struct extent_buffer * |
1451 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1452 | { | |
1453 | struct tree_mod_elem *tm; | |
30b0463a JS |
1454 | struct extent_buffer *eb = NULL; |
1455 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1456 | struct extent_buffer *old; |
a95236d9 | 1457 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1458 | u64 old_generation = 0; |
a95236d9 | 1459 | u64 logical; |
834328a8 | 1460 | u32 blocksize; |
5d9e75c4 | 1461 | |
30b0463a JS |
1462 | eb_root = btrfs_read_lock_root_node(root); |
1463 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1464 | if (!tm) |
30b0463a | 1465 | return eb_root; |
5d9e75c4 | 1466 | |
a95236d9 JS |
1467 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1468 | old_root = &tm->old_root; | |
1469 | old_generation = tm->generation; | |
1470 | logical = old_root->logical; | |
1471 | } else { | |
30b0463a | 1472 | logical = eb_root->start; |
a95236d9 | 1473 | } |
5d9e75c4 | 1474 | |
a95236d9 | 1475 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1476 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1477 | btrfs_tree_read_unlock(eb_root); |
1478 | free_extent_buffer(eb_root); | |
834328a8 | 1479 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1480 | old = read_tree_block(root, logical, blocksize, 0); |
fae7f21c | 1481 | if (WARN_ON(!old || !extent_buffer_uptodate(old))) { |
416bc658 | 1482 | free_extent_buffer(old); |
efe120a0 FH |
1483 | btrfs_warn(root->fs_info, |
1484 | "failed to read tree block %llu from get_old_root", logical); | |
834328a8 | 1485 | } else { |
7bfdcf7f LB |
1486 | eb = btrfs_clone_extent_buffer(old); |
1487 | free_extent_buffer(old); | |
834328a8 JS |
1488 | } |
1489 | } else if (old_root) { | |
30b0463a JS |
1490 | btrfs_tree_read_unlock(eb_root); |
1491 | free_extent_buffer(eb_root); | |
28da9fb4 | 1492 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1493 | } else { |
9ec72677 | 1494 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1495 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1496 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1497 | free_extent_buffer(eb_root); |
834328a8 JS |
1498 | } |
1499 | ||
8ba97a15 JS |
1500 | if (!eb) |
1501 | return NULL; | |
d6381084 | 1502 | extent_buffer_get(eb); |
8ba97a15 | 1503 | btrfs_tree_read_lock(eb); |
a95236d9 | 1504 | if (old_root) { |
5d9e75c4 JS |
1505 | btrfs_set_header_bytenr(eb, eb->start); |
1506 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1507 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1508 | btrfs_set_header_level(eb, old_root->level); |
1509 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1510 | } |
28da9fb4 | 1511 | if (tm) |
f1ca7e98 | 1512 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1513 | else |
1514 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1515 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1516 | |
1517 | return eb; | |
1518 | } | |
1519 | ||
5b6602e7 JS |
1520 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1521 | { | |
1522 | struct tree_mod_elem *tm; | |
1523 | int level; | |
30b0463a | 1524 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1525 | |
30b0463a | 1526 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1527 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1528 | level = tm->old_root.level; | |
1529 | } else { | |
30b0463a | 1530 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1531 | } |
30b0463a | 1532 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1533 | |
1534 | return level; | |
1535 | } | |
1536 | ||
5d4f98a2 YZ |
1537 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1538 | struct btrfs_root *root, | |
1539 | struct extent_buffer *buf) | |
1540 | { | |
f1ebcc74 LB |
1541 | /* ensure we can see the force_cow */ |
1542 | smp_rmb(); | |
1543 | ||
1544 | /* | |
1545 | * We do not need to cow a block if | |
1546 | * 1) this block is not created or changed in this transaction; | |
1547 | * 2) this block does not belong to TREE_RELOC tree; | |
1548 | * 3) the root is not forced COW. | |
1549 | * | |
1550 | * What is forced COW: | |
1551 | * when we create snapshot during commiting the transaction, | |
1552 | * after we've finished coping src root, we must COW the shared | |
1553 | * block to ensure the metadata consistency. | |
1554 | */ | |
5d4f98a2 YZ |
1555 | if (btrfs_header_generation(buf) == trans->transid && |
1556 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1557 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1558 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1559 | !root->force_cow) | |
5d4f98a2 YZ |
1560 | return 0; |
1561 | return 1; | |
1562 | } | |
1563 | ||
d352ac68 CM |
1564 | /* |
1565 | * cows a single block, see __btrfs_cow_block for the real work. | |
1566 | * This version of it has extra checks so that a block isn't cow'd more than | |
1567 | * once per transaction, as long as it hasn't been written yet | |
1568 | */ | |
d397712b | 1569 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1570 | struct btrfs_root *root, struct extent_buffer *buf, |
1571 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1572 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1573 | { |
1574 | u64 search_start; | |
f510cfec | 1575 | int ret; |
dc17ff8f | 1576 | |
31b1a2bd JL |
1577 | if (trans->transaction != root->fs_info->running_transaction) |
1578 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1579 | trans->transid, |
6702ed49 | 1580 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1581 | |
1582 | if (trans->transid != root->fs_info->generation) | |
1583 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1584 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1585 | |
5d4f98a2 | 1586 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1587 | *cow_ret = buf; |
1588 | return 0; | |
1589 | } | |
c487685d | 1590 | |
0b86a832 | 1591 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1592 | |
1593 | if (parent) | |
1594 | btrfs_set_lock_blocking(parent); | |
1595 | btrfs_set_lock_blocking(buf); | |
1596 | ||
f510cfec | 1597 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1598 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1599 | |
1600 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1601 | ||
f510cfec | 1602 | return ret; |
6702ed49 CM |
1603 | } |
1604 | ||
d352ac68 CM |
1605 | /* |
1606 | * helper function for defrag to decide if two blocks pointed to by a | |
1607 | * node are actually close by | |
1608 | */ | |
6b80053d | 1609 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1610 | { |
6b80053d | 1611 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1612 | return 1; |
6b80053d | 1613 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1614 | return 1; |
1615 | return 0; | |
1616 | } | |
1617 | ||
081e9573 CM |
1618 | /* |
1619 | * compare two keys in a memcmp fashion | |
1620 | */ | |
1621 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1622 | { | |
1623 | struct btrfs_key k1; | |
1624 | ||
1625 | btrfs_disk_key_to_cpu(&k1, disk); | |
1626 | ||
20736aba | 1627 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1628 | } |
1629 | ||
f3465ca4 JB |
1630 | /* |
1631 | * same as comp_keys only with two btrfs_key's | |
1632 | */ | |
5d4f98a2 | 1633 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1634 | { |
1635 | if (k1->objectid > k2->objectid) | |
1636 | return 1; | |
1637 | if (k1->objectid < k2->objectid) | |
1638 | return -1; | |
1639 | if (k1->type > k2->type) | |
1640 | return 1; | |
1641 | if (k1->type < k2->type) | |
1642 | return -1; | |
1643 | if (k1->offset > k2->offset) | |
1644 | return 1; | |
1645 | if (k1->offset < k2->offset) | |
1646 | return -1; | |
1647 | return 0; | |
1648 | } | |
081e9573 | 1649 | |
d352ac68 CM |
1650 | /* |
1651 | * this is used by the defrag code to go through all the | |
1652 | * leaves pointed to by a node and reallocate them so that | |
1653 | * disk order is close to key order | |
1654 | */ | |
6702ed49 | 1655 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1656 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1657 | int start_slot, u64 *last_ret, |
a6b6e75e | 1658 | struct btrfs_key *progress) |
6702ed49 | 1659 | { |
6b80053d | 1660 | struct extent_buffer *cur; |
6702ed49 | 1661 | u64 blocknr; |
ca7a79ad | 1662 | u64 gen; |
e9d0b13b CM |
1663 | u64 search_start = *last_ret; |
1664 | u64 last_block = 0; | |
6702ed49 CM |
1665 | u64 other; |
1666 | u32 parent_nritems; | |
6702ed49 CM |
1667 | int end_slot; |
1668 | int i; | |
1669 | int err = 0; | |
f2183bde | 1670 | int parent_level; |
6b80053d CM |
1671 | int uptodate; |
1672 | u32 blocksize; | |
081e9573 CM |
1673 | int progress_passed = 0; |
1674 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1675 | |
5708b959 | 1676 | parent_level = btrfs_header_level(parent); |
5708b959 | 1677 | |
6c1500f2 JL |
1678 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1679 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1680 | |
6b80053d | 1681 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1682 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1683 | end_slot = parent_nritems; |
1684 | ||
1685 | if (parent_nritems == 1) | |
1686 | return 0; | |
1687 | ||
b4ce94de CM |
1688 | btrfs_set_lock_blocking(parent); |
1689 | ||
6702ed49 CM |
1690 | for (i = start_slot; i < end_slot; i++) { |
1691 | int close = 1; | |
a6b6e75e | 1692 | |
081e9573 CM |
1693 | btrfs_node_key(parent, &disk_key, i); |
1694 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1695 | continue; | |
1696 | ||
1697 | progress_passed = 1; | |
6b80053d | 1698 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1699 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1700 | if (last_block == 0) |
1701 | last_block = blocknr; | |
5708b959 | 1702 | |
6702ed49 | 1703 | if (i > 0) { |
6b80053d CM |
1704 | other = btrfs_node_blockptr(parent, i - 1); |
1705 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1706 | } |
0ef3e66b | 1707 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1708 | other = btrfs_node_blockptr(parent, i + 1); |
1709 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1710 | } |
e9d0b13b CM |
1711 | if (close) { |
1712 | last_block = blocknr; | |
6702ed49 | 1713 | continue; |
e9d0b13b | 1714 | } |
6702ed49 | 1715 | |
6b80053d CM |
1716 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1717 | if (cur) | |
b9fab919 | 1718 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1719 | else |
1720 | uptodate = 0; | |
5708b959 | 1721 | if (!cur || !uptodate) { |
6b80053d CM |
1722 | if (!cur) { |
1723 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1724 | blocksize, gen); |
416bc658 JB |
1725 | if (!cur || !extent_buffer_uptodate(cur)) { |
1726 | free_extent_buffer(cur); | |
97d9a8a4 | 1727 | return -EIO; |
416bc658 | 1728 | } |
6b80053d | 1729 | } else if (!uptodate) { |
018642a1 TI |
1730 | err = btrfs_read_buffer(cur, gen); |
1731 | if (err) { | |
1732 | free_extent_buffer(cur); | |
1733 | return err; | |
1734 | } | |
f2183bde | 1735 | } |
6702ed49 | 1736 | } |
e9d0b13b | 1737 | if (search_start == 0) |
6b80053d | 1738 | search_start = last_block; |
e9d0b13b | 1739 | |
e7a84565 | 1740 | btrfs_tree_lock(cur); |
b4ce94de | 1741 | btrfs_set_lock_blocking(cur); |
6b80053d | 1742 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1743 | &cur, search_start, |
6b80053d | 1744 | min(16 * blocksize, |
9fa8cfe7 | 1745 | (end_slot - i) * blocksize)); |
252c38f0 | 1746 | if (err) { |
e7a84565 | 1747 | btrfs_tree_unlock(cur); |
6b80053d | 1748 | free_extent_buffer(cur); |
6702ed49 | 1749 | break; |
252c38f0 | 1750 | } |
e7a84565 CM |
1751 | search_start = cur->start; |
1752 | last_block = cur->start; | |
f2183bde | 1753 | *last_ret = search_start; |
e7a84565 CM |
1754 | btrfs_tree_unlock(cur); |
1755 | free_extent_buffer(cur); | |
6702ed49 CM |
1756 | } |
1757 | return err; | |
1758 | } | |
1759 | ||
74123bd7 CM |
1760 | /* |
1761 | * The leaf data grows from end-to-front in the node. | |
1762 | * this returns the address of the start of the last item, | |
1763 | * which is the stop of the leaf data stack | |
1764 | */ | |
123abc88 | 1765 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1766 | struct extent_buffer *leaf) |
be0e5c09 | 1767 | { |
5f39d397 | 1768 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1769 | if (nr == 0) |
123abc88 | 1770 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1771 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1772 | } |
1773 | ||
aa5d6bed | 1774 | |
74123bd7 | 1775 | /* |
5f39d397 CM |
1776 | * search for key in the extent_buffer. The items start at offset p, |
1777 | * and they are item_size apart. There are 'max' items in p. | |
1778 | * | |
74123bd7 CM |
1779 | * the slot in the array is returned via slot, and it points to |
1780 | * the place where you would insert key if it is not found in | |
1781 | * the array. | |
1782 | * | |
1783 | * slot may point to max if the key is bigger than all of the keys | |
1784 | */ | |
e02119d5 CM |
1785 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1786 | unsigned long p, | |
1787 | int item_size, struct btrfs_key *key, | |
1788 | int max, int *slot) | |
be0e5c09 CM |
1789 | { |
1790 | int low = 0; | |
1791 | int high = max; | |
1792 | int mid; | |
1793 | int ret; | |
479965d6 | 1794 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1795 | struct btrfs_disk_key unaligned; |
1796 | unsigned long offset; | |
5f39d397 CM |
1797 | char *kaddr = NULL; |
1798 | unsigned long map_start = 0; | |
1799 | unsigned long map_len = 0; | |
479965d6 | 1800 | int err; |
be0e5c09 | 1801 | |
d397712b | 1802 | while (low < high) { |
be0e5c09 | 1803 | mid = (low + high) / 2; |
5f39d397 CM |
1804 | offset = p + mid * item_size; |
1805 | ||
a6591715 | 1806 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1807 | (offset + sizeof(struct btrfs_disk_key)) > |
1808 | map_start + map_len) { | |
934d375b CM |
1809 | |
1810 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1811 | sizeof(struct btrfs_disk_key), |
a6591715 | 1812 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1813 | |
1814 | if (!err) { | |
1815 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1816 | map_start); | |
1817 | } else { | |
1818 | read_extent_buffer(eb, &unaligned, | |
1819 | offset, sizeof(unaligned)); | |
1820 | tmp = &unaligned; | |
1821 | } | |
5f39d397 | 1822 | |
5f39d397 CM |
1823 | } else { |
1824 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1825 | map_start); | |
1826 | } | |
be0e5c09 CM |
1827 | ret = comp_keys(tmp, key); |
1828 | ||
1829 | if (ret < 0) | |
1830 | low = mid + 1; | |
1831 | else if (ret > 0) | |
1832 | high = mid; | |
1833 | else { | |
1834 | *slot = mid; | |
1835 | return 0; | |
1836 | } | |
1837 | } | |
1838 | *slot = low; | |
1839 | return 1; | |
1840 | } | |
1841 | ||
97571fd0 CM |
1842 | /* |
1843 | * simple bin_search frontend that does the right thing for | |
1844 | * leaves vs nodes | |
1845 | */ | |
5f39d397 CM |
1846 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1847 | int level, int *slot) | |
be0e5c09 | 1848 | { |
f775738f | 1849 | if (level == 0) |
5f39d397 CM |
1850 | return generic_bin_search(eb, |
1851 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1852 | sizeof(struct btrfs_item), |
5f39d397 | 1853 | key, btrfs_header_nritems(eb), |
7518a238 | 1854 | slot); |
f775738f | 1855 | else |
5f39d397 CM |
1856 | return generic_bin_search(eb, |
1857 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1858 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1859 | key, btrfs_header_nritems(eb), |
7518a238 | 1860 | slot); |
be0e5c09 CM |
1861 | } |
1862 | ||
5d4f98a2 YZ |
1863 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1864 | int level, int *slot) | |
1865 | { | |
1866 | return bin_search(eb, key, level, slot); | |
1867 | } | |
1868 | ||
f0486c68 YZ |
1869 | static void root_add_used(struct btrfs_root *root, u32 size) |
1870 | { | |
1871 | spin_lock(&root->accounting_lock); | |
1872 | btrfs_set_root_used(&root->root_item, | |
1873 | btrfs_root_used(&root->root_item) + size); | |
1874 | spin_unlock(&root->accounting_lock); | |
1875 | } | |
1876 | ||
1877 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1878 | { | |
1879 | spin_lock(&root->accounting_lock); | |
1880 | btrfs_set_root_used(&root->root_item, | |
1881 | btrfs_root_used(&root->root_item) - size); | |
1882 | spin_unlock(&root->accounting_lock); | |
1883 | } | |
1884 | ||
d352ac68 CM |
1885 | /* given a node and slot number, this reads the blocks it points to. The |
1886 | * extent buffer is returned with a reference taken (but unlocked). | |
1887 | * NULL is returned on error. | |
1888 | */ | |
e02119d5 | 1889 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1890 | struct extent_buffer *parent, int slot) |
bb803951 | 1891 | { |
ca7a79ad | 1892 | int level = btrfs_header_level(parent); |
416bc658 JB |
1893 | struct extent_buffer *eb; |
1894 | ||
bb803951 CM |
1895 | if (slot < 0) |
1896 | return NULL; | |
5f39d397 | 1897 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1898 | return NULL; |
ca7a79ad CM |
1899 | |
1900 | BUG_ON(level == 0); | |
1901 | ||
416bc658 JB |
1902 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1903 | btrfs_level_size(root, level - 1), | |
1904 | btrfs_node_ptr_generation(parent, slot)); | |
1905 | if (eb && !extent_buffer_uptodate(eb)) { | |
1906 | free_extent_buffer(eb); | |
1907 | eb = NULL; | |
1908 | } | |
1909 | ||
1910 | return eb; | |
bb803951 CM |
1911 | } |
1912 | ||
d352ac68 CM |
1913 | /* |
1914 | * node level balancing, used to make sure nodes are in proper order for | |
1915 | * item deletion. We balance from the top down, so we have to make sure | |
1916 | * that a deletion won't leave an node completely empty later on. | |
1917 | */ | |
e02119d5 | 1918 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1919 | struct btrfs_root *root, |
1920 | struct btrfs_path *path, int level) | |
bb803951 | 1921 | { |
5f39d397 CM |
1922 | struct extent_buffer *right = NULL; |
1923 | struct extent_buffer *mid; | |
1924 | struct extent_buffer *left = NULL; | |
1925 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1926 | int ret = 0; |
1927 | int wret; | |
1928 | int pslot; | |
bb803951 | 1929 | int orig_slot = path->slots[level]; |
79f95c82 | 1930 | u64 orig_ptr; |
bb803951 CM |
1931 | |
1932 | if (level == 0) | |
1933 | return 0; | |
1934 | ||
5f39d397 | 1935 | mid = path->nodes[level]; |
b4ce94de | 1936 | |
bd681513 CM |
1937 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1938 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1939 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1940 | ||
1d4f8a0c | 1941 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1942 | |
a05a9bb1 | 1943 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1944 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1945 | pslot = path->slots[level + 1]; |
1946 | } | |
bb803951 | 1947 | |
40689478 CM |
1948 | /* |
1949 | * deal with the case where there is only one pointer in the root | |
1950 | * by promoting the node below to a root | |
1951 | */ | |
5f39d397 CM |
1952 | if (!parent) { |
1953 | struct extent_buffer *child; | |
bb803951 | 1954 | |
5f39d397 | 1955 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1956 | return 0; |
1957 | ||
1958 | /* promote the child to a root */ | |
5f39d397 | 1959 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1960 | if (!child) { |
1961 | ret = -EROFS; | |
1962 | btrfs_std_error(root->fs_info, ret); | |
1963 | goto enospc; | |
1964 | } | |
1965 | ||
925baedd | 1966 | btrfs_tree_lock(child); |
b4ce94de | 1967 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1968 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1969 | if (ret) { |
1970 | btrfs_tree_unlock(child); | |
1971 | free_extent_buffer(child); | |
1972 | goto enospc; | |
1973 | } | |
2f375ab9 | 1974 | |
90f8d62e | 1975 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1976 | rcu_assign_pointer(root->node, child); |
925baedd | 1977 | |
0b86a832 | 1978 | add_root_to_dirty_list(root); |
925baedd | 1979 | btrfs_tree_unlock(child); |
b4ce94de | 1980 | |
925baedd | 1981 | path->locks[level] = 0; |
bb803951 | 1982 | path->nodes[level] = NULL; |
5f39d397 | 1983 | clean_tree_block(trans, root, mid); |
925baedd | 1984 | btrfs_tree_unlock(mid); |
bb803951 | 1985 | /* once for the path */ |
5f39d397 | 1986 | free_extent_buffer(mid); |
f0486c68 YZ |
1987 | |
1988 | root_sub_used(root, mid->len); | |
5581a51a | 1989 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1990 | /* once for the root ptr */ |
3083ee2e | 1991 | free_extent_buffer_stale(mid); |
f0486c68 | 1992 | return 0; |
bb803951 | 1993 | } |
5f39d397 | 1994 | if (btrfs_header_nritems(mid) > |
123abc88 | 1995 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1996 | return 0; |
1997 | ||
5f39d397 CM |
1998 | left = read_node_slot(root, parent, pslot - 1); |
1999 | if (left) { | |
925baedd | 2000 | btrfs_tree_lock(left); |
b4ce94de | 2001 | btrfs_set_lock_blocking(left); |
5f39d397 | 2002 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 2003 | parent, pslot - 1, &left); |
