Commit | Line | Data |
---|---|---|
6cbd5570 | 1 | /* |
d352ac68 | 2 | * Copyright (C) 2007,2008 Oracle. All rights reserved. |
6cbd5570 CM |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | ||
a6b6e75e | 19 | #include <linux/sched.h> |
5a0e3ad6 | 20 | #include <linux/slab.h> |
bd989ba3 | 21 | #include <linux/rbtree.h> |
eb60ceac CM |
22 | #include "ctree.h" |
23 | #include "disk-io.h" | |
7f5c1516 | 24 | #include "transaction.h" |
5f39d397 | 25 | #include "print-tree.h" |
925baedd | 26 | #include "locking.h" |
9a8dd150 | 27 | |
e089f05c CM |
28 | static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root |
29 | *root, struct btrfs_path *path, int level); | |
30 | static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root | |
d4dbff95 | 31 | *root, struct btrfs_key *ins_key, |
cc0c5538 | 32 | struct btrfs_path *path, int data_size, int extend); |
5f39d397 CM |
33 | static int push_node_left(struct btrfs_trans_handle *trans, |
34 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 35 | struct extent_buffer *src, int empty); |
5f39d397 CM |
36 | static int balance_node_right(struct btrfs_trans_handle *trans, |
37 | struct btrfs_root *root, | |
38 | struct extent_buffer *dst_buf, | |
39 | struct extent_buffer *src_buf); | |
afe5fea7 TI |
40 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
41 | int level, int slot); | |
f230475e JS |
42 | static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, |
43 | struct extent_buffer *eb); | |
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. | |
477 | */ | |
478 | static noinline int | |
479 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
480 | { | |
481 | struct rb_root *tm_root; | |
482 | struct rb_node **new; | |
483 | struct rb_node *parent = NULL; | |
484 | struct tree_mod_elem *cur; | |
c8cc6341 JB |
485 | int ret = 0; |
486 | ||
487 | BUG_ON(!tm); | |
488 | ||
489 | tree_mod_log_write_lock(fs_info); | |
490 | if (list_empty(&fs_info->tree_mod_seq_list)) { | |
491 | tree_mod_log_write_unlock(fs_info); | |
492 | /* | |
493 | * Ok we no longer care about logging modifications, free up tm | |
494 | * and return 0. Any callers shouldn't be using tm after | |
495 | * calling tree_mod_log_insert, but if they do we can just | |
496 | * change this to return a special error code to let the callers | |
497 | * do their own thing. | |
498 | */ | |
499 | kfree(tm); | |
500 | return 0; | |
501 | } | |
bd989ba3 | 502 | |
c8cc6341 JB |
503 | spin_lock(&fs_info->tree_mod_seq_lock); |
504 | tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info); | |
505 | spin_unlock(&fs_info->tree_mod_seq_lock); | |
bd989ba3 | 506 | |
bd989ba3 JS |
507 | tm_root = &fs_info->tree_mod_log; |
508 | new = &tm_root->rb_node; | |
509 | while (*new) { | |
510 | cur = container_of(*new, struct tree_mod_elem, node); | |
511 | parent = *new; | |
512 | if (cur->index < tm->index) | |
513 | new = &((*new)->rb_left); | |
514 | else if (cur->index > tm->index) | |
515 | new = &((*new)->rb_right); | |
097b8a7c | 516 | else if (cur->seq < tm->seq) |
bd989ba3 | 517 | new = &((*new)->rb_left); |
097b8a7c | 518 | else if (cur->seq > tm->seq) |
bd989ba3 JS |
519 | new = &((*new)->rb_right); |
520 | else { | |
c8cc6341 | 521 | ret = -EEXIST; |
bd989ba3 | 522 | kfree(tm); |
c8cc6341 | 523 | goto out; |
bd989ba3 JS |
524 | } |
525 | } | |
526 | ||
527 | rb_link_node(&tm->node, parent, new); | |
528 | rb_insert_color(&tm->node, tm_root); | |
c8cc6341 JB |
529 | out: |
530 | tree_mod_log_write_unlock(fs_info); | |
531 | return ret; | |
bd989ba3 JS |
532 | } |
533 | ||
097b8a7c JS |
534 | /* |
535 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
536 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
537 | * this until all tree mod log insertions are recorded in the rb tree and then | |
538 | * call tree_mod_log_write_unlock() to release. | |
539 | */ | |
e9b7fd4d JS |
540 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
541 | struct extent_buffer *eb) { | |
542 | smp_mb(); | |
543 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
544 | return 1; | |
097b8a7c JS |
545 | if (eb && btrfs_header_level(eb) == 0) |
546 | return 1; | |
e9b7fd4d JS |
547 | return 0; |
548 | } | |
549 | ||
097b8a7c JS |
550 | static inline int |
551 | __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, | |
552 | struct extent_buffer *eb, int slot, | |
553 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 554 | { |
097b8a7c | 555 | struct tree_mod_elem *tm; |
bd989ba3 | 556 | |
c8cc6341 JB |
557 | tm = kzalloc(sizeof(*tm), flags); |
558 | if (!tm) | |
559 | return -ENOMEM; | |
bd989ba3 JS |
560 | |
561 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
562 | if (op != MOD_LOG_KEY_ADD) { | |
563 | btrfs_node_key(eb, &tm->key, slot); | |
564 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
565 | } | |
566 | tm->op = op; | |
567 | tm->slot = slot; | |
568 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
569 | ||
097b8a7c JS |
570 | return __tree_mod_log_insert(fs_info, tm); |
571 | } | |
572 | ||
573 | static noinline int | |
c8cc6341 JB |
574 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, |
575 | struct extent_buffer *eb, int slot, | |
576 | enum mod_log_op op, gfp_t flags) | |
097b8a7c | 577 | { |
097b8a7c JS |
578 | if (tree_mod_dont_log(fs_info, eb)) |
579 | return 0; | |
580 | ||
c8cc6341 | 581 | return __tree_mod_log_insert_key(fs_info, eb, slot, op, flags); |
097b8a7c JS |
582 | } |
583 | ||
bd989ba3 JS |
584 | static noinline int |
585 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
586 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
587 | int nr_items, gfp_t flags) | |
588 | { | |
589 | struct tree_mod_elem *tm; | |
590 | int ret; | |
591 | int i; | |
592 | ||
f395694c JS |
593 | if (tree_mod_dont_log(fs_info, eb)) |
594 | return 0; | |
bd989ba3 | 595 | |
01763a2e JS |
596 | /* |
597 | * When we override something during the move, we log these removals. | |
598 | * This can only happen when we move towards the beginning of the | |
599 | * buffer, i.e. dst_slot < src_slot. | |
600 | */ | |
bd989ba3 | 601 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
c8cc6341 JB |
602 | ret = __tree_mod_log_insert_key(fs_info, eb, i + dst_slot, |
603 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, GFP_NOFS); | |
bd989ba3 JS |
604 | BUG_ON(ret < 0); |
605 | } | |
606 | ||
c8cc6341 JB |
607 | tm = kzalloc(sizeof(*tm), flags); |
608 | if (!tm) | |
609 | return -ENOMEM; | |
f395694c | 610 | |
bd989ba3 JS |
611 | tm->index = eb->start >> PAGE_CACHE_SHIFT; |
612 | tm->slot = src_slot; | |
613 | tm->move.dst_slot = dst_slot; | |
614 | tm->move.nr_items = nr_items; | |
615 | tm->op = MOD_LOG_MOVE_KEYS; | |
616 | ||
c8cc6341 | 617 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
618 | } |
619 | ||
097b8a7c JS |
620 | static inline void |
621 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
622 | { | |
623 | int i; | |
624 | u32 nritems; | |
625 | int ret; | |
626 | ||
b12a3b1e CM |
627 | if (btrfs_header_level(eb) == 0) |
628 | return; | |
629 | ||
097b8a7c JS |
630 | nritems = btrfs_header_nritems(eb); |
631 | for (i = nritems - 1; i >= 0; i--) { | |
c8cc6341 JB |
632 | ret = __tree_mod_log_insert_key(fs_info, eb, i, |
633 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, GFP_NOFS); | |
097b8a7c JS |
634 | BUG_ON(ret < 0); |
635 | } | |
636 | } | |
637 | ||
bd989ba3 JS |
638 | static noinline int |
639 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
640 | struct extent_buffer *old_root, | |
90f8d62e JS |
641 | struct extent_buffer *new_root, gfp_t flags, |
642 | int log_removal) | |
bd989ba3 JS |
643 | { |
644 | struct tree_mod_elem *tm; | |
bd989ba3 | 645 | |
097b8a7c JS |
646 | if (tree_mod_dont_log(fs_info, NULL)) |
647 | return 0; | |
648 | ||
90f8d62e JS |
649 | if (log_removal) |
650 | __tree_mod_log_free_eb(fs_info, old_root); | |
d9abbf1c | 651 | |
c8cc6341 JB |
652 | tm = kzalloc(sizeof(*tm), flags); |
653 | if (!tm) | |
654 | return -ENOMEM; | |
bd989ba3 JS |
655 | |
656 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
657 | tm->old_root.logical = old_root->start; | |
658 | tm->old_root.level = btrfs_header_level(old_root); | |
659 | tm->generation = btrfs_header_generation(old_root); | |
660 | tm->op = MOD_LOG_ROOT_REPLACE; | |
661 | ||
c8cc6341 | 662 | return __tree_mod_log_insert(fs_info, tm); |
bd989ba3 JS |
663 | } |
664 | ||
665 | static struct tree_mod_elem * | |
666 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
667 | int smallest) | |
668 | { | |
669 | struct rb_root *tm_root; | |
670 | struct rb_node *node; | |
671 | struct tree_mod_elem *cur = NULL; | |
672 | struct tree_mod_elem *found = NULL; | |
673 | u64 index = start >> PAGE_CACHE_SHIFT; | |
674 | ||
097b8a7c | 675 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
676 | tm_root = &fs_info->tree_mod_log; |
677 | node = tm_root->rb_node; | |
678 | while (node) { | |
679 | cur = container_of(node, struct tree_mod_elem, node); | |
680 | if (cur->index < index) { | |
681 | node = node->rb_left; | |
682 | } else if (cur->index > index) { | |
683 | node = node->rb_right; | |
097b8a7c | 684 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
685 | node = node->rb_left; |
686 | } else if (!smallest) { | |
687 | /* we want the node with the highest seq */ | |
688 | if (found) | |
097b8a7c | 689 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
690 | found = cur; |
691 | node = node->rb_left; | |
097b8a7c | 692 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
693 | /* we want the node with the smallest seq */ |
694 | if (found) | |
097b8a7c | 695 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
696 | found = cur; |
697 | node = node->rb_right; | |
698 | } else { | |
699 | found = cur; | |
700 | break; | |
701 | } | |
702 | } | |
097b8a7c | 703 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
704 | |
705 | return found; | |
706 | } | |
707 | ||
708 | /* | |
709 | * this returns the element from the log with the smallest time sequence | |
710 | * value that's in the log (the oldest log item). any element with a time | |
711 | * sequence lower than min_seq will be ignored. | |
712 | */ | |
713 | static struct tree_mod_elem * | |
714 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
715 | u64 min_seq) | |
716 | { | |
717 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
718 | } | |
719 | ||
720 | /* | |
721 | * this returns the element from the log with the largest time sequence | |
722 | * value that's in the log (the most recent log item). any element with | |
723 | * a time sequence lower than min_seq will be ignored. | |
724 | */ | |
725 | static struct tree_mod_elem * | |
726 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
727 | { | |
728 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
729 | } | |
730 | ||
097b8a7c | 731 | static noinline void |
bd989ba3 JS |
732 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
733 | struct extent_buffer *src, unsigned long dst_offset, | |
90f8d62e | 734 | unsigned long src_offset, int nr_items) |
bd989ba3 JS |
735 | { |
736 | int ret; | |
737 | int i; | |
738 | ||
e9b7fd4d | 739 | if (tree_mod_dont_log(fs_info, NULL)) |
bd989ba3 JS |
740 | return; |
741 | ||
c8cc6341 | 742 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) |
bd989ba3 JS |
743 | return; |
744 | ||
bd989ba3 | 745 | for (i = 0; i < nr_items; i++) { |
c8cc6341 | 746 | ret = __tree_mod_log_insert_key(fs_info, src, |
90f8d62e | 747 | i + src_offset, |
c8cc6341 | 748 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
90f8d62e | 749 | BUG_ON(ret < 0); |
c8cc6341 | 750 | ret = __tree_mod_log_insert_key(fs_info, dst, |
097b8a7c | 751 | i + dst_offset, |
c8cc6341 JB |
752 | MOD_LOG_KEY_ADD, |
753 | GFP_NOFS); | |
bd989ba3 JS |
754 | BUG_ON(ret < 0); |
755 | } | |
756 | } | |
757 | ||
758 | static inline void | |
759 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
760 | int dst_offset, int src_offset, int nr_items) | |
761 | { | |
762 | int ret; | |
763 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
764 | nr_items, GFP_NOFS); | |
765 | BUG_ON(ret < 0); | |
766 | } | |
767 | ||
097b8a7c | 768 | static noinline void |
bd989ba3 | 769 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 770 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
771 | { |
772 | int ret; | |
773 | ||
c8cc6341 JB |
774 | ret = __tree_mod_log_insert_key(fs_info, eb, slot, |
775 | MOD_LOG_KEY_REPLACE, | |
776 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
bd989ba3 JS |
777 | BUG_ON(ret < 0); |
778 | } | |
779 | ||
097b8a7c JS |
780 | static noinline void |
781 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
bd989ba3 | 782 | { |
e9b7fd4d | 783 | if (tree_mod_dont_log(fs_info, eb)) |
bd989ba3 | 784 | return; |
097b8a7c | 785 | __tree_mod_log_free_eb(fs_info, eb); |
bd989ba3 JS |
786 | } |
787 | ||
097b8a7c | 788 | static noinline void |
bd989ba3 | 789 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
90f8d62e JS |
790 | struct extent_buffer *new_root_node, |
791 | int log_removal) | |
bd989ba3 JS |
792 | { |
793 | int ret; | |
bd989ba3 | 794 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
90f8d62e | 795 | new_root_node, GFP_NOFS, log_removal); |
bd989ba3 JS |
796 | BUG_ON(ret < 0); |
797 | } | |
798 | ||
5d4f98a2 YZ |
799 | /* |
800 | * check if the tree block can be shared by multiple trees | |
801 | */ | |
802 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
803 | struct extent_buffer *buf) | |
804 | { | |
805 | /* | |
806 | * Tree blocks not in refernece counted trees and tree roots | |
807 | * are never shared. If a block was allocated after the last | |
808 | * snapshot and the block was not allocated by tree relocation, | |
809 | * we know the block is not shared. | |
810 | */ | |
811 | if (root->ref_cows && | |
812 | buf != root->node && buf != root->commit_root && | |
813 | (btrfs_header_generation(buf) <= | |
814 | btrfs_root_last_snapshot(&root->root_item) || | |
815 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
816 | return 1; | |
817 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
818 | if (root->ref_cows && | |
819 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
820 | return 1; | |
821 | #endif | |
822 | return 0; | |
823 | } | |
824 | ||
825 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
826 | struct btrfs_root *root, | |
827 | struct extent_buffer *buf, | |
f0486c68 YZ |
828 | struct extent_buffer *cow, |
829 | int *last_ref) | |
5d4f98a2 YZ |
830 | { |
831 | u64 refs; | |
832 | u64 owner; | |
833 | u64 flags; | |
834 | u64 new_flags = 0; | |
835 | int ret; | |
836 | ||
837 | /* | |
838 | * Backrefs update rules: | |
839 | * | |
840 | * Always use full backrefs for extent pointers in tree block | |
841 | * allocated by tree relocation. | |
842 | * | |
843 | * If a shared tree block is no longer referenced by its owner | |
844 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
845 | * use full backrefs for extent pointers in tree block. | |
846 | * | |
847 | * If a tree block is been relocating | |
848 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
849 | * use full backrefs for extent pointers in tree block. | |
850 | * The reason for this is some operations (such as drop tree) | |
851 | * are only allowed for blocks use full backrefs. | |
852 | */ | |
853 | ||
854 | if (btrfs_block_can_be_shared(root, buf)) { | |
855 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
3173a18f JB |
856 | btrfs_header_level(buf), 1, |
857 | &refs, &flags); | |
be1a5564 MF |
858 | if (ret) |
859 | return ret; | |
e5df9573 MF |
860 | if (refs == 0) { |
861 | ret = -EROFS; | |
862 | btrfs_std_error(root->fs_info, ret); | |
863 | return ret; | |
864 | } | |
5d4f98a2 YZ |
865 | } else { |
866 | refs = 1; | |
867 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
868 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
869 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
870 | else | |
871 | flags = 0; | |
872 | } | |
873 | ||
874 | owner = btrfs_header_owner(buf); | |
875 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
876 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
877 | ||
878 | if (refs > 1) { | |
879 | if ((owner == root->root_key.objectid || | |
880 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
881 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 882 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 883 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
884 | |
885 | if (root->root_key.objectid == | |
886 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 887 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 888 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 889 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 890 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
891 | } |
892 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
893 | } else { | |
894 | ||
895 | if (root->root_key.objectid == | |
896 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 897 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 898 | else |
66d7e7f0 | 899 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 900 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
901 | } |
902 | if (new_flags != 0) { | |
b1c79e09 JB |
903 | int level = btrfs_header_level(buf); |
904 | ||
5d4f98a2 YZ |
905 | ret = btrfs_set_disk_extent_flags(trans, root, |
906 | buf->start, | |
907 | buf->len, | |
b1c79e09 | 908 | new_flags, level, 0); |
be1a5564 MF |
909 | if (ret) |
910 | return ret; | |
5d4f98a2 YZ |
911 | } |
912 | } else { | |
913 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
914 | if (root->root_key.objectid == | |
915 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 916 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 917 | else |
66d7e7f0 | 918 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 919 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 920 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 921 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
922 | } |
923 | clean_tree_block(trans, root, buf); | |
f0486c68 | 924 | *last_ref = 1; |
5d4f98a2 YZ |
925 | } |
926 | return 0; | |
927 | } | |
928 | ||
d352ac68 | 929 | /* |
d397712b CM |
930 | * does the dirty work in cow of a single block. The parent block (if |
931 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
932 | * dirty and returned locked. If you modify the block it needs to be marked | |
933 | * dirty again. | |
d352ac68 CM |
934 | * |
935 | * search_start -- an allocation hint for the new block | |
936 | * | |
d397712b CM |
937 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
938 | * bytes the allocator should try to find free next to the block it returns. | |
939 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 940 | */ |
d397712b | 941 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
942 | struct btrfs_root *root, |
943 | struct extent_buffer *buf, | |
944 | struct extent_buffer *parent, int parent_slot, | |
945 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 946 | u64 search_start, u64 empty_size) |
02217ed2 | 947 | { |
5d4f98a2 | 948 | struct btrfs_disk_key disk_key; |
5f39d397 | 949 | struct extent_buffer *cow; |
be1a5564 | 950 | int level, ret; |
f0486c68 | 951 | int last_ref = 0; |
925baedd | 952 | int unlock_orig = 0; |
5d4f98a2 | 953 | u64 parent_start; |
7bb86316 | 954 | |
925baedd CM |
955 | if (*cow_ret == buf) |
956 | unlock_orig = 1; | |
957 | ||
b9447ef8 | 958 | btrfs_assert_tree_locked(buf); |
925baedd | 959 | |
7bb86316 CM |
960 | WARN_ON(root->ref_cows && trans->transid != |
961 | root->fs_info->running_transaction->transid); | |
6702ed49 | 962 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 963 | |
7bb86316 | 964 | level = btrfs_header_level(buf); |
31840ae1 | 965 | |
5d4f98a2 YZ |
966 | if (level == 0) |
967 | btrfs_item_key(buf, &disk_key, 0); | |
968 | else | |
969 | btrfs_node_key(buf, &disk_key, 0); | |
970 | ||
971 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
972 | if (parent) | |
973 | parent_start = parent->start; | |
974 | else | |
975 | parent_start = 0; | |
976 | } else | |
977 | parent_start = 0; | |
978 | ||
979 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
980 | root->root_key.objectid, &disk_key, | |
5581a51a | 981 | level, search_start, empty_size); |
54aa1f4d CM |
982 | if (IS_ERR(cow)) |
983 | return PTR_ERR(cow); | |
6702ed49 | 984 | |
b4ce94de CM |
985 | /* cow is set to blocking by btrfs_init_new_buffer */ |
986 | ||
5f39d397 | 987 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 988 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 989 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
990 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
991 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
992 | BTRFS_HEADER_FLAG_RELOC); | |
993 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
994 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
995 | else | |
996 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 997 | |
0a4e5586 | 998 | write_extent_buffer(cow, root->fs_info->fsid, btrfs_header_fsid(), |
2b82032c YZ |
999 | BTRFS_FSID_SIZE); |
1000 | ||
be1a5564 | 1001 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1002 | if (ret) { |
79787eaa | 1003 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1004 | return ret; |
1005 | } | |
1a40e23b | 1006 | |
83d4cfd4 JB |
1007 | if (root->ref_cows) { |
1008 | ret = btrfs_reloc_cow_block(trans, root, buf, cow); | |
1009 | if (ret) | |
1010 | return ret; | |
1011 | } | |
3fd0a558 | 1012 | |
02217ed2 | 1013 | if (buf == root->node) { |
925baedd | 1014 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1015 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1016 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1017 | parent_start = buf->start; | |
1018 | else | |
1019 | parent_start = 0; | |
925baedd | 1020 | |
5f39d397 | 1021 | extent_buffer_get(cow); |
90f8d62e | 1022 | tree_mod_log_set_root_pointer(root, cow, 1); |
240f62c8 | 1023 | rcu_assign_pointer(root->node, cow); |
925baedd | 1024 | |
f0486c68 | 1025 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1026 | last_ref); |
5f39d397 | 1027 | free_extent_buffer(buf); |
0b86a832 | 1028 | add_root_to_dirty_list(root); |
02217ed2 | 1029 | } else { |
5d4f98a2 YZ |
1030 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1031 | parent_start = parent->start; | |
1032 | else | |
1033 | parent_start = 0; | |
1034 | ||
1035 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e | 1036 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
c8cc6341 | 1037 | MOD_LOG_KEY_REPLACE, GFP_NOFS); |
5f39d397 | 1038 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1039 | cow->start); |
74493f7a CM |
1040 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1041 | trans->transid); | |
d6025579 | 1042 | btrfs_mark_buffer_dirty(parent); |
7fb7d76f JB |
1043 | if (last_ref) |
1044 | tree_mod_log_free_eb(root->fs_info, buf); | |
f0486c68 | 1045 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1046 | last_ref); |
02217ed2 | 1047 | } |
925baedd CM |
1048 | if (unlock_orig) |
1049 | btrfs_tree_unlock(buf); | |
3083ee2e | 1050 | free_extent_buffer_stale(buf); |
ccd467d6 | 1051 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1052 | *cow_ret = cow; |
02217ed2 CM |
1053 | return 0; |
1054 | } | |
1055 | ||
5d9e75c4 JS |
1056 | /* |
1057 | * returns the logical address of the oldest predecessor of the given root. | |
1058 | * entries older than time_seq are ignored. | |
1059 | */ | |
1060 | static struct tree_mod_elem * | |
1061 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
30b0463a | 1062 | struct extent_buffer *eb_root, u64 time_seq) |
5d9e75c4 JS |
1063 | { |
1064 | struct tree_mod_elem *tm; | |
1065 | struct tree_mod_elem *found = NULL; | |
30b0463a | 1066 | u64 root_logical = eb_root->start; |
5d9e75c4 JS |
1067 | int looped = 0; |
1068 | ||
1069 | if (!time_seq) | |
35a3621b | 1070 | return NULL; |
5d9e75c4 JS |
1071 | |
1072 | /* | |
1073 | * the very last operation that's logged for a root is the replacement | |
1074 | * operation (if it is replaced at all). this has the index of the *new* | |
1075 | * root, making it the very first operation that's logged for this root. | |
1076 | */ | |
1077 | while (1) { | |
1078 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1079 | time_seq); | |
1080 | if (!looped && !tm) | |
35a3621b | 1081 | return NULL; |
5d9e75c4 | 1082 | /* |
28da9fb4 JS |
1083 | * if there are no tree operation for the oldest root, we simply |
1084 | * return it. this should only happen if that (old) root is at | |
1085 | * level 0. | |
5d9e75c4 | 1086 | */ |
28da9fb4 JS |
1087 | if (!tm) |
1088 | break; | |
5d9e75c4 | 1089 | |
28da9fb4 JS |
1090 | /* |
1091 | * if there's an operation that's not a root replacement, we | |
1092 | * found the oldest version of our root. normally, we'll find a | |
1093 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1094 | */ | |
5d9e75c4 JS |
