]>
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); | |
143bede5 | 40 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
0e411ece | 41 | struct btrfs_path *path, 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); | |
44 | struct extent_buffer *read_old_tree_block(struct btrfs_root *root, u64 bytenr, | |
45 | u32 blocksize, u64 parent_transid, | |
46 | u64 time_seq); | |
47 | struct extent_buffer *btrfs_find_old_tree_block(struct btrfs_root *root, | |
48 | u64 bytenr, u32 blocksize, | |
49 | u64 time_seq); | |
d97e63b6 | 50 | |
df24a2b9 | 51 | struct btrfs_path *btrfs_alloc_path(void) |
2c90e5d6 | 52 | { |
df24a2b9 | 53 | struct btrfs_path *path; |
e00f7308 | 54 | path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS); |
df24a2b9 | 55 | return path; |
2c90e5d6 CM |
56 | } |
57 | ||
b4ce94de CM |
58 | /* |
59 | * set all locked nodes in the path to blocking locks. This should | |
60 | * be done before scheduling | |
61 | */ | |
62 | noinline void btrfs_set_path_blocking(struct btrfs_path *p) | |
63 | { | |
64 | int i; | |
65 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { | |
bd681513 CM |
66 | if (!p->nodes[i] || !p->locks[i]) |
67 | continue; | |
68 | btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
69 | if (p->locks[i] == BTRFS_READ_LOCK) | |
70 | p->locks[i] = BTRFS_READ_LOCK_BLOCKING; | |
71 | else if (p->locks[i] == BTRFS_WRITE_LOCK) | |
72 | p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING; | |
b4ce94de CM |
73 | } |
74 | } | |
75 | ||
76 | /* | |
77 | * reset all the locked nodes in the patch to spinning locks. | |
4008c04a CM |
78 | * |
79 | * held is used to keep lockdep happy, when lockdep is enabled | |
80 | * we set held to a blocking lock before we go around and | |
81 | * retake all the spinlocks in the path. You can safely use NULL | |
82 | * for held | |
b4ce94de | 83 | */ |
4008c04a | 84 | noinline void btrfs_clear_path_blocking(struct btrfs_path *p, |
bd681513 | 85 | struct extent_buffer *held, int held_rw) |
b4ce94de CM |
86 | { |
87 | int i; | |
4008c04a CM |
88 | |
89 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
90 | /* lockdep really cares that we take all of these spinlocks | |
91 | * in the right order. If any of the locks in the path are not | |
92 | * currently blocking, it is going to complain. So, make really | |
93 | * really sure by forcing the path to blocking before we clear | |
94 | * the path blocking. | |
95 | */ | |
bd681513 CM |
96 | if (held) { |
97 | btrfs_set_lock_blocking_rw(held, held_rw); | |
98 | if (held_rw == BTRFS_WRITE_LOCK) | |
99 | held_rw = BTRFS_WRITE_LOCK_BLOCKING; | |
100 | else if (held_rw == BTRFS_READ_LOCK) | |
101 | held_rw = BTRFS_READ_LOCK_BLOCKING; | |
102 | } | |
4008c04a CM |
103 | btrfs_set_path_blocking(p); |
104 | #endif | |
105 | ||
106 | for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) { | |
bd681513 CM |
107 | if (p->nodes[i] && p->locks[i]) { |
108 | btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]); | |
109 | if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING) | |
110 | p->locks[i] = BTRFS_WRITE_LOCK; | |
111 | else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING) | |
112 | p->locks[i] = BTRFS_READ_LOCK; | |
113 | } | |
b4ce94de | 114 | } |
4008c04a CM |
115 | |
116 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
117 | if (held) | |
bd681513 | 118 | btrfs_clear_lock_blocking_rw(held, held_rw); |
4008c04a | 119 | #endif |