54aa1f4d CM |
2004 | if (wret) { |
2005 | ret = wret; | |
2006 | goto enospc; | |
2007 | } | |
2cc58cf2 | 2008 | } |
5f39d397 CM |
2009 | right = read_node_slot(root, parent, pslot + 1); |
2010 | if (right) { | |
925baedd | 2011 | btrfs_tree_lock(right); |
b4ce94de | 2012 | btrfs_set_lock_blocking(right); |
5f39d397 | 2013 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 2014 | parent, pslot + 1, &right); |
2cc58cf2 CM |
2015 | if (wret) { |
2016 | ret = wret; | |
2017 | goto enospc; | |
2018 | } | |
2019 | } | |
2020 | ||
2021 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
2022 | if (left) { |
2023 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 2024 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
2025 | if (wret < 0) |
2026 | ret = wret; | |
bb803951 | 2027 | } |
79f95c82 CM |
2028 | |
2029 | /* | |
2030 | * then try to empty the right most buffer into the middle | |
2031 | */ | |
5f39d397 | 2032 | if (right) { |
971a1f66 | 2033 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 2034 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 2035 | ret = wret; |
5f39d397 | 2036 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 2037 | clean_tree_block(trans, root, right); |
925baedd | 2038 | btrfs_tree_unlock(right); |
afe5fea7 | 2039 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 2040 | root_sub_used(root, right->len); |
5581a51a | 2041 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 2042 | free_extent_buffer_stale(right); |
f0486c68 | 2043 | right = NULL; |
bb803951 | 2044 | } else { |
5f39d397 CM |
2045 | struct btrfs_disk_key right_key; |
2046 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 2047 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2048 | pslot + 1, 0); |
5f39d397 CM |
2049 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
2050 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
2051 | } |
2052 | } | |
5f39d397 | 2053 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
2054 | /* |
2055 | * we're not allowed to leave a node with one item in the | |
2056 | * tree during a delete. A deletion from lower in the tree | |
2057 | * could try to delete the only pointer in this node. | |
2058 | * So, pull some keys from the left. | |
2059 | * There has to be a left pointer at this point because | |
2060 | * otherwise we would have pulled some pointers from the | |
2061 | * right | |
2062 | */ | |
305a26af MF |
2063 | if (!left) { |
2064 | ret = -EROFS; | |
2065 | btrfs_std_error(root->fs_info, ret); | |
2066 | goto enospc; | |
2067 | } | |
5f39d397 | 2068 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 2069 | if (wret < 0) { |
79f95c82 | 2070 | ret = wret; |
54aa1f4d CM |
2071 | goto enospc; |
2072 | } | |
bce4eae9 CM |
2073 | if (wret == 1) { |
2074 | wret = push_node_left(trans, root, left, mid, 1); | |
2075 | if (wret < 0) | |
2076 | ret = wret; | |
2077 | } | |
79f95c82 CM |
2078 | BUG_ON(wret == 1); |
2079 | } | |
5f39d397 | 2080 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 2081 | clean_tree_block(trans, root, mid); |
925baedd | 2082 | btrfs_tree_unlock(mid); |
afe5fea7 | 2083 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 2084 | root_sub_used(root, mid->len); |
5581a51a | 2085 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 2086 | free_extent_buffer_stale(mid); |
f0486c68 | 2087 | mid = NULL; |
79f95c82 CM |
2088 | } else { |
2089 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
2090 | struct btrfs_disk_key mid_key; |
2091 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 2092 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 2093 | pslot, 0); |
5f39d397 CM |
2094 | btrfs_set_node_key(parent, &mid_key, pslot); |
2095 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 2096 | } |
bb803951 | 2097 | |
79f95c82 | 2098 | /* update the path */ |
5f39d397 CM |
2099 | if (left) { |
2100 | if (btrfs_header_nritems(left) > orig_slot) { | |
2101 | extent_buffer_get(left); | |
925baedd | 2102 | /* left was locked after cow */ |
5f39d397 | 2103 | path->nodes[level] = left; |
bb803951 CM |
2104 | path->slots[level + 1] -= 1; |
2105 | path->slots[level] = orig_slot; | |
925baedd CM |
2106 | if (mid) { |
2107 | btrfs_tree_unlock(mid); | |
5f39d397 | 2108 | free_extent_buffer(mid); |
925baedd | 2109 | } |
bb803951 | 2110 | } else { |
5f39d397 | 2111 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
2112 | path->slots[level] = orig_slot; |
2113 | } | |
2114 | } | |
79f95c82 | 2115 | /* double check we haven't messed things up */ |
e20d96d6 | 2116 | if (orig_ptr != |
5f39d397 | 2117 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 2118 | BUG(); |
54aa1f4d | 2119 | enospc: |
925baedd CM |
2120 | if (right) { |
2121 | btrfs_tree_unlock(right); | |
5f39d397 | 2122 | free_extent_buffer(right); |
925baedd CM |
2123 | } |
2124 | if (left) { | |
2125 | if (path->nodes[level] != left) | |
2126 | btrfs_tree_unlock(left); | |
5f39d397 | 2127 | free_extent_buffer(left); |
925baedd | 2128 | } |
bb803951 CM |
2129 | return ret; |
2130 | } | |
2131 | ||
d352ac68 CM |
2132 | /* Node balancing for insertion. Here we only split or push nodes around |
2133 | * when they are completely full. This is also done top down, so we | |
2134 | * have to be pessimistic. | |
2135 | */ | |
d397712b | 2136 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
2137 | struct btrfs_root *root, |
2138 | struct btrfs_path *path, int level) | |
e66f709b | 2139 | { |
5f39d397 CM |
2140 | struct extent_buffer *right = NULL; |
2141 | struct extent_buffer *mid; | |
2142 | struct extent_buffer *left = NULL; | |
2143 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
2144 | int ret = 0; |
2145 | int wret; | |
2146 | int pslot; | |
2147 | int orig_slot = path->slots[level]; | |
e66f709b CM |
2148 | |
2149 | if (level == 0) | |
2150 | return 1; | |
2151 | ||
5f39d397 | 2152 | mid = path->nodes[level]; |
7bb86316 | 2153 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 2154 | |
a05a9bb1 | 2155 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 2156 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
2157 | pslot = path->slots[level + 1]; |
2158 | } | |
e66f709b | 2159 | |
5f39d397 | 2160 | if (!parent) |
e66f709b | 2161 | return 1; |
e66f709b | 2162 | |
5f39d397 | 2163 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
2164 | |
2165 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 2166 | if (left) { |
e66f709b | 2167 | u32 left_nr; |
925baedd CM |
2168 | |
2169 | btrfs_tree_lock(left); | |
b4ce94de CM |
2170 | btrfs_set_lock_blocking(left); |
2171 | ||
5f39d397 | 2172 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
2173 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2174 | wret = 1; | |
2175 | } else { | |
5f39d397 | 2176 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 2177 | pslot - 1, &left); |
54aa1f4d CM |
2178 | if (ret) |
2179 | wret = 1; | |
2180 | else { | |
54aa1f4d | 2181 | wret = push_node_left(trans, root, |
971a1f66 | 2182 | left, mid, 0); |
54aa1f4d | 2183 | } |
33ade1f8 | 2184 | } |
e66f709b CM |
2185 | if (wret < 0) |
2186 | ret = wret; | |
2187 | if (wret == 0) { | |
5f39d397 | 2188 | struct btrfs_disk_key disk_key; |
e66f709b | 2189 | orig_slot += left_nr; |
5f39d397 | 2190 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 2191 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2192 | pslot, 0); |
5f39d397 CM |
2193 | btrfs_set_node_key(parent, &disk_key, pslot); |
2194 | btrfs_mark_buffer_dirty(parent); | |
2195 | if (btrfs_header_nritems(left) > orig_slot) { | |
2196 | path->nodes[level] = left; | |
e66f709b CM |
2197 | path->slots[level + 1] -= 1; |
2198 | path->slots[level] = orig_slot; | |
925baedd | 2199 | btrfs_tree_unlock(mid); |
5f39d397 | 2200 | free_extent_buffer(mid); |
e66f709b CM |
2201 | } else { |
2202 | orig_slot -= | |
5f39d397 | 2203 | btrfs_header_nritems(left); |
e66f709b | 2204 | path->slots[level] = orig_slot; |
925baedd | 2205 | btrfs_tree_unlock(left); |
5f39d397 | 2206 | free_extent_buffer(left); |
e66f709b | 2207 | } |
e66f709b CM |
2208 | return 0; |
2209 | } | |
925baedd | 2210 | btrfs_tree_unlock(left); |
5f39d397 | 2211 | free_extent_buffer(left); |
e66f709b | 2212 | } |
925baedd | 2213 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2214 | |
2215 | /* | |
2216 | * then try to empty the right most buffer into the middle | |
2217 | */ | |
5f39d397 | 2218 | if (right) { |
33ade1f8 | 2219 | u32 right_nr; |
b4ce94de | 2220 | |
925baedd | 2221 | btrfs_tree_lock(right); |
b4ce94de CM |
2222 | btrfs_set_lock_blocking(right); |
2223 | ||
5f39d397 | 2224 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2225 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2226 | wret = 1; | |
2227 | } else { | |
5f39d397 CM |
2228 | ret = btrfs_cow_block(trans, root, right, |
2229 | parent, pslot + 1, | |
9fa8cfe7 | 2230 | &right); |
54aa1f4d CM |
2231 | if (ret) |
2232 | wret = 1; | |
2233 | else { | |
54aa1f4d | 2234 | wret = balance_node_right(trans, root, |
5f39d397 | 2235 | right, mid); |
54aa1f4d | 2236 | } |
33ade1f8 | 2237 | } |
e66f709b CM |
2238 | if (wret < 0) |
2239 | ret = wret; | |
2240 | if (wret == 0) { | |
5f39d397 CM |
2241 | struct btrfs_disk_key disk_key; |
2242 | ||
2243 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2244 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2245 | pslot + 1, 0); |
5f39d397 CM |
2246 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2247 | btrfs_mark_buffer_dirty(parent); | |
2248 | ||
2249 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2250 | path->nodes[level] = right; | |
e66f709b CM |
2251 | path->slots[level + 1] += 1; |
2252 | path->slots[level] = orig_slot - | |
5f39d397 | 2253 | btrfs_header_nritems(mid); |
925baedd | 2254 | btrfs_tree_unlock(mid); |
5f39d397 | 2255 | free_extent_buffer(mid); |
e66f709b | 2256 | } else { |
925baedd | 2257 | btrfs_tree_unlock(right); |
5f39d397 | 2258 | free_extent_buffer(right); |
e66f709b | 2259 | } |
e66f709b CM |
2260 | return 0; |
2261 | } | |
925baedd | 2262 | btrfs_tree_unlock(right); |
5f39d397 | 2263 | free_extent_buffer(right); |
e66f709b | 2264 | } |
e66f709b CM |
2265 | return 1; |
2266 | } | |
2267 | ||
3c69faec | 2268 | /* |
d352ac68 CM |
2269 | * readahead one full node of leaves, finding things that are close |
2270 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2271 | */ |
c8c42864 CM |
2272 | static void reada_for_search(struct btrfs_root *root, |
2273 | struct btrfs_path *path, | |
2274 | int level, int slot, u64 objectid) | |
3c69faec | 2275 | { |
5f39d397 | 2276 | struct extent_buffer *node; |
01f46658 | 2277 | struct btrfs_disk_key disk_key; |
3c69faec | 2278 | u32 nritems; |
3c69faec | 2279 | u64 search; |
a7175319 | 2280 | u64 target; |
6b80053d | 2281 | u64 nread = 0; |
cb25c2ea | 2282 | u64 gen; |
3c69faec | 2283 | int direction = path->reada; |
5f39d397 | 2284 | struct extent_buffer *eb; |
6b80053d CM |
2285 | u32 nr; |
2286 | u32 blocksize; | |
2287 | u32 nscan = 0; | |
db94535d | 2288 | |
a6b6e75e | 2289 | if (level != 1) |
6702ed49 CM |
2290 | return; |
2291 | ||
2292 | if (!path->nodes[level]) | |
3c69faec CM |
2293 | return; |
2294 | ||
5f39d397 | 2295 | node = path->nodes[level]; |
925baedd | 2296 | |
3c69faec | 2297 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2298 | blocksize = btrfs_level_size(root, level - 1); |
2299 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2300 | if (eb) { |
2301 | free_extent_buffer(eb); | |
3c69faec CM |
2302 | return; |
2303 | } | |
2304 | ||
a7175319 | 2305 | target = search; |
6b80053d | 2306 | |
5f39d397 | 2307 | nritems = btrfs_header_nritems(node); |
6b80053d | 2308 | nr = slot; |
25b8b936 | 2309 | |
d397712b | 2310 | while (1) { |
6b80053d CM |
2311 | if (direction < 0) { |
2312 | if (nr == 0) | |
2313 | break; | |
2314 | nr--; | |
2315 | } else if (direction > 0) { | |
2316 | nr++; | |
2317 | if (nr >= nritems) | |
2318 | break; | |
3c69faec | 2319 | } |
01f46658 CM |
2320 | if (path->reada < 0 && objectid) { |
2321 | btrfs_node_key(node, &disk_key, nr); | |
2322 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2323 | break; | |
2324 | } | |
6b80053d | 2325 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2326 | if ((search <= target && target - search <= 65536) || |
2327 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2328 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2329 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2330 | nread += blocksize; |
2331 | } | |
2332 | nscan++; | |
a7175319 | 2333 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2334 | break; |
3c69faec CM |
2335 | } |
2336 | } | |
925baedd | 2337 | |
0b08851f JB |
2338 | static noinline void reada_for_balance(struct btrfs_root *root, |
2339 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2340 | { |
2341 | int slot; | |
2342 | int nritems; | |
2343 | struct extent_buffer *parent; | |
2344 | struct extent_buffer *eb; | |
2345 | u64 gen; | |
2346 | u64 block1 = 0; | |
2347 | u64 block2 = 0; | |
b4ce94de CM |
2348 | int blocksize; |
2349 | ||
8c594ea8 | 2350 | parent = path->nodes[level + 1]; |
b4ce94de | 2351 | if (!parent) |
0b08851f | 2352 | return; |
b4ce94de CM |
2353 | |
2354 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2355 | slot = path->slots[level + 1]; |
b4ce94de CM |
2356 | blocksize = btrfs_level_size(root, level); |
2357 | ||
2358 | if (slot > 0) { | |
2359 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2360 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2361 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2362 | /* |
2363 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2364 | * don't want to return eagain here. That will loop | |
2365 | * forever | |
2366 | */ | |
2367 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2368 | block1 = 0; |
2369 | free_extent_buffer(eb); | |
2370 | } | |
8c594ea8 | 2371 | if (slot + 1 < nritems) { |
b4ce94de CM |
2372 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2373 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2374 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2375 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2376 | block2 = 0; |
2377 | free_extent_buffer(eb); | |
2378 | } | |
8c594ea8 | 2379 | |
0b08851f JB |
2380 | if (block1) |
2381 | readahead_tree_block(root, block1, blocksize, 0); | |
2382 | if (block2) | |
2383 | readahead_tree_block(root, block2, blocksize, 0); | |
b4ce94de CM |
2384 | } |
2385 | ||
2386 | ||
d352ac68 | 2387 | /* |
d397712b CM |
2388 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2389 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2390 | * operations on the tree might require changing key pointers higher up in the | |
2391 | * tree. | |
d352ac68 | 2392 | * |
d397712b CM |
2393 | * callers might also have set path->keep_locks, which tells this code to keep |
2394 | * the lock if the path points to the last slot in the block. This is part of | |
2395 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2396 | * |
d397712b CM |
2397 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2398 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2399 | */ |
e02119d5 | 2400 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2401 | int lowest_unlock, int min_write_lock_level, |
2402 | int *write_lock_level) | |
925baedd CM |
2403 | { |
2404 | int i; | |
2405 | int skip_level = level; | |
051e1b9f | 2406 | int no_skips = 0; |
925baedd CM |
2407 | struct extent_buffer *t; |
2408 | ||
2409 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2410 | if (!path->nodes[i]) | |
2411 | break; | |
2412 | if (!path->locks[i]) | |
2413 | break; | |
051e1b9f | 2414 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2415 | skip_level = i + 1; |
2416 | continue; | |
2417 | } | |
051e1b9f | 2418 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2419 | u32 nritems; |
2420 | t = path->nodes[i]; | |
2421 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2422 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2423 | skip_level = i + 1; |
2424 | continue; | |
2425 | } | |
2426 | } | |
051e1b9f CM |
2427 | if (skip_level < i && i >= lowest_unlock) |
2428 | no_skips = 1; | |
2429 | ||
925baedd CM |
2430 | t = path->nodes[i]; |
2431 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2432 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2433 | path->locks[i] = 0; |
f7c79f30 CM |
2434 | if (write_lock_level && |
2435 | i > min_write_lock_level && | |
2436 | i <= *write_lock_level) { | |
2437 | *write_lock_level = i - 1; | |
2438 | } | |
925baedd CM |
2439 | } |
2440 | } | |
2441 | } | |
2442 | ||
b4ce94de CM |
2443 | /* |
2444 | * This releases any locks held in the path starting at level and | |
2445 | * going all the way up to the root. | |
2446 | * | |
2447 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2448 | * corner cases, such as COW of the block at slot zero in the node. This | |
2449 | * ignores those rules, and it should only be called when there are no | |
2450 | * more updates to be done higher up in the tree. | |
2451 | */ | |
2452 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2453 | { | |
2454 | int i; | |
2455 | ||
09a2a8f9 | 2456 | if (path->keep_locks) |
b4ce94de CM |
2457 | return; |
2458 | ||
2459 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2460 | if (!path->nodes[i]) | |
12f4dacc | 2461 | continue; |
b4ce94de | 2462 | if (!path->locks[i]) |
12f4dacc | 2463 | continue; |
bd681513 | 2464 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2465 | path->locks[i] = 0; |
2466 | } | |
2467 | } | |
2468 | ||
c8c42864 CM |
2469 | /* |
2470 | * helper function for btrfs_search_slot. The goal is to find a block | |
2471 | * in cache without setting the path to blocking. If we find the block | |
2472 | * we return zero and the path is unchanged. | |
2473 | * | |
2474 | * If we can't find the block, we set the path blocking and do some | |
2475 | * reada. -EAGAIN is returned and the search must be repeated. | |
2476 | */ | |
2477 | static int | |
2478 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2479 | struct btrfs_root *root, struct btrfs_path *p, | |
2480 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2481 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2482 | { |
2483 | u64 blocknr; | |
2484 | u64 gen; | |
2485 | u32 blocksize; | |
2486 | struct extent_buffer *b = *eb_ret; | |
2487 | struct extent_buffer *tmp; | |
76a05b35 | 2488 | int ret; |
c8c42864 CM |
2489 | |
2490 | blocknr = btrfs_node_blockptr(b, slot); | |
2491 | gen = btrfs_node_ptr_generation(b, slot); | |
2492 | blocksize = btrfs_level_size(root, level - 1); | |
2493 | ||
2494 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2495 | if (tmp) { |
b9fab919 | 2496 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2497 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2498 | *eb_ret = tmp; | |
2499 | return 0; | |
2500 | } | |
2501 | ||
2502 | /* the pages were up to date, but we failed | |
2503 | * the generation number check. Do a full | |
2504 | * read for the generation number that is correct. | |
2505 | * We must do this without dropping locks so | |
2506 | * we can trust our generation number | |
2507 | */ | |
2508 | btrfs_set_path_blocking(p); | |
2509 | ||
2510 | /* now we're allowed to do a blocking uptodate check */ | |
2511 | ret = btrfs_read_buffer(tmp, gen); | |
2512 | if (!ret) { | |
2513 | *eb_ret = tmp; | |
2514 | return 0; | |
cb44921a | 2515 | } |
bdf7c00e JB |
2516 | free_extent_buffer(tmp); |
2517 | btrfs_release_path(p); | |
2518 | return -EIO; | |
c8c42864 CM |
2519 | } |
2520 | ||
2521 | /* | |
2522 | * reduce lock contention at high levels | |
2523 | * of the btree by dropping locks before | |
76a05b35 CM |
2524 | * we read. Don't release the lock on the current |
2525 | * level because we need to walk this node to figure | |
2526 | * out which blocks to read. | |
c8c42864 | 2527 | */ |
8c594ea8 CM |
2528 | btrfs_unlock_up_safe(p, level + 1); |
2529 | btrfs_set_path_blocking(p); | |
2530 | ||
cb44921a | 2531 | free_extent_buffer(tmp); |
c8c42864 CM |
2532 | if (p->reada) |
2533 | reada_for_search(root, p, level, slot, key->objectid); | |
2534 | ||
b3b4aa74 | 2535 | btrfs_release_path(p); |
76a05b35 CM |
2536 | |
2537 | ret = -EAGAIN; | |
5bdd3536 | 2538 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2539 | if (tmp) { |
2540 | /* | |
2541 | * If the read above didn't mark this buffer up to date, | |
2542 | * it will never end up being up to date. Set ret to EIO now | |
2543 | * and give up so that our caller doesn't loop forever | |
2544 | * on our EAGAINs. | |
2545 | */ | |
b9fab919 | 2546 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2547 | ret = -EIO; |
c8c42864 | 2548 | free_extent_buffer(tmp); |
76a05b35 CM |
2549 | } |
2550 | return ret; | |
c8c42864 CM |
2551 | } |
2552 | ||
2553 | /* | |
2554 | * helper function for btrfs_search_slot. This does all of the checks | |
2555 | * for node-level blocks and does any balancing required based on | |
2556 | * the ins_len. | |
2557 | * | |
2558 | * If no extra work was required, zero is returned. If we had to | |
2559 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2560 | * start over | |