1095 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1096 | break; | |
1097 | ||
1098 | found = tm; | |
1099 | root_logical = tm->old_root.logical; | |
5d9e75c4 JS |
1100 | looped = 1; |
1101 | } | |
1102 | ||
a95236d9 JS |
1103 | /* if there's no old root to return, return what we found instead */ |
1104 | if (!found) | |
1105 | found = tm; | |
1106 | ||
5d9e75c4 JS |
1107 | return found; |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1112 | * previous operations will be rewinded (until we reach something older than | |
1113 | * time_seq). | |
1114 | */ | |
1115 | static void | |
f1ca7e98 JB |
1116 | __tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, |
1117 | u64 time_seq, struct tree_mod_elem *first_tm) | |
5d9e75c4 JS |
1118 | { |
1119 | u32 n; | |
1120 | struct rb_node *next; | |
1121 | struct tree_mod_elem *tm = first_tm; | |
1122 | unsigned long o_dst; | |
1123 | unsigned long o_src; | |
1124 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1125 | ||
1126 | n = btrfs_header_nritems(eb); | |
f1ca7e98 | 1127 | tree_mod_log_read_lock(fs_info); |
097b8a7c | 1128 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1129 | /* |
1130 | * all the operations are recorded with the operator used for | |
1131 | * the modification. as we're going backwards, we do the | |
1132 | * opposite of each operation here. | |
1133 | */ | |
1134 | switch (tm->op) { | |
1135 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1136 | BUG_ON(tm->slot < n); | |
1c697d4a | 1137 | /* Fallthrough */ |
95c80bb1 | 1138 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1139 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1140 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1141 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1142 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1143 | tm->generation); | |
4c3e6969 | 1144 | n++; |
5d9e75c4 JS |
1145 | break; |
1146 | case MOD_LOG_KEY_REPLACE: | |
1147 | BUG_ON(tm->slot >= n); | |
1148 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1149 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1150 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1151 | tm->generation); | |
1152 | break; | |
1153 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1154 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1155 | n--; |
1156 | break; | |
1157 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1158 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1159 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1160 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1161 | tm->move.nr_items * p_size); |
1162 | break; | |
1163 | case MOD_LOG_ROOT_REPLACE: | |
1164 | /* | |
1165 | * this operation is special. for roots, this must be | |
1166 | * handled explicitly before rewinding. | |
1167 | * for non-roots, this operation may exist if the node | |
1168 | * was a root: root A -> child B; then A gets empty and | |
1169 | * B is promoted to the new root. in the mod log, we'll | |
1170 | * have a root-replace operation for B, a tree block | |
1171 | * that is no root. we simply ignore that operation. | |
1172 | */ | |
1173 | break; | |
1174 | } | |
1175 | next = rb_next(&tm->node); | |
1176 | if (!next) | |
1177 | break; | |
1178 | tm = container_of(next, struct tree_mod_elem, node); | |
1179 | if (tm->index != first_tm->index) | |
1180 | break; | |
1181 | } | |
f1ca7e98 | 1182 | tree_mod_log_read_unlock(fs_info); |
5d9e75c4 JS |
1183 | btrfs_set_header_nritems(eb, n); |
1184 | } | |
1185 | ||
47fb091f JS |
1186 | /* |
1187 | * Called with eb read locked. If the buffer cannot be rewinded, the same buffer | |
1188 | * is returned. If rewind operations happen, a fresh buffer is returned. The | |
1189 | * returned buffer is always read-locked. If the returned buffer is not the | |
1190 | * input buffer, the lock on the input buffer is released and the input buffer | |
1191 | * is freed (its refcount is decremented). | |
1192 | */ | |
5d9e75c4 | 1193 | static struct extent_buffer * |
9ec72677 JB |
1194 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct btrfs_path *path, |
1195 | struct extent_buffer *eb, u64 time_seq) | |
5d9e75c4 JS |
1196 | { |
1197 | struct extent_buffer *eb_rewin; | |
1198 | struct tree_mod_elem *tm; | |
1199 | ||
1200 | if (!time_seq) | |
1201 | return eb; | |
1202 | ||
1203 | if (btrfs_header_level(eb) == 0) | |
1204 | return eb; | |
1205 | ||
1206 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1207 | if (!tm) | |
1208 | return eb; | |
1209 | ||
9ec72677 JB |
1210 | btrfs_set_path_blocking(path); |
1211 | btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); | |
1212 | ||
5d9e75c4 JS |
1213 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1214 | BUG_ON(tm->slot != 0); | |
1215 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1216 | fs_info->tree_root->nodesize); | |
db7f3436 | 1217 | if (!eb_rewin) { |
9ec72677 | 1218 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1219 | free_extent_buffer(eb); |
1220 | return NULL; | |
1221 | } | |
5d9e75c4 JS |
1222 | btrfs_set_header_bytenr(eb_rewin, eb->start); |
1223 | btrfs_set_header_backref_rev(eb_rewin, | |
1224 | btrfs_header_backref_rev(eb)); | |
1225 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1226 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1227 | } else { |
1228 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
db7f3436 | 1229 | if (!eb_rewin) { |
9ec72677 | 1230 | btrfs_tree_read_unlock_blocking(eb); |
db7f3436 JB |
1231 | free_extent_buffer(eb); |
1232 | return NULL; | |
1233 | } | |
5d9e75c4 JS |
1234 | } |
1235 | ||
9ec72677 JB |
1236 | btrfs_clear_path_blocking(path, NULL, BTRFS_READ_LOCK); |
1237 | btrfs_tree_read_unlock_blocking(eb); | |
5d9e75c4 JS |
1238 | free_extent_buffer(eb); |
1239 | ||
47fb091f JS |
1240 | extent_buffer_get(eb_rewin); |
1241 | btrfs_tree_read_lock(eb_rewin); | |
f1ca7e98 | 1242 | __tree_mod_log_rewind(fs_info, eb_rewin, time_seq, tm); |
57911b8b | 1243 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1244 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1245 | |
1246 | return eb_rewin; | |
1247 | } | |
1248 | ||
8ba97a15 JS |
1249 | /* |
1250 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1251 | * value. If there are no changes, the current root->root_node is returned. If | |
1252 | * anything changed in between, there's a fresh buffer allocated on which the | |
1253 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1254 | * Returns NULL on error (with no locks held). | |
1255 | */ | |
5d9e75c4 JS |
1256 | static inline struct extent_buffer * |
1257 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1258 | { | |
1259 | struct tree_mod_elem *tm; | |
30b0463a JS |
1260 | struct extent_buffer *eb = NULL; |
1261 | struct extent_buffer *eb_root; | |
7bfdcf7f | 1262 | struct extent_buffer *old; |
a95236d9 | 1263 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1264 | u64 old_generation = 0; |
a95236d9 | 1265 | u64 logical; |
834328a8 | 1266 | u32 blocksize; |
5d9e75c4 | 1267 | |
30b0463a JS |
1268 | eb_root = btrfs_read_lock_root_node(root); |
1269 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); | |
5d9e75c4 | 1270 | if (!tm) |
30b0463a | 1271 | return eb_root; |
5d9e75c4 | 1272 | |
a95236d9 JS |
1273 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1274 | old_root = &tm->old_root; | |
1275 | old_generation = tm->generation; | |
1276 | logical = old_root->logical; | |
1277 | } else { | |
30b0463a | 1278 | logical = eb_root->start; |
a95236d9 | 1279 | } |
5d9e75c4 | 1280 | |
a95236d9 | 1281 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 | 1282 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
30b0463a JS |
1283 | btrfs_tree_read_unlock(eb_root); |
1284 | free_extent_buffer(eb_root); | |
834328a8 | 1285 | blocksize = btrfs_level_size(root, old_root->level); |
7bfdcf7f | 1286 | old = read_tree_block(root, logical, blocksize, 0); |
fae7f21c | 1287 | if (WARN_ON(!old || !extent_buffer_uptodate(old))) { |
416bc658 | 1288 | free_extent_buffer(old); |
834328a8 JS |
1289 | pr_warn("btrfs: failed to read tree block %llu from get_old_root\n", |
1290 | logical); | |
834328a8 | 1291 | } else { |
7bfdcf7f LB |
1292 | eb = btrfs_clone_extent_buffer(old); |
1293 | free_extent_buffer(old); | |
834328a8 JS |
1294 | } |
1295 | } else if (old_root) { | |
30b0463a JS |
1296 | btrfs_tree_read_unlock(eb_root); |
1297 | free_extent_buffer(eb_root); | |
28da9fb4 | 1298 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1299 | } else { |
9ec72677 | 1300 | btrfs_set_lock_blocking_rw(eb_root, BTRFS_READ_LOCK); |
30b0463a | 1301 | eb = btrfs_clone_extent_buffer(eb_root); |
9ec72677 | 1302 | btrfs_tree_read_unlock_blocking(eb_root); |
30b0463a | 1303 | free_extent_buffer(eb_root); |
834328a8 JS |
1304 | } |
1305 | ||
8ba97a15 JS |
1306 | if (!eb) |
1307 | return NULL; | |
d6381084 | 1308 | extent_buffer_get(eb); |
8ba97a15 | 1309 | btrfs_tree_read_lock(eb); |
a95236d9 | 1310 | if (old_root) { |
5d9e75c4 JS |
1311 | btrfs_set_header_bytenr(eb, eb->start); |
1312 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
30b0463a | 1313 | btrfs_set_header_owner(eb, btrfs_header_owner(eb_root)); |
a95236d9 JS |
1314 | btrfs_set_header_level(eb, old_root->level); |
1315 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1316 | } |
28da9fb4 | 1317 | if (tm) |
f1ca7e98 | 1318 | __tree_mod_log_rewind(root->fs_info, eb, time_seq, tm); |
28da9fb4 JS |
1319 | else |
1320 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1321 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1322 | |
1323 | return eb; | |
1324 | } | |
1325 | ||
5b6602e7 JS |
1326 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1327 | { | |
1328 | struct tree_mod_elem *tm; | |
1329 | int level; | |
30b0463a | 1330 | struct extent_buffer *eb_root = btrfs_root_node(root); |
5b6602e7 | 1331 | |
30b0463a | 1332 | tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq); |
5b6602e7 JS |
1333 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { |
1334 | level = tm->old_root.level; | |
1335 | } else { | |
30b0463a | 1336 | level = btrfs_header_level(eb_root); |
5b6602e7 | 1337 | } |
30b0463a | 1338 | free_extent_buffer(eb_root); |
5b6602e7 JS |
1339 | |
1340 | return level; | |
1341 | } | |
1342 | ||
5d4f98a2 YZ |
1343 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1344 | struct btrfs_root *root, | |
1345 | struct extent_buffer *buf) | |
1346 | { | |
f1ebcc74 LB |
1347 | /* ensure we can see the force_cow */ |
1348 | smp_rmb(); | |
1349 | ||
1350 | /* | |
1351 | * We do not need to cow a block if | |
1352 | * 1) this block is not created or changed in this transaction; | |
1353 | * 2) this block does not belong to TREE_RELOC tree; | |
1354 | * 3) the root is not forced COW. | |
1355 | * | |
1356 | * What is forced COW: | |
1357 | * when we create snapshot during commiting the transaction, | |
1358 | * after we've finished coping src root, we must COW the shared | |
1359 | * block to ensure the metadata consistency. | |
1360 | */ | |
5d4f98a2 YZ |
1361 | if (btrfs_header_generation(buf) == trans->transid && |
1362 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1363 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1364 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1365 | !root->force_cow) | |
5d4f98a2 YZ |
1366 | return 0; |
1367 | return 1; | |
1368 | } | |
1369 | ||
d352ac68 CM |
1370 | /* |
1371 | * cows a single block, see __btrfs_cow_block for the real work. | |
1372 | * This version of it has extra checks so that a block isn't cow'd more than | |
1373 | * once per transaction, as long as it hasn't been written yet | |
1374 | */ | |
d397712b | 1375 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1376 | struct btrfs_root *root, struct extent_buffer *buf, |
1377 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1378 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1379 | { |
1380 | u64 search_start; | |
f510cfec | 1381 | int ret; |
dc17ff8f | 1382 | |
31b1a2bd JL |
1383 | if (trans->transaction != root->fs_info->running_transaction) |
1384 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1385 | trans->transid, |
6702ed49 | 1386 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1387 | |
1388 | if (trans->transid != root->fs_info->generation) | |
1389 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
c1c9ff7c | 1390 | trans->transid, root->fs_info->generation); |
dc17ff8f | 1391 | |
5d4f98a2 | 1392 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1393 | *cow_ret = buf; |
1394 | return 0; | |
1395 | } | |
c487685d | 1396 | |
0b86a832 | 1397 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1398 | |
1399 | if (parent) | |
1400 | btrfs_set_lock_blocking(parent); | |
1401 | btrfs_set_lock_blocking(buf); | |
1402 | ||
f510cfec | 1403 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1404 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1405 | |
1406 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1407 | ||
f510cfec | 1408 | return ret; |
6702ed49 CM |
1409 | } |
1410 | ||
d352ac68 CM |
1411 | /* |
1412 | * helper function for defrag to decide if two blocks pointed to by a | |
1413 | * node are actually close by | |
1414 | */ | |
6b80053d | 1415 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1416 | { |
6b80053d | 1417 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1418 | return 1; |
6b80053d | 1419 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1420 | return 1; |
1421 | return 0; | |
1422 | } | |
1423 | ||
081e9573 CM |
1424 | /* |
1425 | * compare two keys in a memcmp fashion | |
1426 | */ | |
1427 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1428 | { | |
1429 | struct btrfs_key k1; | |
1430 | ||
1431 | btrfs_disk_key_to_cpu(&k1, disk); | |
1432 | ||
20736aba | 1433 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1434 | } |
1435 | ||
f3465ca4 JB |
1436 | /* |
1437 | * same as comp_keys only with two btrfs_key's | |
1438 | */ | |
5d4f98a2 | 1439 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1440 | { |
1441 | if (k1->objectid > k2->objectid) | |
1442 | return 1; | |
1443 | if (k1->objectid < k2->objectid) | |
1444 | return -1; | |
1445 | if (k1->type > k2->type) | |
1446 | return 1; | |
1447 | if (k1->type < k2->type) | |
1448 | return -1; | |
1449 | if (k1->offset > k2->offset) | |
1450 | return 1; | |
1451 | if (k1->offset < k2->offset) | |
1452 | return -1; | |
1453 | return 0; | |
1454 | } | |
081e9573 | 1455 | |
d352ac68 CM |
1456 | /* |
1457 | * this is used by the defrag code to go through all the | |
1458 | * leaves pointed to by a node and reallocate them so that | |
1459 | * disk order is close to key order | |
1460 | */ | |
6702ed49 | 1461 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1462 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1463 | int start_slot, u64 *last_ret, |
a6b6e75e | 1464 | struct btrfs_key *progress) |
6702ed49 | 1465 | { |
6b80053d | 1466 | struct extent_buffer *cur; |
6702ed49 | 1467 | u64 blocknr; |
ca7a79ad | 1468 | u64 gen; |
e9d0b13b CM |
1469 | u64 search_start = *last_ret; |
1470 | u64 last_block = 0; | |
6702ed49 CM |
1471 | u64 other; |
1472 | u32 parent_nritems; | |
6702ed49 CM |
1473 | int end_slot; |
1474 | int i; | |
1475 | int err = 0; | |
f2183bde | 1476 | int parent_level; |
6b80053d CM |
1477 | int uptodate; |
1478 | u32 blocksize; | |
081e9573 CM |
1479 | int progress_passed = 0; |
1480 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1481 | |
5708b959 | 1482 | parent_level = btrfs_header_level(parent); |
5708b959 | 1483 | |
6c1500f2 JL |
1484 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1485 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1486 | |
6b80053d | 1487 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1488 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1489 | end_slot = parent_nritems; |
1490 | ||
1491 | if (parent_nritems == 1) | |
1492 | return 0; | |
1493 | ||
b4ce94de CM |
1494 | btrfs_set_lock_blocking(parent); |
1495 | ||
6702ed49 CM |
1496 | for (i = start_slot; i < end_slot; i++) { |
1497 | int close = 1; | |
a6b6e75e | 1498 | |
081e9573 CM |
1499 | btrfs_node_key(parent, &disk_key, i); |
1500 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1501 | continue; | |
1502 | ||
1503 | progress_passed = 1; | |
6b80053d | 1504 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1505 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1506 | if (last_block == 0) |
1507 | last_block = blocknr; | |
5708b959 | 1508 | |
6702ed49 | 1509 | if (i > 0) { |
6b80053d CM |
1510 | other = btrfs_node_blockptr(parent, i - 1); |
1511 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1512 | } |
0ef3e66b | 1513 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1514 | other = btrfs_node_blockptr(parent, i + 1); |
1515 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1516 | } |
e9d0b13b CM |
1517 | if (close) { |
1518 | last_block = blocknr; | |
6702ed49 | 1519 | continue; |
e9d0b13b | 1520 | } |
6702ed49 | 1521 | |
6b80053d CM |
1522 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1523 | if (cur) | |
b9fab919 | 1524 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1525 | else |
1526 | uptodate = 0; | |
5708b959 | 1527 | if (!cur || !uptodate) { |
6b80053d CM |
1528 | if (!cur) { |
1529 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1530 | blocksize, gen); |
416bc658 JB |
1531 | if (!cur || !extent_buffer_uptodate(cur)) { |
1532 | free_extent_buffer(cur); | |
97d9a8a4 | 1533 | return -EIO; |
416bc658 | 1534 | } |
6b80053d | 1535 | } else if (!uptodate) { |
018642a1 TI |
1536 | err = btrfs_read_buffer(cur, gen); |
1537 | if (err) { | |
1538 | free_extent_buffer(cur); | |
1539 | return err; | |
1540 | } | |
f2183bde | 1541 | } |
6702ed49 | 1542 | } |
e9d0b13b | 1543 | if (search_start == 0) |
6b80053d | 1544 | search_start = last_block; |
e9d0b13b | 1545 | |
e7a84565 | 1546 | btrfs_tree_lock(cur); |
b4ce94de | 1547 | btrfs_set_lock_blocking(cur); |
6b80053d | 1548 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1549 | &cur, search_start, |
6b80053d | 1550 | min(16 * blocksize, |
9fa8cfe7 | 1551 | (end_slot - i) * blocksize)); |
252c38f0 | 1552 | if (err) { |
e7a84565 | 1553 | btrfs_tree_unlock(cur); |
6b80053d | 1554 | free_extent_buffer(cur); |
6702ed49 | 1555 | break; |
252c38f0 | 1556 | } |
e7a84565 CM |
1557 | search_start = cur->start; |
1558 | last_block = cur->start; | |
f2183bde | 1559 | *last_ret = search_start; |
e7a84565 CM |
1560 | btrfs_tree_unlock(cur); |
1561 | free_extent_buffer(cur); | |
6702ed49 CM |
1562 | } |
1563 | return err; | |
1564 | } | |
1565 | ||
74123bd7 CM |
1566 | /* |
1567 | * The leaf data grows from end-to-front in the node. | |
1568 | * this returns the address of the start of the last item, | |
1569 | * which is the stop of the leaf data stack | |
1570 | */ | |
123abc88 | 1571 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1572 | struct extent_buffer *leaf) |
be0e5c09 | 1573 | { |
5f39d397 | 1574 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1575 | if (nr == 0) |
123abc88 | 1576 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1577 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1578 | } |
1579 | ||
aa5d6bed | 1580 | |
74123bd7 | 1581 | /* |
5f39d397 CM |
1582 | * search for key in the extent_buffer. The items start at offset p, |
1583 | * and they are item_size apart. There are 'max' items in p. | |
1584 | * | |
74123bd7 CM |
1585 | * the slot in the array is returned via slot, and it points to |
1586 | * the place where you would insert key if it is not found in | |
1587 | * the array. | |
1588 | * | |
1589 | * slot may point to max if the key is bigger than all of the keys | |
1590 | */ | |
e02119d5 CM |
1591 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1592 | unsigned long p, | |
1593 | int item_size, struct btrfs_key *key, | |
1594 | int max, int *slot) | |
be0e5c09 CM |
1595 | { |
1596 | int low = 0; | |
1597 | int high = max; | |
1598 | int mid; | |
1599 | int ret; | |
479965d6 | 1600 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1601 | struct btrfs_disk_key unaligned; |
1602 | unsigned long offset; | |
5f39d397 CM |
1603 | char *kaddr = NULL; |
1604 | unsigned long map_start = 0; | |
1605 | unsigned long map_len = 0; | |
479965d6 | 1606 | int err; |
be0e5c09 | 1607 | |
d397712b | 1608 | while (low < high) { |
be0e5c09 | 1609 | mid = (low + high) / 2; |
5f39d397 CM |
1610 | offset = p + mid * item_size; |
1611 | ||
a6591715 | 1612 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1613 | (offset + sizeof(struct btrfs_disk_key)) > |
1614 | map_start + map_len) { | |
934d375b CM |
1615 | |
1616 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1617 | sizeof(struct btrfs_disk_key), |
a6591715 | 1618 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1619 | |
1620 | if (!err) { | |
1621 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1622 | map_start); | |
1623 | } else { | |
1624 | read_extent_buffer(eb, &unaligned, | |
1625 | offset, sizeof(unaligned)); | |
1626 | tmp = &unaligned; | |
1627 | } | |
5f39d397 | 1628 | |
5f39d397 CM |
1629 | } else { |
1630 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1631 | map_start); | |
1632 | } | |
be0e5c09 CM |
1633 | ret = comp_keys(tmp, key); |
1634 | ||
1635 | if (ret < 0) | |
1636 | low = mid + 1; | |
1637 | else if (ret > 0) | |
1638 | high = mid; | |
1639 | else { | |
1640 | *slot = mid; | |
1641 | return 0; | |
1642 | } | |
1643 | } | |
1644 | *slot = low; | |
1645 | return 1; | |
1646 | } | |
1647 | ||
97571fd0 CM |
1648 | /* |
1649 | * simple bin_search frontend that does the right thing for | |
1650 | * leaves vs nodes | |
1651 | */ | |
5f39d397 CM |
1652 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1653 | int level, int *slot) | |
be0e5c09 | 1654 | { |
f775738f | 1655 | if (level == 0) |
5f39d397 CM |
1656 | return generic_bin_search(eb, |
1657 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1658 | sizeof(struct btrfs_item), |
5f39d397 | 1659 | key, btrfs_header_nritems(eb), |
7518a238 | 1660 | slot); |
f775738f | 1661 | else |
5f39d397 CM |
1662 | return generic_bin_search(eb, |
1663 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1664 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1665 | key, btrfs_header_nritems(eb), |
7518a238 | 1666 | slot); |
be0e5c09 CM |
1667 | } |
1668 | ||
5d4f98a2 YZ |
1669 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1670 | int level, int *slot) | |
1671 | { | |
1672 | return bin_search(eb, key, level, slot); | |
1673 | } | |
1674 | ||
f0486c68 YZ |
1675 | static void root_add_used(struct btrfs_root *root, u32 size) |
1676 | { | |
1677 | spin_lock(&root->accounting_lock); | |
1678 | btrfs_set_root_used(&root->root_item, | |
1679 | btrfs_root_used(&root->root_item) + size); | |
1680 | spin_unlock(&root->accounting_lock); | |
1681 | } | |
1682 | ||
1683 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1684 | { | |
1685 | spin_lock(&root->accounting_lock); | |
1686 | btrfs_set_root_used(&root->root_item, | |
1687 | btrfs_root_used(&root->root_item) - size); | |
1688 | spin_unlock(&root->accounting_lock); | |
1689 | } | |
1690 | ||
d352ac68 CM |
1691 | /* given a node and slot number, this reads the blocks it points to. The |
1692 | * extent buffer is returned with a reference taken (but unlocked). | |
1693 | * NULL is returned on error. | |
1694 | */ | |
e02119d5 | 1695 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1696 | struct extent_buffer *parent, int slot) |
bb803951 | 1697 | { |
ca7a79ad | 1698 | int level = btrfs_header_level(parent); |
416bc658 JB |
1699 | struct extent_buffer *eb; |
1700 | ||
bb803951 CM |
1701 | if (slot < 0) |
1702 | return NULL; | |
5f39d397 | 1703 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1704 | return NULL; |
ca7a79ad CM |
1705 | |
1706 | BUG_ON(level == 0); | |
1707 | ||
416bc658 JB |
1708 | eb = read_tree_block(root, btrfs_node_blockptr(parent, slot), |
1709 | btrfs_level_size(root, level - 1), | |
1710 | btrfs_node_ptr_generation(parent, slot)); | |
1711 | if (eb && !extent_buffer_uptodate(eb)) { | |
1712 | free_extent_buffer(eb); | |
1713 | eb = NULL; | |
1714 | } | |
1715 | ||
1716 | return eb; | |
bb803951 CM |
1717 | } |
1718 | ||
d352ac68 CM |
1719 | /* |
1720 | * node level balancing, used to make sure nodes are in proper order for | |
1721 | * item deletion. We balance from the top down, so we have to make sure | |
1722 | * that a deletion won't leave an node completely empty later on. | |
1723 | */ | |
e02119d5 | 1724 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1725 | struct btrfs_root *root, |
1726 | struct btrfs_path *path, int level) | |
bb803951 | 1727 | { |
5f39d397 CM |
1728 | struct extent_buffer *right = NULL; |
1729 | struct extent_buffer *mid; | |
1730 | struct extent_buffer *left = NULL; | |
1731 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1732 | int ret = 0; |
1733 | int wret; | |
1734 | int pslot; | |
bb803951 | 1735 | int orig_slot = path->slots[level]; |