b4ce94de CM |
120 | } |
121 | ||
d352ac68 | 122 | /* this also releases the path */ |
df24a2b9 | 123 | void btrfs_free_path(struct btrfs_path *p) |
be0e5c09 | 124 | { |
ff175d57 JJ |
125 | if (!p) |
126 | return; | |
b3b4aa74 | 127 | btrfs_release_path(p); |
df24a2b9 | 128 | kmem_cache_free(btrfs_path_cachep, p); |
be0e5c09 CM |
129 | } |
130 | ||
d352ac68 CM |
131 | /* |
132 | * path release drops references on the extent buffers in the path | |
133 | * and it drops any locks held by this path | |
134 | * | |
135 | * It is safe to call this on paths that no locks or extent buffers held. | |
136 | */ | |
b3b4aa74 | 137 | noinline void btrfs_release_path(struct btrfs_path *p) |
eb60ceac CM |
138 | { |
139 | int i; | |
a2135011 | 140 | |
234b63a0 | 141 | for (i = 0; i < BTRFS_MAX_LEVEL; i++) { |
3f157a2f | 142 | p->slots[i] = 0; |
eb60ceac | 143 | if (!p->nodes[i]) |
925baedd CM |
144 | continue; |
145 | if (p->locks[i]) { | |
bd681513 | 146 | btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]); |
925baedd CM |
147 | p->locks[i] = 0; |
148 | } | |
5f39d397 | 149 | free_extent_buffer(p->nodes[i]); |
3f157a2f | 150 | p->nodes[i] = NULL; |
eb60ceac CM |
151 | } |
152 | } | |
153 | ||
d352ac68 CM |
154 | /* |
155 | * safely gets a reference on the root node of a tree. A lock | |
156 | * is not taken, so a concurrent writer may put a different node | |
157 | * at the root of the tree. See btrfs_lock_root_node for the | |
158 | * looping required. | |
159 | * | |
160 | * The extent buffer returned by this has a reference taken, so | |
161 | * it won't disappear. It may stop being the root of the tree | |
162 | * at any time because there are no locks held. | |
163 | */ | |
925baedd CM |
164 | struct extent_buffer *btrfs_root_node(struct btrfs_root *root) |
165 | { | |
166 | struct extent_buffer *eb; | |
240f62c8 | 167 | |
3083ee2e JB |
168 | while (1) { |
169 | rcu_read_lock(); | |
170 | eb = rcu_dereference(root->node); | |
171 | ||
172 | /* | |
173 | * RCU really hurts here, we could free up the root node because | |
174 | * it was cow'ed but we may not get the new root node yet so do | |
175 | * the inc_not_zero dance and if it doesn't work then | |
176 | * synchronize_rcu and try again. | |
177 | */ | |
178 | if (atomic_inc_not_zero(&eb->refs)) { | |
179 | rcu_read_unlock(); | |
180 | break; | |
181 | } | |
182 | rcu_read_unlock(); | |
183 | synchronize_rcu(); | |
184 | } | |
925baedd CM |
185 | return eb; |
186 | } | |
187 | ||
d352ac68 CM |
188 | /* loop around taking references on and locking the root node of the |
189 | * tree until you end up with a lock on the root. A locked buffer | |
190 | * is returned, with a reference held. | |
191 | */ | |
925baedd CM |
192 | struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root) |
193 | { | |
194 | struct extent_buffer *eb; | |
195 | ||
d397712b | 196 | while (1) { |
925baedd CM |
197 | eb = btrfs_root_node(root); |
198 | btrfs_tree_lock(eb); | |
240f62c8 | 199 | if (eb == root->node) |
925baedd | 200 | break; |
925baedd CM |
201 | btrfs_tree_unlock(eb); |
202 | free_extent_buffer(eb); | |
203 | } | |
204 | return eb; | |
205 | } | |
206 | ||
bd681513 CM |
207 | /* loop around taking references on and locking the root node of the |
208 | * tree until you end up with a lock on the root. A locked buffer | |
209 | * is returned, with a reference held. | |
210 | */ | |
211 | struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root) | |