2561 | */ | |
2562 | static int | |
2563 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2564 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2565 | struct extent_buffer *b, int level, int ins_len, |
2566 | int *write_lock_level) | |
c8c42864 CM |
2567 | { |
2568 | int ret; | |
2569 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2570 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2571 | int sret; | |
2572 | ||
bd681513 CM |
2573 | if (*write_lock_level < level + 1) { |
2574 | *write_lock_level = level + 1; | |
2575 | btrfs_release_path(p); | |
2576 | goto again; | |
2577 | } | |
2578 | ||
c8c42864 | 2579 | btrfs_set_path_blocking(p); |
0b08851f | 2580 | reada_for_balance(root, p, level); |
c8c42864 | 2581 | sret = split_node(trans, root, p, level); |
bd681513 | 2582 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2583 | |
2584 | BUG_ON(sret > 0); | |
2585 | if (sret) { | |
2586 | ret = sret; | |
2587 | goto done; | |
2588 | } | |
2589 | b = p->nodes[level]; | |
2590 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2591 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2592 | int sret; |
2593 | ||
bd681513 CM |
2594 | if (*write_lock_level < level + 1) { |
2595 | *write_lock_level = level + 1; | |
2596 | btrfs_release_path(p); | |
2597 | goto again; | |
2598 | } | |
2599 | ||
c8c42864 | 2600 | btrfs_set_path_blocking(p); |
0b08851f | 2601 | reada_for_balance(root, p, level); |
c8c42864 | 2602 | sret = balance_level(trans, root, p, level); |
bd681513 | 2603 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2604 | |
2605 | if (sret) { | |
2606 | ret = sret; | |
2607 | goto done; | |
2608 | } | |
2609 | b = p->nodes[level]; | |
2610 | if (!b) { | |
b3b4aa74 | 2611 | btrfs_release_path(p); |
c8c42864 CM |
2612 | goto again; |
2613 | } | |
2614 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2615 | } | |
2616 | return 0; | |
2617 | ||
2618 | again: | |
2619 | ret = -EAGAIN; | |
2620 | done: | |
2621 | return ret; | |
2622 | } | |
2623 | ||
d7396f07 FDBM |
2624 | static void key_search_validate(struct extent_buffer *b, |
2625 | struct btrfs_key *key, | |
2626 | int level) | |
2627 | { | |
2628 | #ifdef CONFIG_BTRFS_ASSERT | |
2629 | struct btrfs_disk_key disk_key; | |
2630 | ||
2631 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2632 | ||
2633 | if (level == 0) | |
2634 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2635 | offsetof(struct btrfs_leaf, items[0].key), | |
2636 | sizeof(disk_key))); | |
2637 | else | |
2638 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2639 | offsetof(struct btrfs_node, ptrs[0].key), | |
2640 | sizeof(disk_key))); | |
2641 | #endif | |
2642 | } | |
2643 | ||
2644 | static int key_search(struct extent_buffer *b, struct btrfs_key *key, | |
2645 | int level, int *prev_cmp, int *slot) | |
2646 | { | |
2647 | if (*prev_cmp != 0) { | |
2648 | *prev_cmp = bin_search(b, key, level, slot); | |
2649 | return *prev_cmp; | |
2650 | } | |
2651 | ||
2652 | key_search_validate(b, key, level); | |
2653 | *slot = 0; | |
2654 | ||
2655 | return 0; | |
2656 | } | |
2657 | ||
3f870c28 | 2658 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path, |
e33d5c3d KN |
2659 | u64 iobjectid, u64 ioff, u8 key_type, |
2660 | struct btrfs_key *found_key) | |
2661 | { | |
2662 | int ret; | |
2663 | struct btrfs_key key; | |
2664 | struct extent_buffer *eb; | |
3f870c28 | 2665 | struct btrfs_path *path; |
e33d5c3d KN |
2666 | |
2667 | key.type = key_type; | |
2668 | key.objectid = iobjectid; | |
2669 | key.offset = ioff; | |
2670 | ||
3f870c28 KN |
2671 | if (found_path == NULL) { |
2672 | path = btrfs_alloc_path(); | |
2673 | if (!path) | |
2674 | return -ENOMEM; | |
2675 | } else | |
2676 | path = found_path; | |
2677 | ||
e33d5c3d | 2678 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); |
3f870c28 KN |
2679 | if ((ret < 0) || (found_key == NULL)) { |
2680 | if (path != found_path) | |
2681 | btrfs_free_path(path); | |
e33d5c3d | 2682 | return ret; |
3f870c28 | 2683 | } |
e33d5c3d KN |
2684 | |
2685 | eb = path->nodes[0]; | |
2686 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2687 | ret = btrfs_next_leaf(fs_root, path); | |
2688 | if (ret) | |
2689 | return ret; | |
2690 | eb = path->nodes[0]; | |
2691 | } | |
2692 | ||
2693 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2694 | if (found_key->type != key.type || | |
2695 | found_key->objectid != key.objectid) | |
2696 | return 1; | |
2697 | ||
2698 | return 0; | |
2699 | } | |
2700 | ||
74123bd7 CM |
2701 | /* |
2702 | * look for key in the tree. path is filled in with nodes along the way | |
2703 | * if key is found, we return zero and you can find the item in the leaf | |
2704 | * level of the path (level 0) | |
2705 | * | |
2706 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2707 | * be inserted, and 1 is returned. If there are other errors during the |
2708 | * search a negative error number is returned. | |
97571fd0 CM |
2709 | * |
2710 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2711 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2712 | * possible) | |
74123bd7 | 2713 | */ |
e089f05c CM |
2714 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2715 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2716 | ins_len, int cow) | |
be0e5c09 | 2717 | { |
5f39d397 | 2718 | struct extent_buffer *b; |
be0e5c09 CM |
2719 | int slot; |
2720 | int ret; | |
33c66f43 | 2721 | int err; |
be0e5c09 | 2722 | int level; |
925baedd | 2723 | int lowest_unlock = 1; |
bd681513 CM |
2724 | int root_lock; |
2725 | /* everything at write_lock_level or lower must be write locked */ | |
2726 | int write_lock_level = 0; | |
9f3a7427 | 2727 | u8 lowest_level = 0; |
f7c79f30 | 2728 | int min_write_lock_level; |
d7396f07 | 2729 | int prev_cmp; |
9f3a7427 | 2730 | |
6702ed49 | 2731 | lowest_level = p->lowest_level; |
323ac95b | 2732 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2733 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2734 | |
bd681513 | 2735 | if (ins_len < 0) { |
925baedd | 2736 | lowest_unlock = 2; |
65b51a00 | 2737 | |
bd681513 CM |
2738 | /* when we are removing items, we might have to go up to level |
2739 | * two as we update tree pointers Make sure we keep write | |
2740 | * for those levels as well | |
2741 | */ | |
2742 | write_lock_level = 2; | |
2743 | } else if (ins_len > 0) { | |
2744 | /* | |
2745 | * for inserting items, make sure we have a write lock on | |
2746 | * level 1 so we can update keys | |
2747 | */ | |
2748 | write_lock_level = 1; | |
2749 | } | |
2750 | ||
2751 | if (!cow) | |
2752 | write_lock_level = -1; | |
2753 | ||
09a2a8f9 | 2754 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2755 | write_lock_level = BTRFS_MAX_LEVEL; |
2756 | ||
f7c79f30 CM |
2757 | min_write_lock_level = write_lock_level; |
2758 | ||
bb803951 | 2759 | again: |
d7396f07 | 2760 | prev_cmp = -1; |
bd681513 CM |
2761 | /* |
2762 | * we try very hard to do read locks on the root | |
2763 | */ | |
2764 | root_lock = BTRFS_READ_LOCK; | |
2765 | level = 0; | |
5d4f98a2 | 2766 | if (p->search_commit_root) { |
bd681513 CM |
2767 | /* |
2768 | * the commit roots are read only | |
2769 | * so we always do read locks | |
2770 | */ | |
5d4f98a2 YZ |
2771 | b = root->commit_root; |
2772 | extent_buffer_get(b); | |
bd681513 | 2773 | level = btrfs_header_level(b); |
5d4f98a2 | 2774 | if (!p->skip_locking) |
bd681513 | 2775 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2776 | } else { |
bd681513 | 2777 | if (p->skip_locking) { |
5d4f98a2 | 2778 | b = btrfs_root_node(root); |
bd681513 CM |
2779 | level = btrfs_header_level(b); |
2780 | } else { | |
2781 | /* we don't know the level of the root node | |
2782 | * until we actually have it read locked | |
2783 | */ | |
2784 | b = btrfs_read_lock_root_node(root); | |
2785 | level = btrfs_header_level(b); | |
2786 | if (level <= write_lock_level) { | |
2787 | /* whoops, must trade for write lock */ | |
2788 | btrfs_tree_read_unlock(b); | |
2789 | free_extent_buffer(b); | |
2790 | b = btrfs_lock_root_node(root); | |
2791 | root_lock = BTRFS_WRITE_LOCK; | |
2792 | ||
2793 | /* the level might have changed, check again */ | |
2794 | level = btrfs_header_level(b); | |
2795 | } | |
2796 | } | |
5d4f98a2 | 2797 | } |
bd681513 CM |
2798 | p->nodes[level] = b; |
2799 | if (!p->skip_locking) | |
2800 | p->locks[level] = root_lock; | |
925baedd | 2801 | |
eb60ceac | 2802 | while (b) { |
5f39d397 | 2803 | level = btrfs_header_level(b); |
65b51a00 CM |
2804 | |
2805 | /* | |
2806 | * setup the path here so we can release it under lock | |
2807 | * contention with the cow code | |
2808 | */ | |
02217ed2 | 2809 | if (cow) { |
c8c42864 CM |
2810 | /* |
2811 | * if we don't really need to cow this block | |
2812 | * then we don't want to set the path blocking, | |
2813 | * so we test it here | |
2814 | */ | |
5d4f98a2 | 2815 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2816 | goto cow_done; |
5d4f98a2 | 2817 | |
b4ce94de CM |
2818 | btrfs_set_path_blocking(p); |
2819 | ||
bd681513 CM |
2820 | /* |
2821 | * must have write locks on this node and the | |
2822 | * parent | |
2823 | */ | |
5124e00e JB |
2824 | if (level > write_lock_level || |
2825 | (level + 1 > write_lock_level && | |
2826 | level + 1 < BTRFS_MAX_LEVEL && | |
2827 | p->nodes[level + 1])) { | |
bd681513 CM |
2828 | write_lock_level = level + 1; |
2829 | btrfs_release_path(p); | |
2830 | goto again; | |
2831 | } | |
2832 | ||
33c66f43 YZ |
2833 | err = btrfs_cow_block(trans, root, b, |
2834 | p->nodes[level + 1], | |
2835 | p->slots[level + 1], &b); | |
2836 | if (err) { | |
33c66f43 | 2837 | ret = err; |
65b51a00 | 2838 | goto done; |
54aa1f4d | 2839 | } |
02217ed2 | 2840 | } |
65b51a00 | 2841 | cow_done: |
02217ed2 | 2842 | BUG_ON(!cow && ins_len); |
65b51a00 | 2843 | |
eb60ceac | 2844 | p->nodes[level] = b; |
bd681513 | 2845 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2846 | |
2847 | /* | |
2848 | * we have a lock on b and as long as we aren't changing | |
2849 | * the tree, there is no way to for the items in b to change. | |
2850 | * It is safe to drop the lock on our parent before we | |
2851 | * go through the expensive btree search on b. | |
2852 | * | |
2853 | * If cow is true, then we might be changing slot zero, | |
2854 | * which may require changing the parent. So, we can't | |
2855 | * drop the lock until after we know which slot we're | |
2856 | * operating on. | |
2857 | */ | |
2858 | if (!cow) | |
2859 | btrfs_unlock_up_safe(p, level + 1); | |
2860 | ||
d7396f07 | 2861 | ret = key_search(b, key, level, &prev_cmp, &slot); |
b4ce94de | 2862 | |
5f39d397 | 2863 | if (level != 0) { |
33c66f43 YZ |
2864 | int dec = 0; |
2865 | if (ret && slot > 0) { | |
2866 | dec = 1; | |
be0e5c09 | 2867 | slot -= 1; |
33c66f43 | 2868 | } |
be0e5c09 | 2869 | p->slots[level] = slot; |
33c66f43 | 2870 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2871 | ins_len, &write_lock_level); |
33c66f43 | 2872 | if (err == -EAGAIN) |
c8c42864 | 2873 | goto again; |
33c66f43 YZ |
2874 | if (err) { |
2875 | ret = err; | |
c8c42864 | 2876 | goto done; |
33c66f43 | 2877 | } |
c8c42864 CM |
2878 | b = p->nodes[level]; |
2879 | slot = p->slots[level]; | |
b4ce94de | 2880 | |
bd681513 CM |
2881 | /* |
2882 | * slot 0 is special, if we change the key | |
2883 | * we have to update the parent pointer | |
2884 | * which means we must have a write lock | |
2885 | * on the parent | |
2886 | */ | |
2887 | if (slot == 0 && cow && | |
2888 | write_lock_level < level + 1) { | |
2889 | write_lock_level = level + 1; | |
2890 | btrfs_release_path(p); | |
2891 | goto again; | |
2892 | } | |
2893 | ||
f7c79f30 CM |
2894 | unlock_up(p, level, lowest_unlock, |
2895 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2896 | |
925baedd | 2897 | if (level == lowest_level) { |
33c66f43 YZ |
2898 | if (dec) |
2899 | p->slots[level]++; | |
5b21f2ed | 2900 | goto done; |
925baedd | 2901 | } |
ca7a79ad | 2902 | |
33c66f43 | 2903 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2904 | &b, level, slot, key, 0); |
33c66f43 | 2905 | if (err == -EAGAIN) |
c8c42864 | 2906 | goto again; |
33c66f43 YZ |
2907 | if (err) { |
2908 | ret = err; | |
76a05b35 | 2909 | goto done; |
33c66f43 | 2910 | } |
76a05b35 | 2911 | |
b4ce94de | 2912 | if (!p->skip_locking) { |
bd681513 CM |
2913 | level = btrfs_header_level(b); |
2914 | if (level <= write_lock_level) { | |
2915 | err = btrfs_try_tree_write_lock(b); | |
2916 | if (!err) { | |
2917 | btrfs_set_path_blocking(p); | |
2918 | btrfs_tree_lock(b); | |
2919 | btrfs_clear_path_blocking(p, b, | |
2920 | BTRFS_WRITE_LOCK); | |
2921 | } | |
2922 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2923 | } else { | |
2924 | err = btrfs_try_tree_read_lock(b); | |
2925 | if (!err) { | |
2926 | btrfs_set_path_blocking(p); | |
2927 | btrfs_tree_read_lock(b); | |
2928 | btrfs_clear_path_blocking(p, b, | |
2929 | BTRFS_READ_LOCK); | |
2930 | } | |
2931 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2932 | } |
bd681513 | 2933 | p->nodes[level] = b; |
b4ce94de | 2934 | } |
be0e5c09 CM |
2935 | } else { |
2936 | p->slots[level] = slot; | |
87b29b20 YZ |
2937 | if (ins_len > 0 && |
2938 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2939 | if (write_lock_level < 1) { |
2940 | write_lock_level = 1; | |
2941 | btrfs_release_path(p); | |
2942 | goto again; | |
2943 | } | |
2944 | ||
b4ce94de | 2945 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2946 | err = split_leaf(trans, root, key, |
2947 | p, ins_len, ret == 0); | |
bd681513 | 2948 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2949 | |
33c66f43 YZ |
2950 | BUG_ON(err > 0); |
2951 | if (err) { | |
2952 | ret = err; | |
65b51a00 CM |
2953 | goto done; |
2954 | } | |
5c680ed6 | 2955 | } |
459931ec | 2956 | if (!p->search_for_split) |
f7c79f30 CM |
2957 | unlock_up(p, level, lowest_unlock, |
2958 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2959 | goto done; |
be0e5c09 CM |
2960 | } |
2961 | } | |
65b51a00 CM |
2962 | ret = 1; |
2963 | done: | |
b4ce94de CM |
2964 | /* |
2965 | * we don't really know what they plan on doing with the path | |
2966 | * from here on, so for now just mark it as blocking | |
2967 | */ | |
b9473439 CM |
2968 | if (!p->leave_spinning) |
2969 | btrfs_set_path_blocking(p); | |
76a05b35 | 2970 | if (ret < 0) |
b3b4aa74 | 2971 | btrfs_release_path(p); |
65b51a00 | 2972 | return ret; |
be0e5c09 CM |
2973 | } |
2974 | ||
5d9e75c4 JS |
2975 | /* |
2976 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2977 | * current state of the tree together with the operations recorded in the tree | |
2978 | * modification log to search for the key in a previous version of this tree, as | |
2979 | * denoted by the time_seq parameter. | |
2980 | * | |
2981 | * Naturally, there is no support for insert, delete or cow operations. | |
2982 | * | |
2983 | * The resulting path and return value will be set up as if we called | |
2984 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2985 | */ | |
2986 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2987 | struct btrfs_path *p, u64 time_seq) | |
2988 | { | |
2989 | struct extent_buffer *b; | |
2990 | int slot; | |
2991 | int ret; | |
2992 | int err; | |
2993 | int level; | |
2994 | int lowest_unlock = 1; | |
2995 | u8 lowest_level = 0; | |
d4b4087c | 2996 | int prev_cmp = -1; |
5d9e75c4 JS |
2997 | |
2998 | lowest_level = p->lowest_level; | |
2999 | WARN_ON(p->nodes[0] != NULL); | |
3000 | ||
3001 | if (p->search_commit_root) { | |
3002 | BUG_ON(time_seq); | |
3003 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3004 | } | |
3005 | ||
3006 | again: | |
5d9e75c4 | 3007 | b = get_old_root(root, time_seq); |
5d9e75c4 | 3008 | level = btrfs_header_level(b); |
5d9e75c4 JS |
3009 | p->locks[level] = BTRFS_READ_LOCK; |
3010 | ||
3011 | while (b) { | |
3012 | level = btrfs_header_level(b); | |
3013 | p->nodes[level] = b; | |
3014 | btrfs_clear_path_blocking(p, NULL, 0); | |
3015 | ||
3016 | /* | |
3017 | * we have a lock on b and as long as we aren't changing | |
3018 | * the tree, there is no way to for the items in b to change. | |
3019 | * It is safe to drop the lock on our parent before we | |
3020 | * go through the expensive btree search on b. | |
3021 | */ | |
3022 | btrfs_unlock_up_safe(p, level + 1); | |
3023 | ||
d4b4087c JB |
3024 | /* |
3025 | * Since we can unwind eb's we want to do a real search every | |
3026 | * time. | |
3027 | */ | |
3028 | prev_cmp = -1; | |
d7396f07 | 3029 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
3030 | |
3031 | if (level != 0) { | |
3032 | int dec = 0; | |
3033 | if (ret && slot > 0) { | |
3034 | dec = 1; | |
3035 | slot -= 1; | |
3036 | } | |
3037 | p->slots[level] = slot; | |
3038 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3039 | ||
3040 | if (level == lowest_level) { | |
3041 | if (dec) | |
3042 | p->slots[level]++; | |
3043 | goto done; | |
3044 | } | |
3045 | ||
3046 | err = read_block_for_search(NULL, root, p, &b, level, | |
3047 | slot, key, time_seq); | |
3048 | if (err == -EAGAIN) | |
3049 | goto again; | |
3050 | if (err) { | |
3051 | ret = err; | |
3052 | goto done; | |
3053 | } | |
3054 | ||
3055 | level = btrfs_header_level(b); | |
3056 | err = btrfs_try_tree_read_lock(b); | |
3057 | if (!err) { | |
3058 | btrfs_set_path_blocking(p); | |
3059 | btrfs_tree_read_lock(b); | |
3060 | btrfs_clear_path_blocking(p, b, | |
3061 | BTRFS_READ_LOCK); | |
3062 | } | |
9ec72677 | 3063 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
3064 | if (!b) { |
3065 | ret = -ENOMEM; | |
3066 | goto done; | |
3067 | } | |
5d9e75c4 JS |
3068 | p->locks[level] = BTRFS_READ_LOCK; |
3069 | p->nodes[level] = b; | |
5d9e75c4 JS |
3070 | } else { |
3071 | p->slots[level] = slot; | |
3072 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
3073 | goto done; | |
3074 | } | |
3075 | } | |
3076 | ret = 1; | |
3077 | done: | |
3078 | if (!p->leave_spinning) | |
3079 | btrfs_set_path_blocking(p); | |
3080 | if (ret < 0) | |
3081 | btrfs_release_path(p); | |
3082 | ||
3083 | return ret; | |
3084 | } | |
3085 | ||
2f38b3e1 AJ |
3086 | /* |
3087 | * helper to use instead of search slot if no exact match is needed but | |
3088 | * instead the next or previous item should be returned. | |
3089 | * When find_higher is true, the next higher item is returned, the next lower | |
3090 | * otherwise. | |
3091 | * When return_any and find_higher are both true, and no higher item is found, | |
3092 | * return the next lower instead. | |
3093 | * When return_any is true and find_higher is false, and no lower item is found, | |
3094 | * return the next higher instead. | |
3095 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
3096 | * < 0 on error | |
3097 | */ | |
3098 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
3099 | struct btrfs_key *key, struct btrfs_path *p, | |
3100 | int find_higher, int return_any) | |
3101 | { | |
3102 | int ret; | |
3103 | struct extent_buffer *leaf; | |
3104 | ||
3105 | again: | |
3106 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
3107 | if (ret <= 0) | |
3108 | return ret; | |
3109 | /* | |
3110 | * a return value of 1 means the path is at the position where the | |
3111 | * item should be inserted. Normally this is the next bigger item, | |
3112 | * but in case the previous item is the last in a leaf, path points | |
3113 | * to the first free slot in the previous leaf, i.e. at an invalid | |
3114 | * item. | |
3115 | */ | |
3116 | leaf = p->nodes[0]; | |
3117 | ||
3118 | if (find_higher) { | |
3119 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
3120 | ret = btrfs_next_leaf(root, p); | |
3121 | if (ret <= 0) | |
3122 | return ret; | |
3123 | if (!return_any) | |
3124 | return 1; | |
3125 | /* | |
3126 | * no higher item found, return the next | |
3127 | * lower instead | |
3128 | */ | |
3129 | return_any = 0; | |
3130 | find_higher = 0; | |
3131 | btrfs_release_path(p); | |
3132 | goto again; | |
3133 | } | |
3134 | } else { | |
e6793769 AJ |
3135 | if (p->slots[0] == 0) { |
3136 | ret = btrfs_prev_leaf(root, p); | |
3137 | if (ret < 0) | |
3138 | return ret; | |
3139 | if (!ret) { | |
3140 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
3141 | return 0; | |
2f38b3e1 | 3142 | } |
e6793769 AJ |
3143 | if (!return_any) |
3144 | return 1; | |
3145 | /* | |
3146 | * no lower item found, return the next | |
3147 | * higher instead | |
3148 | */ | |
3149 | return_any = 0; | |
3150 | find_higher = 1; | |
3151 | btrfs_release_path(p); | |
3152 | goto again; | |
3153 | } else { | |
2f38b3e1 AJ |
3154 | --p->slots[0]; |
3155 | } | |
3156 | } | |
3157 | return 0; | |
3158 | } | |
3159 | ||
74123bd7 CM |
3160 | /* |
3161 | * adjust the pointers going up the tree, starting at level | |
3162 | * making sure the right key of each node is points to 'key'. | |
3163 | * This is used after shifting pointers to the left, so it stops | |
3164 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
3165 | * higher levels | |
aa5d6bed | 3166 | * |
74123bd7 | 3167 | */ |