79f95c82 | 1736 | u64 orig_ptr; |
bb803951 CM |
1737 | |
1738 | if (level == 0) | |
1739 | return 0; | |
1740 | ||
5f39d397 | 1741 | mid = path->nodes[level]; |
b4ce94de | 1742 | |
bd681513 CM |
1743 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1744 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1745 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1746 | ||
1d4f8a0c | 1747 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1748 | |
a05a9bb1 | 1749 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1750 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1751 | pslot = path->slots[level + 1]; |
1752 | } | |
bb803951 | 1753 | |
40689478 CM |
1754 | /* |
1755 | * deal with the case where there is only one pointer in the root | |
1756 | * by promoting the node below to a root | |
1757 | */ | |
5f39d397 CM |
1758 | if (!parent) { |
1759 | struct extent_buffer *child; | |
bb803951 | 1760 | |
5f39d397 | 1761 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1762 | return 0; |
1763 | ||
1764 | /* promote the child to a root */ | |
5f39d397 | 1765 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1766 | if (!child) { |
1767 | ret = -EROFS; | |
1768 | btrfs_std_error(root->fs_info, ret); | |
1769 | goto enospc; | |
1770 | } | |
1771 | ||
925baedd | 1772 | btrfs_tree_lock(child); |
b4ce94de | 1773 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1774 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1775 | if (ret) { |
1776 | btrfs_tree_unlock(child); | |
1777 | free_extent_buffer(child); | |
1778 | goto enospc; | |
1779 | } | |
2f375ab9 | 1780 | |
90f8d62e | 1781 | tree_mod_log_set_root_pointer(root, child, 1); |
240f62c8 | 1782 | rcu_assign_pointer(root->node, child); |
925baedd | 1783 | |
0b86a832 | 1784 | add_root_to_dirty_list(root); |
925baedd | 1785 | btrfs_tree_unlock(child); |
b4ce94de | 1786 | |
925baedd | 1787 | path->locks[level] = 0; |
bb803951 | 1788 | path->nodes[level] = NULL; |
5f39d397 | 1789 | clean_tree_block(trans, root, mid); |
925baedd | 1790 | btrfs_tree_unlock(mid); |
bb803951 | 1791 | /* once for the path */ |
5f39d397 | 1792 | free_extent_buffer(mid); |
f0486c68 YZ |
1793 | |
1794 | root_sub_used(root, mid->len); | |
5581a51a | 1795 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1796 | /* once for the root ptr */ |
3083ee2e | 1797 | free_extent_buffer_stale(mid); |
f0486c68 | 1798 | return 0; |
bb803951 | 1799 | } |
5f39d397 | 1800 | if (btrfs_header_nritems(mid) > |
123abc88 | 1801 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1802 | return 0; |
1803 | ||
5f39d397 CM |
1804 | left = read_node_slot(root, parent, pslot - 1); |
1805 | if (left) { | |
925baedd | 1806 | btrfs_tree_lock(left); |
b4ce94de | 1807 | btrfs_set_lock_blocking(left); |
5f39d397 | 1808 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1809 | parent, pslot - 1, &left); |
54aa1f4d CM |
1810 | if (wret) { |
1811 | ret = wret; | |
1812 | goto enospc; | |
1813 | } | |
2cc58cf2 | 1814 | } |
5f39d397 CM |
1815 | right = read_node_slot(root, parent, pslot + 1); |
1816 | if (right) { | |
925baedd | 1817 | btrfs_tree_lock(right); |
b4ce94de | 1818 | btrfs_set_lock_blocking(right); |
5f39d397 | 1819 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1820 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1821 | if (wret) { |
1822 | ret = wret; | |
1823 | goto enospc; | |
1824 | } | |
1825 | } | |
1826 | ||
1827 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1828 | if (left) { |
1829 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1830 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1831 | if (wret < 0) |
1832 | ret = wret; | |
bb803951 | 1833 | } |
79f95c82 CM |
1834 | |
1835 | /* | |
1836 | * then try to empty the right most buffer into the middle | |
1837 | */ | |
5f39d397 | 1838 | if (right) { |
971a1f66 | 1839 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1840 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1841 | ret = wret; |
5f39d397 | 1842 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1843 | clean_tree_block(trans, root, right); |
925baedd | 1844 | btrfs_tree_unlock(right); |
afe5fea7 | 1845 | del_ptr(root, path, level + 1, pslot + 1); |
f0486c68 | 1846 | root_sub_used(root, right->len); |
5581a51a | 1847 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1848 | free_extent_buffer_stale(right); |
f0486c68 | 1849 | right = NULL; |
bb803951 | 1850 | } else { |
5f39d397 CM |
1851 | struct btrfs_disk_key right_key; |
1852 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 1853 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1854 | pslot + 1, 0); |
5f39d397 CM |
1855 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1856 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1857 | } |
1858 | } | |
5f39d397 | 1859 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1860 | /* |
1861 | * we're not allowed to leave a node with one item in the | |
1862 | * tree during a delete. A deletion from lower in the tree | |
1863 | * could try to delete the only pointer in this node. | |
1864 | * So, pull some keys from the left. | |
1865 | * There has to be a left pointer at this point because | |
1866 | * otherwise we would have pulled some pointers from the | |
1867 | * right | |
1868 | */ | |
305a26af MF |
1869 | if (!left) { |
1870 | ret = -EROFS; | |
1871 | btrfs_std_error(root->fs_info, ret); | |
1872 | goto enospc; | |
1873 | } | |
5f39d397 | 1874 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1875 | if (wret < 0) { |
79f95c82 | 1876 | ret = wret; |
54aa1f4d CM |
1877 | goto enospc; |
1878 | } | |
bce4eae9 CM |
1879 | if (wret == 1) { |
1880 | wret = push_node_left(trans, root, left, mid, 1); | |
1881 | if (wret < 0) | |
1882 | ret = wret; | |
1883 | } | |
79f95c82 CM |
1884 | BUG_ON(wret == 1); |
1885 | } | |
5f39d397 | 1886 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1887 | clean_tree_block(trans, root, mid); |
925baedd | 1888 | btrfs_tree_unlock(mid); |
afe5fea7 | 1889 | del_ptr(root, path, level + 1, pslot); |
f0486c68 | 1890 | root_sub_used(root, mid->len); |
5581a51a | 1891 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1892 | free_extent_buffer_stale(mid); |
f0486c68 | 1893 | mid = NULL; |
79f95c82 CM |
1894 | } else { |
1895 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1896 | struct btrfs_disk_key mid_key; |
1897 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 1898 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 1899 | pslot, 0); |
5f39d397 CM |
1900 | btrfs_set_node_key(parent, &mid_key, pslot); |
1901 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1902 | } |
bb803951 | 1903 | |
79f95c82 | 1904 | /* update the path */ |
5f39d397 CM |
1905 | if (left) { |
1906 | if (btrfs_header_nritems(left) > orig_slot) { | |
1907 | extent_buffer_get(left); | |
925baedd | 1908 | /* left was locked after cow */ |
5f39d397 | 1909 | path->nodes[level] = left; |
bb803951 CM |
1910 | path->slots[level + 1] -= 1; |
1911 | path->slots[level] = orig_slot; | |
925baedd CM |
1912 | if (mid) { |
1913 | btrfs_tree_unlock(mid); | |
5f39d397 | 1914 | free_extent_buffer(mid); |
925baedd | 1915 | } |
bb803951 | 1916 | } else { |
5f39d397 | 1917 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1918 | path->slots[level] = orig_slot; |
1919 | } | |
1920 | } | |
79f95c82 | 1921 | /* double check we haven't messed things up */ |
e20d96d6 | 1922 | if (orig_ptr != |
5f39d397 | 1923 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1924 | BUG(); |
54aa1f4d | 1925 | enospc: |
925baedd CM |
1926 | if (right) { |
1927 | btrfs_tree_unlock(right); | |
5f39d397 | 1928 | free_extent_buffer(right); |
925baedd CM |
1929 | } |
1930 | if (left) { | |
1931 | if (path->nodes[level] != left) | |
1932 | btrfs_tree_unlock(left); | |
5f39d397 | 1933 | free_extent_buffer(left); |
925baedd | 1934 | } |
bb803951 CM |
1935 | return ret; |
1936 | } | |
1937 | ||
d352ac68 CM |
1938 | /* Node balancing for insertion. Here we only split or push nodes around |
1939 | * when they are completely full. This is also done top down, so we | |
1940 | * have to be pessimistic. | |
1941 | */ | |
d397712b | 1942 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1943 | struct btrfs_root *root, |
1944 | struct btrfs_path *path, int level) | |
e66f709b | 1945 | { |
5f39d397 CM |
1946 | struct extent_buffer *right = NULL; |
1947 | struct extent_buffer *mid; | |
1948 | struct extent_buffer *left = NULL; | |
1949 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1950 | int ret = 0; |
1951 | int wret; | |
1952 | int pslot; | |
1953 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1954 | |
1955 | if (level == 0) | |
1956 | return 1; | |
1957 | ||
5f39d397 | 1958 | mid = path->nodes[level]; |
7bb86316 | 1959 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1960 | |
a05a9bb1 | 1961 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1962 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1963 | pslot = path->slots[level + 1]; |
1964 | } | |
e66f709b | 1965 | |
5f39d397 | 1966 | if (!parent) |
e66f709b | 1967 | return 1; |
e66f709b | 1968 | |
5f39d397 | 1969 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1970 | |
1971 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1972 | if (left) { |
e66f709b | 1973 | u32 left_nr; |
925baedd CM |
1974 | |
1975 | btrfs_tree_lock(left); | |
b4ce94de CM |
1976 | btrfs_set_lock_blocking(left); |
1977 | ||
5f39d397 | 1978 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1979 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1980 | wret = 1; | |
1981 | } else { | |
5f39d397 | 1982 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1983 | pslot - 1, &left); |
54aa1f4d CM |
1984 | if (ret) |
1985 | wret = 1; | |
1986 | else { | |
54aa1f4d | 1987 | wret = push_node_left(trans, root, |
971a1f66 | 1988 | left, mid, 0); |
54aa1f4d | 1989 | } |
33ade1f8 | 1990 | } |
e66f709b CM |
1991 | if (wret < 0) |
1992 | ret = wret; | |
1993 | if (wret == 0) { | |
5f39d397 | 1994 | struct btrfs_disk_key disk_key; |
e66f709b | 1995 | orig_slot += left_nr; |
5f39d397 | 1996 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 1997 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1998 | pslot, 0); |
5f39d397 CM |
1999 | btrfs_set_node_key(parent, &disk_key, pslot); |
2000 | btrfs_mark_buffer_dirty(parent); | |
2001 | if (btrfs_header_nritems(left) > orig_slot) { | |
2002 | path->nodes[level] = left; | |
e66f709b CM |
2003 | path->slots[level + 1] -= 1; |
2004 | path->slots[level] = orig_slot; | |
925baedd | 2005 | btrfs_tree_unlock(mid); |
5f39d397 | 2006 | free_extent_buffer(mid); |
e66f709b CM |
2007 | } else { |
2008 | orig_slot -= | |
5f39d397 | 2009 | btrfs_header_nritems(left); |
e66f709b | 2010 | path->slots[level] = orig_slot; |
925baedd | 2011 | btrfs_tree_unlock(left); |
5f39d397 | 2012 | free_extent_buffer(left); |
e66f709b | 2013 | } |
e66f709b CM |
2014 | return 0; |
2015 | } | |
925baedd | 2016 | btrfs_tree_unlock(left); |
5f39d397 | 2017 | free_extent_buffer(left); |
e66f709b | 2018 | } |
925baedd | 2019 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
2020 | |
2021 | /* | |
2022 | * then try to empty the right most buffer into the middle | |
2023 | */ | |
5f39d397 | 2024 | if (right) { |
33ade1f8 | 2025 | u32 right_nr; |
b4ce94de | 2026 | |
925baedd | 2027 | btrfs_tree_lock(right); |
b4ce94de CM |
2028 | btrfs_set_lock_blocking(right); |
2029 | ||
5f39d397 | 2030 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2031 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2032 | wret = 1; | |
2033 | } else { | |
5f39d397 CM |
2034 | ret = btrfs_cow_block(trans, root, right, |
2035 | parent, pslot + 1, | |
9fa8cfe7 | 2036 | &right); |
54aa1f4d CM |
2037 | if (ret) |
2038 | wret = 1; | |
2039 | else { | |
54aa1f4d | 2040 | wret = balance_node_right(trans, root, |
5f39d397 | 2041 | right, mid); |
54aa1f4d | 2042 | } |
33ade1f8 | 2043 | } |
e66f709b CM |
2044 | if (wret < 0) |
2045 | ret = wret; | |
2046 | if (wret == 0) { | |
5f39d397 CM |
2047 | struct btrfs_disk_key disk_key; |
2048 | ||
2049 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2050 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2051 | pslot + 1, 0); |
5f39d397 CM |
2052 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2053 | btrfs_mark_buffer_dirty(parent); | |
2054 | ||
2055 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2056 | path->nodes[level] = right; | |
e66f709b CM |
2057 | path->slots[level + 1] += 1; |
2058 | path->slots[level] = orig_slot - | |
5f39d397 | 2059 | btrfs_header_nritems(mid); |
925baedd | 2060 | btrfs_tree_unlock(mid); |
5f39d397 | 2061 | free_extent_buffer(mid); |
e66f709b | 2062 | } else { |
925baedd | 2063 | btrfs_tree_unlock(right); |
5f39d397 | 2064 | free_extent_buffer(right); |
e66f709b | 2065 | } |
e66f709b CM |
2066 | return 0; |
2067 | } | |
925baedd | 2068 | btrfs_tree_unlock(right); |
5f39d397 | 2069 | free_extent_buffer(right); |
e66f709b | 2070 | } |
e66f709b CM |
2071 | return 1; |
2072 | } | |
2073 | ||
3c69faec | 2074 | /* |
d352ac68 CM |
2075 | * readahead one full node of leaves, finding things that are close |
2076 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2077 | */ |
c8c42864 CM |
2078 | static void reada_for_search(struct btrfs_root *root, |
2079 | struct btrfs_path *path, | |
2080 | int level, int slot, u64 objectid) | |
3c69faec | 2081 | { |
5f39d397 | 2082 | struct extent_buffer *node; |
01f46658 | 2083 | struct btrfs_disk_key disk_key; |
3c69faec | 2084 | u32 nritems; |
3c69faec | 2085 | u64 search; |
a7175319 | 2086 | u64 target; |
6b80053d | 2087 | u64 nread = 0; |
cb25c2ea | 2088 | u64 gen; |
3c69faec | 2089 | int direction = path->reada; |
5f39d397 | 2090 | struct extent_buffer *eb; |
6b80053d CM |
2091 | u32 nr; |
2092 | u32 blocksize; | |
2093 | u32 nscan = 0; | |
db94535d | 2094 | |
a6b6e75e | 2095 | if (level != 1) |
6702ed49 CM |
2096 | return; |
2097 | ||
2098 | if (!path->nodes[level]) | |
3c69faec CM |
2099 | return; |
2100 | ||
5f39d397 | 2101 | node = path->nodes[level]; |
925baedd | 2102 | |
3c69faec | 2103 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2104 | blocksize = btrfs_level_size(root, level - 1); |
2105 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2106 | if (eb) { |
2107 | free_extent_buffer(eb); | |
3c69faec CM |
2108 | return; |
2109 | } | |
2110 | ||
a7175319 | 2111 | target = search; |
6b80053d | 2112 | |
5f39d397 | 2113 | nritems = btrfs_header_nritems(node); |
6b80053d | 2114 | nr = slot; |
25b8b936 | 2115 | |
d397712b | 2116 | while (1) { |
6b80053d CM |
2117 | if (direction < 0) { |
2118 | if (nr == 0) | |
2119 | break; | |
2120 | nr--; | |
2121 | } else if (direction > 0) { | |
2122 | nr++; | |
2123 | if (nr >= nritems) | |
2124 | break; | |
3c69faec | 2125 | } |
01f46658 CM |
2126 | if (path->reada < 0 && objectid) { |
2127 | btrfs_node_key(node, &disk_key, nr); | |
2128 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2129 | break; | |
2130 | } | |
6b80053d | 2131 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2132 | if ((search <= target && target - search <= 65536) || |
2133 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2134 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2135 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2136 | nread += blocksize; |
2137 | } | |
2138 | nscan++; | |
a7175319 | 2139 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2140 | break; |
3c69faec CM |
2141 | } |
2142 | } | |
925baedd | 2143 | |
0b08851f JB |
2144 | static noinline void reada_for_balance(struct btrfs_root *root, |
2145 | struct btrfs_path *path, int level) | |
b4ce94de CM |
2146 | { |
2147 | int slot; | |
2148 | int nritems; | |
2149 | struct extent_buffer *parent; | |
2150 | struct extent_buffer *eb; | |
2151 | u64 gen; | |
2152 | u64 block1 = 0; | |
2153 | u64 block2 = 0; | |
b4ce94de CM |
2154 | int blocksize; |
2155 | ||
8c594ea8 | 2156 | parent = path->nodes[level + 1]; |
b4ce94de | 2157 | if (!parent) |
0b08851f | 2158 | return; |
b4ce94de CM |
2159 | |
2160 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2161 | slot = path->slots[level + 1]; |
b4ce94de CM |
2162 | blocksize = btrfs_level_size(root, level); |
2163 | ||
2164 | if (slot > 0) { | |
2165 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2166 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2167 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2168 | /* |
2169 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2170 | * don't want to return eagain here. That will loop | |
2171 | * forever | |
2172 | */ | |
2173 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2174 | block1 = 0; |
2175 | free_extent_buffer(eb); | |
2176 | } | |
8c594ea8 | 2177 | if (slot + 1 < nritems) { |
b4ce94de CM |
2178 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2179 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2180 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2181 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2182 | block2 = 0; |
2183 | free_extent_buffer(eb); | |
2184 | } | |
8c594ea8 | 2185 | |
0b08851f JB |
2186 | if (block1) |
2187 | readahead_tree_block(root, block1, blocksize, 0); | |
2188 | if (block2) | |
2189 | readahead_tree_block(root, block2, blocksize, 0); | |
b4ce94de CM |
2190 | } |
2191 | ||
2192 | ||
d352ac68 | 2193 | /* |
d397712b CM |
2194 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2195 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2196 | * operations on the tree might require changing key pointers higher up in the | |
2197 | * tree. | |
d352ac68 | 2198 | * |
d397712b CM |
2199 | * callers might also have set path->keep_locks, which tells this code to keep |
2200 | * the lock if the path points to the last slot in the block. This is part of | |
2201 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2202 | * |
d397712b CM |
2203 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2204 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2205 | */ |
e02119d5 | 2206 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2207 | int lowest_unlock, int min_write_lock_level, |
2208 | int *write_lock_level) | |
925baedd CM |
2209 | { |
2210 | int i; | |
2211 | int skip_level = level; | |
051e1b9f | 2212 | int no_skips = 0; |
925baedd CM |
2213 | struct extent_buffer *t; |
2214 | ||
2215 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2216 | if (!path->nodes[i]) | |
2217 | break; | |
2218 | if (!path->locks[i]) | |
2219 | break; | |
051e1b9f | 2220 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2221 | skip_level = i + 1; |
2222 | continue; | |
2223 | } | |
051e1b9f | 2224 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2225 | u32 nritems; |
2226 | t = path->nodes[i]; | |
2227 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2228 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2229 | skip_level = i + 1; |
2230 | continue; | |
2231 | } | |
2232 | } | |
051e1b9f CM |
2233 | if (skip_level < i && i >= lowest_unlock) |
2234 | no_skips = 1; | |
2235 | ||
925baedd CM |
2236 | t = path->nodes[i]; |
2237 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2238 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2239 | path->locks[i] = 0; |
f7c79f30 CM |
2240 | if (write_lock_level && |
2241 | i > min_write_lock_level && | |
2242 | i <= *write_lock_level) { | |
2243 | *write_lock_level = i - 1; | |
2244 | } | |
925baedd CM |
2245 | } |
2246 | } | |
2247 | } | |
2248 | ||
b4ce94de CM |
2249 | /* |
2250 | * This releases any locks held in the path starting at level and | |
2251 | * going all the way up to the root. | |
2252 | * | |
2253 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2254 | * corner cases, such as COW of the block at slot zero in the node. This | |
2255 | * ignores those rules, and it should only be called when there are no | |
2256 | * more updates to be done higher up in the tree. | |
2257 | */ | |
2258 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2259 | { | |
2260 | int i; | |
2261 | ||
09a2a8f9 | 2262 | if (path->keep_locks) |
b4ce94de CM |
2263 | return; |
2264 | ||
2265 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2266 | if (!path->nodes[i]) | |
12f4dacc | 2267 | continue; |
b4ce94de | 2268 | if (!path->locks[i]) |
12f4dacc | 2269 | continue; |
bd681513 | 2270 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2271 | path->locks[i] = 0; |
2272 | } | |
2273 | } | |
2274 | ||
c8c42864 CM |
2275 | /* |
2276 | * helper function for btrfs_search_slot. The goal is to find a block | |
2277 | * in cache without setting the path to blocking. If we find the block | |
2278 | * we return zero and the path is unchanged. | |
2279 | * | |
2280 | * If we can't find the block, we set the path blocking and do some | |
2281 | * reada. -EAGAIN is returned and the search must be repeated. | |
2282 | */ | |
2283 | static int | |
2284 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2285 | struct btrfs_root *root, struct btrfs_path *p, | |
2286 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2287 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2288 | { |
2289 | u64 blocknr; | |
2290 | u64 gen; | |
2291 | u32 blocksize; | |
2292 | struct extent_buffer *b = *eb_ret; | |
2293 | struct extent_buffer *tmp; | |
76a05b35 | 2294 | int ret; |
c8c42864 CM |
2295 | |
2296 | blocknr = btrfs_node_blockptr(b, slot); | |
2297 | gen = btrfs_node_ptr_generation(b, slot); | |
2298 | blocksize = btrfs_level_size(root, level - 1); | |
2299 | ||
2300 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2301 | if (tmp) { |
b9fab919 | 2302 | /* first we do an atomic uptodate check */ |
bdf7c00e JB |
2303 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { |
2304 | *eb_ret = tmp; | |
2305 | return 0; | |
2306 | } | |
2307 | ||
2308 | /* the pages were up to date, but we failed | |
2309 | * the generation number check. Do a full | |
2310 | * read for the generation number that is correct. | |
2311 | * We must do this without dropping locks so | |
2312 | * we can trust our generation number | |
2313 | */ | |
2314 | btrfs_set_path_blocking(p); | |
2315 | ||
2316 | /* now we're allowed to do a blocking uptodate check */ | |
2317 | ret = btrfs_read_buffer(tmp, gen); | |
2318 | if (!ret) { | |
2319 | *eb_ret = tmp; | |
2320 | return 0; | |
cb44921a | 2321 | } |
bdf7c00e JB |
2322 | free_extent_buffer(tmp); |
2323 | btrfs_release_path(p); | |
2324 | return -EIO; | |
c8c42864 CM |
2325 | } |
2326 | ||
2327 | /* | |
2328 | * reduce lock contention at high levels | |
2329 | * of the btree by dropping locks before | |
76a05b35 CM |
2330 | * we read. Don't release the lock on the current |
2331 | * level because we need to walk this node to figure | |
2332 | * out which blocks to read. | |
c8c42864 | 2333 | */ |
8c594ea8 CM |
2334 | btrfs_unlock_up_safe(p, level + 1); |
2335 | btrfs_set_path_blocking(p); | |
2336 | ||
cb44921a | 2337 | free_extent_buffer(tmp); |
c8c42864 CM |
2338 | if (p->reada) |
2339 | reada_for_search(root, p, level, slot, key->objectid); | |
2340 | ||
b3b4aa74 | 2341 | btrfs_release_path(p); |
76a05b35 CM |
2342 | |
2343 | ret = -EAGAIN; | |
5bdd3536 | 2344 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2345 | if (tmp) { |
2346 | /* | |
2347 | * If the read above didn't mark this buffer up to date, | |
2348 | * it will never end up being up to date. Set ret to EIO now | |
2349 | * and give up so that our caller doesn't loop forever | |
2350 | * on our EAGAINs. | |
2351 | */ | |
b9fab919 | 2352 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2353 | ret = -EIO; |
c8c42864 | 2354 | free_extent_buffer(tmp); |
76a05b35 CM |
2355 | } |
2356 | return ret; | |
c8c42864 CM |
2357 | } |
2358 | ||
2359 | /* | |
2360 | * helper function for btrfs_search_slot. This does all of the checks | |
2361 | * for node-level blocks and does any balancing required based on | |
2362 | * the ins_len. | |
2363 | * | |
2364 | * If no extra work was required, zero is returned. If we had to | |
2365 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2366 | * start over | |
2367 | */ | |
2368 | static int | |
2369 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2370 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2371 | struct extent_buffer *b, int level, int ins_len, |
2372 | int *write_lock_level) | |
c8c42864 CM |
2373 | { |
2374 | int ret; | |
2375 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2376 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2377 | int sret; | |
2378 | ||
bd681513 CM |
2379 | if (*write_lock_level < level + 1) { |
2380 | *write_lock_level = level + 1; | |
2381 | btrfs_release_path(p); | |
2382 | goto again; | |
2383 | } | |
2384 | ||
c8c42864 | 2385 | btrfs_set_path_blocking(p); |
0b08851f | 2386 | reada_for_balance(root, p, level); |
c8c42864 | 2387 | sret = split_node(trans, root, p, level); |
bd681513 | 2388 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2389 | |
2390 | BUG_ON(sret > 0); | |
2391 | if (sret) { | |
2392 | ret = sret; | |
2393 | goto done; | |
2394 | } | |