212 | { | |
213 | struct extent_buffer *eb; | |
214 | ||
215 | while (1) { | |
216 | eb = btrfs_root_node(root); | |
217 | btrfs_tree_read_lock(eb); | |
218 | if (eb == root->node) | |
219 | break; | |
220 | btrfs_tree_read_unlock(eb); | |
221 | free_extent_buffer(eb); | |
222 | } | |
223 | return eb; | |
224 | } | |
225 | ||
d352ac68 CM |
226 | /* cowonly root (everything not a reference counted cow subvolume), just get |
227 | * put onto a simple dirty list. transaction.c walks this to make sure they | |
228 | * get properly updated on disk. | |
229 | */ | |
0b86a832 CM |
230 | static void add_root_to_dirty_list(struct btrfs_root *root) |
231 | { | |
e5846fc6 | 232 | spin_lock(&root->fs_info->trans_lock); |
0b86a832 CM |
233 | if (root->track_dirty && list_empty(&root->dirty_list)) { |
234 | list_add(&root->dirty_list, | |
235 | &root->fs_info->dirty_cowonly_roots); | |
236 | } | |
e5846fc6 | 237 | spin_unlock(&root->fs_info->trans_lock); |
0b86a832 CM |
238 | } |
239 | ||
d352ac68 CM |
240 | /* |
241 | * used by snapshot creation to make a copy of a root for a tree with | |
242 | * a given objectid. The buffer with the new root node is returned in | |
243 | * cow_ret, and this func returns zero on success or a negative error code. | |
244 | */ | |
be20aa9d CM |
245 | int btrfs_copy_root(struct btrfs_trans_handle *trans, |
246 | struct btrfs_root *root, | |
247 | struct extent_buffer *buf, | |
248 | struct extent_buffer **cow_ret, u64 new_root_objectid) | |
249 | { | |
250 | struct extent_buffer *cow; | |
be20aa9d CM |
251 | int ret = 0; |
252 | int level; | |
5d4f98a2 | 253 | struct btrfs_disk_key disk_key; |
be20aa9d CM |
254 | |
255 | WARN_ON(root->ref_cows && trans->transid != | |
256 | root->fs_info->running_transaction->transid); | |
257 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); | |
258 | ||
259 | level = btrfs_header_level(buf); | |
5d4f98a2 YZ |
260 | if (level == 0) |
261 | btrfs_item_key(buf, &disk_key, 0); | |
262 | else | |
263 | btrfs_node_key(buf, &disk_key, 0); | |
31840ae1 | 264 | |
5d4f98a2 YZ |
265 | cow = btrfs_alloc_free_block(trans, root, buf->len, 0, |
266 | new_root_objectid, &disk_key, level, | |
5581a51a | 267 | buf->start, 0); |
5d4f98a2 | 268 | if (IS_ERR(cow)) |
be20aa9d CM |
269 | return PTR_ERR(cow); |
270 | ||
271 | copy_extent_buffer(cow, buf, 0, 0, cow->len); | |
272 | btrfs_set_header_bytenr(cow, cow->start); | |
273 | btrfs_set_header_generation(cow, trans->transid); | |
5d4f98a2 YZ |
274 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
275 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
276 | BTRFS_HEADER_FLAG_RELOC); | |
277 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) | |
278 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
279 | else | |
280 | btrfs_set_header_owner(cow, new_root_objectid); | |
be20aa9d | 281 | |
2b82032c YZ |
282 | write_extent_buffer(cow, root->fs_info->fsid, |
283 | (unsigned long)btrfs_header_fsid(cow), | |
284 | BTRFS_FSID_SIZE); | |
285 | ||
be20aa9d | 286 | WARN_ON(btrfs_header_generation(buf) > trans->transid); |
5d4f98a2 | 287 | if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID) |
66d7e7f0 | 288 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 289 | else |
66d7e7f0 | 290 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
4aec2b52 | 291 | |
be20aa9d CM |
292 | if (ret) |
293 | return ret; | |
294 | ||
295 | btrfs_mark_buffer_dirty(cow); | |