d6a0a126 | 3168 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 3169 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
3170 | { |
3171 | int i; | |
5f39d397 CM |
3172 | struct extent_buffer *t; |
3173 | ||
234b63a0 | 3174 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 3175 | int tslot = path->slots[i]; |
eb60ceac | 3176 | if (!path->nodes[i]) |
be0e5c09 | 3177 | break; |
5f39d397 | 3178 | t = path->nodes[i]; |
32adf090 | 3179 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 3180 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 3181 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
3182 | if (tslot != 0) |
3183 | break; | |
3184 | } | |
3185 | } | |
3186 | ||
31840ae1 ZY |
3187 | /* |
3188 | * update item key. | |
3189 | * | |
3190 | * This function isn't completely safe. It's the caller's responsibility | |
3191 | * that the new key won't break the order | |
3192 | */ | |
afe5fea7 | 3193 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 3194 | struct btrfs_key *new_key) |
31840ae1 ZY |
3195 | { |
3196 | struct btrfs_disk_key disk_key; | |
3197 | struct extent_buffer *eb; | |
3198 | int slot; | |
3199 | ||
3200 | eb = path->nodes[0]; | |
3201 | slot = path->slots[0]; | |
3202 | if (slot > 0) { | |
3203 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3204 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3205 | } |
3206 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3207 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3208 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3209 | } |
3210 | ||
3211 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3212 | btrfs_set_item_key(eb, &disk_key, slot); | |
3213 | btrfs_mark_buffer_dirty(eb); | |
3214 | if (slot == 0) | |
d6a0a126 | 3215 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
3216 | } |
3217 | ||
74123bd7 CM |
3218 | /* |
3219 | * try to push data from one node into the next node left in the | |
79f95c82 | 3220 | * tree. |
aa5d6bed CM |
3221 | * |
3222 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3223 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3224 | */ |
98ed5174 CM |
3225 | static int push_node_left(struct btrfs_trans_handle *trans, |
3226 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3227 | struct extent_buffer *src, int empty) |
be0e5c09 | 3228 | { |
be0e5c09 | 3229 | int push_items = 0; |
bb803951 CM |
3230 | int src_nritems; |
3231 | int dst_nritems; | |
aa5d6bed | 3232 | int ret = 0; |
be0e5c09 | 3233 | |
5f39d397 CM |
3234 | src_nritems = btrfs_header_nritems(src); |
3235 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3236 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
3237 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3238 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3239 | |
bce4eae9 | 3240 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3241 | return 1; |
3242 | ||
d397712b | 3243 | if (push_items <= 0) |
be0e5c09 CM |
3244 | return 1; |
3245 | ||
bce4eae9 | 3246 | if (empty) { |
971a1f66 | 3247 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3248 | if (push_items < src_nritems) { |
3249 | /* leave at least 8 pointers in the node if | |
3250 | * we aren't going to empty it | |
3251 | */ | |
3252 | if (src_nritems - push_items < 8) { | |
3253 | if (push_items <= 8) | |
3254 | return 1; | |
3255 | push_items -= 8; | |
3256 | } | |
3257 | } | |
3258 | } else | |
3259 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3260 | |
5de865ee FDBM |
3261 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
3262 | push_items); | |
3263 | if (ret) { | |
3264 | btrfs_abort_transaction(trans, root, ret); | |
3265 | return ret; | |
3266 | } | |
5f39d397 CM |
3267 | copy_extent_buffer(dst, src, |
3268 | btrfs_node_key_ptr_offset(dst_nritems), | |
3269 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3270 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3271 | |
bb803951 | 3272 | if (push_items < src_nritems) { |
57911b8b JS |
3273 | /* |
3274 | * don't call tree_mod_log_eb_move here, key removal was already | |
3275 | * fully logged by tree_mod_log_eb_copy above. | |
3276 | */ | |
5f39d397 CM |
3277 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3278 | btrfs_node_key_ptr_offset(push_items), | |
3279 | (src_nritems - push_items) * | |
3280 | sizeof(struct btrfs_key_ptr)); | |
3281 | } | |
3282 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3283 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3284 | btrfs_mark_buffer_dirty(src); | |
3285 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3286 | |
79f95c82 CM |
3287 | return ret; |
3288 | } | |
3289 | ||
3290 | /* | |
3291 | * try to push data from one node into the next node right in the | |
3292 | * tree. | |
3293 | * | |
3294 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3295 | * error, and > 0 if there was no room in the right hand block. | |
3296 | * | |
3297 | * this will only push up to 1/2 the contents of the left node over | |
3298 | */ | |
5f39d397 CM |
3299 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3300 | struct btrfs_root *root, | |
3301 | struct extent_buffer *dst, | |
3302 | struct extent_buffer *src) | |
79f95c82 | 3303 | { |
79f95c82 CM |
3304 | int push_items = 0; |
3305 | int max_push; | |
3306 | int src_nritems; | |
3307 | int dst_nritems; | |
3308 | int ret = 0; | |
79f95c82 | 3309 | |
7bb86316 CM |
3310 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3311 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3312 | ||
5f39d397 CM |
3313 | src_nritems = btrfs_header_nritems(src); |
3314 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3315 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3316 | if (push_items <= 0) |
79f95c82 | 3317 | return 1; |
bce4eae9 | 3318 | |
d397712b | 3319 | if (src_nritems < 4) |
bce4eae9 | 3320 | return 1; |
79f95c82 CM |
3321 | |
3322 | max_push = src_nritems / 2 + 1; | |
3323 | /* don't try to empty the node */ | |
d397712b | 3324 | if (max_push >= src_nritems) |
79f95c82 | 3325 | return 1; |
252c38f0 | 3326 | |
79f95c82 CM |
3327 | if (max_push < push_items) |
3328 | push_items = max_push; | |
3329 | ||
f230475e | 3330 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3331 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3332 | btrfs_node_key_ptr_offset(0), | |
3333 | (dst_nritems) * | |
3334 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3335 | |
5de865ee FDBM |
3336 | ret = tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
3337 | src_nritems - push_items, push_items); | |
3338 | if (ret) { | |
3339 | btrfs_abort_transaction(trans, root, ret); | |
3340 | return ret; | |
3341 | } | |
5f39d397 CM |
3342 | copy_extent_buffer(dst, src, |
3343 | btrfs_node_key_ptr_offset(0), | |
3344 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3345 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3346 | |
5f39d397 CM |
3347 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3348 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3349 | |
5f39d397 CM |
3350 | btrfs_mark_buffer_dirty(src); |
3351 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3352 | |
aa5d6bed | 3353 | return ret; |
be0e5c09 CM |
3354 | } |
3355 | ||
97571fd0 CM |
3356 | /* |
3357 | * helper function to insert a new root level in the tree. | |
3358 | * A new node is allocated, and a single item is inserted to | |
3359 | * point to the existing root | |
aa5d6bed CM |
3360 | * |
3361 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3362 | */ |
d397712b | 3363 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3364 | struct btrfs_root *root, |
fdd99c72 | 3365 | struct btrfs_path *path, int level) |
5c680ed6 | 3366 | { |
7bb86316 | 3367 | u64 lower_gen; |
5f39d397 CM |
3368 | struct extent_buffer *lower; |
3369 | struct extent_buffer *c; | |
925baedd | 3370 | struct extent_buffer *old; |
5f39d397 | 3371 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3372 | |
3373 | BUG_ON(path->nodes[level]); | |
3374 | BUG_ON(path->nodes[level-1] != root->node); | |
3375 | ||
7bb86316 CM |
3376 | lower = path->nodes[level-1]; |
3377 | if (level == 1) | |
3378 | btrfs_item_key(lower, &lower_key, 0); | |
3379 | else | |
3380 | btrfs_node_key(lower, &lower_key, 0); | |
3381 | ||
31840ae1 | 3382 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3383 | root->root_key.objectid, &lower_key, |
5581a51a | 3384 | level, root->node->start, 0); |
5f39d397 CM |
3385 | if (IS_ERR(c)) |
3386 | return PTR_ERR(c); | |
925baedd | 3387 | |
f0486c68 YZ |
3388 | root_add_used(root, root->nodesize); |
3389 | ||
5d4f98a2 | 3390 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3391 | btrfs_set_header_nritems(c, 1); |
3392 | btrfs_set_header_level(c, level); | |
db94535d | 3393 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3394 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3395 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3396 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3397 | |
0a4e5586 | 3398 | write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(), |
5f39d397 | 3399 | BTRFS_FSID_SIZE); |
e17cade2 CM |
3400 | |
3401 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 3402 | btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE); |
e17cade2 | 3403 | |
5f39d397 | 3404 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3405 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3406 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3407 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3408 | |
3409 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3410 | |
5f39d397 | 3411 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3412 | |
925baedd | 3413 | old = root->node; |
fdd99c72 | 3414 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3415 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3416 | |
3417 | /* the super has an extra ref to root->node */ | |
3418 | free_extent_buffer(old); | |
3419 | ||
0b86a832 | 3420 | add_root_to_dirty_list(root); |
5f39d397 CM |
3421 | extent_buffer_get(c); |
3422 | path->nodes[level] = c; | |
bd681513 | 3423 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3424 | path->slots[level] = 0; |
3425 | return 0; | |
3426 | } | |
3427 | ||
74123bd7 CM |
3428 | /* |
3429 | * worker function to insert a single pointer in a node. | |
3430 | * the node should have enough room for the pointer already | |
97571fd0 | 3431 | * |
74123bd7 CM |
3432 | * slot and level indicate where you want the key to go, and |
3433 | * blocknr is the block the key points to. | |
3434 | */ | |
143bede5 JM |
3435 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3436 | struct btrfs_root *root, struct btrfs_path *path, | |
3437 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3438 | int slot, int level) |
74123bd7 | 3439 | { |
5f39d397 | 3440 | struct extent_buffer *lower; |
74123bd7 | 3441 | int nritems; |
f3ea38da | 3442 | int ret; |
5c680ed6 CM |
3443 | |
3444 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3445 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3446 | lower = path->nodes[level]; |
3447 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3448 | BUG_ON(slot > nritems); |
143bede5 | 3449 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3450 | if (slot != nritems) { |
c3e06965 | 3451 | if (level) |
f3ea38da JS |
3452 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3453 | slot, nritems - slot); | |
5f39d397 CM |
3454 | memmove_extent_buffer(lower, |
3455 | btrfs_node_key_ptr_offset(slot + 1), | |
3456 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3457 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3458 | } |
c3e06965 | 3459 | if (level) { |
f3ea38da | 3460 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3461 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3462 | BUG_ON(ret < 0); |
3463 | } | |
5f39d397 | 3464 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3465 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3466 | WARN_ON(trans->transid == 0); |
3467 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3468 | btrfs_set_header_nritems(lower, nritems + 1); |
3469 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3470 | } |
3471 | ||
97571fd0 CM |
3472 | /* |
3473 | * split the node at the specified level in path in two. | |
3474 | * The path is corrected to point to the appropriate node after the split | |
3475 | * | |
3476 | * Before splitting this tries to make some room in the node by pushing | |
3477 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3478 | * |
3479 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3480 | */ |
e02119d5 CM |
3481 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3482 | struct btrfs_root *root, | |
3483 | struct btrfs_path *path, int level) | |
be0e5c09 | 3484 | { |
5f39d397 CM |
3485 | struct extent_buffer *c; |
3486 | struct extent_buffer *split; | |
3487 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3488 | int mid; |
5c680ed6 | 3489 | int ret; |
7518a238 | 3490 | u32 c_nritems; |
eb60ceac | 3491 | |
5f39d397 | 3492 | c = path->nodes[level]; |
7bb86316 | 3493 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3494 | if (c == root->node) { |
d9abbf1c | 3495 | /* |
90f8d62e JS |
3496 | * trying to split the root, lets make a new one |
3497 | * | |
fdd99c72 | 3498 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3499 | * insert_new_root, because that root buffer will be kept as a |
3500 | * normal node. We are going to log removal of half of the | |
3501 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3502 | * tree lock on the buffer, which is why we cannot race with | |
3503 | * other tree_mod_log users. | |
d9abbf1c | 3504 | */ |
fdd99c72 | 3505 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3506 | if (ret) |
3507 | return ret; | |
b3612421 | 3508 | } else { |
e66f709b | 3509 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3510 | c = path->nodes[level]; |
3511 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3512 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3513 | return 0; |
54aa1f4d CM |
3514 | if (ret < 0) |
3515 | return ret; | |
be0e5c09 | 3516 | } |
e66f709b | 3517 | |
5f39d397 | 3518 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3519 | mid = (c_nritems + 1) / 2; |
3520 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3521 | |
5d4f98a2 | 3522 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3523 | root->root_key.objectid, |
5581a51a | 3524 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3525 | if (IS_ERR(split)) |
3526 | return PTR_ERR(split); | |
3527 | ||
f0486c68 YZ |
3528 | root_add_used(root, root->nodesize); |
3529 | ||
5d4f98a2 | 3530 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3531 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3532 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3533 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3534 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3535 | btrfs_set_header_owner(split, root->root_key.objectid); |
3536 | write_extent_buffer(split, root->fs_info->fsid, | |
0a4e5586 | 3537 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 | 3538 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
b308bc2f | 3539 | btrfs_header_chunk_tree_uuid(split), |
e17cade2 | 3540 | BTRFS_UUID_SIZE); |
54aa1f4d | 3541 | |
5de865ee FDBM |
3542 | ret = tree_mod_log_eb_copy(root->fs_info, split, c, 0, |
3543 | mid, c_nritems - mid); | |
3544 | if (ret) { | |
3545 | btrfs_abort_transaction(trans, root, ret); | |
3546 | return ret; | |
3547 | } | |
5f39d397 CM |
3548 | copy_extent_buffer(split, c, |
3549 | btrfs_node_key_ptr_offset(0), | |
3550 | btrfs_node_key_ptr_offset(mid), | |
3551 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3552 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3553 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3554 | ret = 0; |
3555 | ||
5f39d397 CM |
3556 | btrfs_mark_buffer_dirty(c); |
3557 | btrfs_mark_buffer_dirty(split); | |
3558 | ||
143bede5 | 3559 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3560 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3561 | |
5de08d7d | 3562 | if (path->slots[level] >= mid) { |
5c680ed6 | 3563 | path->slots[level] -= mid; |
925baedd | 3564 | btrfs_tree_unlock(c); |
5f39d397 CM |
3565 | free_extent_buffer(c); |
3566 | path->nodes[level] = split; | |
5c680ed6 CM |
3567 | path->slots[level + 1] += 1; |
3568 | } else { | |
925baedd | 3569 | btrfs_tree_unlock(split); |
5f39d397 | 3570 | free_extent_buffer(split); |
be0e5c09 | 3571 | } |
aa5d6bed | 3572 | return ret; |
be0e5c09 CM |
3573 | } |
3574 | ||
74123bd7 CM |
3575 | /* |
3576 | * how many bytes are required to store the items in a leaf. start | |
3577 | * and nr indicate which items in the leaf to check. This totals up the | |
3578 | * space used both by the item structs and the item data | |
3579 | */ | |
5f39d397 | 3580 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3581 | { |
41be1f3b JB |
3582 | struct btrfs_item *start_item; |
3583 | struct btrfs_item *end_item; | |
3584 | struct btrfs_map_token token; | |
be0e5c09 | 3585 | int data_len; |
5f39d397 | 3586 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3587 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3588 | |
3589 | if (!nr) | |
3590 | return 0; | |
41be1f3b | 3591 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3592 | start_item = btrfs_item_nr(start); |
3593 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3594 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3595 | btrfs_token_item_size(l, start_item, &token); | |
3596 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3597 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3598 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3599 | return data_len; |
3600 | } | |
3601 | ||
d4dbff95 CM |
3602 | /* |
3603 | * The space between the end of the leaf items and | |
3604 | * the start of the leaf data. IOW, how much room | |
3605 | * the leaf has left for both items and data | |
3606 | */ | |
d397712b | 3607 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3608 | struct extent_buffer *leaf) |
d4dbff95 | 3609 | { |
5f39d397 CM |
3610 | int nritems = btrfs_header_nritems(leaf); |
3611 | int ret; | |
3612 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3613 | if (ret < 0) { | |
efe120a0 FH |
3614 | btrfs_crit(root->fs_info, |
3615 | "leaf free space ret %d, leaf data size %lu, used %d nritems %d", | |
ae2f5411 | 3616 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3617 | leaf_space_used(leaf, 0, nritems), nritems); |
3618 | } | |
3619 | return ret; | |
d4dbff95 CM |
3620 | } |
3621 | ||
99d8f83c CM |
3622 | /* |
3623 | * min slot controls the lowest index we're willing to push to the | |
3624 | * right. We'll push up to and including min_slot, but no lower | |
3625 | */ | |
44871b1b CM |
3626 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3627 | struct btrfs_root *root, | |
3628 | struct btrfs_path *path, | |
3629 | int data_size, int empty, | |
3630 | struct extent_buffer *right, | |
99d8f83c CM |
3631 | int free_space, u32 left_nritems, |
3632 | u32 min_slot) | |
00ec4c51 | 3633 | { |
5f39d397 | 3634 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3635 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3636 | struct btrfs_map_token token; |
5f39d397 | 3637 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3638 | int slot; |
34a38218 | 3639 | u32 i; |
00ec4c51 CM |
3640 | int push_space = 0; |
3641 | int push_items = 0; | |
0783fcfc | 3642 | struct btrfs_item *item; |
34a38218 | 3643 | u32 nr; |
7518a238 | 3644 | u32 right_nritems; |
5f39d397 | 3645 | u32 data_end; |
db94535d | 3646 | u32 this_item_size; |
00ec4c51 | 3647 | |
cfed81a0 CM |
3648 | btrfs_init_map_token(&token); |
3649 | ||
34a38218 CM |
3650 | if (empty) |
3651 | nr = 0; | |
3652 | else | |
99d8f83c | 3653 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3654 | |
31840ae1 | 3655 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3656 | push_space += data_size; |
31840ae1 | 3657 | |
44871b1b | 3658 | slot = path->slots[1]; |
34a38218 CM |
3659 | i = left_nritems - 1; |
3660 | while (i >= nr) { | |
dd3cc16b | 3661 | item = btrfs_item_nr(i); |
db94535d | 3662 | |
31840ae1 ZY |
3663 | if (!empty && push_items > 0) { |
3664 | if (path->slots[0] > i) | |
3665 | break; | |
3666 | if (path->slots[0] == i) { | |
3667 | int space = btrfs_leaf_free_space(root, left); | |
3668 | if (space + push_space * 2 > free_space) | |
3669 | break; | |
3670 | } | |
3671 | } | |
3672 | ||
00ec4c51 | 3673 | if (path->slots[0] == i) |
87b29b20 | 3674 | push_space += data_size; |
db94535d | 3675 | |
db94535d CM |
3676 | this_item_size = btrfs_item_size(left, item); |
3677 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3678 | break; |
31840ae1 | 3679 | |
00ec4c51 | 3680 | push_items++; |
db94535d | 3681 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3682 | if (i == 0) |
3683 | break; | |
3684 | i--; | |
db94535d | 3685 | } |
5f39d397 | 3686 | |
925baedd CM |
3687 | if (push_items == 0) |
3688 | goto out_unlock; | |
5f39d397 | 3689 | |
6c1500f2 | 3690 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3691 | |