2395 | b = p->nodes[level]; | |
2396 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2397 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2398 | int sret; |
2399 | ||
bd681513 CM |
2400 | if (*write_lock_level < level + 1) { |
2401 | *write_lock_level = level + 1; | |
2402 | btrfs_release_path(p); | |
2403 | goto again; | |
2404 | } | |
2405 | ||
c8c42864 | 2406 | btrfs_set_path_blocking(p); |
0b08851f | 2407 | reada_for_balance(root, p, level); |
c8c42864 | 2408 | sret = balance_level(trans, root, p, level); |
bd681513 | 2409 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2410 | |
2411 | if (sret) { | |
2412 | ret = sret; | |
2413 | goto done; | |
2414 | } | |
2415 | b = p->nodes[level]; | |
2416 | if (!b) { | |
b3b4aa74 | 2417 | btrfs_release_path(p); |
c8c42864 CM |
2418 | goto again; |
2419 | } | |
2420 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2421 | } | |
2422 | return 0; | |
2423 | ||
2424 | again: | |
2425 | ret = -EAGAIN; | |
2426 | done: | |
2427 | return ret; | |
2428 | } | |
2429 | ||
d7396f07 FDBM |
2430 | static void key_search_validate(struct extent_buffer *b, |
2431 | struct btrfs_key *key, | |
2432 | int level) | |
2433 | { | |
2434 | #ifdef CONFIG_BTRFS_ASSERT | |
2435 | struct btrfs_disk_key disk_key; | |
2436 | ||
2437 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2438 | ||
2439 | if (level == 0) | |
2440 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2441 | offsetof(struct btrfs_leaf, items[0].key), | |
2442 | sizeof(disk_key))); | |
2443 | else | |
2444 | ASSERT(!memcmp_extent_buffer(b, &disk_key, | |
2445 | offsetof(struct btrfs_node, ptrs[0].key), | |
2446 | sizeof(disk_key))); | |
2447 | #endif | |
2448 | } | |
2449 | ||
2450 | static int key_search(struct extent_buffer *b, struct btrfs_key *key, | |
2451 | int level, int *prev_cmp, int *slot) | |
2452 | { | |
2453 | if (*prev_cmp != 0) { | |
2454 | *prev_cmp = bin_search(b, key, level, slot); | |
2455 | return *prev_cmp; | |
2456 | } | |
2457 | ||
2458 | key_search_validate(b, key, level); | |
2459 | *slot = 0; | |
2460 | ||
2461 | return 0; | |
2462 | } | |
2463 | ||
3f870c28 | 2464 | int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path, |
e33d5c3d KN |
2465 | u64 iobjectid, u64 ioff, u8 key_type, |
2466 | struct btrfs_key *found_key) | |
2467 | { | |
2468 | int ret; | |
2469 | struct btrfs_key key; | |
2470 | struct extent_buffer *eb; | |
3f870c28 | 2471 | struct btrfs_path *path; |
e33d5c3d KN |
2472 | |
2473 | key.type = key_type; | |
2474 | key.objectid = iobjectid; | |
2475 | key.offset = ioff; | |
2476 | ||
3f870c28 KN |
2477 | if (found_path == NULL) { |
2478 | path = btrfs_alloc_path(); | |
2479 | if (!path) | |
2480 | return -ENOMEM; | |
2481 | } else | |
2482 | path = found_path; | |
2483 | ||
e33d5c3d | 2484 | ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0); |
3f870c28 KN |
2485 | if ((ret < 0) || (found_key == NULL)) { |
2486 | if (path != found_path) | |
2487 | btrfs_free_path(path); | |
e33d5c3d | 2488 | return ret; |
3f870c28 | 2489 | } |
e33d5c3d KN |
2490 | |
2491 | eb = path->nodes[0]; | |
2492 | if (ret && path->slots[0] >= btrfs_header_nritems(eb)) { | |
2493 | ret = btrfs_next_leaf(fs_root, path); | |
2494 | if (ret) | |
2495 | return ret; | |
2496 | eb = path->nodes[0]; | |
2497 | } | |
2498 | ||
2499 | btrfs_item_key_to_cpu(eb, found_key, path->slots[0]); | |
2500 | if (found_key->type != key.type || | |
2501 | found_key->objectid != key.objectid) | |
2502 | return 1; | |
2503 | ||
2504 | return 0; | |
2505 | } | |
2506 | ||
74123bd7 CM |
2507 | /* |
2508 | * look for key in the tree. path is filled in with nodes along the way | |
2509 | * if key is found, we return zero and you can find the item in the leaf | |
2510 | * level of the path (level 0) | |
2511 | * | |
2512 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2513 | * be inserted, and 1 is returned. If there are other errors during the |
2514 | * search a negative error number is returned. | |
97571fd0 CM |
2515 | * |
2516 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2517 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2518 | * possible) | |
74123bd7 | 2519 | */ |
e089f05c CM |
2520 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2521 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2522 | ins_len, int cow) | |
be0e5c09 | 2523 | { |
5f39d397 | 2524 | struct extent_buffer *b; |
be0e5c09 CM |
2525 | int slot; |
2526 | int ret; | |
33c66f43 | 2527 | int err; |
be0e5c09 | 2528 | int level; |
925baedd | 2529 | int lowest_unlock = 1; |
bd681513 CM |
2530 | int root_lock; |
2531 | /* everything at write_lock_level or lower must be write locked */ | |
2532 | int write_lock_level = 0; | |
9f3a7427 | 2533 | u8 lowest_level = 0; |
f7c79f30 | 2534 | int min_write_lock_level; |
d7396f07 | 2535 | int prev_cmp; |
9f3a7427 | 2536 | |
6702ed49 | 2537 | lowest_level = p->lowest_level; |
323ac95b | 2538 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2539 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2540 | |
bd681513 | 2541 | if (ins_len < 0) { |
925baedd | 2542 | lowest_unlock = 2; |
65b51a00 | 2543 | |
bd681513 CM |
2544 | /* when we are removing items, we might have to go up to level |
2545 | * two as we update tree pointers Make sure we keep write | |
2546 | * for those levels as well | |
2547 | */ | |
2548 | write_lock_level = 2; | |
2549 | } else if (ins_len > 0) { | |
2550 | /* | |
2551 | * for inserting items, make sure we have a write lock on | |
2552 | * level 1 so we can update keys | |
2553 | */ | |
2554 | write_lock_level = 1; | |
2555 | } | |
2556 | ||
2557 | if (!cow) | |
2558 | write_lock_level = -1; | |
2559 | ||
09a2a8f9 | 2560 | if (cow && (p->keep_locks || p->lowest_level)) |
bd681513 CM |
2561 | write_lock_level = BTRFS_MAX_LEVEL; |
2562 | ||
f7c79f30 CM |
2563 | min_write_lock_level = write_lock_level; |
2564 | ||
bb803951 | 2565 | again: |
d7396f07 | 2566 | prev_cmp = -1; |
bd681513 CM |
2567 | /* |
2568 | * we try very hard to do read locks on the root | |
2569 | */ | |
2570 | root_lock = BTRFS_READ_LOCK; | |
2571 | level = 0; | |
5d4f98a2 | 2572 | if (p->search_commit_root) { |
bd681513 CM |
2573 | /* |
2574 | * the commit roots are read only | |
2575 | * so we always do read locks | |
2576 | */ | |
5d4f98a2 YZ |
2577 | b = root->commit_root; |
2578 | extent_buffer_get(b); | |
bd681513 | 2579 | level = btrfs_header_level(b); |
5d4f98a2 | 2580 | if (!p->skip_locking) |
bd681513 | 2581 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2582 | } else { |
bd681513 | 2583 | if (p->skip_locking) { |
5d4f98a2 | 2584 | b = btrfs_root_node(root); |
bd681513 CM |
2585 | level = btrfs_header_level(b); |
2586 | } else { | |
2587 | /* we don't know the level of the root node | |
2588 | * until we actually have it read locked | |
2589 | */ | |
2590 | b = btrfs_read_lock_root_node(root); | |
2591 | level = btrfs_header_level(b); | |
2592 | if (level <= write_lock_level) { | |
2593 | /* whoops, must trade for write lock */ | |
2594 | btrfs_tree_read_unlock(b); | |
2595 | free_extent_buffer(b); | |
2596 | b = btrfs_lock_root_node(root); | |
2597 | root_lock = BTRFS_WRITE_LOCK; | |
2598 | ||
2599 | /* the level might have changed, check again */ | |
2600 | level = btrfs_header_level(b); | |
2601 | } | |
2602 | } | |
5d4f98a2 | 2603 | } |
bd681513 CM |
2604 | p->nodes[level] = b; |
2605 | if (!p->skip_locking) | |
2606 | p->locks[level] = root_lock; | |
925baedd | 2607 | |
eb60ceac | 2608 | while (b) { |
5f39d397 | 2609 | level = btrfs_header_level(b); |
65b51a00 CM |
2610 | |
2611 | /* | |
2612 | * setup the path here so we can release it under lock | |
2613 | * contention with the cow code | |
2614 | */ | |
02217ed2 | 2615 | if (cow) { |
c8c42864 CM |
2616 | /* |
2617 | * if we don't really need to cow this block | |
2618 | * then we don't want to set the path blocking, | |
2619 | * so we test it here | |
2620 | */ | |
5d4f98a2 | 2621 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2622 | goto cow_done; |
5d4f98a2 | 2623 | |
b4ce94de CM |
2624 | btrfs_set_path_blocking(p); |
2625 | ||
bd681513 CM |
2626 | /* |
2627 | * must have write locks on this node and the | |
2628 | * parent | |
2629 | */ | |
5124e00e JB |
2630 | if (level > write_lock_level || |
2631 | (level + 1 > write_lock_level && | |
2632 | level + 1 < BTRFS_MAX_LEVEL && | |
2633 | p->nodes[level + 1])) { | |
bd681513 CM |
2634 | write_lock_level = level + 1; |
2635 | btrfs_release_path(p); | |
2636 | goto again; | |
2637 | } | |
2638 | ||
33c66f43 YZ |
2639 | err = btrfs_cow_block(trans, root, b, |
2640 | p->nodes[level + 1], | |
2641 | p->slots[level + 1], &b); | |
2642 | if (err) { | |
33c66f43 | 2643 | ret = err; |
65b51a00 | 2644 | goto done; |
54aa1f4d | 2645 | } |
02217ed2 | 2646 | } |
65b51a00 | 2647 | cow_done: |
02217ed2 | 2648 | BUG_ON(!cow && ins_len); |
65b51a00 | 2649 | |
eb60ceac | 2650 | p->nodes[level] = b; |
bd681513 | 2651 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2652 | |
2653 | /* | |
2654 | * we have a lock on b and as long as we aren't changing | |
2655 | * the tree, there is no way to for the items in b to change. | |
2656 | * It is safe to drop the lock on our parent before we | |
2657 | * go through the expensive btree search on b. | |
2658 | * | |
2659 | * If cow is true, then we might be changing slot zero, | |
2660 | * which may require changing the parent. So, we can't | |
2661 | * drop the lock until after we know which slot we're | |
2662 | * operating on. | |
2663 | */ | |
2664 | if (!cow) | |
2665 | btrfs_unlock_up_safe(p, level + 1); | |
2666 | ||
d7396f07 | 2667 | ret = key_search(b, key, level, &prev_cmp, &slot); |
b4ce94de | 2668 | |
5f39d397 | 2669 | if (level != 0) { |
33c66f43 YZ |
2670 | int dec = 0; |
2671 | if (ret && slot > 0) { | |
2672 | dec = 1; | |
be0e5c09 | 2673 | slot -= 1; |
33c66f43 | 2674 | } |
be0e5c09 | 2675 | p->slots[level] = slot; |
33c66f43 | 2676 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2677 | ins_len, &write_lock_level); |
33c66f43 | 2678 | if (err == -EAGAIN) |
c8c42864 | 2679 | goto again; |
33c66f43 YZ |
2680 | if (err) { |
2681 | ret = err; | |
c8c42864 | 2682 | goto done; |
33c66f43 | 2683 | } |
c8c42864 CM |
2684 | b = p->nodes[level]; |
2685 | slot = p->slots[level]; | |
b4ce94de | 2686 | |
bd681513 CM |
2687 | /* |
2688 | * slot 0 is special, if we change the key | |
2689 | * we have to update the parent pointer | |
2690 | * which means we must have a write lock | |
2691 | * on the parent | |
2692 | */ | |
2693 | if (slot == 0 && cow && | |
2694 | write_lock_level < level + 1) { | |
2695 | write_lock_level = level + 1; | |
2696 | btrfs_release_path(p); | |
2697 | goto again; | |
2698 | } | |
2699 | ||
f7c79f30 CM |
2700 | unlock_up(p, level, lowest_unlock, |
2701 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2702 | |
925baedd | 2703 | if (level == lowest_level) { |
33c66f43 YZ |
2704 | if (dec) |
2705 | p->slots[level]++; | |
5b21f2ed | 2706 | goto done; |
925baedd | 2707 | } |
ca7a79ad | 2708 | |
33c66f43 | 2709 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2710 | &b, level, slot, key, 0); |
33c66f43 | 2711 | if (err == -EAGAIN) |
c8c42864 | 2712 | goto again; |
33c66f43 YZ |
2713 | if (err) { |
2714 | ret = err; | |
76a05b35 | 2715 | goto done; |
33c66f43 | 2716 | } |
76a05b35 | 2717 | |
b4ce94de | 2718 | if (!p->skip_locking) { |
bd681513 CM |
2719 | level = btrfs_header_level(b); |
2720 | if (level <= write_lock_level) { | |
2721 | err = btrfs_try_tree_write_lock(b); | |
2722 | if (!err) { | |
2723 | btrfs_set_path_blocking(p); | |
2724 | btrfs_tree_lock(b); | |
2725 | btrfs_clear_path_blocking(p, b, | |
2726 | BTRFS_WRITE_LOCK); | |
2727 | } | |
2728 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2729 | } else { | |
2730 | err = btrfs_try_tree_read_lock(b); | |
2731 | if (!err) { | |
2732 | btrfs_set_path_blocking(p); | |
2733 | btrfs_tree_read_lock(b); | |
2734 | btrfs_clear_path_blocking(p, b, | |
2735 | BTRFS_READ_LOCK); | |
2736 | } | |
2737 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2738 | } |
bd681513 | 2739 | p->nodes[level] = b; |
b4ce94de | 2740 | } |
be0e5c09 CM |
2741 | } else { |
2742 | p->slots[level] = slot; | |
87b29b20 YZ |
2743 | if (ins_len > 0 && |
2744 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2745 | if (write_lock_level < 1) { |
2746 | write_lock_level = 1; | |
2747 | btrfs_release_path(p); | |
2748 | goto again; | |
2749 | } | |
2750 | ||
b4ce94de | 2751 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2752 | err = split_leaf(trans, root, key, |
2753 | p, ins_len, ret == 0); | |
bd681513 | 2754 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2755 | |
33c66f43 YZ |
2756 | BUG_ON(err > 0); |
2757 | if (err) { | |
2758 | ret = err; | |
65b51a00 CM |
2759 | goto done; |
2760 | } | |
5c680ed6 | 2761 | } |
459931ec | 2762 | if (!p->search_for_split) |
f7c79f30 CM |
2763 | unlock_up(p, level, lowest_unlock, |
2764 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2765 | goto done; |
be0e5c09 CM |
2766 | } |
2767 | } | |
65b51a00 CM |
2768 | ret = 1; |
2769 | done: | |
b4ce94de CM |
2770 | /* |
2771 | * we don't really know what they plan on doing with the path | |
2772 | * from here on, so for now just mark it as blocking | |
2773 | */ | |
b9473439 CM |
2774 | if (!p->leave_spinning) |
2775 | btrfs_set_path_blocking(p); | |
76a05b35 | 2776 | if (ret < 0) |
b3b4aa74 | 2777 | btrfs_release_path(p); |
65b51a00 | 2778 | return ret; |
be0e5c09 CM |
2779 | } |
2780 | ||
5d9e75c4 JS |
2781 | /* |
2782 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2783 | * current state of the tree together with the operations recorded in the tree | |
2784 | * modification log to search for the key in a previous version of this tree, as | |
2785 | * denoted by the time_seq parameter. | |
2786 | * | |
2787 | * Naturally, there is no support for insert, delete or cow operations. | |
2788 | * | |
2789 | * The resulting path and return value will be set up as if we called | |
2790 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2791 | */ | |
2792 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2793 | struct btrfs_path *p, u64 time_seq) | |
2794 | { | |
2795 | struct extent_buffer *b; | |
2796 | int slot; | |
2797 | int ret; | |
2798 | int err; | |
2799 | int level; | |
2800 | int lowest_unlock = 1; | |
2801 | u8 lowest_level = 0; | |
d4b4087c | 2802 | int prev_cmp = -1; |
5d9e75c4 JS |
2803 | |
2804 | lowest_level = p->lowest_level; | |
2805 | WARN_ON(p->nodes[0] != NULL); | |
2806 | ||
2807 | if (p->search_commit_root) { | |
2808 | BUG_ON(time_seq); | |
2809 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2810 | } | |
2811 | ||
2812 | again: | |
5d9e75c4 | 2813 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2814 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2815 | p->locks[level] = BTRFS_READ_LOCK; |
2816 | ||
2817 | while (b) { | |
2818 | level = btrfs_header_level(b); | |
2819 | p->nodes[level] = b; | |
2820 | btrfs_clear_path_blocking(p, NULL, 0); | |
2821 | ||
2822 | /* | |
2823 | * we have a lock on b and as long as we aren't changing | |
2824 | * the tree, there is no way to for the items in b to change. | |
2825 | * It is safe to drop the lock on our parent before we | |
2826 | * go through the expensive btree search on b. | |
2827 | */ | |
2828 | btrfs_unlock_up_safe(p, level + 1); | |
2829 | ||
d4b4087c JB |
2830 | /* |
2831 | * Since we can unwind eb's we want to do a real search every | |
2832 | * time. | |
2833 | */ | |
2834 | prev_cmp = -1; | |
d7396f07 | 2835 | ret = key_search(b, key, level, &prev_cmp, &slot); |
5d9e75c4 JS |
2836 | |
2837 | if (level != 0) { | |
2838 | int dec = 0; | |
2839 | if (ret && slot > 0) { | |
2840 | dec = 1; | |
2841 | slot -= 1; | |
2842 | } | |
2843 | p->slots[level] = slot; | |
2844 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2845 | ||
2846 | if (level == lowest_level) { | |
2847 | if (dec) | |
2848 | p->slots[level]++; | |
2849 | goto done; | |
2850 | } | |
2851 | ||
2852 | err = read_block_for_search(NULL, root, p, &b, level, | |
2853 | slot, key, time_seq); | |
2854 | if (err == -EAGAIN) | |
2855 | goto again; | |
2856 | if (err) { | |
2857 | ret = err; | |
2858 | goto done; | |
2859 | } | |
2860 | ||
2861 | level = btrfs_header_level(b); | |
2862 | err = btrfs_try_tree_read_lock(b); | |
2863 | if (!err) { | |
2864 | btrfs_set_path_blocking(p); | |
2865 | btrfs_tree_read_lock(b); | |
2866 | btrfs_clear_path_blocking(p, b, | |
2867 | BTRFS_READ_LOCK); | |
2868 | } | |
9ec72677 | 2869 | b = tree_mod_log_rewind(root->fs_info, p, b, time_seq); |
db7f3436 JB |
2870 | if (!b) { |
2871 | ret = -ENOMEM; | |
2872 | goto done; | |
2873 | } | |
5d9e75c4 JS |
2874 | p->locks[level] = BTRFS_READ_LOCK; |
2875 | p->nodes[level] = b; | |
5d9e75c4 JS |
2876 | } else { |
2877 | p->slots[level] = slot; | |
2878 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2879 | goto done; | |
2880 | } | |
2881 | } | |
2882 | ret = 1; | |
2883 | done: | |
2884 | if (!p->leave_spinning) | |
2885 | btrfs_set_path_blocking(p); | |
2886 | if (ret < 0) | |
2887 | btrfs_release_path(p); | |
2888 | ||
2889 | return ret; | |
2890 | } | |
2891 | ||
2f38b3e1 AJ |
2892 | /* |
2893 | * helper to use instead of search slot if no exact match is needed but | |
2894 | * instead the next or previous item should be returned. | |
2895 | * When find_higher is true, the next higher item is returned, the next lower | |
2896 | * otherwise. | |
2897 | * When return_any and find_higher are both true, and no higher item is found, | |
2898 | * return the next lower instead. | |
2899 | * When return_any is true and find_higher is false, and no lower item is found, | |
2900 | * return the next higher instead. | |
2901 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2902 | * < 0 on error | |
2903 | */ | |
2904 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
2905 | struct btrfs_key *key, struct btrfs_path *p, | |
2906 | int find_higher, int return_any) | |
2907 | { | |
2908 | int ret; | |
2909 | struct extent_buffer *leaf; | |
2910 | ||
2911 | again: | |
2912 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2913 | if (ret <= 0) | |
2914 | return ret; | |
2915 | /* | |
2916 | * a return value of 1 means the path is at the position where the | |
2917 | * item should be inserted. Normally this is the next bigger item, | |
2918 | * but in case the previous item is the last in a leaf, path points | |
2919 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2920 | * item. | |
2921 | */ | |
2922 | leaf = p->nodes[0]; | |
2923 | ||
2924 | if (find_higher) { | |
2925 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2926 | ret = btrfs_next_leaf(root, p); | |
2927 | if (ret <= 0) | |
2928 | return ret; | |
2929 | if (!return_any) | |
2930 | return 1; | |
2931 | /* | |
2932 | * no higher item found, return the next | |
2933 | * lower instead | |
2934 | */ | |
2935 | return_any = 0; | |
2936 | find_higher = 0; | |
2937 | btrfs_release_path(p); | |
2938 | goto again; | |
2939 | } | |
2940 | } else { | |
e6793769 AJ |
2941 | if (p->slots[0] == 0) { |
2942 | ret = btrfs_prev_leaf(root, p); | |
2943 | if (ret < 0) | |
2944 | return ret; | |
2945 | if (!ret) { | |
2946 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
2947 | return 0; | |
2f38b3e1 | 2948 | } |
e6793769 AJ |
2949 | if (!return_any) |
2950 | return 1; | |
2951 | /* | |
2952 | * no lower item found, return the next | |
2953 | * higher instead | |
2954 | */ | |
2955 | return_any = 0; | |
2956 | find_higher = 1; | |
2957 | btrfs_release_path(p); | |
2958 | goto again; | |
2959 | } else { | |
2f38b3e1 AJ |
2960 | --p->slots[0]; |
2961 | } | |
2962 | } | |
2963 | return 0; | |
2964 | } | |
2965 | ||
74123bd7 CM |
2966 | /* |
2967 | * adjust the pointers going up the tree, starting at level | |
2968 | * making sure the right key of each node is points to 'key'. | |
2969 | * This is used after shifting pointers to the left, so it stops | |
2970 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2971 | * higher levels | |
aa5d6bed | 2972 | * |
74123bd7 | 2973 | */ |
d6a0a126 | 2974 | static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 2975 | struct btrfs_disk_key *key, int level) |
be0e5c09 CM |
2976 | { |
2977 | int i; | |
5f39d397 CM |
2978 | struct extent_buffer *t; |
2979 | ||
234b63a0 | 2980 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2981 | int tslot = path->slots[i]; |
eb60ceac | 2982 | if (!path->nodes[i]) |
be0e5c09 | 2983 | break; |
5f39d397 | 2984 | t = path->nodes[i]; |
32adf090 | 2985 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 2986 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2987 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2988 | if (tslot != 0) |
2989 | break; | |
2990 | } | |
2991 | } | |
2992 | ||
31840ae1 ZY |
2993 | /* |
2994 | * update item key. | |
2995 | * | |
2996 | * This function isn't completely safe. It's the caller's responsibility | |
2997 | * that the new key won't break the order | |
2998 | */ | |
afe5fea7 | 2999 | void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 3000 | struct btrfs_key *new_key) |
31840ae1 ZY |
3001 | { |
3002 | struct btrfs_disk_key disk_key; | |
3003 | struct extent_buffer *eb; | |
3004 | int slot; | |
3005 | ||
3006 | eb = path->nodes[0]; | |
3007 | slot = path->slots[0]; | |
3008 | if (slot > 0) { | |
3009 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 3010 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
3011 | } |
3012 | if (slot < btrfs_header_nritems(eb) - 1) { | |
3013 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 3014 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
3015 | } |
3016 | ||
3017 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
3018 | btrfs_set_item_key(eb, &disk_key, slot); | |
3019 | btrfs_mark_buffer_dirty(eb); | |
3020 | if (slot == 0) | |
d6a0a126 | 3021 | fixup_low_keys(root, path, &disk_key, 1); |
31840ae1 ZY |
3022 | } |
3023 | ||
74123bd7 CM |
3024 | /* |
3025 | * try to push data from one node into the next node left in the | |
79f95c82 | 3026 | * tree. |
aa5d6bed CM |
3027 | * |
3028 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
3029 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 3030 | */ |
98ed5174 CM |
3031 | static int push_node_left(struct btrfs_trans_handle *trans, |
3032 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 3033 | struct extent_buffer *src, int empty) |
be0e5c09 | 3034 | { |
be0e5c09 | 3035 | int push_items = 0; |
bb803951 CM |
3036 | int src_nritems; |
3037 | int dst_nritems; | |
aa5d6bed | 3038 | int ret = 0; |
be0e5c09 | 3039 | |
5f39d397 CM |
3040 | src_nritems = btrfs_header_nritems(src); |
3041 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3042 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
3043 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3044 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 3045 | |
bce4eae9 | 3046 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
3047 | return 1; |
3048 | ||
d397712b | 3049 | if (push_items <= 0) |
be0e5c09 CM |
3050 | return 1; |
3051 | ||
bce4eae9 | 3052 | if (empty) { |
971a1f66 | 3053 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
3054 | if (push_items < src_nritems) { |
3055 | /* leave at least 8 pointers in the node if | |
3056 | * we aren't going to empty it | |
3057 | */ | |
3058 | if (src_nritems - push_items < 8) { | |
3059 | if (push_items <= 8) | |
3060 | return 1; | |
3061 | push_items -= 8; | |
3062 | } | |
3063 | } | |
3064 | } else | |
3065 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 3066 | |
f230475e | 3067 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
90f8d62e | 3068 | push_items); |
5f39d397 CM |
3069 | copy_extent_buffer(dst, src, |
3070 | btrfs_node_key_ptr_offset(dst_nritems), | |
3071 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3072 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3073 | |
bb803951 | 3074 | if (push_items < src_nritems) { |
57911b8b JS |
3075 | /* |
3076 | * don't call tree_mod_log_eb_move here, key removal was already | |
3077 | * fully logged by tree_mod_log_eb_copy above. | |
3078 | */ | |
5f39d397 CM |
3079 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3080 | btrfs_node_key_ptr_offset(push_items), | |
3081 | (src_nritems - push_items) * | |
3082 | sizeof(struct btrfs_key_ptr)); | |
3083 | } | |
3084 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3085 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3086 | btrfs_mark_buffer_dirty(src); | |
3087 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3088 | |
79f95c82 CM |
3089 | return ret; |
3090 | } | |
3091 | ||
3092 | /* | |
3093 | * try to push data from one node into the next node right in the | |
3094 | * tree. | |
3095 | * | |
3096 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3097 | * error, and > 0 if there was no room in the right hand block. | |
3098 | * | |
3099 | * this will only push up to 1/2 the contents of the left node over | |