296 | *cow_ret = cow; | |
297 | return 0; | |
298 | } | |
299 | ||
bd989ba3 JS |
300 | enum mod_log_op { |
301 | MOD_LOG_KEY_REPLACE, | |
302 | MOD_LOG_KEY_ADD, | |
303 | MOD_LOG_KEY_REMOVE, | |
304 | MOD_LOG_KEY_REMOVE_WHILE_FREEING, | |
305 | MOD_LOG_KEY_REMOVE_WHILE_MOVING, | |
306 | MOD_LOG_MOVE_KEYS, | |
307 | MOD_LOG_ROOT_REPLACE, | |
308 | }; | |
309 | ||
310 | struct tree_mod_move { | |
311 | int dst_slot; | |
312 | int nr_items; | |
313 | }; | |
314 | ||
315 | struct tree_mod_root { | |
316 | u64 logical; | |
317 | u8 level; | |
318 | }; | |
319 | ||
320 | struct tree_mod_elem { | |
321 | struct rb_node node; | |
322 | u64 index; /* shifted logical */ | |
097b8a7c | 323 | u64 seq; |
bd989ba3 JS |
324 | enum mod_log_op op; |
325 | ||
326 | /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */ | |
327 | int slot; | |
328 | ||
329 | /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */ | |
330 | u64 generation; | |
331 | ||
332 | /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */ | |
333 | struct btrfs_disk_key key; | |
334 | u64 blockptr; | |
335 | ||
336 | /* this is used for op == MOD_LOG_MOVE_KEYS */ | |
337 | struct tree_mod_move move; | |
338 | ||
339 | /* this is used for op == MOD_LOG_ROOT_REPLACE */ | |
340 | struct tree_mod_root old_root; | |
341 | }; | |
342 | ||
097b8a7c | 343 | static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info) |
bd989ba3 | 344 | { |
097b8a7c | 345 | read_lock(&fs_info->tree_mod_log_lock); |
bd989ba3 JS |
346 | } |
347 | ||
097b8a7c JS |
348 | static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info) |
349 | { | |
350 | read_unlock(&fs_info->tree_mod_log_lock); | |
351 | } | |
352 | ||
353 | static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info) | |
354 | { | |
355 | write_lock(&fs_info->tree_mod_log_lock); | |
356 | } | |
357 | ||
358 | static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info) | |
359 | { | |
360 | write_unlock(&fs_info->tree_mod_log_lock); | |
361 | } | |
362 | ||
363 | /* | |
364 | * This adds a new blocker to the tree mod log's blocker list if the @elem | |
365 | * passed does not already have a sequence number set. So when a caller expects | |
366 | * to record tree modifications, it should ensure to set elem->seq to zero | |
367 | * before calling btrfs_get_tree_mod_seq. | |
368 | * Returns a fresh, unused tree log modification sequence number, even if no new | |
369 | * blocker was added. | |
370 | */ | |
371 | u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
372 | struct seq_list *elem) | |
bd989ba3 | 373 | { |
097b8a7c JS |
374 | u64 seq; |
375 | ||
376 | tree_mod_log_write_lock(fs_info); | |
bd989ba3 | 377 | spin_lock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
378 | if (!elem->seq) { |
379 | elem->seq = btrfs_inc_tree_mod_seq(fs_info); | |
380 | list_add_tail(&elem->list, &fs_info->tree_mod_seq_list); | |
381 | } | |
382 | seq = btrfs_inc_tree_mod_seq(fs_info); | |
bd989ba3 | 383 | spin_unlock(&fs_info->tree_mod_seq_lock); |
097b8a7c JS |
384 | tree_mod_log_write_unlock(fs_info); |
385 | ||
386 | return seq; | |
bd989ba3 JS |
387 | } |
388 | ||
389 | void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info, | |
390 | struct seq_list *elem) | |
391 | { | |
392 | struct rb_root *tm_root; | |
393 | struct rb_node *node; | |
394 | struct rb_node *next; | |
395 | struct seq_list *cur_elem; | |
396 | struct tree_mod_elem *tm; | |