00ec4c51 | 3692 | /* push left to right */ |
5f39d397 | 3693 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3694 | |
5f39d397 | 3695 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3696 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3697 | |
00ec4c51 | 3698 | /* make room in the right data area */ |
5f39d397 CM |
3699 | data_end = leaf_data_end(root, right); |
3700 | memmove_extent_buffer(right, | |
3701 | btrfs_leaf_data(right) + data_end - push_space, | |
3702 | btrfs_leaf_data(right) + data_end, | |
3703 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3704 | ||
00ec4c51 | 3705 | /* copy from the left data area */ |
5f39d397 | 3706 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3707 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3708 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3709 | push_space); | |
5f39d397 CM |
3710 | |
3711 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3712 | btrfs_item_nr_offset(0), | |
3713 | right_nritems * sizeof(struct btrfs_item)); | |
3714 | ||
00ec4c51 | 3715 | /* copy the items from left to right */ |
5f39d397 CM |
3716 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3717 | btrfs_item_nr_offset(left_nritems - push_items), | |
3718 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3719 | |
3720 | /* update the item pointers */ | |
7518a238 | 3721 | right_nritems += push_items; |
5f39d397 | 3722 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3723 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3724 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3725 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3726 | push_space -= btrfs_token_item_size(right, item, &token); |
3727 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3728 | } |
3729 | ||
7518a238 | 3730 | left_nritems -= push_items; |
5f39d397 | 3731 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3732 | |
34a38218 CM |
3733 | if (left_nritems) |
3734 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3735 | else |
3736 | clean_tree_block(trans, root, left); | |
3737 | ||
5f39d397 | 3738 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3739 | |
5f39d397 CM |
3740 | btrfs_item_key(right, &disk_key, 0); |
3741 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3742 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3743 | |
00ec4c51 | 3744 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3745 | if (path->slots[0] >= left_nritems) { |
3746 | path->slots[0] -= left_nritems; | |
925baedd CM |
3747 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3748 | clean_tree_block(trans, root, path->nodes[0]); | |
3749 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3750 | free_extent_buffer(path->nodes[0]); |
3751 | path->nodes[0] = right; | |
00ec4c51 CM |
3752 | path->slots[1] += 1; |
3753 | } else { | |
925baedd | 3754 | btrfs_tree_unlock(right); |
5f39d397 | 3755 | free_extent_buffer(right); |
00ec4c51 CM |
3756 | } |
3757 | return 0; | |
925baedd CM |
3758 | |
3759 | out_unlock: | |
3760 | btrfs_tree_unlock(right); | |
3761 | free_extent_buffer(right); | |
3762 | return 1; | |
00ec4c51 | 3763 | } |
925baedd | 3764 | |
44871b1b CM |
3765 | /* |
3766 | * push some data in the path leaf to the right, trying to free up at | |
3767 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3768 | * | |
3769 | * returns 1 if the push failed because the other node didn't have enough | |
3770 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3771 | * |
3772 | * this will push starting from min_slot to the end of the leaf. It won't | |
3773 | * push any slot lower than min_slot | |
44871b1b CM |
3774 | */ |
3775 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3776 | *root, struct btrfs_path *path, |
3777 | int min_data_size, int data_size, | |
3778 | int empty, u32 min_slot) | |
44871b1b CM |
3779 | { |
3780 | struct extent_buffer *left = path->nodes[0]; | |
3781 | struct extent_buffer *right; | |
3782 | struct extent_buffer *upper; | |
3783 | int slot; | |
3784 | int free_space; | |
3785 | u32 left_nritems; | |
3786 | int ret; | |
3787 | ||
3788 | if (!path->nodes[1]) | |
3789 | return 1; | |
3790 | ||
3791 | slot = path->slots[1]; | |
3792 | upper = path->nodes[1]; | |
3793 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3794 | return 1; | |
3795 | ||
3796 | btrfs_assert_tree_locked(path->nodes[1]); | |
3797 | ||
3798 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3799 | if (right == NULL) |
3800 | return 1; | |
3801 | ||
44871b1b CM |
3802 | btrfs_tree_lock(right); |
3803 | btrfs_set_lock_blocking(right); | |
3804 | ||
3805 | free_space = btrfs_leaf_free_space(root, right); | |
3806 | if (free_space < data_size) | |
3807 | goto out_unlock; | |
3808 | ||
3809 | /* cow and double check */ | |
3810 | ret = btrfs_cow_block(trans, root, right, upper, | |
3811 | slot + 1, &right); | |
3812 | if (ret) | |
3813 | goto out_unlock; | |
3814 | ||
3815 | free_space = btrfs_leaf_free_space(root, right); | |
3816 | if (free_space < data_size) | |
3817 | goto out_unlock; | |
3818 | ||
3819 | left_nritems = btrfs_header_nritems(left); | |
3820 | if (left_nritems == 0) | |
3821 | goto out_unlock; | |
3822 | ||
2ef1fed2 FDBM |
3823 | if (path->slots[0] == left_nritems && !empty) { |
3824 | /* Key greater than all keys in the leaf, right neighbor has | |
3825 | * enough room for it and we're not emptying our leaf to delete | |
3826 | * it, therefore use right neighbor to insert the new item and | |
3827 | * no need to touch/dirty our left leaft. */ | |
3828 | btrfs_tree_unlock(left); | |
3829 | free_extent_buffer(left); | |
3830 | path->nodes[0] = right; | |
3831 | path->slots[0] = 0; | |
3832 | path->slots[1]++; | |
3833 | return 0; | |
3834 | } | |
3835 | ||
99d8f83c CM |
3836 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3837 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3838 | out_unlock: |
3839 | btrfs_tree_unlock(right); | |
3840 | free_extent_buffer(right); | |
3841 | return 1; | |
3842 | } | |
3843 | ||
74123bd7 CM |
3844 | /* |
3845 | * push some data in the path leaf to the left, trying to free up at | |
3846 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3847 | * |
3848 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3849 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3850 | * items | |
74123bd7 | 3851 | */ |
44871b1b CM |
3852 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3853 | struct btrfs_root *root, | |
3854 | struct btrfs_path *path, int data_size, | |
3855 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3856 | int free_space, u32 right_nritems, |
3857 | u32 max_slot) | |
be0e5c09 | 3858 | { |
5f39d397 CM |
3859 | struct btrfs_disk_key disk_key; |
3860 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3861 | int i; |
be0e5c09 CM |
3862 | int push_space = 0; |
3863 | int push_items = 0; | |
0783fcfc | 3864 | struct btrfs_item *item; |
7518a238 | 3865 | u32 old_left_nritems; |
34a38218 | 3866 | u32 nr; |
aa5d6bed | 3867 | int ret = 0; |
db94535d CM |
3868 | u32 this_item_size; |
3869 | u32 old_left_item_size; | |
cfed81a0 CM |
3870 | struct btrfs_map_token token; |
3871 | ||
3872 | btrfs_init_map_token(&token); | |
be0e5c09 | 3873 | |
34a38218 | 3874 | if (empty) |
99d8f83c | 3875 | nr = min(right_nritems, max_slot); |
34a38218 | 3876 | else |
99d8f83c | 3877 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3878 | |
3879 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3880 | item = btrfs_item_nr(i); |
db94535d | 3881 | |
31840ae1 ZY |
3882 | if (!empty && push_items > 0) { |
3883 | if (path->slots[0] < i) | |
3884 | break; | |
3885 | if (path->slots[0] == i) { | |
3886 | int space = btrfs_leaf_free_space(root, right); | |
3887 | if (space + push_space * 2 > free_space) | |
3888 | break; | |
3889 | } | |
3890 | } | |
3891 | ||
be0e5c09 | 3892 | if (path->slots[0] == i) |
87b29b20 | 3893 | push_space += data_size; |
db94535d CM |
3894 | |
3895 | this_item_size = btrfs_item_size(right, item); | |
3896 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3897 | break; |
db94535d | 3898 | |
be0e5c09 | 3899 | push_items++; |
db94535d CM |
3900 | push_space += this_item_size + sizeof(*item); |
3901 | } | |
3902 | ||
be0e5c09 | 3903 | if (push_items == 0) { |
925baedd CM |
3904 | ret = 1; |
3905 | goto out; | |
be0e5c09 | 3906 | } |
fae7f21c | 3907 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3908 | |
be0e5c09 | 3909 | /* push data from right to left */ |
5f39d397 CM |
3910 | copy_extent_buffer(left, right, |
3911 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3912 | btrfs_item_nr_offset(0), | |
3913 | push_items * sizeof(struct btrfs_item)); | |
3914 | ||
123abc88 | 3915 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3916 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3917 | |
3918 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3919 | leaf_data_end(root, left) - push_space, |
3920 | btrfs_leaf_data(right) + | |
5f39d397 | 3921 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3922 | push_space); |
5f39d397 | 3923 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3924 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3925 | |
db94535d | 3926 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3927 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3928 | u32 ioff; |
db94535d | 3929 | |
dd3cc16b | 3930 | item = btrfs_item_nr(i); |
db94535d | 3931 | |
cfed81a0 CM |
3932 | ioff = btrfs_token_item_offset(left, item, &token); |
3933 | btrfs_set_token_item_offset(left, item, | |
3934 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3935 | &token); | |
be0e5c09 | 3936 | } |
5f39d397 | 3937 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3938 | |
3939 | /* fixup right node */ | |
31b1a2bd JL |
3940 | if (push_items > right_nritems) |
3941 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3942 | right_nritems); |
34a38218 CM |
3943 | |
3944 | if (push_items < right_nritems) { | |
3945 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3946 | leaf_data_end(root, right); | |
3947 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3948 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3949 | btrfs_leaf_data(right) + | |
3950 | leaf_data_end(root, right), push_space); | |
3951 | ||
3952 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3953 | btrfs_item_nr_offset(push_items), |
3954 | (btrfs_header_nritems(right) - push_items) * | |
3955 | sizeof(struct btrfs_item)); | |
34a38218 | 3956 | } |
eef1c494 Y |
3957 | right_nritems -= push_items; |
3958 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3959 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 3960 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3961 | item = btrfs_item_nr(i); |
db94535d | 3962 | |
cfed81a0 CM |
3963 | push_space = push_space - btrfs_token_item_size(right, |
3964 | item, &token); | |
3965 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3966 | } |
eb60ceac | 3967 | |
5f39d397 | 3968 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3969 | if (right_nritems) |
3970 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3971 | else |
3972 | clean_tree_block(trans, root, right); | |
098f59c2 | 3973 | |
5f39d397 | 3974 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3975 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3976 | |
3977 | /* then fixup the leaf pointer in the path */ | |
3978 | if (path->slots[0] < push_items) { | |
3979 | path->slots[0] += old_left_nritems; | |
925baedd | 3980 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3981 | free_extent_buffer(path->nodes[0]); |
3982 | path->nodes[0] = left; | |
be0e5c09 CM |
3983 | path->slots[1] -= 1; |
3984 | } else { | |
925baedd | 3985 | btrfs_tree_unlock(left); |
5f39d397 | 3986 | free_extent_buffer(left); |
be0e5c09 CM |
3987 | path->slots[0] -= push_items; |
3988 | } | |
eb60ceac | 3989 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3990 | return ret; |
925baedd CM |
3991 | out: |
3992 | btrfs_tree_unlock(left); | |
3993 | free_extent_buffer(left); | |
3994 | return ret; | |
be0e5c09 CM |
3995 | } |
3996 | ||
44871b1b CM |
3997 | /* |
3998 | * push some data in the path leaf to the left, trying to free up at | |
3999 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
4000 | * |
4001 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
4002 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
4003 | * items | |
44871b1b CM |
4004 | */ |
4005 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
4006 | *root, struct btrfs_path *path, int min_data_size, |
4007 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
4008 | { |
4009 | struct extent_buffer *right = path->nodes[0]; | |
4010 | struct extent_buffer *left; | |
4011 | int slot; | |
4012 | int free_space; | |
4013 | u32 right_nritems; | |
4014 | int ret = 0; | |
4015 | ||
4016 | slot = path->slots[1]; | |
4017 | if (slot == 0) | |
4018 | return 1; | |
4019 | if (!path->nodes[1]) | |
4020 | return 1; | |
4021 | ||
4022 | right_nritems = btrfs_header_nritems(right); | |
4023 | if (right_nritems == 0) | |
4024 | return 1; | |
4025 | ||
4026 | btrfs_assert_tree_locked(path->nodes[1]); | |
4027 | ||
4028 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
4029 | if (left == NULL) |
4030 | return 1; | |
4031 | ||
44871b1b CM |
4032 | btrfs_tree_lock(left); |
4033 | btrfs_set_lock_blocking(left); | |
4034 | ||
4035 | free_space = btrfs_leaf_free_space(root, left); | |
4036 | if (free_space < data_size) { | |
4037 | ret = 1; | |
4038 | goto out; | |
4039 | } | |
4040 | ||
4041 | /* cow and double check */ | |
4042 | ret = btrfs_cow_block(trans, root, left, | |
4043 | path->nodes[1], slot - 1, &left); | |
4044 | if (ret) { | |
4045 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
4046 | if (ret == -ENOSPC) |
4047 | ret = 1; | |
44871b1b CM |
4048 | goto out; |
4049 | } | |
4050 | ||
4051 | free_space = btrfs_leaf_free_space(root, left); | |
4052 | if (free_space < data_size) { | |
4053 | ret = 1; | |
4054 | goto out; | |
4055 | } | |
4056 | ||
99d8f83c CM |
4057 | return __push_leaf_left(trans, root, path, min_data_size, |
4058 | empty, left, free_space, right_nritems, | |
4059 | max_slot); | |
44871b1b CM |
4060 | out: |
4061 | btrfs_tree_unlock(left); | |
4062 | free_extent_buffer(left); | |
4063 | return ret; | |
4064 | } | |
4065 | ||
4066 | /* | |
4067 | * split the path's leaf in two, making sure there is at least data_size | |
4068 | * available for the resulting leaf level of the path. | |
44871b1b | 4069 | */ |
143bede5 JM |
4070 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
4071 | struct btrfs_root *root, | |
4072 | struct btrfs_path *path, | |
4073 | struct extent_buffer *l, | |
4074 | struct extent_buffer *right, | |
4075 | int slot, int mid, int nritems) | |
44871b1b CM |
4076 | { |
4077 | int data_copy_size; | |
4078 | int rt_data_off; | |
4079 | int i; | |
44871b1b | 4080 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
4081 | struct btrfs_map_token token; |
4082 | ||
4083 | btrfs_init_map_token(&token); | |
44871b1b CM |
4084 | |
4085 | nritems = nritems - mid; | |
4086 | btrfs_set_header_nritems(right, nritems); | |
4087 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
4088 | ||
4089 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
4090 | btrfs_item_nr_offset(mid), | |
4091 | nritems * sizeof(struct btrfs_item)); | |
4092 | ||
4093 | copy_extent_buffer(right, l, | |
4094 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
4095 | data_copy_size, btrfs_leaf_data(l) + | |
4096 | leaf_data_end(root, l), data_copy_size); | |
4097 | ||
4098 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
4099 | btrfs_item_end_nr(l, mid); | |
4100 | ||
4101 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 4102 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
4103 | u32 ioff; |
4104 | ||
cfed81a0 CM |
4105 | ioff = btrfs_token_item_offset(right, item, &token); |
4106 | btrfs_set_token_item_offset(right, item, | |
4107 | ioff + rt_data_off, &token); | |
44871b1b CM |
4108 | } |
4109 | ||
44871b1b | 4110 | btrfs_set_header_nritems(l, mid); |
44871b1b | 4111 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 4112 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4113 | path->slots[1] + 1, 1); |
44871b1b CM |
4114 | |
4115 | btrfs_mark_buffer_dirty(right); | |
4116 | btrfs_mark_buffer_dirty(l); | |
4117 | BUG_ON(path->slots[0] != slot); | |
4118 | ||
44871b1b CM |
4119 | if (mid <= slot) { |
4120 | btrfs_tree_unlock(path->nodes[0]); | |
4121 | free_extent_buffer(path->nodes[0]); | |
4122 | path->nodes[0] = right; | |
4123 | path->slots[0] -= mid; | |
4124 | path->slots[1] += 1; | |
4125 | } else { | |
4126 | btrfs_tree_unlock(right); | |
4127 | free_extent_buffer(right); | |
4128 | } | |
4129 | ||
4130 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
4131 | } |
4132 | ||
99d8f83c CM |
4133 | /* |
4134 | * double splits happen when we need to insert a big item in the middle | |
4135 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
4136 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
4137 | * A B C | |
4138 | * | |
4139 | * We avoid this by trying to push the items on either side of our target | |
4140 | * into the adjacent leaves. If all goes well we can avoid the double split | |
4141 | * completely. | |
4142 | */ | |
4143 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
4144 | struct btrfs_root *root, | |
4145 | struct btrfs_path *path, | |
4146 | int data_size) | |
4147 | { | |
4148 | int ret; | |
4149 | int progress = 0; | |
4150 | int slot; | |
4151 | u32 nritems; | |
5a4267ca | 4152 | int space_needed = data_size; |
99d8f83c CM |
4153 | |
4154 | slot = path->slots[0]; | |
5a4267ca FDBM |
4155 | if (slot < btrfs_header_nritems(path->nodes[0])) |
4156 | space_needed -= btrfs_leaf_free_space(root, path->nodes[0]); | |
99d8f83c CM |
4157 | |
4158 | /* | |
4159 | * try to push all the items after our slot into the | |
4160 | * right leaf | |
4161 | */ | |
5a4267ca | 4162 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4163 | if (ret < 0) |
4164 | return ret; | |
4165 | ||
4166 | if (ret == 0) | |
4167 | progress++; | |
4168 | ||
4169 | nritems = btrfs_header_nritems(path->nodes[0]); | |
4170 | /* | |
4171 | * our goal is to get our slot at the start or end of a leaf. If | |
4172 | * we've done so we're done | |
4173 | */ | |
4174 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
4175 | return 0; | |
4176 | ||
4177 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4178 | return 0; | |
4179 | ||
4180 | /* try to push all the items before our slot into the next leaf */ | |
4181 | slot = path->slots[0]; | |
5a4267ca | 4182 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
4183 | if (ret < 0) |
4184 | return ret; | |
4185 | ||
4186 | if (ret == 0) | |
4187 | progress++; | |
4188 | ||
4189 | if (progress) | |
4190 | return 0; | |
4191 | return 1; | |
4192 | } | |
4193 | ||
74123bd7 CM |
4194 | /* |
4195 | * split the path's leaf in two, making sure there is at least data_size | |
4196 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
4197 | * |
4198 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 4199 | */ |
e02119d5 CM |
4200 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
4201 | struct btrfs_root *root, | |
4202 | struct btrfs_key *ins_key, | |
4203 | struct btrfs_path *path, int data_size, | |
4204 | int extend) | |
be0e5c09 | 4205 | { |
5d4f98a2 | 4206 | struct btrfs_disk_key disk_key; |
5f39d397 | 4207 | struct extent_buffer *l; |
7518a238 | 4208 | u32 nritems; |
eb60ceac CM |
4209 | int mid; |
4210 | int slot; | |
5f39d397 | 4211 | struct extent_buffer *right; |
d4dbff95 | 4212 | int ret = 0; |
aa5d6bed | 4213 | int wret; |
5d4f98a2 | 4214 | int split; |
cc0c5538 | 4215 | int num_doubles = 0; |
99d8f83c | 4216 | int tried_avoid_double = 0; |
aa5d6bed | 4217 | |
a5719521 YZ |
4218 | l = path->nodes[0]; |
4219 | slot = path->slots[0]; | |
4220 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
4221 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
4222 | return -EOVERFLOW; | |
4223 | ||
40689478 | 4224 | /* first try to make some room by pushing left and right */ |
33157e05 | 4225 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4226 | int space_needed = data_size; |
4227 | ||
4228 | if (slot < btrfs_header_nritems(l)) | |
4229 | space_needed -= btrfs_leaf_free_space(root, l); | |
4230 | ||
4231 | wret = push_leaf_right(trans, root, path, space_needed, | |
4232 | space_needed, 0, 0); | |
d397712b | 4233 | if (wret < 0) |
eaee50e8 | 4234 | return wret; |
3685f791 | 4235 | if (wret) { |
5a4267ca FDBM |
4236 | wret = push_leaf_left(trans, root, path, space_needed, |
4237 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4238 | if (wret < 0) |
4239 | return wret; | |
4240 | } | |
4241 | l = path->nodes[0]; | |
aa5d6bed | 4242 | |
3685f791 | 4243 | /* did the pushes work? */ |
87b29b20 | 4244 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 4245 | return 0; |
3326d1b0 | 4246 | } |
aa5d6bed | 4247 | |
5c680ed6 | 4248 | if (!path->nodes[1]) { |
fdd99c72 | 4249 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4250 | if (ret) |