3100 | */ | |
5f39d397 CM |
3101 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3102 | struct btrfs_root *root, | |
3103 | struct extent_buffer *dst, | |
3104 | struct extent_buffer *src) | |
79f95c82 | 3105 | { |
79f95c82 CM |
3106 | int push_items = 0; |
3107 | int max_push; | |
3108 | int src_nritems; | |
3109 | int dst_nritems; | |
3110 | int ret = 0; | |
79f95c82 | 3111 | |
7bb86316 CM |
3112 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3113 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3114 | ||
5f39d397 CM |
3115 | src_nritems = btrfs_header_nritems(src); |
3116 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3117 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3118 | if (push_items <= 0) |
79f95c82 | 3119 | return 1; |
bce4eae9 | 3120 | |
d397712b | 3121 | if (src_nritems < 4) |
bce4eae9 | 3122 | return 1; |
79f95c82 CM |
3123 | |
3124 | max_push = src_nritems / 2 + 1; | |
3125 | /* don't try to empty the node */ | |
d397712b | 3126 | if (max_push >= src_nritems) |
79f95c82 | 3127 | return 1; |
252c38f0 | 3128 | |
79f95c82 CM |
3129 | if (max_push < push_items) |
3130 | push_items = max_push; | |
3131 | ||
f230475e | 3132 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3133 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3134 | btrfs_node_key_ptr_offset(0), | |
3135 | (dst_nritems) * | |
3136 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3137 | |
f230475e | 3138 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
90f8d62e | 3139 | src_nritems - push_items, push_items); |
5f39d397 CM |
3140 | copy_extent_buffer(dst, src, |
3141 | btrfs_node_key_ptr_offset(0), | |
3142 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3143 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3144 | |
5f39d397 CM |
3145 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3146 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3147 | |
5f39d397 CM |
3148 | btrfs_mark_buffer_dirty(src); |
3149 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3150 | |
aa5d6bed | 3151 | return ret; |
be0e5c09 CM |
3152 | } |
3153 | ||
97571fd0 CM |
3154 | /* |
3155 | * helper function to insert a new root level in the tree. | |
3156 | * A new node is allocated, and a single item is inserted to | |
3157 | * point to the existing root | |
aa5d6bed CM |
3158 | * |
3159 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3160 | */ |
d397712b | 3161 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 | 3162 | struct btrfs_root *root, |
fdd99c72 | 3163 | struct btrfs_path *path, int level) |
5c680ed6 | 3164 | { |
7bb86316 | 3165 | u64 lower_gen; |
5f39d397 CM |
3166 | struct extent_buffer *lower; |
3167 | struct extent_buffer *c; | |
925baedd | 3168 | struct extent_buffer *old; |
5f39d397 | 3169 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3170 | |
3171 | BUG_ON(path->nodes[level]); | |
3172 | BUG_ON(path->nodes[level-1] != root->node); | |
3173 | ||
7bb86316 CM |
3174 | lower = path->nodes[level-1]; |
3175 | if (level == 1) | |
3176 | btrfs_item_key(lower, &lower_key, 0); | |
3177 | else | |
3178 | btrfs_node_key(lower, &lower_key, 0); | |
3179 | ||
31840ae1 | 3180 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3181 | root->root_key.objectid, &lower_key, |
5581a51a | 3182 | level, root->node->start, 0); |
5f39d397 CM |
3183 | if (IS_ERR(c)) |
3184 | return PTR_ERR(c); | |
925baedd | 3185 | |
f0486c68 YZ |
3186 | root_add_used(root, root->nodesize); |
3187 | ||
5d4f98a2 | 3188 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3189 | btrfs_set_header_nritems(c, 1); |
3190 | btrfs_set_header_level(c, level); | |
db94535d | 3191 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3192 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3193 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3194 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 | 3195 | |
0a4e5586 | 3196 | write_extent_buffer(c, root->fs_info->fsid, btrfs_header_fsid(), |
5f39d397 | 3197 | BTRFS_FSID_SIZE); |
e17cade2 CM |
3198 | |
3199 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 3200 | btrfs_header_chunk_tree_uuid(c), BTRFS_UUID_SIZE); |
e17cade2 | 3201 | |
5f39d397 | 3202 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3203 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3204 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3205 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3206 | |
3207 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3208 | |
5f39d397 | 3209 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3210 | |
925baedd | 3211 | old = root->node; |
fdd99c72 | 3212 | tree_mod_log_set_root_pointer(root, c, 0); |
240f62c8 | 3213 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3214 | |
3215 | /* the super has an extra ref to root->node */ | |
3216 | free_extent_buffer(old); | |
3217 | ||
0b86a832 | 3218 | add_root_to_dirty_list(root); |
5f39d397 CM |
3219 | extent_buffer_get(c); |
3220 | path->nodes[level] = c; | |
bd681513 | 3221 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3222 | path->slots[level] = 0; |
3223 | return 0; | |
3224 | } | |
3225 | ||
74123bd7 CM |
3226 | /* |
3227 | * worker function to insert a single pointer in a node. | |
3228 | * the node should have enough room for the pointer already | |
97571fd0 | 3229 | * |
74123bd7 CM |
3230 | * slot and level indicate where you want the key to go, and |
3231 | * blocknr is the block the key points to. | |
3232 | */ | |
143bede5 JM |
3233 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3234 | struct btrfs_root *root, struct btrfs_path *path, | |
3235 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3236 | int slot, int level) |
74123bd7 | 3237 | { |
5f39d397 | 3238 | struct extent_buffer *lower; |
74123bd7 | 3239 | int nritems; |
f3ea38da | 3240 | int ret; |
5c680ed6 CM |
3241 | |
3242 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3243 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3244 | lower = path->nodes[level]; |
3245 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3246 | BUG_ON(slot > nritems); |
143bede5 | 3247 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3248 | if (slot != nritems) { |
c3e06965 | 3249 | if (level) |
f3ea38da JS |
3250 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3251 | slot, nritems - slot); | |
5f39d397 CM |
3252 | memmove_extent_buffer(lower, |
3253 | btrfs_node_key_ptr_offset(slot + 1), | |
3254 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3255 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3256 | } |
c3e06965 | 3257 | if (level) { |
f3ea38da | 3258 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
c8cc6341 | 3259 | MOD_LOG_KEY_ADD, GFP_NOFS); |
f3ea38da JS |
3260 | BUG_ON(ret < 0); |
3261 | } | |
5f39d397 | 3262 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3263 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3264 | WARN_ON(trans->transid == 0); |
3265 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3266 | btrfs_set_header_nritems(lower, nritems + 1); |
3267 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3268 | } |
3269 | ||
97571fd0 CM |
3270 | /* |
3271 | * split the node at the specified level in path in two. | |
3272 | * The path is corrected to point to the appropriate node after the split | |
3273 | * | |
3274 | * Before splitting this tries to make some room in the node by pushing | |
3275 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3276 | * |
3277 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3278 | */ |
e02119d5 CM |
3279 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3280 | struct btrfs_root *root, | |
3281 | struct btrfs_path *path, int level) | |
be0e5c09 | 3282 | { |
5f39d397 CM |
3283 | struct extent_buffer *c; |
3284 | struct extent_buffer *split; | |
3285 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3286 | int mid; |
5c680ed6 | 3287 | int ret; |
7518a238 | 3288 | u32 c_nritems; |
eb60ceac | 3289 | |
5f39d397 | 3290 | c = path->nodes[level]; |
7bb86316 | 3291 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3292 | if (c == root->node) { |
d9abbf1c | 3293 | /* |
90f8d62e JS |
3294 | * trying to split the root, lets make a new one |
3295 | * | |
fdd99c72 | 3296 | * tree mod log: We don't log_removal old root in |
90f8d62e JS |
3297 | * insert_new_root, because that root buffer will be kept as a |
3298 | * normal node. We are going to log removal of half of the | |
3299 | * elements below with tree_mod_log_eb_copy. We're holding a | |
3300 | * tree lock on the buffer, which is why we cannot race with | |
3301 | * other tree_mod_log users. | |
d9abbf1c | 3302 | */ |
fdd99c72 | 3303 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3304 | if (ret) |
3305 | return ret; | |
b3612421 | 3306 | } else { |
e66f709b | 3307 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3308 | c = path->nodes[level]; |
3309 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3310 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3311 | return 0; |
54aa1f4d CM |
3312 | if (ret < 0) |
3313 | return ret; | |
be0e5c09 | 3314 | } |
e66f709b | 3315 | |
5f39d397 | 3316 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3317 | mid = (c_nritems + 1) / 2; |
3318 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3319 | |
5d4f98a2 | 3320 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3321 | root->root_key.objectid, |
5581a51a | 3322 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3323 | if (IS_ERR(split)) |
3324 | return PTR_ERR(split); | |
3325 | ||
f0486c68 YZ |
3326 | root_add_used(root, root->nodesize); |
3327 | ||
5d4f98a2 | 3328 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3329 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3330 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3331 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3332 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3333 | btrfs_set_header_owner(split, root->root_key.objectid); |
3334 | write_extent_buffer(split, root->fs_info->fsid, | |
0a4e5586 | 3335 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 | 3336 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
b308bc2f | 3337 | btrfs_header_chunk_tree_uuid(split), |
e17cade2 | 3338 | BTRFS_UUID_SIZE); |
54aa1f4d | 3339 | |
90f8d62e | 3340 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
3341 | copy_extent_buffer(split, c, |
3342 | btrfs_node_key_ptr_offset(0), | |
3343 | btrfs_node_key_ptr_offset(mid), | |
3344 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3345 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3346 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3347 | ret = 0; |
3348 | ||
5f39d397 CM |
3349 | btrfs_mark_buffer_dirty(c); |
3350 | btrfs_mark_buffer_dirty(split); | |
3351 | ||
143bede5 | 3352 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3353 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3354 | |
5de08d7d | 3355 | if (path->slots[level] >= mid) { |
5c680ed6 | 3356 | path->slots[level] -= mid; |
925baedd | 3357 | btrfs_tree_unlock(c); |
5f39d397 CM |
3358 | free_extent_buffer(c); |
3359 | path->nodes[level] = split; | |
5c680ed6 CM |
3360 | path->slots[level + 1] += 1; |
3361 | } else { | |
925baedd | 3362 | btrfs_tree_unlock(split); |
5f39d397 | 3363 | free_extent_buffer(split); |
be0e5c09 | 3364 | } |
aa5d6bed | 3365 | return ret; |
be0e5c09 CM |
3366 | } |
3367 | ||
74123bd7 CM |
3368 | /* |
3369 | * how many bytes are required to store the items in a leaf. start | |
3370 | * and nr indicate which items in the leaf to check. This totals up the | |
3371 | * space used both by the item structs and the item data | |
3372 | */ | |
5f39d397 | 3373 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3374 | { |
41be1f3b JB |
3375 | struct btrfs_item *start_item; |
3376 | struct btrfs_item *end_item; | |
3377 | struct btrfs_map_token token; | |
be0e5c09 | 3378 | int data_len; |
5f39d397 | 3379 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3380 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3381 | |
3382 | if (!nr) | |
3383 | return 0; | |
41be1f3b | 3384 | btrfs_init_map_token(&token); |
dd3cc16b RK |
3385 | start_item = btrfs_item_nr(start); |
3386 | end_item = btrfs_item_nr(end); | |
41be1f3b JB |
3387 | data_len = btrfs_token_item_offset(l, start_item, &token) + |
3388 | btrfs_token_item_size(l, start_item, &token); | |
3389 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3390 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3391 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3392 | return data_len; |
3393 | } | |
3394 | ||
d4dbff95 CM |
3395 | /* |
3396 | * The space between the end of the leaf items and | |
3397 | * the start of the leaf data. IOW, how much room | |
3398 | * the leaf has left for both items and data | |
3399 | */ | |
d397712b | 3400 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3401 | struct extent_buffer *leaf) |
d4dbff95 | 3402 | { |
5f39d397 CM |
3403 | int nritems = btrfs_header_nritems(leaf); |
3404 | int ret; | |
3405 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3406 | if (ret < 0) { | |
d397712b CM |
3407 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
3408 | "used %d nritems %d\n", | |
ae2f5411 | 3409 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3410 | leaf_space_used(leaf, 0, nritems), nritems); |
3411 | } | |
3412 | return ret; | |
d4dbff95 CM |
3413 | } |
3414 | ||
99d8f83c CM |
3415 | /* |
3416 | * min slot controls the lowest index we're willing to push to the | |
3417 | * right. We'll push up to and including min_slot, but no lower | |
3418 | */ | |
44871b1b CM |
3419 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3420 | struct btrfs_root *root, | |
3421 | struct btrfs_path *path, | |
3422 | int data_size, int empty, | |
3423 | struct extent_buffer *right, | |
99d8f83c CM |
3424 | int free_space, u32 left_nritems, |
3425 | u32 min_slot) | |
00ec4c51 | 3426 | { |
5f39d397 | 3427 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3428 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3429 | struct btrfs_map_token token; |
5f39d397 | 3430 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3431 | int slot; |
34a38218 | 3432 | u32 i; |
00ec4c51 CM |
3433 | int push_space = 0; |
3434 | int push_items = 0; | |
0783fcfc | 3435 | struct btrfs_item *item; |
34a38218 | 3436 | u32 nr; |
7518a238 | 3437 | u32 right_nritems; |
5f39d397 | 3438 | u32 data_end; |
db94535d | 3439 | u32 this_item_size; |
00ec4c51 | 3440 | |
cfed81a0 CM |
3441 | btrfs_init_map_token(&token); |
3442 | ||
34a38218 CM |
3443 | if (empty) |
3444 | nr = 0; | |
3445 | else | |
99d8f83c | 3446 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3447 | |
31840ae1 | 3448 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3449 | push_space += data_size; |
31840ae1 | 3450 | |
44871b1b | 3451 | slot = path->slots[1]; |
34a38218 CM |
3452 | i = left_nritems - 1; |
3453 | while (i >= nr) { | |
dd3cc16b | 3454 | item = btrfs_item_nr(i); |
db94535d | 3455 | |
31840ae1 ZY |
3456 | if (!empty && push_items > 0) { |
3457 | if (path->slots[0] > i) | |
3458 | break; | |
3459 | if (path->slots[0] == i) { | |
3460 | int space = btrfs_leaf_free_space(root, left); | |
3461 | if (space + push_space * 2 > free_space) | |
3462 | break; | |
3463 | } | |
3464 | } | |
3465 | ||
00ec4c51 | 3466 | if (path->slots[0] == i) |
87b29b20 | 3467 | push_space += data_size; |
db94535d | 3468 | |
db94535d CM |
3469 | this_item_size = btrfs_item_size(left, item); |
3470 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3471 | break; |
31840ae1 | 3472 | |
00ec4c51 | 3473 | push_items++; |
db94535d | 3474 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3475 | if (i == 0) |
3476 | break; | |
3477 | i--; | |
db94535d | 3478 | } |
5f39d397 | 3479 | |
925baedd CM |
3480 | if (push_items == 0) |
3481 | goto out_unlock; | |
5f39d397 | 3482 | |
6c1500f2 | 3483 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3484 | |
00ec4c51 | 3485 | /* push left to right */ |
5f39d397 | 3486 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3487 | |
5f39d397 | 3488 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3489 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3490 | |
00ec4c51 | 3491 | /* make room in the right data area */ |
5f39d397 CM |
3492 | data_end = leaf_data_end(root, right); |
3493 | memmove_extent_buffer(right, | |
3494 | btrfs_leaf_data(right) + data_end - push_space, | |
3495 | btrfs_leaf_data(right) + data_end, | |
3496 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3497 | ||
00ec4c51 | 3498 | /* copy from the left data area */ |
5f39d397 | 3499 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3500 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3501 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3502 | push_space); | |
5f39d397 CM |
3503 | |
3504 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3505 | btrfs_item_nr_offset(0), | |
3506 | right_nritems * sizeof(struct btrfs_item)); | |
3507 | ||
00ec4c51 | 3508 | /* copy the items from left to right */ |
5f39d397 CM |
3509 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3510 | btrfs_item_nr_offset(left_nritems - push_items), | |
3511 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3512 | |
3513 | /* update the item pointers */ | |
7518a238 | 3514 | right_nritems += push_items; |
5f39d397 | 3515 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3516 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3517 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3518 | item = btrfs_item_nr(i); |
cfed81a0 CM |
3519 | push_space -= btrfs_token_item_size(right, item, &token); |
3520 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3521 | } |
3522 | ||
7518a238 | 3523 | left_nritems -= push_items; |
5f39d397 | 3524 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3525 | |
34a38218 CM |
3526 | if (left_nritems) |
3527 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3528 | else |
3529 | clean_tree_block(trans, root, left); | |
3530 | ||
5f39d397 | 3531 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3532 | |
5f39d397 CM |
3533 | btrfs_item_key(right, &disk_key, 0); |
3534 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3535 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3536 | |
00ec4c51 | 3537 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3538 | if (path->slots[0] >= left_nritems) { |
3539 | path->slots[0] -= left_nritems; | |
925baedd CM |
3540 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3541 | clean_tree_block(trans, root, path->nodes[0]); | |
3542 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3543 | free_extent_buffer(path->nodes[0]); |
3544 | path->nodes[0] = right; | |
00ec4c51 CM |
3545 | path->slots[1] += 1; |
3546 | } else { | |
925baedd | 3547 | btrfs_tree_unlock(right); |
5f39d397 | 3548 | free_extent_buffer(right); |
00ec4c51 CM |
3549 | } |
3550 | return 0; | |
925baedd CM |
3551 | |
3552 | out_unlock: | |
3553 | btrfs_tree_unlock(right); | |
3554 | free_extent_buffer(right); | |
3555 | return 1; | |
00ec4c51 | 3556 | } |
925baedd | 3557 | |
44871b1b CM |
3558 | /* |
3559 | * push some data in the path leaf to the right, trying to free up at | |
3560 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3561 | * | |
3562 | * returns 1 if the push failed because the other node didn't have enough | |
3563 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3564 | * |
3565 | * this will push starting from min_slot to the end of the leaf. It won't | |
3566 | * push any slot lower than min_slot | |
44871b1b CM |
3567 | */ |
3568 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3569 | *root, struct btrfs_path *path, |
3570 | int min_data_size, int data_size, | |
3571 | int empty, u32 min_slot) | |
44871b1b CM |
3572 | { |
3573 | struct extent_buffer *left = path->nodes[0]; | |
3574 | struct extent_buffer *right; | |
3575 | struct extent_buffer *upper; | |
3576 | int slot; | |
3577 | int free_space; | |
3578 | u32 left_nritems; | |
3579 | int ret; | |
3580 | ||
3581 | if (!path->nodes[1]) | |
3582 | return 1; | |
3583 | ||
3584 | slot = path->slots[1]; | |
3585 | upper = path->nodes[1]; | |
3586 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3587 | return 1; | |
3588 | ||
3589 | btrfs_assert_tree_locked(path->nodes[1]); | |
3590 | ||
3591 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3592 | if (right == NULL) |
3593 | return 1; | |
3594 | ||
44871b1b CM |
3595 | btrfs_tree_lock(right); |
3596 | btrfs_set_lock_blocking(right); | |
3597 | ||
3598 | free_space = btrfs_leaf_free_space(root, right); | |
3599 | if (free_space < data_size) | |
3600 | goto out_unlock; | |
3601 | ||
3602 | /* cow and double check */ | |
3603 | ret = btrfs_cow_block(trans, root, right, upper, | |
3604 | slot + 1, &right); | |
3605 | if (ret) | |
3606 | goto out_unlock; | |
3607 | ||
3608 | free_space = btrfs_leaf_free_space(root, right); | |
3609 | if (free_space < data_size) | |
3610 | goto out_unlock; | |
3611 | ||
3612 | left_nritems = btrfs_header_nritems(left); | |
3613 | if (left_nritems == 0) | |
3614 | goto out_unlock; | |
3615 | ||
99d8f83c CM |
3616 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3617 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3618 | out_unlock: |
3619 | btrfs_tree_unlock(right); | |
3620 | free_extent_buffer(right); | |
3621 | return 1; | |
3622 | } | |
3623 | ||
74123bd7 CM |
3624 | /* |
3625 | * push some data in the path leaf to the left, trying to free up at | |
3626 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3627 | * |
3628 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3629 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3630 | * items | |
74123bd7 | 3631 | */ |
44871b1b CM |
3632 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3633 | struct btrfs_root *root, | |
3634 | struct btrfs_path *path, int data_size, | |
3635 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3636 | int free_space, u32 right_nritems, |
3637 | u32 max_slot) | |
be0e5c09 | 3638 | { |
5f39d397 CM |
3639 | struct btrfs_disk_key disk_key; |
3640 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3641 | int i; |
be0e5c09 CM |
3642 | int push_space = 0; |
3643 | int push_items = 0; | |
0783fcfc | 3644 | struct btrfs_item *item; |
7518a238 | 3645 | u32 old_left_nritems; |
34a38218 | 3646 | u32 nr; |
aa5d6bed | 3647 | int ret = 0; |
db94535d CM |
3648 | u32 this_item_size; |
3649 | u32 old_left_item_size; | |
cfed81a0 CM |
3650 | struct btrfs_map_token token; |
3651 | ||
3652 | btrfs_init_map_token(&token); | |
be0e5c09 | 3653 | |
34a38218 | 3654 | if (empty) |
99d8f83c | 3655 | nr = min(right_nritems, max_slot); |
34a38218 | 3656 | else |
99d8f83c | 3657 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3658 | |
3659 | for (i = 0; i < nr; i++) { | |
dd3cc16b | 3660 | item = btrfs_item_nr(i); |
db94535d | 3661 | |
31840ae1 ZY |
3662 | if (!empty && push_items > 0) { |
3663 | if (path->slots[0] < i) | |
3664 | break; | |
3665 | if (path->slots[0] == i) { | |
3666 | int space = btrfs_leaf_free_space(root, right); | |
3667 | if (space + push_space * 2 > free_space) | |
3668 | break; | |
3669 | } | |
3670 | } | |
3671 | ||
be0e5c09 | 3672 | if (path->slots[0] == i) |
87b29b20 | 3673 | push_space += data_size; |
db94535d CM |
3674 | |
3675 | this_item_size = btrfs_item_size(right, item); | |
3676 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3677 | break; |
db94535d | 3678 | |
be0e5c09 | 3679 | push_items++; |
db94535d CM |
3680 | push_space += this_item_size + sizeof(*item); |
3681 | } | |
3682 | ||
be0e5c09 | 3683 | if (push_items == 0) { |
925baedd CM |
3684 | ret = 1; |
3685 | goto out; | |
be0e5c09 | 3686 | } |
fae7f21c | 3687 | WARN_ON(!empty && push_items == btrfs_header_nritems(right)); |
5f39d397 | 3688 | |
be0e5c09 | 3689 | /* push data from right to left */ |
5f39d397 CM |
3690 | copy_extent_buffer(left, right, |
3691 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3692 | btrfs_item_nr_offset(0), | |
3693 | push_items * sizeof(struct btrfs_item)); | |
3694 | ||
123abc88 | 3695 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3696 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3697 | |
3698 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3699 | leaf_data_end(root, left) - push_space, |
3700 | btrfs_leaf_data(right) + | |
5f39d397 | 3701 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3702 | push_space); |
5f39d397 | 3703 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3704 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3705 | |
db94535d | 3706 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3707 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3708 | u32 ioff; |
db94535d | 3709 | |
dd3cc16b | 3710 | item = btrfs_item_nr(i); |
db94535d | 3711 | |
cfed81a0 CM |
3712 | ioff = btrfs_token_item_offset(left, item, &token); |
3713 | btrfs_set_token_item_offset(left, item, | |
3714 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3715 | &token); | |
be0e5c09 | 3716 | } |