4251 | return ret; | |
4252 | } | |
cc0c5538 | 4253 | again: |
5d4f98a2 | 4254 | split = 1; |
cc0c5538 | 4255 | l = path->nodes[0]; |
eb60ceac | 4256 | slot = path->slots[0]; |
5f39d397 | 4257 | nritems = btrfs_header_nritems(l); |
d397712b | 4258 | mid = (nritems + 1) / 2; |
54aa1f4d | 4259 | |
5d4f98a2 YZ |
4260 | if (mid <= slot) { |
4261 | if (nritems == 1 || | |
4262 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
4263 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4264 | if (slot >= nritems) { | |
4265 | split = 0; | |
4266 | } else { | |
4267 | mid = slot; | |
4268 | if (mid != nritems && | |
4269 | leaf_space_used(l, mid, nritems - mid) + | |
4270 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4271 | if (data_size && !tried_avoid_double) |
4272 | goto push_for_double; | |
5d4f98a2 YZ |
4273 | split = 2; |
4274 | } | |
4275 | } | |
4276 | } | |
4277 | } else { | |
4278 | if (leaf_space_used(l, 0, mid) + data_size > | |
4279 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4280 | if (!extend && data_size && slot == 0) { | |
4281 | split = 0; | |
4282 | } else if ((extend || !data_size) && slot == 0) { | |
4283 | mid = 1; | |
4284 | } else { | |
4285 | mid = slot; | |
4286 | if (mid != nritems && | |
4287 | leaf_space_used(l, mid, nritems - mid) + | |
4288 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4289 | if (data_size && !tried_avoid_double) |
4290 | goto push_for_double; | |
67871254 | 4291 | split = 2; |
5d4f98a2 YZ |
4292 | } |
4293 | } | |
4294 | } | |
4295 | } | |
4296 | ||
4297 | if (split == 0) | |
4298 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4299 | else | |
4300 | btrfs_item_key(l, &disk_key, mid); | |
4301 | ||
4302 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 4303 | root->root_key.objectid, |
5581a51a | 4304 | &disk_key, 0, l->start, 0); |
f0486c68 | 4305 | if (IS_ERR(right)) |
5f39d397 | 4306 | return PTR_ERR(right); |
f0486c68 YZ |
4307 | |
4308 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4309 | |
4310 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4311 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4312 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4313 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4314 | btrfs_set_header_owner(right, root->root_key.objectid); |
4315 | btrfs_set_header_level(right, 0); | |
4316 | write_extent_buffer(right, root->fs_info->fsid, | |
0a4e5586 | 4317 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 CM |
4318 | |
4319 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 4320 | btrfs_header_chunk_tree_uuid(right), |
e17cade2 | 4321 | BTRFS_UUID_SIZE); |
44871b1b | 4322 | |
5d4f98a2 YZ |
4323 | if (split == 0) { |
4324 | if (mid <= slot) { | |
4325 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4326 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4327 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4328 | btrfs_tree_unlock(path->nodes[0]); |
4329 | free_extent_buffer(path->nodes[0]); | |
4330 | path->nodes[0] = right; | |
4331 | path->slots[0] = 0; | |
4332 | path->slots[1] += 1; | |
4333 | } else { | |
4334 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4335 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4336 | path->slots[1], 1); |
5d4f98a2 YZ |
4337 | btrfs_tree_unlock(path->nodes[0]); |
4338 | free_extent_buffer(path->nodes[0]); | |
4339 | path->nodes[0] = right; | |
4340 | path->slots[0] = 0; | |
143bede5 | 4341 | if (path->slots[1] == 0) |
d6a0a126 | 4342 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4343 | } |
5d4f98a2 YZ |
4344 | btrfs_mark_buffer_dirty(right); |
4345 | return ret; | |
d4dbff95 | 4346 | } |
74123bd7 | 4347 | |
143bede5 | 4348 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4349 | |
5d4f98a2 | 4350 | if (split == 2) { |
cc0c5538 CM |
4351 | BUG_ON(num_doubles != 0); |
4352 | num_doubles++; | |
4353 | goto again; | |
a429e513 | 4354 | } |
44871b1b | 4355 | |
143bede5 | 4356 | return 0; |
99d8f83c CM |
4357 | |
4358 | push_for_double: | |
4359 | push_for_double_split(trans, root, path, data_size); | |
4360 | tried_avoid_double = 1; | |
4361 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4362 | return 0; | |
4363 | goto again; | |
be0e5c09 CM |
4364 | } |
4365 | ||
ad48fd75 YZ |
4366 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4367 | struct btrfs_root *root, | |
4368 | struct btrfs_path *path, int ins_len) | |
459931ec | 4369 | { |
ad48fd75 | 4370 | struct btrfs_key key; |
459931ec | 4371 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4372 | struct btrfs_file_extent_item *fi; |
4373 | u64 extent_len = 0; | |
4374 | u32 item_size; | |
4375 | int ret; | |
459931ec CM |
4376 | |
4377 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4378 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4379 | ||
4380 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4381 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4382 | ||
4383 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4384 | return 0; | |
459931ec CM |
4385 | |
4386 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4387 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4388 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4389 | struct btrfs_file_extent_item); | |
4390 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4391 | } | |
b3b4aa74 | 4392 | btrfs_release_path(path); |
459931ec | 4393 | |
459931ec | 4394 | path->keep_locks = 1; |
ad48fd75 YZ |
4395 | path->search_for_split = 1; |
4396 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4397 | path->search_for_split = 0; |
ad48fd75 YZ |
4398 | if (ret < 0) |
4399 | goto err; | |
459931ec | 4400 | |
ad48fd75 YZ |
4401 | ret = -EAGAIN; |
4402 | leaf = path->nodes[0]; | |
459931ec | 4403 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4404 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4405 | goto err; | |
4406 | ||
109f6aef CM |
4407 | /* the leaf has changed, it now has room. return now */ |
4408 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4409 | goto err; | |
4410 | ||
ad48fd75 YZ |
4411 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4412 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4413 | struct btrfs_file_extent_item); | |
4414 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4415 | goto err; | |
459931ec CM |
4416 | } |
4417 | ||
b9473439 | 4418 | btrfs_set_path_blocking(path); |
ad48fd75 | 4419 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4420 | if (ret) |
4421 | goto err; | |
459931ec | 4422 | |
ad48fd75 | 4423 | path->keep_locks = 0; |
b9473439 | 4424 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4425 | return 0; |
4426 | err: | |
4427 | path->keep_locks = 0; | |
4428 | return ret; | |
4429 | } | |
4430 | ||
4431 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4432 | struct btrfs_root *root, | |
4433 | struct btrfs_path *path, | |
4434 | struct btrfs_key *new_key, | |
4435 | unsigned long split_offset) | |
4436 | { | |
4437 | struct extent_buffer *leaf; | |
4438 | struct btrfs_item *item; | |
4439 | struct btrfs_item *new_item; | |
4440 | int slot; | |
4441 | char *buf; | |
4442 | u32 nritems; | |
4443 | u32 item_size; | |
4444 | u32 orig_offset; | |
4445 | struct btrfs_disk_key disk_key; | |
4446 | ||
b9473439 CM |
4447 | leaf = path->nodes[0]; |
4448 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4449 | ||
b4ce94de CM |
4450 | btrfs_set_path_blocking(path); |
4451 | ||
dd3cc16b | 4452 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4453 | orig_offset = btrfs_item_offset(leaf, item); |
4454 | item_size = btrfs_item_size(leaf, item); | |
4455 | ||
459931ec | 4456 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4457 | if (!buf) |
4458 | return -ENOMEM; | |
4459 | ||
459931ec CM |
4460 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4461 | path->slots[0]), item_size); | |
459931ec | 4462 | |
ad48fd75 | 4463 | slot = path->slots[0] + 1; |
459931ec | 4464 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4465 | if (slot != nritems) { |
4466 | /* shift the items */ | |
4467 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4468 | btrfs_item_nr_offset(slot), |
4469 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4470 | } |
4471 | ||
4472 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4473 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4474 | ||
dd3cc16b | 4475 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4476 | |
4477 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4478 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4479 | ||
4480 | btrfs_set_item_offset(leaf, item, | |
4481 | orig_offset + item_size - split_offset); | |
4482 | btrfs_set_item_size(leaf, item, split_offset); | |
4483 | ||
4484 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4485 | ||
4486 | /* write the data for the start of the original item */ | |
4487 | write_extent_buffer(leaf, buf, | |
4488 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4489 | split_offset); | |
4490 | ||
4491 | /* write the data for the new item */ | |
4492 | write_extent_buffer(leaf, buf + split_offset, | |
4493 | btrfs_item_ptr_offset(leaf, slot), | |
4494 | item_size - split_offset); | |
4495 | btrfs_mark_buffer_dirty(leaf); | |
4496 | ||
ad48fd75 | 4497 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4498 | kfree(buf); |
ad48fd75 YZ |
4499 | return 0; |
4500 | } | |
4501 | ||
4502 | /* | |
4503 | * This function splits a single item into two items, | |
4504 | * giving 'new_key' to the new item and splitting the | |
4505 | * old one at split_offset (from the start of the item). | |
4506 | * | |
4507 | * The path may be released by this operation. After | |
4508 | * the split, the path is pointing to the old item. The | |
4509 | * new item is going to be in the same node as the old one. | |
4510 | * | |
4511 | * Note, the item being split must be smaller enough to live alone on | |
4512 | * a tree block with room for one extra struct btrfs_item | |
4513 | * | |
4514 | * This allows us to split the item in place, keeping a lock on the | |
4515 | * leaf the entire time. | |
4516 | */ | |
4517 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4518 | struct btrfs_root *root, | |
4519 | struct btrfs_path *path, | |
4520 | struct btrfs_key *new_key, | |
4521 | unsigned long split_offset) | |
4522 | { | |
4523 | int ret; | |
4524 | ret = setup_leaf_for_split(trans, root, path, | |
4525 | sizeof(struct btrfs_item)); | |
4526 | if (ret) | |
4527 | return ret; | |
4528 | ||
4529 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4530 | return ret; |
4531 | } | |
4532 | ||
ad48fd75 YZ |
4533 | /* |
4534 | * This function duplicate a item, giving 'new_key' to the new item. | |
4535 | * It guarantees both items live in the same tree leaf and the new item | |
4536 | * is contiguous with the original item. | |
4537 | * | |
4538 | * This allows us to split file extent in place, keeping a lock on the | |
4539 | * leaf the entire time. | |
4540 | */ | |
4541 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4542 | struct btrfs_root *root, | |
4543 | struct btrfs_path *path, | |
4544 | struct btrfs_key *new_key) | |
4545 | { | |
4546 | struct extent_buffer *leaf; | |
4547 | int ret; | |
4548 | u32 item_size; | |
4549 | ||
4550 | leaf = path->nodes[0]; | |
4551 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4552 | ret = setup_leaf_for_split(trans, root, path, | |
4553 | item_size + sizeof(struct btrfs_item)); | |
4554 | if (ret) | |
4555 | return ret; | |
4556 | ||
4557 | path->slots[0]++; | |
afe5fea7 | 4558 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4559 | item_size, item_size + |
4560 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4561 | leaf = path->nodes[0]; |
4562 | memcpy_extent_buffer(leaf, | |
4563 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4564 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4565 | item_size); | |
4566 | return 0; | |
4567 | } | |
4568 | ||
d352ac68 CM |
4569 | /* |
4570 | * make the item pointed to by the path smaller. new_size indicates | |
4571 | * how small to make it, and from_end tells us if we just chop bytes | |
4572 | * off the end of the item or if we shift the item to chop bytes off | |
4573 | * the front. | |
4574 | */ | |
afe5fea7 | 4575 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4576 | u32 new_size, int from_end) |
b18c6685 | 4577 | { |
b18c6685 | 4578 | int slot; |
5f39d397 CM |
4579 | struct extent_buffer *leaf; |
4580 | struct btrfs_item *item; | |
b18c6685 CM |
4581 | u32 nritems; |
4582 | unsigned int data_end; | |
4583 | unsigned int old_data_start; | |
4584 | unsigned int old_size; | |
4585 | unsigned int size_diff; | |
4586 | int i; | |
cfed81a0 CM |
4587 | struct btrfs_map_token token; |
4588 | ||
4589 | btrfs_init_map_token(&token); | |
b18c6685 | 4590 | |
5f39d397 | 4591 | leaf = path->nodes[0]; |
179e29e4 CM |
4592 | slot = path->slots[0]; |
4593 | ||
4594 | old_size = btrfs_item_size_nr(leaf, slot); | |
4595 | if (old_size == new_size) | |
143bede5 | 4596 | return; |
b18c6685 | 4597 | |
5f39d397 | 4598 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4599 | data_end = leaf_data_end(root, leaf); |
4600 | ||
5f39d397 | 4601 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4602 | |
b18c6685 CM |
4603 | size_diff = old_size - new_size; |
4604 | ||
4605 | BUG_ON(slot < 0); | |
4606 | BUG_ON(slot >= nritems); | |
4607 | ||
4608 | /* | |
4609 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4610 | */ | |
4611 | /* first correct the data pointers */ | |
4612 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4613 | u32 ioff; |
dd3cc16b | 4614 | item = btrfs_item_nr(i); |
db94535d | 4615 | |
cfed81a0 CM |
4616 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4617 | btrfs_set_token_item_offset(leaf, item, | |
4618 | ioff + size_diff, &token); | |
b18c6685 | 4619 | } |
db94535d | 4620 | |
b18c6685 | 4621 | /* shift the data */ |
179e29e4 CM |
4622 | if (from_end) { |
4623 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4624 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4625 | data_end, old_data_start + new_size - data_end); | |
4626 | } else { | |
4627 | struct btrfs_disk_key disk_key; | |
4628 | u64 offset; | |
4629 | ||
4630 | btrfs_item_key(leaf, &disk_key, slot); | |
4631 | ||
4632 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4633 | unsigned long ptr; | |
4634 | struct btrfs_file_extent_item *fi; | |
4635 | ||
4636 | fi = btrfs_item_ptr(leaf, slot, | |
4637 | struct btrfs_file_extent_item); | |
4638 | fi = (struct btrfs_file_extent_item *)( | |
4639 | (unsigned long)fi - size_diff); | |
4640 | ||
4641 | if (btrfs_file_extent_type(leaf, fi) == | |
4642 | BTRFS_FILE_EXTENT_INLINE) { | |
4643 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4644 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4645 | (unsigned long)fi, |
4646 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4647 | disk_bytenr)); |
4648 | } | |
4649 | } | |
4650 | ||
4651 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4652 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4653 | data_end, old_data_start - data_end); | |
4654 | ||
4655 | offset = btrfs_disk_key_offset(&disk_key); | |
4656 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4657 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4658 | if (slot == 0) | |
d6a0a126 | 4659 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4660 | } |
5f39d397 | 4661 | |
dd3cc16b | 4662 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4663 | btrfs_set_item_size(leaf, item, new_size); |
4664 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4665 | |
5f39d397 CM |
4666 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4667 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4668 | BUG(); |
5f39d397 | 4669 | } |
b18c6685 CM |
4670 | } |
4671 | ||
d352ac68 | 4672 | /* |
8f69dbd2 | 4673 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4674 | */ |
4b90c680 | 4675 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4676 | u32 data_size) |
6567e837 | 4677 | { |
6567e837 | 4678 | int slot; |
5f39d397 CM |
4679 | struct extent_buffer *leaf; |
4680 | struct btrfs_item *item; | |
6567e837 CM |
4681 | u32 nritems; |
4682 | unsigned int data_end; | |
4683 | unsigned int old_data; | |
4684 | unsigned int old_size; | |
4685 | int i; | |
cfed81a0 CM |
4686 | struct btrfs_map_token token; |
4687 | ||
4688 | btrfs_init_map_token(&token); | |
6567e837 | 4689 | |
5f39d397 | 4690 | leaf = path->nodes[0]; |
6567e837 | 4691 | |
5f39d397 | 4692 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4693 | data_end = leaf_data_end(root, leaf); |
4694 | ||
5f39d397 CM |
4695 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4696 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4697 | BUG(); |
5f39d397 | 4698 | } |
6567e837 | 4699 | slot = path->slots[0]; |
5f39d397 | 4700 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4701 | |
4702 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4703 | if (slot >= nritems) { |
4704 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4705 | btrfs_crit(root->fs_info, "slot %d too large, nritems %d", |
d397712b | 4706 | slot, nritems); |
3326d1b0 CM |
4707 | BUG_ON(1); |
4708 | } | |
6567e837 CM |
4709 | |
4710 | /* | |
4711 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4712 | */ | |
4713 | /* first correct the data pointers */ | |
4714 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4715 | u32 ioff; |
dd3cc16b | 4716 | item = btrfs_item_nr(i); |
db94535d | 4717 | |
cfed81a0 CM |
4718 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4719 | btrfs_set_token_item_offset(leaf, item, | |
4720 | ioff - data_size, &token); | |
6567e837 | 4721 | } |
5f39d397 | 4722 | |
6567e837 | 4723 | /* shift the data */ |
5f39d397 | 4724 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4725 | data_end - data_size, btrfs_leaf_data(leaf) + |
4726 | data_end, old_data - data_end); | |
5f39d397 | 4727 | |
6567e837 | 4728 | data_end = old_data; |
5f39d397 | 4729 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4730 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4731 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4732 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4733 | |
5f39d397 CM |
4734 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4735 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4736 | BUG(); |
5f39d397 | 4737 | } |
6567e837 CM |
4738 | } |
4739 | ||
74123bd7 | 4740 | /* |
44871b1b CM |
4741 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4742 | * to save stack depth by doing the bulk of the work in a function | |
4743 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4744 | */ |
afe5fea7 | 4745 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4746 | struct btrfs_key *cpu_key, u32 *data_size, |
4747 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4748 | { |
5f39d397 | 4749 | struct btrfs_item *item; |
9c58309d | 4750 | int i; |
7518a238 | 4751 | u32 nritems; |
be0e5c09 | 4752 | unsigned int data_end; |
e2fa7227 | 4753 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4754 | struct extent_buffer *leaf; |
4755 | int slot; | |
cfed81a0 CM |
4756 | struct btrfs_map_token token; |
4757 | ||
4758 | btrfs_init_map_token(&token); | |
e2fa7227 | 4759 | |
5f39d397 | 4760 | leaf = path->nodes[0]; |
44871b1b | 4761 | slot = path->slots[0]; |
74123bd7 | 4762 | |
5f39d397 | 4763 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4764 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4765 | |
f25956cc | 4766 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4767 | btrfs_print_leaf(root, leaf); |
efe120a0 | 4768 | btrfs_crit(root->fs_info, "not enough freespace need %u have %d", |
9c58309d | 4769 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4770 | BUG(); |
d4dbff95 | 4771 | } |
5f39d397 | 4772 | |
be0e5c09 | 4773 | if (slot != nritems) { |
5f39d397 | 4774 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4775 | |
5f39d397 CM |
4776 | if (old_data < data_end) { |
4777 | btrfs_print_leaf(root, leaf); | |
efe120a0 | 4778 | btrfs_crit(root->fs_info, "slot %d old_data %d data_end %d", |
5f39d397 CM |
4779 | slot, old_data, data_end); |
4780 | BUG_ON(1); | |
4781 | } | |
be0e5c09 CM |
4782 | /* |
4783 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4784 | */ | |
4785 | /* first correct the data pointers */ | |
0783fcfc | 4786 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4787 | u32 ioff; |
db94535d | 4788 | |
dd3cc16b | 4789 | item = btrfs_item_nr( i); |
cfed81a0 CM |
4790 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4791 | btrfs_set_token_item_offset(leaf, item, | |
4792 | ioff - total_data, &token); | |
0783fcfc | 4793 | } |
be0e5c09 | 4794 | /* shift the items */ |
9c58309d | 4795 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4796 | btrfs_item_nr_offset(slot), |
d6025579 | 4797 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4798 | |
4799 | /* shift the data */ | |
5f39d397 | 4800 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4801 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4802 | data_end, old_data - data_end); |