5f39d397 | 3717 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3718 | |
3719 | /* fixup right node */ | |
31b1a2bd JL |
3720 | if (push_items > right_nritems) |
3721 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3722 | right_nritems); |
34a38218 CM |
3723 | |
3724 | if (push_items < right_nritems) { | |
3725 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3726 | leaf_data_end(root, right); | |
3727 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3728 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3729 | btrfs_leaf_data(right) + | |
3730 | leaf_data_end(root, right), push_space); | |
3731 | ||
3732 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3733 | btrfs_item_nr_offset(push_items), |
3734 | (btrfs_header_nritems(right) - push_items) * | |
3735 | sizeof(struct btrfs_item)); | |
34a38218 | 3736 | } |
eef1c494 Y |
3737 | right_nritems -= push_items; |
3738 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3739 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 3740 | for (i = 0; i < right_nritems; i++) { |
dd3cc16b | 3741 | item = btrfs_item_nr(i); |
db94535d | 3742 | |
cfed81a0 CM |
3743 | push_space = push_space - btrfs_token_item_size(right, |
3744 | item, &token); | |
3745 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3746 | } |
eb60ceac | 3747 | |
5f39d397 | 3748 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3749 | if (right_nritems) |
3750 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3751 | else |
3752 | clean_tree_block(trans, root, right); | |
098f59c2 | 3753 | |
5f39d397 | 3754 | btrfs_item_key(right, &disk_key, 0); |
d6a0a126 | 3755 | fixup_low_keys(root, path, &disk_key, 1); |
be0e5c09 CM |
3756 | |
3757 | /* then fixup the leaf pointer in the path */ | |
3758 | if (path->slots[0] < push_items) { | |
3759 | path->slots[0] += old_left_nritems; | |
925baedd | 3760 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3761 | free_extent_buffer(path->nodes[0]); |
3762 | path->nodes[0] = left; | |
be0e5c09 CM |
3763 | path->slots[1] -= 1; |
3764 | } else { | |
925baedd | 3765 | btrfs_tree_unlock(left); |
5f39d397 | 3766 | free_extent_buffer(left); |
be0e5c09 CM |
3767 | path->slots[0] -= push_items; |
3768 | } | |
eb60ceac | 3769 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3770 | return ret; |
925baedd CM |
3771 | out: |
3772 | btrfs_tree_unlock(left); | |
3773 | free_extent_buffer(left); | |
3774 | return ret; | |
be0e5c09 CM |
3775 | } |
3776 | ||
44871b1b CM |
3777 | /* |
3778 | * push some data in the path leaf to the left, trying to free up at | |
3779 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3780 | * |
3781 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3782 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3783 | * items | |
44871b1b CM |
3784 | */ |
3785 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3786 | *root, struct btrfs_path *path, int min_data_size, |
3787 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3788 | { |
3789 | struct extent_buffer *right = path->nodes[0]; | |
3790 | struct extent_buffer *left; | |
3791 | int slot; | |
3792 | int free_space; | |
3793 | u32 right_nritems; | |
3794 | int ret = 0; | |
3795 | ||
3796 | slot = path->slots[1]; | |
3797 | if (slot == 0) | |
3798 | return 1; | |
3799 | if (!path->nodes[1]) | |
3800 | return 1; | |
3801 | ||
3802 | right_nritems = btrfs_header_nritems(right); | |
3803 | if (right_nritems == 0) | |
3804 | return 1; | |
3805 | ||
3806 | btrfs_assert_tree_locked(path->nodes[1]); | |
3807 | ||
3808 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3809 | if (left == NULL) |
3810 | return 1; | |
3811 | ||
44871b1b CM |
3812 | btrfs_tree_lock(left); |
3813 | btrfs_set_lock_blocking(left); | |
3814 | ||
3815 | free_space = btrfs_leaf_free_space(root, left); | |
3816 | if (free_space < data_size) { | |
3817 | ret = 1; | |
3818 | goto out; | |
3819 | } | |
3820 | ||
3821 | /* cow and double check */ | |
3822 | ret = btrfs_cow_block(trans, root, left, | |
3823 | path->nodes[1], slot - 1, &left); | |
3824 | if (ret) { | |
3825 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3826 | if (ret == -ENOSPC) |
3827 | ret = 1; | |
44871b1b CM |
3828 | goto out; |
3829 | } | |
3830 | ||
3831 | free_space = btrfs_leaf_free_space(root, left); | |
3832 | if (free_space < data_size) { | |
3833 | ret = 1; | |
3834 | goto out; | |
3835 | } | |
3836 | ||
99d8f83c CM |
3837 | return __push_leaf_left(trans, root, path, min_data_size, |
3838 | empty, left, free_space, right_nritems, | |
3839 | max_slot); | |
44871b1b CM |
3840 | out: |
3841 | btrfs_tree_unlock(left); | |
3842 | free_extent_buffer(left); | |
3843 | return ret; | |
3844 | } | |
3845 | ||
3846 | /* | |
3847 | * split the path's leaf in two, making sure there is at least data_size | |
3848 | * available for the resulting leaf level of the path. | |
44871b1b | 3849 | */ |
143bede5 JM |
3850 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3851 | struct btrfs_root *root, | |
3852 | struct btrfs_path *path, | |
3853 | struct extent_buffer *l, | |
3854 | struct extent_buffer *right, | |
3855 | int slot, int mid, int nritems) | |
44871b1b CM |
3856 | { |
3857 | int data_copy_size; | |
3858 | int rt_data_off; | |
3859 | int i; | |
44871b1b | 3860 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3861 | struct btrfs_map_token token; |
3862 | ||
3863 | btrfs_init_map_token(&token); | |
44871b1b CM |
3864 | |
3865 | nritems = nritems - mid; | |
3866 | btrfs_set_header_nritems(right, nritems); | |
3867 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3868 | ||
3869 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3870 | btrfs_item_nr_offset(mid), | |
3871 | nritems * sizeof(struct btrfs_item)); | |
3872 | ||
3873 | copy_extent_buffer(right, l, | |
3874 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3875 | data_copy_size, btrfs_leaf_data(l) + | |
3876 | leaf_data_end(root, l), data_copy_size); | |
3877 | ||
3878 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3879 | btrfs_item_end_nr(l, mid); | |
3880 | ||
3881 | for (i = 0; i < nritems; i++) { | |
dd3cc16b | 3882 | struct btrfs_item *item = btrfs_item_nr(i); |
44871b1b CM |
3883 | u32 ioff; |
3884 | ||
cfed81a0 CM |
3885 | ioff = btrfs_token_item_offset(right, item, &token); |
3886 | btrfs_set_token_item_offset(right, item, | |
3887 | ioff + rt_data_off, &token); | |
44871b1b CM |
3888 | } |
3889 | ||
44871b1b | 3890 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3891 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3892 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 3893 | path->slots[1] + 1, 1); |
44871b1b CM |
3894 | |
3895 | btrfs_mark_buffer_dirty(right); | |
3896 | btrfs_mark_buffer_dirty(l); | |
3897 | BUG_ON(path->slots[0] != slot); | |
3898 | ||
44871b1b CM |
3899 | if (mid <= slot) { |
3900 | btrfs_tree_unlock(path->nodes[0]); | |
3901 | free_extent_buffer(path->nodes[0]); | |
3902 | path->nodes[0] = right; | |
3903 | path->slots[0] -= mid; | |
3904 | path->slots[1] += 1; | |
3905 | } else { | |
3906 | btrfs_tree_unlock(right); | |
3907 | free_extent_buffer(right); | |
3908 | } | |
3909 | ||
3910 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3911 | } |
3912 | ||
99d8f83c CM |
3913 | /* |
3914 | * double splits happen when we need to insert a big item in the middle | |
3915 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3916 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3917 | * A B C | |
3918 | * | |
3919 | * We avoid this by trying to push the items on either side of our target | |
3920 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3921 | * completely. | |
3922 | */ | |
3923 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3924 | struct btrfs_root *root, | |
3925 | struct btrfs_path *path, | |
3926 | int data_size) | |
3927 | { | |
3928 | int ret; | |
3929 | int progress = 0; | |
3930 | int slot; | |
3931 | u32 nritems; | |
5a4267ca | 3932 | int space_needed = data_size; |
99d8f83c CM |
3933 | |
3934 | slot = path->slots[0]; | |
5a4267ca FDBM |
3935 | if (slot < btrfs_header_nritems(path->nodes[0])) |
3936 | space_needed -= btrfs_leaf_free_space(root, path->nodes[0]); | |
99d8f83c CM |
3937 | |
3938 | /* | |
3939 | * try to push all the items after our slot into the | |
3940 | * right leaf | |
3941 | */ | |
5a4267ca | 3942 | ret = push_leaf_right(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3943 | if (ret < 0) |
3944 | return ret; | |
3945 | ||
3946 | if (ret == 0) | |
3947 | progress++; | |
3948 | ||
3949 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3950 | /* | |
3951 | * our goal is to get our slot at the start or end of a leaf. If | |
3952 | * we've done so we're done | |
3953 | */ | |
3954 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3955 | return 0; | |
3956 | ||
3957 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3958 | return 0; | |
3959 | ||
3960 | /* try to push all the items before our slot into the next leaf */ | |
3961 | slot = path->slots[0]; | |
5a4267ca | 3962 | ret = push_leaf_left(trans, root, path, 1, space_needed, 0, slot); |
99d8f83c CM |
3963 | if (ret < 0) |
3964 | return ret; | |
3965 | ||
3966 | if (ret == 0) | |
3967 | progress++; | |
3968 | ||
3969 | if (progress) | |
3970 | return 0; | |
3971 | return 1; | |
3972 | } | |
3973 | ||
74123bd7 CM |
3974 | /* |
3975 | * split the path's leaf in two, making sure there is at least data_size | |
3976 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3977 | * |
3978 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3979 | */ |
e02119d5 CM |
3980 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3981 | struct btrfs_root *root, | |
3982 | struct btrfs_key *ins_key, | |
3983 | struct btrfs_path *path, int data_size, | |
3984 | int extend) | |
be0e5c09 | 3985 | { |
5d4f98a2 | 3986 | struct btrfs_disk_key disk_key; |
5f39d397 | 3987 | struct extent_buffer *l; |
7518a238 | 3988 | u32 nritems; |
eb60ceac CM |
3989 | int mid; |
3990 | int slot; | |
5f39d397 | 3991 | struct extent_buffer *right; |
d4dbff95 | 3992 | int ret = 0; |
aa5d6bed | 3993 | int wret; |
5d4f98a2 | 3994 | int split; |
cc0c5538 | 3995 | int num_doubles = 0; |
99d8f83c | 3996 | int tried_avoid_double = 0; |
aa5d6bed | 3997 | |
a5719521 YZ |
3998 | l = path->nodes[0]; |
3999 | slot = path->slots[0]; | |
4000 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
4001 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
4002 | return -EOVERFLOW; | |
4003 | ||
40689478 | 4004 | /* first try to make some room by pushing left and right */ |
33157e05 | 4005 | if (data_size && path->nodes[1]) { |
5a4267ca FDBM |
4006 | int space_needed = data_size; |
4007 | ||
4008 | if (slot < btrfs_header_nritems(l)) | |
4009 | space_needed -= btrfs_leaf_free_space(root, l); | |
4010 | ||
4011 | wret = push_leaf_right(trans, root, path, space_needed, | |
4012 | space_needed, 0, 0); | |
d397712b | 4013 | if (wret < 0) |
eaee50e8 | 4014 | return wret; |
3685f791 | 4015 | if (wret) { |
5a4267ca FDBM |
4016 | wret = push_leaf_left(trans, root, path, space_needed, |
4017 | space_needed, 0, (u32)-1); | |
3685f791 CM |
4018 | if (wret < 0) |
4019 | return wret; | |
4020 | } | |
4021 | l = path->nodes[0]; | |
aa5d6bed | 4022 | |
3685f791 | 4023 | /* did the pushes work? */ |
87b29b20 | 4024 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 4025 | return 0; |
3326d1b0 | 4026 | } |
aa5d6bed | 4027 | |
5c680ed6 | 4028 | if (!path->nodes[1]) { |
fdd99c72 | 4029 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
4030 | if (ret) |
4031 | return ret; | |
4032 | } | |
cc0c5538 | 4033 | again: |
5d4f98a2 | 4034 | split = 1; |
cc0c5538 | 4035 | l = path->nodes[0]; |
eb60ceac | 4036 | slot = path->slots[0]; |
5f39d397 | 4037 | nritems = btrfs_header_nritems(l); |
d397712b | 4038 | mid = (nritems + 1) / 2; |
54aa1f4d | 4039 | |
5d4f98a2 YZ |
4040 | if (mid <= slot) { |
4041 | if (nritems == 1 || | |
4042 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
4043 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4044 | if (slot >= nritems) { | |
4045 | split = 0; | |
4046 | } else { | |
4047 | mid = slot; | |
4048 | if (mid != nritems && | |
4049 | leaf_space_used(l, mid, nritems - mid) + | |
4050 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4051 | if (data_size && !tried_avoid_double) |
4052 | goto push_for_double; | |
5d4f98a2 YZ |
4053 | split = 2; |
4054 | } | |
4055 | } | |
4056 | } | |
4057 | } else { | |
4058 | if (leaf_space_used(l, 0, mid) + data_size > | |
4059 | BTRFS_LEAF_DATA_SIZE(root)) { | |
4060 | if (!extend && data_size && slot == 0) { | |
4061 | split = 0; | |
4062 | } else if ((extend || !data_size) && slot == 0) { | |
4063 | mid = 1; | |
4064 | } else { | |
4065 | mid = slot; | |
4066 | if (mid != nritems && | |
4067 | leaf_space_used(l, mid, nritems - mid) + | |
4068 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
4069 | if (data_size && !tried_avoid_double) |
4070 | goto push_for_double; | |
67871254 | 4071 | split = 2; |
5d4f98a2 YZ |
4072 | } |
4073 | } | |
4074 | } | |
4075 | } | |
4076 | ||
4077 | if (split == 0) | |
4078 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
4079 | else | |
4080 | btrfs_item_key(l, &disk_key, mid); | |
4081 | ||
4082 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 4083 | root->root_key.objectid, |
5581a51a | 4084 | &disk_key, 0, l->start, 0); |
f0486c68 | 4085 | if (IS_ERR(right)) |
5f39d397 | 4086 | return PTR_ERR(right); |
f0486c68 YZ |
4087 | |
4088 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4089 | |
4090 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4091 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4092 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4093 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4094 | btrfs_set_header_owner(right, root->root_key.objectid); |
4095 | btrfs_set_header_level(right, 0); | |
4096 | write_extent_buffer(right, root->fs_info->fsid, | |
0a4e5586 | 4097 | btrfs_header_fsid(), BTRFS_FSID_SIZE); |
e17cade2 CM |
4098 | |
4099 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
b308bc2f | 4100 | btrfs_header_chunk_tree_uuid(right), |
e17cade2 | 4101 | BTRFS_UUID_SIZE); |
44871b1b | 4102 | |
5d4f98a2 YZ |
4103 | if (split == 0) { |
4104 | if (mid <= slot) { | |
4105 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4106 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4107 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4108 | btrfs_tree_unlock(path->nodes[0]); |
4109 | free_extent_buffer(path->nodes[0]); | |
4110 | path->nodes[0] = right; | |
4111 | path->slots[0] = 0; | |
4112 | path->slots[1] += 1; | |
4113 | } else { | |
4114 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4115 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4116 | path->slots[1], 1); |
5d4f98a2 YZ |
4117 | btrfs_tree_unlock(path->nodes[0]); |
4118 | free_extent_buffer(path->nodes[0]); | |
4119 | path->nodes[0] = right; | |
4120 | path->slots[0] = 0; | |
143bede5 | 4121 | if (path->slots[1] == 0) |
d6a0a126 | 4122 | fixup_low_keys(root, path, &disk_key, 1); |
d4dbff95 | 4123 | } |
5d4f98a2 YZ |
4124 | btrfs_mark_buffer_dirty(right); |
4125 | return ret; | |
d4dbff95 | 4126 | } |
74123bd7 | 4127 | |
143bede5 | 4128 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4129 | |
5d4f98a2 | 4130 | if (split == 2) { |
cc0c5538 CM |
4131 | BUG_ON(num_doubles != 0); |
4132 | num_doubles++; | |
4133 | goto again; | |
a429e513 | 4134 | } |
44871b1b | 4135 | |
143bede5 | 4136 | return 0; |
99d8f83c CM |
4137 | |
4138 | push_for_double: | |
4139 | push_for_double_split(trans, root, path, data_size); | |
4140 | tried_avoid_double = 1; | |
4141 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4142 | return 0; | |
4143 | goto again; | |
be0e5c09 CM |
4144 | } |
4145 | ||
ad48fd75 YZ |
4146 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4147 | struct btrfs_root *root, | |
4148 | struct btrfs_path *path, int ins_len) | |
459931ec | 4149 | { |
ad48fd75 | 4150 | struct btrfs_key key; |
459931ec | 4151 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4152 | struct btrfs_file_extent_item *fi; |
4153 | u64 extent_len = 0; | |
4154 | u32 item_size; | |
4155 | int ret; | |
459931ec CM |
4156 | |
4157 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4158 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4159 | ||
4160 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4161 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4162 | ||
4163 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4164 | return 0; | |
459931ec CM |
4165 | |
4166 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4167 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4168 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4169 | struct btrfs_file_extent_item); | |
4170 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4171 | } | |
b3b4aa74 | 4172 | btrfs_release_path(path); |
459931ec | 4173 | |
459931ec | 4174 | path->keep_locks = 1; |
ad48fd75 YZ |
4175 | path->search_for_split = 1; |
4176 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4177 | path->search_for_split = 0; |
ad48fd75 YZ |
4178 | if (ret < 0) |
4179 | goto err; | |
459931ec | 4180 | |
ad48fd75 YZ |
4181 | ret = -EAGAIN; |
4182 | leaf = path->nodes[0]; | |
459931ec | 4183 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4184 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4185 | goto err; | |
4186 | ||
109f6aef CM |
4187 | /* the leaf has changed, it now has room. return now */ |
4188 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4189 | goto err; | |
4190 | ||
ad48fd75 YZ |
4191 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4192 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4193 | struct btrfs_file_extent_item); | |
4194 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4195 | goto err; | |
459931ec CM |
4196 | } |
4197 | ||
b9473439 | 4198 | btrfs_set_path_blocking(path); |
ad48fd75 | 4199 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4200 | if (ret) |
4201 | goto err; | |
459931ec | 4202 | |
ad48fd75 | 4203 | path->keep_locks = 0; |
b9473439 | 4204 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4205 | return 0; |
4206 | err: | |
4207 | path->keep_locks = 0; | |
4208 | return ret; | |
4209 | } | |
4210 | ||
4211 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4212 | struct btrfs_root *root, | |
4213 | struct btrfs_path *path, | |
4214 | struct btrfs_key *new_key, | |
4215 | unsigned long split_offset) | |
4216 | { | |
4217 | struct extent_buffer *leaf; | |
4218 | struct btrfs_item *item; | |
4219 | struct btrfs_item *new_item; | |
4220 | int slot; | |
4221 | char *buf; | |
4222 | u32 nritems; | |
4223 | u32 item_size; | |
4224 | u32 orig_offset; | |
4225 | struct btrfs_disk_key disk_key; | |
4226 | ||
b9473439 CM |
4227 | leaf = path->nodes[0]; |
4228 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4229 | ||
b4ce94de CM |
4230 | btrfs_set_path_blocking(path); |
4231 | ||
dd3cc16b | 4232 | item = btrfs_item_nr(path->slots[0]); |
459931ec CM |
4233 | orig_offset = btrfs_item_offset(leaf, item); |
4234 | item_size = btrfs_item_size(leaf, item); | |
4235 | ||
459931ec | 4236 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4237 | if (!buf) |
4238 | return -ENOMEM; | |
4239 | ||
459931ec CM |
4240 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4241 | path->slots[0]), item_size); | |
459931ec | 4242 | |
ad48fd75 | 4243 | slot = path->slots[0] + 1; |
459931ec | 4244 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4245 | if (slot != nritems) { |
4246 | /* shift the items */ | |
4247 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4248 | btrfs_item_nr_offset(slot), |
4249 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4250 | } |
4251 | ||
4252 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4253 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4254 | ||
dd3cc16b | 4255 | new_item = btrfs_item_nr(slot); |
459931ec CM |
4256 | |
4257 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4258 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4259 | ||
4260 | btrfs_set_item_offset(leaf, item, | |
4261 | orig_offset + item_size - split_offset); | |
4262 | btrfs_set_item_size(leaf, item, split_offset); | |
4263 | ||
4264 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4265 | ||
4266 | /* write the data for the start of the original item */ | |
4267 | write_extent_buffer(leaf, buf, | |
4268 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4269 | split_offset); | |
4270 | ||
4271 | /* write the data for the new item */ | |
4272 | write_extent_buffer(leaf, buf + split_offset, | |
4273 | btrfs_item_ptr_offset(leaf, slot), | |
4274 | item_size - split_offset); | |
4275 | btrfs_mark_buffer_dirty(leaf); | |
4276 | ||
ad48fd75 | 4277 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4278 | kfree(buf); |
ad48fd75 YZ |
4279 | return 0; |
4280 | } | |
4281 | ||
4282 | /* | |
4283 | * This function splits a single item into two items, | |
4284 | * giving 'new_key' to the new item and splitting the | |
4285 | * old one at split_offset (from the start of the item). | |
4286 | * | |
4287 | * The path may be released by this operation. After | |
4288 | * the split, the path is pointing to the old item. The | |
4289 | * new item is going to be in the same node as the old one. | |
4290 | * | |
4291 | * Note, the item being split must be smaller enough to live alone on | |
4292 | * a tree block with room for one extra struct btrfs_item | |
4293 | * | |
4294 | * This allows us to split the item in place, keeping a lock on the | |
4295 | * leaf the entire time. | |
4296 | */ | |
4297 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4298 | struct btrfs_root *root, | |
4299 | struct btrfs_path *path, | |
4300 | struct btrfs_key *new_key, | |
4301 | unsigned long split_offset) | |
4302 | { | |
4303 | int ret; | |
4304 | ret = setup_leaf_for_split(trans, root, path, | |
4305 | sizeof(struct btrfs_item)); | |
4306 | if (ret) | |
4307 | return ret; | |
4308 | ||
4309 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4310 | return ret; |
4311 | } | |
4312 | ||
ad48fd75 YZ |
4313 | /* |
4314 | * This function duplicate a item, giving 'new_key' to the new item. | |
4315 | * It guarantees both items live in the same tree leaf and the new item | |
4316 | * is contiguous with the original item. | |
4317 | * | |
4318 | * This allows us to split file extent in place, keeping a lock on the | |
4319 | * leaf the entire time. | |
4320 | */ | |
4321 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4322 | struct btrfs_root *root, | |
4323 | struct btrfs_path *path, | |
4324 | struct btrfs_key *new_key) | |
4325 | { | |
4326 | struct extent_buffer *leaf; | |
4327 | int ret; | |
4328 | u32 item_size; | |
4329 | ||
4330 | leaf = path->nodes[0]; | |
4331 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4332 | ret = setup_leaf_for_split(trans, root, path, | |
4333 | item_size + sizeof(struct btrfs_item)); | |
4334 | if (ret) | |
4335 | return ret; | |
4336 | ||
4337 | path->slots[0]++; | |
afe5fea7 | 4338 | setup_items_for_insert(root, path, new_key, &item_size, |
143bede5 JM |
4339 | item_size, item_size + |
4340 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4341 | leaf = path->nodes[0]; |
4342 | memcpy_extent_buffer(leaf, | |
4343 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4344 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4345 | item_size); | |
4346 | return 0; | |
4347 | } | |
4348 | ||
d352ac68 CM |
4349 | /* |
4350 | * make the item pointed to by the path smaller. new_size indicates | |
4351 | * how small to make it, and from_end tells us if we just chop bytes | |
4352 | * off the end of the item or if we shift the item to chop bytes off | |
4353 | * the front. | |
4354 | */ | |
afe5fea7 | 4355 | void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4356 | u32 new_size, int from_end) |
b18c6685 | 4357 | { |
b18c6685 | 4358 | int slot; |
5f39d397 CM |
4359 | struct extent_buffer *leaf; |
4360 | struct btrfs_item *item; | |
b18c6685 CM |
4361 | u32 nritems; |
4362 | unsigned int data_end; | |
4363 | unsigned int old_data_start; | |
4364 | unsigned int old_size; | |
4365 | unsigned int size_diff; | |
4366 | int i; | |
cfed81a0 CM |
4367 | struct btrfs_map_token token; |
4368 | ||
4369 | btrfs_init_map_token(&token); | |
b18c6685 | 4370 | |
5f39d397 | 4371 | leaf = path->nodes[0]; |
179e29e4 CM |
4372 | slot = path->slots[0]; |
4373 | ||
4374 | old_size = btrfs_item_size_nr(leaf, slot); | |
4375 | if (old_size == new_size) | |
143bede5 | 4376 | return; |
b18c6685 | 4377 | |
5f39d397 | 4378 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4379 | data_end = leaf_data_end(root, leaf); |
4380 | ||
5f39d397 | 4381 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4382 | |
b18c6685 CM |
4383 | size_diff = old_size - new_size; |
4384 | ||
4385 | BUG_ON(slot < 0); | |
4386 | BUG_ON(slot >= nritems); | |
4387 | ||
4388 | /* | |
4389 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4390 | */ | |
4391 | /* first correct the data pointers */ | |
4392 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4393 | u32 ioff; |
dd3cc16b | 4394 | item = btrfs_item_nr(i); |
db94535d | 4395 | |
cfed81a0 CM |
4396 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4397 | btrfs_set_token_item_offset(leaf, item, | |
4398 | ioff + size_diff, &token); | |
b18c6685 | 4399 | } |
db94535d | 4400 | |
b18c6685 | 4401 | /* shift the data */ |
179e29e4 CM |
4402 | if (from_end) { |
4403 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4404 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4405 | data_end, old_data_start + new_size - data_end); | |
4406 | } else { | |
4407 | struct btrfs_disk_key disk_key; | |
4408 | u64 offset; | |
4409 | ||
4410 | btrfs_item_key(leaf, &disk_key, slot); | |
4411 | ||
4412 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4413 | unsigned long ptr; | |
4414 | struct btrfs_file_extent_item *fi; | |
4415 | ||
4416 | fi = btrfs_item_ptr(leaf, slot, | |
4417 | struct btrfs_file_extent_item); | |
4418 | fi = (struct btrfs_file_extent_item *)( | |
4419 | (unsigned long)fi - size_diff); | |
4420 | ||
4421 | if (btrfs_file_extent_type(leaf, fi) == | |
4422 | BTRFS_FILE_EXTENT_INLINE) { | |
4423 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4424 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4425 | (unsigned long)fi, |
4426 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4427 | disk_bytenr)); |
4428 | } | |
4429 | } | |
4430 | ||
4431 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4432 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4433 | data_end, old_data_start - data_end); | |
4434 | ||
4435 | offset = btrfs_disk_key_offset(&disk_key); | |
4436 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4437 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4438 | if (slot == 0) | |
d6a0a126 | 4439 | fixup_low_keys(root, path, &disk_key, 1); |
179e29e4 | 4440 | } |
5f39d397 | 4441 | |
dd3cc16b | 4442 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4443 | btrfs_set_item_size(leaf, item, new_size); |
4444 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4445 | |
5f39d397 CM |
4446 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4447 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4448 | BUG(); |
5f39d397 | 4449 | } |
b18c6685 CM |
4450 | } |
4451 | ||
d352ac68 | 4452 | /* |
8f69dbd2 | 4453 | * make the item pointed to by the path bigger, data_size is the added size. |
d352ac68 | 4454 | */ |
4b90c680 | 4455 | void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 | 4456 | u32 data_size) |
6567e837 | 4457 | { |
6567e837 | 4458 | int slot; |
5f39d397 CM |
4459 | struct extent_buffer *leaf; |
4460 | struct btrfs_item *item; | |
6567e837 CM |
4461 | u32 nritems; |
4462 | unsigned int data_end; | |
4463 | unsigned int old_data; | |
4464 | unsigned int old_size; | |
4465 | int i; | |
cfed81a0 CM |
4466 | struct btrfs_map_token token; |
4467 | ||
4468 | btrfs_init_map_token(&token); | |
6567e837 | 4469 | |
5f39d397 | 4470 | leaf = path->nodes[0]; |
6567e837 | 4471 | |
5f39d397 | 4472 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4473 | data_end = leaf_data_end(root, leaf); |
4474 | ||
5f39d397 CM |
4475 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4476 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4477 | BUG(); |
5f39d397 | 4478 | } |
6567e837 | 4479 | slot = path->slots[0]; |
5f39d397 | 4480 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4481 | |
4482 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4483 | if (slot >= nritems) { |
4484 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
4485 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
4486 | slot, nritems); | |
3326d1b0 CM |
4487 | BUG_ON(1); |
4488 | } | |
6567e837 CM |
4489 | |
4490 | /* | |
4491 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4492 | */ | |
4493 | /* first correct the data pointers */ | |
4494 | for (i = slot; i < nritems; i++) { | |
5f39d397 | 4495 | u32 ioff; |
dd3cc16b | 4496 | item = btrfs_item_nr(i); |
db94535d | 4497 | |
cfed81a0 CM |
4498 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4499 | btrfs_set_token_item_offset(leaf, item, | |
4500 | ioff - data_size, &token); | |
6567e837 | 4501 | } |
5f39d397 | 4502 | |
6567e837 | 4503 | /* shift the data */ |
5f39d397 | 4504 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4505 | data_end - data_size, btrfs_leaf_data(leaf) + |
4506 | data_end, old_data - data_end); | |
5f39d397 | 4507 | |
6567e837 | 4508 | data_end = old_data; |
5f39d397 | 4509 | old_size = btrfs_item_size_nr(leaf, slot); |
dd3cc16b | 4510 | item = btrfs_item_nr(slot); |
5f39d397 CM |
4511 | btrfs_set_item_size(leaf, item, old_size + data_size); |
4512 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4513 | |
5f39d397 CM |
4514 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4515 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4516 | BUG(); |
5f39d397 | 4517 | } |
6567e837 CM |
4518 | } |
4519 | ||
74123bd7 | 4520 | /* |
44871b1b CM |
4521 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4522 | * to save stack depth by doing the bulk of the work in a function | |
4523 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4524 | */ |
afe5fea7 | 4525 | void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path, |
143bede5 JM |
4526 | struct btrfs_key *cpu_key, u32 *data_size, |
4527 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4528 | { |
5f39d397 | 4529 | struct btrfs_item *item; |
9c58309d | 4530 | int i; |
7518a238 | 4531 | u32 nritems; |
be0e5c09 | 4532 | unsigned int data_end; |
e2fa7227 | 4533 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4534 | struct extent_buffer *leaf; |
4535 | int slot; | |
cfed81a0 CM |
4536 | struct btrfs_map_token token; |
4537 | ||
4538 | btrfs_init_map_token(&token); | |
e2fa7227 | 4539 | |
5f39d397 | 4540 | leaf = path->nodes[0]; |
44871b1b | 4541 | slot = path->slots[0]; |
74123bd7 | 4542 | |
5f39d397 | 4543 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4544 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4545 | |
f25956cc | 4546 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4547 | btrfs_print_leaf(root, leaf); |
d397712b | 4548 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4549 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4550 | BUG(); |
d4dbff95 | 4551 | } |
5f39d397 | 4552 | |
be0e5c09 | 4553 | if (slot != nritems) { |
5f39d397 | 4554 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4555 | |
5f39d397 CM |
4556 | if (old_data < data_end) { |
4557 | btrfs_print_leaf(root, leaf); | |
d397712b | 4558 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4559 | slot, old_data, data_end); |
4560 | BUG_ON(1); | |
4561 | } | |
be0e5c09 CM |
4562 | /* |
4563 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4564 | */ | |
4565 | /* first correct the data pointers */ | |
0783fcfc | 4566 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4567 | u32 ioff; |
db94535d | 4568 | |
dd3cc16b | 4569 | item = btrfs_item_nr( i); |
cfed81a0 CM |
4570 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4571 | btrfs_set_token_item_offset(leaf, item, | |
4572 | ioff - total_data, &token); | |
0783fcfc | 4573 | } |
be0e5c09 | 4574 | /* shift the items */ |
9c58309d | 4575 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4576 | btrfs_item_nr_offset(slot), |
d6025579 | 4577 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4578 | |
4579 | /* shift the data */ | |
5f39d397 | 4580 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4581 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4582 | data_end, old_data - data_end); |
be0e5c09 CM |
4583 | data_end = old_data; |
4584 | } | |
5f39d397 | 4585 | |
62e2749e | 4586 | /* setup the item for the new data */ |
9c58309d CM |
4587 | for (i = 0; i < nr; i++) { |
4588 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4589 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
dd3cc16b | 4590 | item = btrfs_item_nr(slot + i); |
cfed81a0 CM |
4591 | btrfs_set_token_item_offset(leaf, item, |
4592 | data_end - data_size[i], &token); | |
9c58309d | 4593 | data_end -= data_size[i]; |
cfed81a0 | 4594 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4595 | } |
44871b1b | 4596 | |
9c58309d | 4597 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4598 | |
5a01a2e3 CM |
4599 | if (slot == 0) { |
4600 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
d6a0a126 | 4601 | fixup_low_keys(root, path, &disk_key, 1); |
5a01a2e3 | 4602 | } |
b9473439 CM |
4603 | btrfs_unlock_up_safe(path, 1); |
4604 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4605 | |
5f39d397 CM |
4606 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4607 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4608 | BUG(); |
5f39d397 | 4609 | } |
44871b1b CM |
4610 | } |
4611 | ||
4612 | /* | |
4613 | * Given a key and some data, insert items into the tree. | |
4614 | * This does all the path init required, making room in the tree if needed. | |
4615 | */ | |
4616 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4617 | struct btrfs_root *root, | |
4618 | struct btrfs_path *path, | |
4619 | struct btrfs_key *cpu_key, u32 *data_size, | |
4620 | int nr) | |
4621 | { | |
44871b1b CM |
4622 | int ret = 0; |
4623 | int slot; | |
4624 | int i; | |
4625 | u32 total_size = 0; | |
4626 | u32 total_data = 0; | |
4627 | ||
4628 | for (i = 0; i < nr; i++) | |
4629 | total_data += data_size[i]; | |
4630 | ||
4631 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4632 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4633 | if (ret == 0) | |
4634 | return -EEXIST; | |
4635 | if (ret < 0) | |
143bede5 | 4636 | return ret; |
44871b1b | 4637 | |
44871b1b CM |
4638 | slot = path->slots[0]; |
4639 | BUG_ON(slot < 0); | |
4640 | ||
afe5fea7 | 4641 | setup_items_for_insert(root, path, cpu_key, data_size, |
44871b1b | 4642 | total_data, total_size, nr); |
143bede5 | 4643 | return 0; |
62e2749e CM |
4644 | } |
4645 | ||
4646 | /* | |
4647 | * Given a key and some data, insert an item into the tree. | |
4648 | * This does all the path init required, making room in the tree if needed. | |
4649 | */ | |
e089f05c CM |
4650 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4651 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4652 | data_size) | |
62e2749e CM |
4653 | { |
4654 | int ret = 0; | |
2c90e5d6 | 4655 | struct btrfs_path *path; |
5f39d397 CM |
4656 | struct extent_buffer *leaf; |
4657 | unsigned long ptr; | |
62e2749e | 4658 | |
2c90e5d6 | 4659 | path = btrfs_alloc_path(); |
db5b493a TI |
4660 | if (!path) |
4661 | return -ENOMEM; | |
2c90e5d6 | 4662 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4663 | if (!ret) { |
5f39d397 CM |
4664 | leaf = path->nodes[0]; |
4665 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4666 | write_extent_buffer(leaf, data, ptr, data_size); | |
4667 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4668 | } |
2c90e5d6 | 4669 | btrfs_free_path(path); |
aa5d6bed | 4670 | return ret; |
be0e5c09 CM |
4671 | } |
4672 | ||
74123bd7 | 4673 | /* |
5de08d7d | 4674 | * delete the pointer from a given node. |
74123bd7 | 4675 | * |
d352ac68 CM |
4676 | * the tree should have been previously balanced so the deletion does not |
4677 | * empty a node. | |
74123bd7 | 4678 | */ |
afe5fea7 TI |
4679 | static void del_ptr(struct btrfs_root *root, struct btrfs_path *path, |
4680 | int level, int slot) | |
be0e5c09 | 4681 | { |
5f39d397 | 4682 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4683 | u32 nritems; |
f3ea38da | 4684 | int ret; |
be0e5c09 | 4685 | |
5f39d397 | 4686 | nritems = btrfs_header_nritems(parent); |
d397712b | 4687 | if (slot != nritems - 1) { |
0e411ece | 4688 | if (level) |
f3ea38da JS |
4689 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4690 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4691 | memmove_extent_buffer(parent, |
4692 | btrfs_node_key_ptr_offset(slot), | |
4693 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4694 | sizeof(struct btrfs_key_ptr) * |
4695 | (nritems - slot - 1)); | |
57ba86c0 CM |
4696 | } else if (level) { |
4697 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
c8cc6341 | 4698 | MOD_LOG_KEY_REMOVE, GFP_NOFS); |
57ba86c0 | 4699 | BUG_ON(ret < 0); |
bb803951 | 4700 | } |
f3ea38da | 4701 | |
7518a238 | 4702 | nritems--; |
5f39d397 | 4703 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4704 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4705 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4706 | /* just turn the root into a leaf and break */ |
5f39d397 | 4707 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4708 | } else if (slot == 0) { |
5f39d397 CM |
4709 | struct btrfs_disk_key disk_key; |
4710 | ||
4711 | btrfs_node_key(parent, &disk_key, 0); | |
d6a0a126 | 4712 | fixup_low_keys(root, path, &disk_key, level + 1); |
be0e5c09 | 4713 | } |
d6025579 | 4714 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4715 | } |
4716 | ||
323ac95b CM |
4717 | /* |
4718 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4719 | * path->nodes[1]. |
323ac95b CM |
4720 | * |
4721 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4722 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4723 | * | |
4724 | * The path must have already been setup for deleting the leaf, including | |
4725 | * all the proper balancing. path->nodes[1] must be locked. | |
4726 | */ | |
143bede5 JM |
4727 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4728 | struct btrfs_root *root, | |
4729 | struct btrfs_path *path, | |
4730 | struct extent_buffer *leaf) | |
323ac95b | 4731 | { |
5d4f98a2 | 4732 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
afe5fea7 | 4733 | del_ptr(root, path, 1, path->slots[1]); |
323ac95b | 4734 | |
4d081c41 CM |
4735 | /* |
4736 | * btrfs_free_extent is expensive, we want to make sure we | |
4737 | * aren't holding any locks when we call it | |
4738 | */ | |
4739 | btrfs_unlock_up_safe(path, 0); | |
4740 | ||
f0486c68 YZ |
4741 | root_sub_used(root, leaf->len); |
4742 | ||
3083ee2e | 4743 | extent_buffer_get(leaf); |
5581a51a | 4744 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4745 | free_extent_buffer_stale(leaf); |
323ac95b | 4746 | } |
74123bd7 CM |
4747 | /* |
4748 | * delete the item at the leaf level in path. If that empties | |
4749 | * the leaf, remove it from the tree | |
4750 | */ | |
85e21bac CM |
4751 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4752 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4753 | { |
5f39d397 CM |
4754 | struct extent_buffer *leaf; |
4755 | struct btrfs_item *item; | |
85e21bac CM |
4756 | int last_off; |
4757 | int dsize = 0; | |
aa5d6bed CM |
4758 | int ret = 0; |
4759 | int wret; | |
85e21bac | 4760 | int i; |
7518a238 | 4761 | u32 nritems; |
cfed81a0 CM |
4762 | struct btrfs_map_token token; |
4763 | ||
4764 | btrfs_init_map_token(&token); | |
be0e5c09 | 4765 | |
5f39d397 | 4766 | leaf = path->nodes[0]; |
85e21bac CM |
4767 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4768 | ||
4769 | for (i = 0; i < nr; i++) | |
4770 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4771 | ||
5f39d397 | 4772 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4773 | |
85e21bac | 4774 | if (slot + nr != nritems) { |
123abc88 | 4775 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4776 | |
4777 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4778 | data_end + dsize, |
4779 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4780 | last_off - data_end); |
5f39d397 | 4781 | |
85e21bac | 4782 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4783 | u32 ioff; |
db94535d | 4784 | |
dd3cc16b | 4785 | item = btrfs_item_nr(i); |
cfed81a0 CM |
4786 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4787 | btrfs_set_token_item_offset(leaf, item, | |
4788 | ioff + dsize, &token); | |
0783fcfc | 4789 | } |
db94535d | 4790 | |
5f39d397 | 4791 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4792 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4793 | sizeof(struct btrfs_item) * |
85e21bac | 4794 | (nritems - slot - nr)); |
be0e5c09 | 4795 | } |
85e21bac CM |
4796 | btrfs_set_header_nritems(leaf, nritems - nr); |
4797 | nritems -= nr; | |
5f39d397 | 4798 | |
74123bd7 | 4799 | /* delete the leaf if we've emptied it */ |
7518a238 | 4800 | if (nritems == 0) { |
5f39d397 CM |
4801 | if (leaf == root->node) { |
4802 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4803 | } else { |
f0486c68 YZ |
4804 | btrfs_set_path_blocking(path); |
4805 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4806 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4807 | } |
be0e5c09 | 4808 | } else { |
7518a238 | 4809 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4810 | if (slot == 0) { |
5f39d397 CM |
4811 | struct btrfs_disk_key disk_key; |
4812 | ||
4813 | btrfs_item_key(leaf, &disk_key, 0); | |
d6a0a126 | 4814 | fixup_low_keys(root, path, &disk_key, 1); |
aa5d6bed | 4815 | } |
aa5d6bed | 4816 | |
74123bd7 | 4817 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4818 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4819 | /* push_leaf_left fixes the path. |
4820 | * make sure the path still points to our leaf | |
4821 | * for possible call to del_ptr below | |
4822 | */ | |
4920c9ac | 4823 | slot = path->slots[1]; |
5f39d397 CM |
4824 | extent_buffer_get(leaf); |
4825 | ||
b9473439 | 4826 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4827 | wret = push_leaf_left(trans, root, path, 1, 1, |
4828 | 1, (u32)-1); | |
54aa1f4d | 4829 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4830 | ret = wret; |
5f39d397 CM |
4831 | |
4832 | if (path->nodes[0] == leaf && | |
4833 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4834 | wret = push_leaf_right(trans, root, path, 1, |
4835 | 1, 1, 0); | |
54aa1f4d | 4836 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4837 | ret = wret; |
4838 | } | |
5f39d397 CM |
4839 | |
4840 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4841 | path->slots[1] = slot; |
143bede5 | 4842 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4843 | free_extent_buffer(leaf); |
143bede5 | 4844 | ret = 0; |
5de08d7d | 4845 | } else { |
925baedd CM |
4846 | /* if we're still in the path, make sure |
4847 | * we're dirty. Otherwise, one of the | |
4848 | * push_leaf functions must have already | |
4849 | * dirtied this buffer | |
4850 | */ | |
4851 | if (path->nodes[0] == leaf) | |
4852 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4853 | free_extent_buffer(leaf); |
be0e5c09 | 4854 | } |
d5719762 | 4855 | } else { |
5f39d397 | 4856 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4857 | } |
4858 | } | |
aa5d6bed | 4859 | return ret; |
be0e5c09 CM |
4860 | } |
4861 | ||
7bb86316 | 4862 | /* |
925baedd | 4863 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4864 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4865 | * returns < 0 on io errors. | |
d352ac68 CM |
4866 | * |
4867 | * This may release the path, and so you may lose any locks held at the | |
4868 | * time you call it. | |
7bb86316 | 4869 | */ |
16e7549f | 4870 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) |
7bb86316 | 4871 | { |
925baedd CM |
4872 | struct btrfs_key key; |
4873 | struct btrfs_disk_key found_key; | |
4874 | int ret; | |
7bb86316 | 4875 | |
925baedd | 4876 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4877 | |
e8b0d724 | 4878 | if (key.offset > 0) { |
925baedd | 4879 | key.offset--; |
e8b0d724 | 4880 | } else if (key.type > 0) { |
925baedd | 4881 | key.type--; |
e8b0d724 FDBM |
4882 | key.offset = (u64)-1; |
4883 | } else if (key.objectid > 0) { | |
925baedd | 4884 | key.objectid--; |
e8b0d724 FDBM |
4885 | key.type = (u8)-1; |
4886 | key.offset = (u64)-1; | |
4887 | } else { | |
925baedd | 4888 | return 1; |
e8b0d724 | 4889 | } |
7bb86316 | 4890 | |
b3b4aa74 | 4891 | btrfs_release_path(path); |
925baedd CM |
4892 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4893 | if (ret < 0) | |
4894 | return ret; | |
4895 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4896 | ret = comp_keys(&found_key, &key); | |
4897 | if (ret < 0) | |
4898 | return 0; | |
4899 | return 1; | |
7bb86316 CM |
4900 | } |
4901 | ||
3f157a2f CM |
4902 | /* |
4903 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4904 | * for nodes or leaves that are have a minimum transaction id. |
4905 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4906 | * |
4907 | * This does not cow, but it does stuff the starting key it finds back | |
4908 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4909 | * key and get a writable path. | |
4910 | * | |
4911 | * This does lock as it descends, and path->keep_locks should be set | |
4912 | * to 1 by the caller. | |
4913 | * | |
4914 | * This honors path->lowest_level to prevent descent past a given level | |
4915 | * of the tree. | |
4916 | * | |
d352ac68 CM |
4917 | * min_trans indicates the oldest transaction that you are interested |
4918 | * in walking through. Any nodes or leaves older than min_trans are | |
4919 | * skipped over (without reading them). | |
4920 | * | |
3f157a2f CM |
4921 | * returns zero if something useful was found, < 0 on error and 1 if there |
4922 | * was nothing in the tree that matched the search criteria. | |
4923 | */ | |
4924 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
de78b51a | 4925 | struct btrfs_path *path, |
3f157a2f CM |
4926 | u64 min_trans) |
4927 | { | |
4928 | struct extent_buffer *cur; | |
4929 | struct btrfs_key found_key; | |
4930 | int slot; | |
9652480b | 4931 | int sret; |
3f157a2f CM |
4932 | u32 nritems; |
4933 | int level; | |
4934 | int ret = 1; | |
4935 | ||
934d375b | 4936 | WARN_ON(!path->keep_locks); |
3f157a2f | 4937 | again: |
bd681513 | 4938 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4939 | level = btrfs_header_level(cur); |
e02119d5 | 4940 | WARN_ON(path->nodes[level]); |
3f157a2f | 4941 | path->nodes[level] = cur; |
bd681513 | 4942 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4943 | |
4944 | if (btrfs_header_generation(cur) < min_trans) { | |
4945 | ret = 1; | |
4946 | goto out; | |
4947 | } | |
d397712b | 4948 | while (1) { |
3f157a2f CM |
4949 | nritems = btrfs_header_nritems(cur); |
4950 | level = btrfs_header_level(cur); | |
9652480b | 4951 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4952 | |
323ac95b CM |
4953 | /* at the lowest level, we're done, setup the path and exit */ |
4954 | if (level == path->lowest_level) { | |
e02119d5 CM |
4955 | if (slot >= nritems) |
4956 | goto find_next_key; | |
3f157a2f CM |
4957 | ret = 0; |
4958 | path->slots[level] = slot; | |
4959 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4960 | goto out; | |
4961 | } | |
9652480b Y |
4962 | if (sret && slot > 0) |
4963 | slot--; | |
3f157a2f | 4964 | /* |
de78b51a ES |
4965 | * check this node pointer against the min_trans parameters. |
4966 | * If it is too old, old, skip to the next one. | |
3f157a2f | 4967 | */ |
d397712b | 4968 | while (slot < nritems) { |
3f157a2f | 4969 | u64 gen; |
e02119d5 | 4970 | |
3f157a2f CM |
4971 | gen = btrfs_node_ptr_generation(cur, slot); |
4972 | if (gen < min_trans) { | |
4973 | slot++; | |
4974 | continue; | |
4975 | } | |
de78b51a | 4976 | break; |
3f157a2f | 4977 | } |
e02119d5 | 4978 | find_next_key: |
3f157a2f CM |
4979 | /* |
4980 | * we didn't find a candidate key in this node, walk forward | |
4981 | * and find another one | |
4982 | */ | |
4983 | if (slot >= nritems) { | |
e02119d5 | 4984 | path->slots[level] = slot; |
b4ce94de | 4985 | btrfs_set_path_blocking(path); |
e02119d5 | 4986 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 4987 | min_trans); |
e02119d5 | 4988 | if (sret == 0) { |
b3b4aa74 | 4989 | btrfs_release_path(path); |
3f157a2f CM |
4990 | goto again; |
4991 | } else { | |
4992 | goto out; | |
4993 | } | |
4994 | } | |
4995 | /* save our key for returning back */ | |
4996 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4997 | path->slots[level] = slot; | |
4998 | if (level == path->lowest_level) { | |
4999 | ret = 0; | |
f7c79f30 | 5000 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
5001 | goto out; |
5002 | } | |
b4ce94de | 5003 | btrfs_set_path_blocking(path); |
3f157a2f | 5004 | cur = read_node_slot(root, cur, slot); |
79787eaa | 5005 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 5006 | |
bd681513 | 5007 | btrfs_tree_read_lock(cur); |
b4ce94de | 5008 | |
bd681513 | 5009 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 5010 | path->nodes[level - 1] = cur; |
f7c79f30 | 5011 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 5012 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
5013 | } |
5014 | out: | |
5015 | if (ret == 0) | |
5016 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 5017 | btrfs_set_path_blocking(path); |
3f157a2f CM |
5018 | return ret; |
5019 | } | |
5020 | ||
7069830a AB |
5021 | static void tree_move_down(struct btrfs_root *root, |