be0e5c09 CM |
4803 | data_end = old_data; |
4804 | } | |
5f39d397 | 4805 | |
62e2749e | 4806 | /* setup the item for the new data */ |
9c58309d CM |
4807 | for (i = 0; i < nr; i++) { |
4808 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4809 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4810 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4811 | btrfs_set_token_item_offset(leaf, item, |
4812 | data_end - data_size[i], &token); | |
9c58309d | 4813 | data_end -= data_size[i]; |
cfed81a0 | 4814 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4815 | } |
44871b1b | 4816 | |
9c58309d | 4817 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4818 | |
5a01a2e3 CM |
4819 | if (slot == 0) { |
4820 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4821 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4822 | } |
b9473439 CM |
4823 | btrfs_unlock_up_safe(path, 1); |
4824 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4825 | |
5f39d397 CM |
4826 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4827 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4828 | BUG(); |
5f39d397 | 4829 | } |
44871b1b CM |
4830 | } |
4831 | ||
4832 | /* | |
4833 | * Given a key and some data, insert items into the tree. | |
4834 | * This does all the path init required, making room in the tree if needed. | |
4835 | */ | |
4836 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4837 | struct btrfs_root *root, | |
4838 | struct btrfs_path *path, | |
4839 | struct btrfs_key *cpu_key, u32 *data_size, | |
4840 | int nr) | |
4841 | { | |
44871b1b CM |
4842 | int ret = 0; |
4843 | int slot; | |
4844 | int i; | |
4845 | u32 total_size = 0; | |
4846 | u32 total_data = 0; | |
4847 | ||
4848 | for (i = 0; i < nr; i++) | |
4849 | total_data += data_size[i]; | |
4850 | ||
4851 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4852 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4853 | if (ret == 0) | |
4854 | return -EEXIST; | |
4855 | if (ret < 0) | |
143bede5 | 4856 | return ret; |
44871b1b | 4857 | |
44871b1b CM |
4858 | slot = path->slots[0]; |
4859 | BUG_ON(slot < 0); | |
4860 | ||
afe5fea7 | 4861 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4862 | total_data, total_size, nr); |
143bede5 | 4863 | return 0; |
62e2749e CM |
4864 | } |
4865 | ||
4866 | /* | |
4867 | * Given a key and some data, insert an item into the tree. | |
4868 | * This does all the path init required, making room in the tree if needed. | |
4869 | */ | |
e089f05c CM |
4870 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4871 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4872 | data_size) | |
62e2749e CM |
4873 | { |
4874 | int ret = 0; | |
2c90e5d6 | 4875 | struct btrfs_path *path; |
5f39d397 CM |
4876 | struct extent_buffer *leaf; |
4877 | unsigned long ptr; | |
62e2749e | 4878 | |
2c90e5d6 | 4879 | path = btrfs_alloc_path(); |
db5b493a TI |
4880 | if (!path) |
4881 | return -ENOMEM; | |
2c90e5d6 | 4882 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4883 | if (!ret) { |
5f39d397 CM |
4884 | leaf = path->nodes[0]; |
4885 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4886 | write_extent_buffer(leaf, data, ptr, data_size); | |
4887 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4888 | } |
2c90e5d6 | 4889 | btrfs_free_path(path); |
aa5d6bed | 4890 | return ret; |
be0e5c09 CM |
4891 | } |
4892 | ||
74123bd7 | 4893 | /* |
5de08d7d | 4894 | * delete the pointer from a given node. |
74123bd7 | 4895 | * |
d352ac68 CM |
4896 | * the tree should have been previously balanced so the deletion does not |
4897 | * empty a node. | |
74123bd7 | 4898 | */ |
afe5fea7 TI |
4899 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4900 | int level, int slot) | |
be0e5c09 | 4901 | { |
5f39d397 | 4902 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4903 | u32 nritems; |
f3ea38da | 4904 | int ret; |
be0e5c09 | 4905 | |
5f39d397 | 4906 | nritems = btrfs_header_nritems(parent); |
d397712b | 4907 | if (slot != nritems - 1) { |
0e411ece | 4908 | if (level) |
f3ea38da JS |
4909 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4910 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4911 | memmove_extent_buffer(parent, |
4912 | btrfs_node_key_ptr_offset(slot), | |
4913 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4914 | sizeof(struct btrfs_key_ptr) * |
4915 | (nritems - slot - 1)); | |
57ba86c0 CM |
4916 | } else if (level) { |
4917 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4918 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4919 | BUG_ON(ret < 0); |
bb803951 | 4920 | } |
f3ea38da | 4921 | |
7518a238 | 4922 | nritems--; |
5f39d397 | 4923 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4924 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4925 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4926 | /* just turn the root into a leaf and break */ |
5f39d397 | 4927 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4928 | } else if (slot == 0) { |
5f39d397 CM |
4929 | struct btrfs_disk_key disk_key; |
4930 | ||
4931 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4932 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4933 | } |
d6025579 | 4934 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4935 | } |
4936 | ||
323ac95b CM |
4937 | /* |
4938 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4939 | * path->nodes[1]. |
323ac95b CM |
4940 | * |
4941 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4942 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4943 | * | |
4944 | * The path must have already been setup for deleting the leaf, including | |
4945 | * all the proper balancing. path->nodes[1] must be locked. | |
4946 | */ | |
143bede5 JM |
4947 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4948 | struct btrfs_root *root, | |
4949 | struct btrfs_path *path, | |
4950 | struct extent_buffer *leaf) | |
323ac95b | 4951 | { |
5d4f98a2 | 4952 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4953 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4954 | |
4d081c41 CM |
4955 | /* |
4956 | * btrfs_free_extent is expensive, we want to make sure we | |
4957 | * aren't holding any locks when we call it | |
4958 | */ | |
4959 | btrfs_unlock_up_safe(path, 0); | |
4960 | ||
f0486c68 YZ |
4961 | root_sub_used(root, leaf->len); |
4962 | ||
3083ee2e | 4963 | extent_buffer_get(leaf); |
5581a51a | 4964 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4965 | free_extent_buffer_stale(leaf); |
323ac95b | 4966 | } |
74123bd7 CM |
4967 | /* |
4968 | * delete the item at the leaf level in path. If that empties | |
4969 | * the leaf, remove it from the tree | |
4970 | */ | |
85e21bac CM |
4971 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4972 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4973 | { |
5f39d397 CM |
4974 | struct extent_buffer *leaf; |
4975 | struct btrfs_item *item; | |
85e21bac CM |
4976 | int last_off; |
4977 | int dsize = 0; | |
aa5d6bed CM |
4978 | int ret = 0; |
4979 | int wret; | |
85e21bac | 4980 | int i; |
7518a238 | 4981 | u32 nritems; |
cfed81a0 CM |
4982 | struct btrfs_map_token token; |
4983 | ||
4984 | btrfs_init_map_token(&token); | |
be0e5c09 | 4985 | |
5f39d397 | 4986 | leaf = path->nodes[0]; |
85e21bac CM |
4987 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4988 | ||
4989 | for (i = 0; i < nr; i++) | |
4990 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4991 | ||
5f39d397 | 4992 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4993 | |
85e21bac | 4994 | if (slot + nr != nritems) { |
123abc88 | 4995 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4996 | |
4997 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4998 | data_end + dsize, |
4999 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 5000 | last_off - data_end); |
5f39d397 | 5001 | |
85e21bac | 5002 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 5003 | u32 ioff; |
db94535d | 5004 | |
dd3cc16b | 5005 | item = btrfs_item_nr(i); |
cfed81a0 CM |
5006 | ioff = btrfs_token_item_offset(leaf, item, &token); |
5007 | btrfs_set_token_item_offset(leaf, item, | |
5008 | ioff + dsize, &token); | |
0783fcfc | 5009 | } |
db94535d | 5010 | |
5f39d397 | 5011 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 5012 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 5013 | sizeof(struct btrfs_item) * |
85e21bac | 5014 | (nritems - slot - nr)); |
be0e5c09 | 5015 | } |
85e21bac CM |
5016 | btrfs_set_header_nritems(leaf, nritems - nr); |
5017 | nritems -= nr; | |
5f39d397 | 5018 | |
74123bd7 | 5019 | /* delete the leaf if we've emptied it */ |
7518a238 | 5020 | if (nritems == 0) { |
5f39d397 CM |
5021 | if (leaf == root->node) { |
5022 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 5023 | } else { |
f0486c68 YZ |
5024 | btrfs_set_path_blocking(path); |
5025 | clean_tree_block(trans, root, leaf); | |
143bede5 | 5026 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 5027 | } |
be0e5c09 | 5028 | } else { |
7518a238 | 5029 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 5030 | if (slot == 0) { |
5f39d397 CM |
5031 | struct btrfs_disk_key disk_key; |
5032 | ||
5033 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 5034 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 5035 | } |
aa5d6bed | 5036 | |
74123bd7 | 5037 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 5038 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
5039 | /* push_leaf_left fixes the path. |
5040 | * make sure the path still points to our leaf | |
5041 | * for possible call to del_ptr below | |
5042 | */ | |
4920c9ac | 5043 | slot = path->slots[1]; |
5f39d397 CM |
5044 | extent_buffer_get(leaf); |
5045 | ||
b9473439 | 5046 | btrfs_set_path_blocking(path); |
99d8f83c CM |
5047 | wret = push_leaf_left(trans, root, path, 1, 1, |
5048 | 1, (u32)-1); | |
54aa1f4d | 5049 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 5050 | ret = wret; |
5f39d397 CM |
5051 | |
5052 | if (path->nodes[0] == leaf && | |
5053 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
5054 | wret = push_leaf_right(trans, root, path, 1, |
5055 | 1, 1, 0); | |
54aa1f4d | 5056 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
5057 | ret = wret; |
5058 | } | |
5f39d397 CM |
5059 | |
5060 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 5061 | path->slots[1] = slot; |
143bede5 | 5062 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 5063 | free_extent_buffer(leaf); |
143bede5 | 5064 | ret = 0; |
5de08d7d | 5065 | } else { |
925baedd CM |
5066 | /* if we're still in the path, make sure |
5067 | * we're dirty. Otherwise, one of the | |
5068 | * push_leaf functions must have already | |
5069 | * dirtied this buffer | |
5070 | */ | |
5071 | if (path->nodes[0] == leaf) | |
5072 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 5073 | free_extent_buffer(leaf); |
be0e5c09 | 5074 | } |
d5719762 | 5075 | } else { |
5f39d397 | 5076 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
5077 | } |
5078 | } | |
aa5d6bed | 5079 | return ret; |
be0e5c09 CM |
5080 | } |
5081 | ||
7bb86316 | 5082 | /* |
925baedd | 5083 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
5084 | * returns 0 if it found something or 1 if there are no lesser leaves. |
5085 | * returns < 0 on io errors. | |
d352ac68 CM |
5086 | * |
5087 | * This may release the path, and so you may lose any locks held at the | |
5088 | * time you call it. | |
7bb86316 | 5089 | */ |
16e7549f | 5090 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 5091 | { |
925baedd CM |
5092 | struct btrfs_key key; |
5093 | struct btrfs_disk_key found_key; | |
5094 | int ret; | |
7bb86316 | 5095 | |
925baedd | 5096 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 5097 | |
e8b0d724 | 5098 | if (key.offset > 0) { |
925baedd | 5099 | key.offset--; |
e8b0d724 | 5100 | } else if (key.type > 0) { |
925baedd | 5101 | key.type--; |
e8b0d724 FDBM |
5102 | key.offset = (u64)-1; |
5103 | } else if (key.objectid > 0) { | |
925baedd | 5104 | key.objectid--; |
e8b0d724 FDBM |
5105 | key.type = (u8)-1; |
5106 | key.offset = (u64)-1; | |
5107 | } else { | |
925baedd | 5108 | return 1; |
e8b0d724 | 5109 | } |
7bb86316 | 5110 | |
b3b4aa74 | 5111 | btrfs_release_path(path); |
925baedd CM |
5112 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
5113 | if (ret < 0) | |
5114 | return ret; | |
5115 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
5116 | ret = comp_keys(&found_key, &key); | |
5117 | if (ret < 0) | |
5118 | return 0; | |
5119 | return 1; | |
7bb86316 CM |
5120 | } |
5121 | ||
3f157a2f CM |
5122 | /* |
5123 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
5124 | * for nodes or leaves that are have a minimum transaction id. |
5125 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
5126 | * |
5127 | * This does not cow, but it does stuff the starting key it finds back | |
5128 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
5129 | * key and get a writable path. | |
5130 | * | |
5131 | * This does lock as it descends, and path->keep_locks should be set | |
5132 | * to 1 by the caller. | |
5133 | * | |
5134 | * This honors path->lowest_level to prevent descent past a given level | |
5135 | * of the tree. | |
5136 | * | |
d352ac68 CM |
5137 | * min_trans indicates the oldest transaction that you are interested |
5138 | * in walking through. Any nodes or leaves older than min_trans are | |
5139 | * skipped over (without reading them). | |
5140 | * | |
3f157a2f CM |
5141 | * returns zero if something useful was found, < 0 on error and 1 if there |
5142 | * was nothing in the tree that matched the search criteria. | |
5143 | */ | |
5144 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 5145 | struct btrfs_path *path, |
3f157a2f CM |
5146 | u64 min_trans) |
5147 | { | |
5148 | struct extent_buffer *cur; | |
5149 | struct btrfs_key found_key; | |
5150 | int slot; | |
9652480b | 5151 | int sret; |
3f157a2f CM |
5152 | u32 nritems; |
5153 | int level; | |
5154 | int ret = 1; | |
5155 | ||
934d375b | 5156 | WARN_ON(!path->keep_locks); |
3f157a2f | 5157 | again: |
bd681513 | 5158 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 5159 | level = btrfs_header_level(cur); |
e02119d5 | 5160 | WARN_ON(path->nodes[level]); |
3f157a2f | 5161 | path->nodes[level] = cur; |
bd681513 | 5162 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
5163 | |
5164 | if (btrfs_header_generation(cur) < min_trans) { | |
5165 | ret = 1; | |
5166 | goto out; | |
5167 | } | |
d397712b | 5168 | while (1) { |
3f157a2f CM |
5169 | nritems = btrfs_header_nritems(cur); |
5170 | level = btrfs_header_level(cur); | |
9652480b | 5171 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 5172 | |
323ac95b CM |
5173 | /* at the lowest level, we're done, setup the path and exit */ |
5174 | if (level == path->lowest_level) { | |
e02119d5 CM |
5175 | if (slot >= nritems) |
5176 | goto find_next_key; | |
3f157a2f CM |
5177 | ret = 0; |
5178 | path->slots[level] = slot; | |
5179 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
5180 | goto out; | |
5181 | } | |
9652480b Y |
5182 | if (sret && slot > 0) |
5183 | slot--; | |
3f157a2f | 5184 | /* |
de78b51a ES |
5185 | * check this node pointer against the min_trans parameters. |
5186 | * If it is too old, old, skip to the next one. | |
3f157a2f | 5187 | */ |
d397712b | 5188 | while (slot < nritems) { |
3f157a2f | 5189 | u64 gen; |
e02119d5 | 5190 | |
3f157a2f CM |
5191 | gen = btrfs_node_ptr_generation(cur, slot); |
5192 | if (gen < min_trans) { | |
5193 | slot++; | |
5194 | continue; | |
5195 | } | |
de78b51a | 5196 | break; |
3f157a2f | 5197 | } |
e02119d5 | 5198 | find_next_key: |
3f157a2f CM |
5199 | /* |
5200 | * we didn't find a candidate key in this node, walk forward | |
5201 | * and find another one | |
5202 | */ | |
5203 | if (slot >= nritems) { | |
e02119d5 | 5204 | path->slots[level] = slot; |
b4ce94de | 5205 | btrfs_set_path_blocking(path); |
e02119d5 | 5206 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 5207 | min_trans); |
e02119d5 | 5208 | if (sret == 0) { |
b3b4aa74 | 5209 | btrfs_release_path(path); |
3f157a2f CM |
5210 | goto again; |
5211 | } else { | |
5212 | goto out; | |
5213 | } | |
5214 | } | |
5215 | /* save our key for returning back */ | |
5216 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
5217 | path->slots[level] = slot; | |
5218 | if (level == path->lowest_level) { | |
5219 | ret = 0; | |
f7c79f30 | 5220 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5221 | goto out; |
5222 | } | |
b4ce94de | 5223 | btrfs_set_path_blocking(path); |
3f157a2f | 5224 | cur = read_node_slot(root, cur, slot); |
79787eaa | 5225 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 5226 | |
bd681513 | 5227 | btrfs_tree_read_lock(cur); |
b4ce94de | 5228 | |
bd681513 | 5229 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5230 | path->nodes[level - 1] = cur; |
f7c79f30 | 5231 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5232 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5233 | } |
5234 | out: | |
5235 | if (ret == 0) | |
5236 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 5237 | btrfs_set_path_blocking(path); |
3f157a2f CM |
5238 | return ret; |
5239 | } | |
5240 | ||
7069830a AB |
5241 | static void tree_move_down(struct btrfs_root *root, |
5242 | struct btrfs_path *path, | |
5243 | int *level, int root_level) | |
5244 | { | |
74dd17fb | 5245 | BUG_ON(*level == 0); |
7069830a AB |
5246 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
5247 | path->slots[*level]); | |
5248 | path->slots[*level - 1] = 0; | |
5249 | (*level)--; | |
5250 | } | |
5251 | ||
5252 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5253 | struct btrfs_path *path, | |
5254 | int *level, int root_level) | |
5255 | { | |
5256 | int ret = 0; | |
5257 | int nritems; | |
5258 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5259 | ||
5260 | path->slots[*level]++; | |
5261 | ||
74dd17fb | 5262 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5263 | if (*level == root_level) |
5264 | return -1; | |
5265 | ||
5266 | /* move upnext */ | |
5267 | path->slots[*level] = 0; | |
5268 | free_extent_buffer(path->nodes[*level]); | |
5269 | path->nodes[*level] = NULL; | |
5270 | (*level)++; | |
5271 | path->slots[*level]++; | |
5272 | ||
5273 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5274 | ret = 1; | |
5275 | } | |
5276 | return ret; | |
5277 | } | |
5278 | ||
5279 | /* | |
5280 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5281 | * or down. | |
5282 | */ | |
5283 | static int tree_advance(struct btrfs_root *root, | |
5284 | struct btrfs_path *path, | |
5285 | int *level, int root_level, | |
5286 | int allow_down, | |
5287 | struct btrfs_key *key) | |
5288 | { | |
5289 | int ret; | |
5290 | ||
5291 | if (*level == 0 || !allow_down) { | |
5292 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5293 | } else { | |
5294 | tree_move_down(root, path, level, root_level); | |
5295 | ret = 0; | |
5296 | } | |
5297 | if (ret >= 0) { | |
5298 | if (*level == 0) | |
5299 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5300 | path->slots[*level]); | |
5301 | else | |
5302 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5303 | path->slots[*level]); | |
5304 | } | |
5305 | return ret; | |
5306 | } | |
5307 | ||
5308 | static int tree_compare_item(struct btrfs_root *left_root, | |
5309 | struct btrfs_path *left_path, | |
5310 | struct btrfs_path *right_path, | |
5311 | char *tmp_buf) | |
5312 | { | |
5313 | int cmp; | |
5314 | int len1, len2; | |
5315 | unsigned long off1, off2; | |
5316 | ||
5317 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5318 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5319 | if (len1 != len2) | |
5320 | return 1; | |
5321 | ||
5322 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5323 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5324 | right_path->slots[0]); | |
5325 | ||
5326 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5327 | ||
5328 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5329 | if (cmp) | |
5330 | return 1; | |
5331 | return 0; | |
5332 | } | |
5333 | ||
5334 | #define ADVANCE 1 | |
5335 | #define ADVANCE_ONLY_NEXT -1 | |
5336 | ||
5337 | /* | |
5338 | * This function compares two trees and calls the provided callback for | |
5339 | * every changed/new/deleted item it finds. | |
5340 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5341 | * the compare pretty fast on snapshotted subvolumes. | |
5342 | * | |
5343 | * This currently works on commit roots only. As commit roots are read only, | |
5344 | * we don't do any locking. The commit roots are protected with transactions. | |
5345 | * Transactions are ended and rejoined when a commit is tried in between. | |
5346 | * | |
5347 | * This function checks for modifications done to the trees while comparing. | |
5348 | * If it detects a change, it aborts immediately. | |
5349 | */ | |
5350 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5351 | struct btrfs_root *right_root, | |
5352 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5353 | { | |
5354 | int ret; | |
5355 | int cmp; | |
5356 | struct btrfs_trans_handle *trans = NULL; | |