5022 | struct btrfs_path *path, | |
5023 | int *level, int root_level) | |
5024 | { | |
74dd17fb | 5025 | BUG_ON(*level == 0); |
7069830a AB |
5026 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
5027 | path->slots[*level]); | |
5028 | path->slots[*level - 1] = 0; | |
5029 | (*level)--; | |
5030 | } | |
5031 | ||
5032 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
5033 | struct btrfs_path *path, | |
5034 | int *level, int root_level) | |
5035 | { | |
5036 | int ret = 0; | |
5037 | int nritems; | |
5038 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5039 | ||
5040 | path->slots[*level]++; | |
5041 | ||
74dd17fb | 5042 | while (path->slots[*level] >= nritems) { |
7069830a AB |
5043 | if (*level == root_level) |
5044 | return -1; | |
5045 | ||
5046 | /* move upnext */ | |
5047 | path->slots[*level] = 0; | |
5048 | free_extent_buffer(path->nodes[*level]); | |
5049 | path->nodes[*level] = NULL; | |
5050 | (*level)++; | |
5051 | path->slots[*level]++; | |
5052 | ||
5053 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
5054 | ret = 1; | |
5055 | } | |
5056 | return ret; | |
5057 | } | |
5058 | ||
5059 | /* | |
5060 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
5061 | * or down. | |
5062 | */ | |
5063 | static int tree_advance(struct btrfs_root *root, | |
5064 | struct btrfs_path *path, | |
5065 | int *level, int root_level, | |
5066 | int allow_down, | |
5067 | struct btrfs_key *key) | |
5068 | { | |
5069 | int ret; | |
5070 | ||
5071 | if (*level == 0 || !allow_down) { | |
5072 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
5073 | } else { | |
5074 | tree_move_down(root, path, level, root_level); | |
5075 | ret = 0; | |
5076 | } | |
5077 | if (ret >= 0) { | |
5078 | if (*level == 0) | |
5079 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5080 | path->slots[*level]); | |
5081 | else | |
5082 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5083 | path->slots[*level]); | |
5084 | } | |
5085 | return ret; | |
5086 | } | |
5087 | ||
5088 | static int tree_compare_item(struct btrfs_root *left_root, | |
5089 | struct btrfs_path *left_path, | |
5090 | struct btrfs_path *right_path, | |
5091 | char *tmp_buf) | |
5092 | { | |
5093 | int cmp; | |
5094 | int len1, len2; | |
5095 | unsigned long off1, off2; | |
5096 | ||
5097 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5098 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5099 | if (len1 != len2) | |
5100 | return 1; | |
5101 | ||
5102 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5103 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5104 | right_path->slots[0]); | |
5105 | ||
5106 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5107 | ||
5108 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5109 | if (cmp) | |
5110 | return 1; | |
5111 | return 0; | |
5112 | } | |
5113 | ||
5114 | #define ADVANCE 1 | |
5115 | #define ADVANCE_ONLY_NEXT -1 | |
5116 | ||
5117 | /* | |
5118 | * This function compares two trees and calls the provided callback for | |
5119 | * every changed/new/deleted item it finds. | |
5120 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5121 | * the compare pretty fast on snapshotted subvolumes. | |
5122 | * | |
5123 | * This currently works on commit roots only. As commit roots are read only, | |
5124 | * we don't do any locking. The commit roots are protected with transactions. | |
5125 | * Transactions are ended and rejoined when a commit is tried in between. | |
5126 | * | |
5127 | * This function checks for modifications done to the trees while comparing. | |
5128 | * If it detects a change, it aborts immediately. | |
5129 | */ | |
5130 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5131 | struct btrfs_root *right_root, | |
5132 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5133 | { | |
5134 | int ret; | |
5135 | int cmp; | |
5136 | struct btrfs_trans_handle *trans = NULL; | |
5137 | struct btrfs_path *left_path = NULL; | |
5138 | struct btrfs_path *right_path = NULL; | |
5139 | struct btrfs_key left_key; | |
5140 | struct btrfs_key right_key; | |
5141 | char *tmp_buf = NULL; | |
5142 | int left_root_level; | |
5143 | int right_root_level; | |
5144 | int left_level; | |
5145 | int right_level; | |
5146 | int left_end_reached; | |
5147 | int right_end_reached; | |
5148 | int advance_left; | |
5149 | int advance_right; | |
5150 | u64 left_blockptr; | |
5151 | u64 right_blockptr; | |
5152 | u64 left_start_ctransid; | |
5153 | u64 right_start_ctransid; | |
5154 | u64 ctransid; | |
5155 | ||
5156 | left_path = btrfs_alloc_path(); | |
5157 | if (!left_path) { | |
5158 | ret = -ENOMEM; | |
5159 | goto out; | |
5160 | } | |
5161 | right_path = btrfs_alloc_path(); | |
5162 | if (!right_path) { | |
5163 | ret = -ENOMEM; | |
5164 | goto out; | |
5165 | } | |
5166 | ||
5167 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5168 | if (!tmp_buf) { | |
5169 | ret = -ENOMEM; | |
5170 | goto out; | |
5171 | } | |
5172 | ||
5173 | left_path->search_commit_root = 1; | |
5174 | left_path->skip_locking = 1; | |
5175 | right_path->search_commit_root = 1; | |
5176 | right_path->skip_locking = 1; | |
5177 | ||
5f3ab90a | 5178 | spin_lock(&left_root->root_item_lock); |
7069830a | 5179 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5180 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5181 | |
5f3ab90a | 5182 | spin_lock(&right_root->root_item_lock); |
7069830a | 5183 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5184 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5185 | |
5186 | trans = btrfs_join_transaction(left_root); | |
5187 | if (IS_ERR(trans)) { | |
5188 | ret = PTR_ERR(trans); | |
5189 | trans = NULL; | |
5190 | goto out; | |
5191 | } | |
5192 | ||
5193 | /* | |
5194 | * Strategy: Go to the first items of both trees. Then do | |
5195 | * | |
5196 | * If both trees are at level 0 | |
5197 | * Compare keys of current items | |
5198 | * If left < right treat left item as new, advance left tree | |
5199 | * and repeat | |
5200 | * If left > right treat right item as deleted, advance right tree | |
5201 | * and repeat | |
5202 | * If left == right do deep compare of items, treat as changed if | |
5203 | * needed, advance both trees and repeat | |
5204 | * If both trees are at the same level but not at level 0 | |
5205 | * Compare keys of current nodes/leafs | |
5206 | * If left < right advance left tree and repeat | |
5207 | * If left > right advance right tree and repeat | |
5208 | * If left == right compare blockptrs of the next nodes/leafs | |
5209 | * If they match advance both trees but stay at the same level | |
5210 | * and repeat | |
5211 | * If they don't match advance both trees while allowing to go | |
5212 | * deeper and repeat | |
5213 | * If tree levels are different | |
5214 | * Advance the tree that needs it and repeat | |
5215 | * | |
5216 | * Advancing a tree means: | |
5217 | * If we are at level 0, try to go to the next slot. If that's not | |
5218 | * possible, go one level up and repeat. Stop when we found a level | |
5219 | * where we could go to the next slot. We may at this point be on a | |
5220 | * node or a leaf. | |
5221 | * | |
5222 | * If we are not at level 0 and not on shared tree blocks, go one | |
5223 | * level deeper. | |
5224 | * | |
5225 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5226 | * the right if possible or go up and right. | |
5227 | */ | |
5228 | ||
5229 | left_level = btrfs_header_level(left_root->commit_root); | |
5230 | left_root_level = left_level; | |
5231 | left_path->nodes[left_level] = left_root->commit_root; | |
5232 | extent_buffer_get(left_path->nodes[left_level]); | |
5233 | ||
5234 | right_level = btrfs_header_level(right_root->commit_root); | |
5235 | right_root_level = right_level; | |
5236 | right_path->nodes[right_level] = right_root->commit_root; | |
5237 | extent_buffer_get(right_path->nodes[right_level]); | |
5238 | ||
5239 | if (left_level == 0) | |
5240 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5241 | &left_key, left_path->slots[left_level]); | |
5242 | else | |
5243 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5244 | &left_key, left_path->slots[left_level]); | |
5245 | if (right_level == 0) | |
5246 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5247 | &right_key, right_path->slots[right_level]); | |
5248 | else | |
5249 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5250 | &right_key, right_path->slots[right_level]); | |
5251 | ||
5252 | left_end_reached = right_end_reached = 0; | |
5253 | advance_left = advance_right = 0; | |
5254 | ||
5255 | while (1) { | |
5256 | /* | |
5257 | * We need to make sure the transaction does not get committed | |
5258 | * while we do anything on commit roots. This means, we need to | |
5259 | * join and leave transactions for every item that we process. | |
5260 | */ | |
5261 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5262 | btrfs_release_path(left_path); | |
5263 | btrfs_release_path(right_path); | |
5264 | ||
5265 | ret = btrfs_end_transaction(trans, left_root); | |
5266 | trans = NULL; | |
5267 | if (ret < 0) | |
5268 | goto out; | |
5269 | } | |
5270 | /* now rejoin the transaction */ | |
5271 | if (!trans) { | |
5272 | trans = btrfs_join_transaction(left_root); | |
5273 | if (IS_ERR(trans)) { | |
5274 | ret = PTR_ERR(trans); | |
5275 | trans = NULL; | |
5276 | goto out; | |
5277 | } | |
5278 | ||
5f3ab90a | 5279 | spin_lock(&left_root->root_item_lock); |
7069830a | 5280 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5281 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5282 | if (ctransid != left_start_ctransid) |
5283 | left_start_ctransid = 0; | |
5284 | ||
5f3ab90a | 5285 | spin_lock(&right_root->root_item_lock); |
7069830a | 5286 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5287 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5288 | if (ctransid != right_start_ctransid) |
5289 | right_start_ctransid = 0; | |
5290 | ||
5291 | if (!left_start_ctransid || !right_start_ctransid) { | |
5292 | WARN(1, KERN_WARNING | |
5293 | "btrfs: btrfs_compare_tree detected " | |
5294 | "a change in one of the trees while " | |
5295 | "iterating. This is probably a " | |
5296 | "bug.\n"); | |
5297 | ret = -EIO; | |
5298 | goto out; | |
5299 | } | |
5300 | ||
5301 | /* | |
5302 | * the commit root may have changed, so start again | |
5303 | * where we stopped | |
5304 | */ | |
5305 | left_path->lowest_level = left_level; | |
5306 | right_path->lowest_level = right_level; | |
5307 | ret = btrfs_search_slot(NULL, left_root, | |
5308 | &left_key, left_path, 0, 0); | |
5309 | if (ret < 0) | |
5310 | goto out; | |
5311 | ret = btrfs_search_slot(NULL, right_root, | |
5312 | &right_key, right_path, 0, 0); | |
5313 | if (ret < 0) | |
5314 | goto out; | |
5315 | } | |
5316 | ||
5317 | if (advance_left && !left_end_reached) { | |
5318 | ret = tree_advance(left_root, left_path, &left_level, | |
5319 | left_root_level, | |
5320 | advance_left != ADVANCE_ONLY_NEXT, | |
5321 | &left_key); | |
5322 | if (ret < 0) | |
5323 | left_end_reached = ADVANCE; | |
5324 | advance_left = 0; | |
5325 | } | |
5326 | if (advance_right && !right_end_reached) { | |
5327 | ret = tree_advance(right_root, right_path, &right_level, | |
5328 | right_root_level, | |
5329 | advance_right != ADVANCE_ONLY_NEXT, | |
5330 | &right_key); | |
5331 | if (ret < 0) | |
5332 | right_end_reached = ADVANCE; | |
5333 | advance_right = 0; | |
5334 | } | |
5335 | ||
5336 | if (left_end_reached && right_end_reached) { | |
5337 | ret = 0; | |
5338 | goto out; | |
5339 | } else if (left_end_reached) { | |
5340 | if (right_level == 0) { | |
5341 | ret = changed_cb(left_root, right_root, | |
5342 | left_path, right_path, | |
5343 | &right_key, | |
5344 | BTRFS_COMPARE_TREE_DELETED, | |
5345 | ctx); | |
5346 | if (ret < 0) | |
5347 | goto out; | |
5348 | } | |
5349 | advance_right = ADVANCE; | |
5350 | continue; | |
5351 | } else if (right_end_reached) { | |
5352 | if (left_level == 0) { | |
5353 | ret = changed_cb(left_root, right_root, | |
5354 | left_path, right_path, | |
5355 | &left_key, | |
5356 | BTRFS_COMPARE_TREE_NEW, | |
5357 | ctx); | |
5358 | if (ret < 0) | |
5359 | goto out; | |
5360 | } | |
5361 | advance_left = ADVANCE; | |
5362 | continue; | |
5363 | } | |
5364 | ||
5365 | if (left_level == 0 && right_level == 0) { | |
5366 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5367 | if (cmp < 0) { | |
5368 | ret = changed_cb(left_root, right_root, | |
5369 | left_path, right_path, | |
5370 | &left_key, | |
5371 | BTRFS_COMPARE_TREE_NEW, | |
5372 | ctx); | |
5373 | if (ret < 0) | |
5374 | goto out; | |
5375 | advance_left = ADVANCE; | |
5376 | } else if (cmp > 0) { | |
5377 | ret = changed_cb(left_root, right_root, | |
5378 | left_path, right_path, | |
5379 | &right_key, | |
5380 | BTRFS_COMPARE_TREE_DELETED, | |
5381 | ctx); | |
5382 | if (ret < 0) | |
5383 | goto out; | |
5384 | advance_right = ADVANCE; | |
5385 | } else { | |
ba5e8f2e JB |
5386 | enum btrfs_compare_tree_result cmp; |
5387 | ||
74dd17fb | 5388 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5389 | ret = tree_compare_item(left_root, left_path, |
5390 | right_path, tmp_buf); | |
ba5e8f2e JB |
5391 | if (ret) |
5392 | cmp = BTRFS_COMPARE_TREE_CHANGED; | |
5393 | else | |
5394 | cmp = BTRFS_COMPARE_TREE_SAME; | |
5395 | ret = changed_cb(left_root, right_root, | |
5396 | left_path, right_path, | |
5397 | &left_key, cmp, ctx); | |
5398 | if (ret < 0) | |
5399 | goto out; | |
7069830a AB |
5400 | advance_left = ADVANCE; |
5401 | advance_right = ADVANCE; | |
5402 | } | |
5403 | } else if (left_level == right_level) { | |
5404 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5405 | if (cmp < 0) { | |
5406 | advance_left = ADVANCE; | |
5407 | } else if (cmp > 0) { | |
5408 | advance_right = ADVANCE; | |
5409 | } else { | |
5410 | left_blockptr = btrfs_node_blockptr( | |
5411 | left_path->nodes[left_level], | |
5412 | left_path->slots[left_level]); | |
5413 | right_blockptr = btrfs_node_blockptr( | |
5414 | right_path->nodes[right_level], | |
5415 | right_path->slots[right_level]); | |
5416 | if (left_blockptr == right_blockptr) { | |
5417 | /* | |
5418 | * As we're on a shared block, don't | |
5419 | * allow to go deeper. | |
5420 | */ | |
5421 | advance_left = ADVANCE_ONLY_NEXT; | |
5422 | advance_right = ADVANCE_ONLY_NEXT; | |
5423 | } else { | |
5424 | advance_left = ADVANCE; | |
5425 | advance_right = ADVANCE; | |
5426 | } | |
5427 | } | |
5428 | } else if (left_level < right_level) { | |
5429 | advance_right = ADVANCE; | |
5430 | } else { | |
5431 | advance_left = ADVANCE; | |
5432 | } | |
5433 | } | |
5434 | ||
5435 | out: | |
5436 | btrfs_free_path(left_path); | |
5437 | btrfs_free_path(right_path); | |
5438 | kfree(tmp_buf); | |
5439 | ||
5440 | if (trans) { | |
5441 | if (!ret) | |
5442 | ret = btrfs_end_transaction(trans, left_root); | |
5443 | else | |
5444 | btrfs_end_transaction(trans, left_root); | |
5445 | } | |
5446 | ||
5447 | return ret; | |
5448 | } | |
5449 | ||
3f157a2f CM |
5450 | /* |
5451 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5452 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5453 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5454 | * |
5455 | * 0 is returned if another key is found, < 0 if there are any errors | |
5456 | * and 1 is returned if there are no higher keys in the tree | |
5457 | * | |
5458 | * path->keep_locks should be set to 1 on the search made before | |
5459 | * calling this function. | |
5460 | */ | |
e7a84565 | 5461 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5462 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5463 | { |
e7a84565 CM |
5464 | int slot; |
5465 | struct extent_buffer *c; | |
5466 | ||
934d375b | 5467 | WARN_ON(!path->keep_locks); |
d397712b | 5468 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5469 | if (!path->nodes[level]) |
5470 | return 1; | |
5471 | ||
5472 | slot = path->slots[level] + 1; | |
5473 | c = path->nodes[level]; | |
3f157a2f | 5474 | next: |
e7a84565 | 5475 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5476 | int ret; |
5477 | int orig_lowest; | |
5478 | struct btrfs_key cur_key; | |
5479 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5480 | !path->nodes[level + 1]) | |
e7a84565 | 5481 | return 1; |
33c66f43 YZ |
5482 | |
5483 | if (path->locks[level + 1]) { | |
5484 | level++; | |
5485 | continue; | |
5486 | } | |
5487 | ||
5488 | slot = btrfs_header_nritems(c) - 1; | |
5489 | if (level == 0) | |
5490 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5491 | else | |
5492 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5493 | ||
5494 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5495 | btrfs_release_path(path); |
33c66f43 YZ |
5496 | path->lowest_level = level; |
5497 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5498 | 0, 0); | |
5499 | path->lowest_level = orig_lowest; | |
5500 | if (ret < 0) | |
5501 | return ret; | |
5502 | ||
5503 | c = path->nodes[level]; | |
5504 | slot = path->slots[level]; | |
5505 | if (ret == 0) | |
5506 | slot++; | |
5507 | goto next; | |
e7a84565 | 5508 | } |
33c66f43 | 5509 | |
e7a84565 CM |
5510 | if (level == 0) |
5511 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5512 | else { |
3f157a2f CM |
5513 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5514 | ||
3f157a2f CM |
5515 | if (gen < min_trans) { |
5516 | slot++; | |
5517 | goto next; | |
5518 | } | |
e7a84565 | 5519 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5520 | } |
e7a84565 CM |
5521 | return 0; |
5522 | } | |
5523 | return 1; | |
5524 | } | |
5525 | ||
97571fd0 | 5526 | /* |
925baedd | 5527 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5528 | * returns 0 if it found something or 1 if there are no greater leaves. |
5529 | * returns < 0 on io errors. | |
97571fd0 | 5530 | */ |
234b63a0 | 5531 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5532 | { |
5533 | return btrfs_next_old_leaf(root, path, 0); | |
5534 | } | |
5535 | ||
5536 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, | |
5537 | u64 time_seq) | |
d97e63b6 CM |
5538 | { |
5539 | int slot; | |
8e73f275 | 5540 | int level; |
5f39d397 | 5541 | struct extent_buffer *c; |
8e73f275 | 5542 | struct extent_buffer *next; |
925baedd CM |
5543 | struct btrfs_key key; |
5544 | u32 nritems; | |
5545 | int ret; | |
8e73f275 | 5546 | int old_spinning = path->leave_spinning; |
bd681513 | 5547 | int next_rw_lock = 0; |
925baedd CM |
5548 | |
5549 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5550 | if (nritems == 0) |
925baedd | 5551 | return 1; |
925baedd | 5552 | |
8e73f275 CM |
5553 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5554 | again: | |
5555 | level = 1; | |
5556 | next = NULL; | |
bd681513 | 5557 | next_rw_lock = 0; |
b3b4aa74 | 5558 | btrfs_release_path(path); |
8e73f275 | 5559 | |
a2135011 | 5560 | path->keep_locks = 1; |
31533fb2 | 5561 | path->leave_spinning = 1; |
8e73f275 | 5562 | |
3d7806ec JS |
5563 | if (time_seq) |
5564 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5565 | else | |
5566 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5567 | path->keep_locks = 0; |
5568 | ||
5569 | if (ret < 0) | |
5570 | return ret; | |
5571 | ||
a2135011 | 5572 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5573 | /* |
5574 | * by releasing the path above we dropped all our locks. A balance | |
5575 | * could have added more items next to the key that used to be | |
5576 | * at the very end of the block. So, check again here and | |
5577 | * advance the path if there are now more items available. | |
5578 | */ | |
a2135011 | 5579 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5580 | if (ret == 0) |
5581 | path->slots[0]++; | |
8e73f275 | 5582 | ret = 0; |
925baedd CM |
5583 | goto done; |
5584 | } | |
d97e63b6 | 5585 | |
d397712b | 5586 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5587 | if (!path->nodes[level]) { |
5588 | ret = 1; | |
5589 | goto done; | |
5590 | } | |
5f39d397 | 5591 | |
d97e63b6 CM |
5592 | slot = path->slots[level] + 1; |
5593 | c = path->nodes[level]; | |
5f39d397 | 5594 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5595 | level++; |
8e73f275 CM |
5596 | if (level == BTRFS_MAX_LEVEL) { |
5597 | ret = 1; | |
5598 | goto done; | |
5599 | } | |
d97e63b6 CM |
5600 | continue; |
5601 | } | |
5f39d397 | 5602 | |
925baedd | 5603 | if (next) { |
bd681513 | 5604 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5605 | free_extent_buffer(next); |
925baedd | 5606 | } |
5f39d397 | 5607 | |
8e73f275 | 5608 | next = c; |
bd681513 | 5609 | next_rw_lock = path->locks[level]; |
8e73f275 | 5610 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5611 | slot, &key, 0); |
8e73f275 CM |
5612 | if (ret == -EAGAIN) |
5613 | goto again; | |
5f39d397 | 5614 | |
76a05b35 | 5615 | if (ret < 0) { |
b3b4aa74 | 5616 | btrfs_release_path(path); |
76a05b35 CM |
5617 | goto done; |
5618 | } | |
5619 | ||
5cd57b2c | 5620 | if (!path->skip_locking) { |
bd681513 | 5621 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5622 | if (!ret && time_seq) { |
5623 | /* | |
5624 | * If we don't get the lock, we may be racing | |
5625 | * with push_leaf_left, holding that lock while | |
5626 | * itself waiting for the leaf we've currently | |
5627 | * locked. To solve this situation, we give up | |
5628 | * on our lock and cycle. | |
5629 | */ | |
cf538830 | 5630 | free_extent_buffer(next); |
d42244a0 JS |
5631 | btrfs_release_path(path); |
5632 | cond_resched(); | |
5633 | goto again; | |
5634 | } | |
8e73f275 CM |
5635 | if (!ret) { |
5636 | btrfs_set_path_blocking(path); | |
bd681513 | 5637 | btrfs_tree_read_lock(next); |
31533fb2 | 5638 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5639 | BTRFS_READ_LOCK); |
8e73f275 | 5640 | } |
31533fb2 | 5641 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5642 | } |
d97e63b6 CM |
5643 | break; |
5644 | } | |
5645 | path->slots[level] = slot; | |
d397712b | 5646 | while (1) { |
d97e63b6 CM |
5647 | level--; |
5648 | c = path->nodes[level]; | |
925baedd | 5649 | if (path->locks[level]) |
bd681513 | 5650 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5651 | |
5f39d397 | 5652 | free_extent_buffer(c); |
d97e63b6 CM |
5653 | path->nodes[level] = next; |
5654 | path->slots[level] = 0; | |
a74a4b97 | 5655 | if (!path->skip_locking) |
bd681513 | 5656 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5657 | if (!level) |
5658 | break; | |
b4ce94de | 5659 | |
8e73f275 | 5660 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5661 | 0, &key, 0); |
8e73f275 CM |
5662 | if (ret == -EAGAIN) |
5663 | goto again; | |
5664 | ||
76a05b35 | 5665 | if (ret < 0) { |
b3b4aa74 | 5666 | btrfs_release_path(path); |
76a05b35 CM |
5667 | goto done; |
5668 | } | |
5669 | ||
5cd57b2c | 5670 | if (!path->skip_locking) { |
bd681513 | 5671 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5672 | if (!ret) { |
5673 | btrfs_set_path_blocking(path); | |
bd681513 | 5674 | btrfs_tree_read_lock(next); |
31533fb2 | 5675 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5676 | BTRFS_READ_LOCK); |
5677 | } | |
31533fb2 | 5678 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5679 | } |
d97e63b6 | 5680 | } |
8e73f275 | 5681 | ret = 0; |
925baedd | 5682 | done: |
f7c79f30 | 5683 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5684 | path->leave_spinning = old_spinning; |
5685 | if (!old_spinning) | |
5686 | btrfs_set_path_blocking(path); | |
5687 | ||
5688 | return ret; | |
d97e63b6 | 5689 | } |
0b86a832 | 5690 | |
3f157a2f CM |
5691 | /* |
5692 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5693 | * searching until it gets past min_objectid or finds an item of 'type' | |
5694 | * | |
5695 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5696 | */ | |
0b86a832 CM |
5697 | int btrfs_previous_item(struct btrfs_root *root, |
5698 | struct btrfs_path *path, u64 min_objectid, | |
5699 | int type) | |
5700 | { | |
5701 | struct btrfs_key found_key; | |
5702 | struct extent_buffer *leaf; | |
e02119d5 | 5703 | u32 nritems; |
0b86a832 CM |
5704 | int ret; |
5705 | ||
d397712b | 5706 | while (1) { |
0b86a832 | 5707 | if (path->slots[0] == 0) { |
b4ce94de | 5708 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5709 | ret = btrfs_prev_leaf(root, path); |
5710 | if (ret != 0) | |
5711 | return ret; | |
5712 | } else { | |
5713 | path->slots[0]--; | |
5714 | } | |
5715 | leaf = path->nodes[0]; | |
e02119d5 CM |
5716 | nritems = btrfs_header_nritems(leaf); |
5717 | if (nritems == 0) | |
5718 | return 1; | |
5719 | if (path->slots[0] == nritems) | |
5720 | path->slots[0]--; | |
5721 | ||
0b86a832 | 5722 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5723 | if (found_key.objectid < min_objectid) |
5724 | break; | |
0a4eefbb YZ |
5725 | if (found_key.type == type) |
5726 | return 0; | |
e02119d5 CM |
5727 | if (found_key.objectid == min_objectid && |
5728 | found_key.type < type) | |
5729 | break; | |
0b86a832 CM |
5730 | } |
5731 | return 1; | |
5732 | } |