5357 | struct btrfs_path *left_path = NULL; | |
5358 | struct btrfs_path *right_path = NULL; | |
5359 | struct btrfs_key left_key; | |
5360 | struct btrfs_key right_key; | |
5361 | char *tmp_buf = NULL; | |
5362 | int left_root_level; | |
5363 | int right_root_level; | |
5364 | int left_level; | |
5365 | int right_level; | |
5366 | int left_end_reached; | |
5367 | int right_end_reached; | |
5368 | int advance_left; | |
5369 | int advance_right; | |
5370 | u64 left_blockptr; | |
5371 | u64 right_blockptr; | |
5372 | u64 left_start_ctransid; | |
5373 | u64 right_start_ctransid; | |
5374 | u64 ctransid; | |
5375 | ||
5376 | left_path = btrfs_alloc_path(); | |
5377 | if (!left_path) { | |
5378 | ret = -ENOMEM; | |
5379 | goto out; | |
5380 | } | |
5381 | right_path = btrfs_alloc_path(); | |
5382 | if (!right_path) { | |
5383 | ret = -ENOMEM; | |
5384 | goto out; | |
5385 | } | |
5386 | ||
5387 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5388 | if (!tmp_buf) { | |
5389 | ret = -ENOMEM; | |
5390 | goto out; | |
5391 | } | |
5392 | ||
5393 | left_path->search_commit_root = 1; | |
5394 | left_path->skip_locking = 1; | |
5395 | right_path->search_commit_root = 1; | |
5396 | right_path->skip_locking = 1; | |
5397 | ||
5f3ab90a | 5398 | spin_lock(&left_root->root_item_lock); |
7069830a | 5399 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5400 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5401 | |
5f3ab90a | 5402 | spin_lock(&right_root->root_item_lock); |
7069830a | 5403 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5404 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5405 | |
5406 | trans = btrfs_join_transaction(left_root); | |
5407 | if (IS_ERR(trans)) { | |
5408 | ret = PTR_ERR(trans); | |
5409 | trans = NULL; | |
5410 | goto out; | |
5411 | } | |
5412 | ||
5413 | /* | |
5414 | * Strategy: Go to the first items of both trees. Then do | |
5415 | * | |
5416 | * If both trees are at level 0 | |
5417 | * Compare keys of current items | |
5418 | * If left < right treat left item as new, advance left tree | |
5419 | * and repeat | |
5420 | * If left > right treat right item as deleted, advance right tree | |
5421 | * and repeat | |
5422 | * If left == right do deep compare of items, treat as changed if | |
5423 | * needed, advance both trees and repeat | |
5424 | * If both trees are at the same level but not at level 0 | |
5425 | * Compare keys of current nodes/leafs | |
5426 | * If left < right advance left tree and repeat | |
5427 | * If left > right advance right tree and repeat | |
5428 | * If left == right compare blockptrs of the next nodes/leafs | |
5429 | * If they match advance both trees but stay at the same level | |
5430 | * and repeat | |
5431 | * If they don't match advance both trees while allowing to go | |
5432 | * deeper and repeat | |
5433 | * If tree levels are different | |
5434 | * Advance the tree that needs it and repeat | |
5435 | * | |
5436 | * Advancing a tree means: | |
5437 | * If we are at level 0, try to go to the next slot. If that's not | |
5438 | * possible, go one level up and repeat. Stop when we found a level | |
5439 | * where we could go to the next slot. We may at this point be on a | |
5440 | * node or a leaf. | |
5441 | * | |
5442 | * If we are not at level 0 and not on shared tree blocks, go one | |
5443 | * level deeper. | |
5444 | * | |
5445 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5446 | * the right if possible or go up and right. | |
5447 | */ | |
5448 | ||
5449 | left_level = btrfs_header_level(left_root->commit_root); | |
5450 | left_root_level = left_level; | |
5451 | left_path->nodes[left_level] = left_root->commit_root; | |
5452 | extent_buffer_get(left_path->nodes[left_level]); | |
5453 | ||
5454 | right_level = btrfs_header_level(right_root->commit_root); | |
5455 | right_root_level = right_level; | |
5456 | right_path->nodes[right_level] = right_root->commit_root; | |
5457 | extent_buffer_get(right_path->nodes[right_level]); | |
5458 | ||
5459 | if (left_level == 0) | |
5460 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5461 | &left_key, left_path->slots[left_level]); | |
5462 | else | |
5463 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5464 | &left_key, left_path->slots[left_level]); | |
5465 | if (right_level == 0) | |
5466 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5467 | &right_key, right_path->slots[right_level]); | |
5468 | else | |
5469 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5470 | &right_key, right_path->slots[right_level]); | |
5471 | ||
5472 | left_end_reached = right_end_reached = 0; | |
5473 | advance_left = advance_right = 0; | |
5474 | ||
5475 | while (1) { | |
5476 | /* | |
5477 | * We need to make sure the transaction does not get committed | |
5478 | * while we do anything on commit roots. This means, we need to | |
5479 | * join and leave transactions for every item that we process. | |
5480 | */ | |
5481 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5482 | btrfs_release_path(left_path); | |
5483 | btrfs_release_path(right_path); | |
5484 | ||
5485 | ret = btrfs_end_transaction(trans, left_root); | |
5486 | trans = NULL; | |
5487 | if (ret < 0) | |
5488 | goto out; | |
5489 | } | |
5490 | /* now rejoin the transaction */ | |
5491 | if (!trans) { | |
5492 | trans = btrfs_join_transaction(left_root); | |
5493 | if (IS_ERR(trans)) { | |
5494 | ret = PTR_ERR(trans); | |
5495 | trans = NULL; | |
5496 | goto out; | |
5497 | } | |
5498 | ||
5f3ab90a | 5499 | spin_lock(&left_root->root_item_lock); |
7069830a | 5500 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5501 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5502 | if (ctransid != left_start_ctransid) |
5503 | left_start_ctransid = 0; | |
5504 | ||
5f3ab90a | 5505 | spin_lock(&right_root->root_item_lock); |
7069830a | 5506 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5507 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5508 | if (ctransid != right_start_ctransid) |
5509 | right_start_ctransid = 0; | |
5510 | ||
5511 | if (!left_start_ctransid || !right_start_ctransid) { | |
5512 | WARN(1, KERN_WARNING | |
efe120a0 | 5513 | "BTRFS: btrfs_compare_tree detected " |
7069830a AB |
5514 | "a change in one of the trees while " |
5515 | "iterating. This is probably a " | |
5516 | "bug.\n"); | |
5517 | ret = -EIO; | |
5518 | goto out; | |
5519 | } | |
5520 | ||
5521 | /* | |
5522 | * the commit root may have changed, so start again | |
5523 | * where we stopped | |
5524 | */ | |
5525 | left_path->lowest_level = left_level; | |
5526 | right_path->lowest_level = right_level; | |
5527 | ret = btrfs_search_slot(NULL, left_root, | |
5528 | &left_key, left_path, 0, 0); | |
5529 | if (ret < 0) | |
5530 | goto out; | |
5531 | ret = btrfs_search_slot(NULL, right_root, | |
5532 | &right_key, right_path, 0, 0); | |
5533 | if (ret < 0) | |
5534 | goto out; | |
5535 | } | |
5536 | ||
5537 | if (advance_left && !left_end_reached) { | |
5538 | ret = tree_advance(left_root, left_path, &left_level, | |
5539 | left_root_level, | |
5540 | advance_left != ADVANCE_ONLY_NEXT, | |
5541 | &left_key); | |
5542 | if (ret < 0) | |
5543 | left_end_reached = ADVANCE; | |
5544 | advance_left = 0; | |
5545 | } | |
5546 | if (advance_right && !right_end_reached) { | |
5547 | ret = tree_advance(right_root, right_path, &right_level, | |
5548 | right_root_level, | |
5549 | advance_right != ADVANCE_ONLY_NEXT, | |
5550 | &right_key); | |
5551 | if (ret < 0) | |
5552 | right_end_reached = ADVANCE; | |
5553 | advance_right = 0; | |
5554 | } | |
5555 | ||
5556 | if (left_end_reached && right_end_reached) { | |
5557 | ret = 0; | |
5558 | goto out; | |
5559 | } else if (left_end_reached) { | |
5560 | if (right_level == 0) { | |
5561 | ret = changed_cb(left_root, right_root, | |
5562 | left_path, right_path, | |
5563 | &right_key, | |
5564 | BTRFS_COMPARE_TREE_DELETED, | |
5565 | ctx); | |
5566 | if (ret < 0) | |
5567 | goto out; | |
5568 | } | |
5569 | advance_right = ADVANCE; | |
5570 | continue; | |
5571 | } else if (right_end_reached) { | |
5572 | if (left_level == 0) { | |
5573 | ret = changed_cb(left_root, right_root, | |
5574 | left_path, right_path, | |
5575 | &left_key, | |
5576 | BTRFS_COMPARE_TREE_NEW, | |
5577 | ctx); | |
5578 | if (ret < 0) | |
5579 | goto out; | |
5580 | } | |
5581 | advance_left = ADVANCE; | |
5582 | continue; | |
5583 | } | |
5584 | ||
5585 | if (left_level == 0 && right_level == 0) { | |
5586 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5587 | if (cmp < 0) { | |
5588 | ret = changed_cb(left_root, right_root, | |
5589 | left_path, right_path, | |
5590 | &left_key, | |
5591 | BTRFS_COMPARE_TREE_NEW, | |
5592 | ctx); | |
5593 | if (ret < 0) | |
5594 | goto out; | |
5595 | advance_left = ADVANCE; | |
5596 | } else if (cmp > 0) { | |
5597 | ret = changed_cb(left_root, right_root, | |
5598 | left_path, right_path, | |
5599 | &right_key, | |
5600 | BTRFS_COMPARE_TREE_DELETED, | |
5601 | ctx); | |
5602 | if (ret < 0) | |
5603 | goto out; | |
5604 | advance_right = ADVANCE; | |
5605 | } else { | |
ba5e8f2e JB |
5606 | enum btrfs_compare_tree_result cmp; |
5607 | ||
74dd17fb | 5608 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5609 | ret = tree_compare_item(left_root, left_path, |
5610 | right_path, tmp_buf); | |
ba5e8f2e JB |
5611 | if (ret) |
5612 | cmp = BTRFS_COMPARE_TREE_CHANGED; | |
5613 | else | |
5614 | cmp = BTRFS_COMPARE_TREE_SAME; | |
5615 | ret = changed_cb(left_root, right_root, | |
5616 | left_path, right_path, | |
5617 | &left_key, cmp, ctx); | |
5618 | if (ret < 0) | |
5619 | goto out; | |
7069830a AB |
5620 | advance_left = ADVANCE; |
5621 | advance_right = ADVANCE; | |
5622 | } | |
5623 | } else if (left_level == right_level) { | |
5624 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5625 | if (cmp < 0) { | |
5626 | advance_left = ADVANCE; | |
5627 | } else if (cmp > 0) { | |
5628 | advance_right = ADVANCE; | |
5629 | } else { | |
5630 | left_blockptr = btrfs_node_blockptr( | |
5631 | left_path->nodes[left_level], | |
5632 | left_path->slots[left_level]); | |
5633 | right_blockptr = btrfs_node_blockptr( | |
5634 | right_path->nodes[right_level], | |
5635 | right_path->slots[right_level]); | |
5636 | if (left_blockptr == right_blockptr) { | |
5637 | /* | |
5638 | * As we're on a shared block, don't | |
5639 | * allow to go deeper. | |
5640 | */ | |
5641 | advance_left = ADVANCE_ONLY_NEXT; | |
5642 | advance_right = ADVANCE_ONLY_NEXT; | |
5643 | } else { | |
5644 | advance_left = ADVANCE; | |
5645 | advance_right = ADVANCE; | |
5646 | } | |
5647 | } | |
5648 | } else if (left_level < right_level) { | |
5649 | advance_right = ADVANCE; | |
5650 | } else { | |
5651 | advance_left = ADVANCE; | |
5652 | } | |
5653 | } | |
5654 | ||
5655 | out: | |
5656 | btrfs_free_path(left_path); | |
5657 | btrfs_free_path(right_path); | |
5658 | kfree(tmp_buf); | |
5659 | ||
5660 | if (trans) { | |
5661 | if (!ret) | |
5662 | ret = btrfs_end_transaction(trans, left_root); | |
5663 | else | |
5664 | btrfs_end_transaction(trans, left_root); | |
5665 | } | |
5666 | ||
5667 | return ret; | |
5668 | } | |
5669 | ||
3f157a2f CM |
5670 | /* |
5671 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5672 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5673 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5674 | * |
5675 | * 0 is returned if another key is found, < 0 if there are any errors | |
5676 | * and 1 is returned if there are no higher keys in the tree | |
5677 | * | |
5678 | * path->keep_locks should be set to 1 on the search made before | |
5679 | * calling this function. | |
5680 | */ | |
e7a84565 | 5681 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5682 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5683 | { |
e7a84565 CM |
5684 | int slot; |
5685 | struct extent_buffer *c; | |
5686 | ||
934d375b | 5687 | WARN_ON(!path->keep_locks); |
d397712b | 5688 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5689 | if (!path->nodes[level]) |
5690 | return 1; | |
5691 | ||
5692 | slot = path->slots[level] + 1; | |
5693 | c = path->nodes[level]; | |
3f157a2f | 5694 | next: |
e7a84565 | 5695 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5696 | int ret; |
5697 | int orig_lowest; | |
5698 | struct btrfs_key cur_key; | |
5699 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5700 | !path->nodes[level + 1]) | |
e7a84565 | 5701 | return 1; |
33c66f43 YZ |
5702 | |
5703 | if (path->locks[level + 1]) { | |
5704 | level++; | |
5705 | continue; | |
5706 | } | |
5707 | ||
5708 | slot = btrfs_header_nritems(c) - 1; | |
5709 | if (level == 0) | |
5710 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5711 | else | |
5712 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5713 | ||
5714 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5715 | btrfs_release_path(path); |
33c66f43 YZ |
5716 | path->lowest_level = level; |
5717 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5718 | 0, 0); | |
5719 | path->lowest_level = orig_lowest; | |
5720 | if (ret < 0) | |
5721 | return ret; | |
5722 | ||
5723 | c = path->nodes[level]; | |
5724 | slot = path->slots[level]; | |
5725 | if (ret == 0) | |
5726 | slot++; | |
5727 | goto next; | |
e7a84565 | 5728 | } |
33c66f43 | 5729 | |
e7a84565 CM |
5730 | if (level == 0) |
5731 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5732 | else { |
3f157a2f CM |
5733 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5734 | ||
3f157a2f CM |
5735 | if (gen < min_trans) { |
5736 | slot++; | |
5737 | goto next; | |
5738 | } | |
e7a84565 | 5739 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5740 | } |
e7a84565 CM |
5741 | return 0; |
5742 | } | |
5743 | return 1; | |
5744 | } | |
5745 | ||
97571fd0 | 5746 | /* |
925baedd | 5747 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5748 | * returns 0 if it found something or 1 if there are no greater leaves. |
5749 | * returns < 0 on io errors. | |
97571fd0 | 5750 | */ |
234b63a0 | 5751 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5752 | { |
5753 | return btrfs_next_old_leaf(root, path, 0); | |
5754 | } | |
5755 | ||
5756 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5757 | u64 time_seq) | |
d97e63b6 CM |
5758 | { |
5759 | int slot; | |
8e73f275 | 5760 | int level; |
5f39d397 | 5761 | struct extent_buffer *c; |
8e73f275 | 5762 | struct extent_buffer *next; |
925baedd CM |
5763 | struct btrfs_key key; |
5764 | u32 nritems; | |
5765 | int ret; | |
8e73f275 | 5766 | int old_spinning = path->leave_spinning; |
bd681513 | 5767 | int next_rw_lock = 0; |
925baedd CM |
5768 | |
5769 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5770 | if (nritems == 0) |
925baedd | 5771 | return 1; |
925baedd | 5772 | |
8e73f275 CM |
5773 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5774 | again: | |
5775 | level = 1; | |
5776 | next = NULL; | |
bd681513 | 5777 | next_rw_lock = 0; |
b3b4aa74 | 5778 | btrfs_release_path(path); |
8e73f275 | 5779 | |
a2135011 | 5780 | path->keep_locks = 1; |
31533fb2 | 5781 | path->leave_spinning = 1; |
8e73f275 | 5782 | |
3d7806ec JS |
5783 | if (time_seq) |
5784 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5785 | else | |
5786 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5787 | path->keep_locks = 0; |
5788 | ||
5789 | if (ret < 0) | |
5790 | return ret; | |
5791 | ||
a2135011 | 5792 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5793 | /* |
5794 | * by releasing the path above we dropped all our locks. A balance | |
5795 | * could have added more items next to the key that used to be | |
5796 | * at the very end of the block. So, check again here and | |
5797 | * advance the path if there are now more items available. | |
5798 | */ | |
a2135011 | 5799 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5800 | if (ret == 0) |
5801 | path->slots[0]++; | |
8e73f275 | 5802 | ret = 0; |
925baedd CM |
5803 | goto done; |
5804 | } | |
d97e63b6 | 5805 | |
d397712b | 5806 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5807 | if (!path->nodes[level]) { |
5808 | ret = 1; | |
5809 | goto done; | |
5810 | } | |
5f39d397 | 5811 | |
d97e63b6 CM |
5812 | slot = path->slots[level] + 1; |
5813 | c = path->nodes[level]; | |
5f39d397 | 5814 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5815 | level++; |
8e73f275 CM |
5816 | if (level == BTRFS_MAX_LEVEL) { |
5817 | ret = 1; | |
5818 | goto done; | |
5819 | } | |
d97e63b6 CM |
5820 | continue; |
5821 | } | |
5f39d397 | 5822 | |
925baedd | 5823 | if (next) { |
bd681513 | 5824 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5825 | free_extent_buffer(next); |
925baedd | 5826 | } |
5f39d397 | 5827 | |
8e73f275 | 5828 | next = c; |
bd681513 | 5829 | next_rw_lock = path->locks[level]; |
8e73f275 | 5830 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5831 | slot, &key, 0); |
8e73f275 CM |
5832 | if (ret == -EAGAIN) |
5833 | goto again; | |
5f39d397 | 5834 | |
76a05b35 | 5835 | if (ret < 0) { |
b3b4aa74 | 5836 | btrfs_release_path(path); |
76a05b35 CM |
5837 | goto done; |
5838 | } | |
5839 | ||
5cd57b2c | 5840 | if (!path->skip_locking) { |
bd681513 | 5841 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5842 | if (!ret && time_seq) { |
5843 | /* | |
5844 | * If we don't get the lock, we may be racing | |
5845 | * with push_leaf_left, holding that lock while | |
5846 | * itself waiting for the leaf we've currently | |
5847 | * locked. To solve this situation, we give up | |
5848 | * on our lock and cycle. | |
5849 | */ | |
cf538830 | 5850 | free_extent_buffer(next); |
d42244a0 JS |
5851 | btrfs_release_path(path); |
5852 | cond_resched(); | |
5853 | goto again; | |
5854 | } | |
8e73f275 CM |
5855 | if (!ret) { |
5856 | btrfs_set_path_blocking(path); | |
bd681513 | 5857 | btrfs_tree_read_lock(next); |
31533fb2 | 5858 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5859 | BTRFS_READ_LOCK); |
8e73f275 | 5860 | } |
31533fb2 | 5861 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5862 | } |
d97e63b6 CM |
5863 | break; |
5864 | } | |
5865 | path->slots[level] = slot; | |
d397712b | 5866 | while (1) { |
d97e63b6 CM |
5867 | level--; |
5868 | c = path->nodes[level]; | |
925baedd | 5869 | if (path->locks[level]) |
bd681513 | 5870 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5871 | |
5f39d397 | 5872 | free_extent_buffer(c); |
d97e63b6 CM |
5873 | path->nodes[level] = next; |
5874 | path->slots[level] = 0; | |
a74a4b97 | 5875 | if (!path->skip_locking) |
bd681513 | 5876 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5877 | if (!level) |
5878 | break; | |
b4ce94de | 5879 | |
8e73f275 | 5880 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5881 | 0, &key, 0); |
8e73f275 CM |
5882 | if (ret == -EAGAIN) |
5883 | goto again; | |
5884 | ||
76a05b35 | 5885 | if (ret < 0) { |
b3b4aa74 | 5886 | btrfs_release_path(path); |
76a05b35 CM |
5887 | goto done; |
5888 | } | |
5889 | ||
5cd57b2c | 5890 | if (!path->skip_locking) { |
bd681513 | 5891 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5892 | if (!ret) { |
5893 | btrfs_set_path_blocking(path); | |
bd681513 | 5894 | btrfs_tree_read_lock(next); |
31533fb2 | 5895 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5896 | BTRFS_READ_LOCK); |
5897 | } | |
31533fb2 | 5898 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5899 | } |
d97e63b6 | 5900 | } |
8e73f275 | 5901 | ret = 0; |
925baedd | 5902 | done: |
f7c79f30 | 5903 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5904 | path->leave_spinning = old_spinning; |
5905 | if (!old_spinning) | |
5906 | btrfs_set_path_blocking(path); | |
5907 | ||
5908 | return ret; | |
d97e63b6 | 5909 | } |
0b86a832 | 5910 | |
3f157a2f CM |
5911 | /* |
5912 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5913 | * searching until it gets past min_objectid or finds an item of 'type' | |
5914 | * | |
5915 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5916 | */ | |
0b86a832 CM |
5917 | int btrfs_previous_item(struct btrfs_root *root, |
5918 | struct btrfs_path *path, u64 min_objectid, | |
5919 | int type) | |
5920 | { | |
5921 | struct btrfs_key found_key; | |
5922 | struct extent_buffer *leaf; | |
e02119d5 | 5923 | u32 nritems; |
0b86a832 CM |
5924 | int ret; |
5925 | ||
d397712b | 5926 | while (1) { |
0b86a832 | 5927 | if (path->slots[0] == 0) { |
b4ce94de | 5928 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5929 | ret = btrfs_prev_leaf(root, path); |
5930 | if (ret != 0) | |
5931 | return ret; | |
5932 | } else { | |
5933 | path->slots[0]--; | |
5934 | } | |
5935 | leaf = path->nodes[0]; | |
e02119d5 CM |
5936 | nritems = btrfs_header_nritems(leaf); |
5937 | if (nritems == 0) | |
5938 | return 1; | |
5939 | if (path->slots[0] == nritems) | |
5940 | path->slots[0]--; | |
5941 | ||
0b86a832 | 5942 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5943 | if (found_key.objectid < min_objectid) |
5944 | break; | |
0a4eefbb YZ |
5945 | if (found_key.type == type) |
5946 | return 0; | |
e02119d5 CM |
5947 | if (found_key.objectid == min_objectid && |
5948 | found_key.type < type) | |
5949 | break; | |
0b86a832 CM |
5950 | } |
5951 | return 1; | |
5952 | } |