397 | u64 min_seq = (u64)-1; | |
398 | u64 seq_putting = elem->seq; | |
399 | ||
400 | if (!seq_putting) | |
401 | return; | |
402 | ||
bd989ba3 JS |
403 | spin_lock(&fs_info->tree_mod_seq_lock); |
404 | list_del(&elem->list); | |
097b8a7c | 405 | elem->seq = 0; |
bd989ba3 JS |
406 | |
407 | list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) { | |
097b8a7c | 408 | if (cur_elem->seq < min_seq) { |
bd989ba3 JS |
409 | if (seq_putting > cur_elem->seq) { |
410 | /* | |
411 | * blocker with lower sequence number exists, we | |
412 | * cannot remove anything from the log | |
413 | */ | |
097b8a7c JS |
414 | spin_unlock(&fs_info->tree_mod_seq_lock); |
415 | return; | |
bd989ba3 JS |
416 | } |
417 | min_seq = cur_elem->seq; | |
418 | } | |
419 | } | |
097b8a7c JS |
420 | spin_unlock(&fs_info->tree_mod_seq_lock); |
421 | ||
bd989ba3 JS |
422 | /* |
423 | * anything that's lower than the lowest existing (read: blocked) | |
424 | * sequence number can be removed from the tree. | |
425 | */ | |
097b8a7c | 426 | tree_mod_log_write_lock(fs_info); |
bd989ba3 JS |
427 | tm_root = &fs_info->tree_mod_log; |
428 | for (node = rb_first(tm_root); node; node = next) { | |
429 | next = rb_next(node); | |
430 | tm = container_of(node, struct tree_mod_elem, node); | |
097b8a7c | 431 | if (tm->seq > min_seq) |
bd989ba3 JS |
432 | continue; |
433 | rb_erase(node, tm_root); | |
bd989ba3 JS |
434 | kfree(tm); |
435 | } | |
097b8a7c | 436 | tree_mod_log_write_unlock(fs_info); |
bd989ba3 JS |
437 | } |
438 | ||
439 | /* | |
440 | * key order of the log: | |
441 | * index -> sequence | |
442 | * | |
443 | * the index is the shifted logical of the *new* root node for root replace | |
444 | * operations, or the shifted logical of the affected block for all other | |
445 | * operations. | |
446 | */ | |
447 | static noinline int | |
448 | __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm) | |
449 | { | |
450 | struct rb_root *tm_root; | |
451 | struct rb_node **new; | |
452 | struct rb_node *parent = NULL; | |
453 | struct tree_mod_elem *cur; | |
bd989ba3 | 454 | |
097b8a7c | 455 | BUG_ON(!tm || !tm->seq); |
bd989ba3 | 456 | |
bd989ba3 JS |
457 | tm_root = &fs_info->tree_mod_log; |
458 | new = &tm_root->rb_node; | |
459 | while (*new) { | |
460 | cur = container_of(*new, struct tree_mod_elem, node); | |
461 | parent = *new; | |
462 | if (cur->index < tm->index) | |
463 | new = &((*new)->rb_left); | |
464 | else if (cur->index > tm->index) | |
465 | new = &((*new)->rb_right); | |
097b8a7c | 466 | else if (cur->seq < tm->seq) |
bd989ba3 | 467 | new = &((*new)->rb_left); |
097b8a7c | 468 | else if (cur->seq > tm->seq) |
bd989ba3 JS |
469 | new = &((*new)->rb_right); |
470 | else { | |
471 | kfree(tm); | |
097b8a7c | 472 | return -EEXIST; |
bd989ba3 JS |
473 | } |
474 | } | |
475 | ||
476 | rb_link_node(&tm->node, parent, new); | |
477 | rb_insert_color(&tm->node, tm_root); | |
097b8a7c | 478 | return 0; |
bd989ba3 JS |
479 | } |
480 | ||
097b8a7c JS |
481 | /* |
482 | * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it | |
483 | * returns zero with the tree_mod_log_lock acquired. The caller must hold | |
484 | * this until all tree mod log insertions are recorded in the rb tree and then | |
485 | * call tree_mod_log_write_unlock() to release. | |
486 | */ | |
e9b7fd4d JS |
487 | static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info, |
488 | struct extent_buffer *eb) { | |
489 | smp_mb(); | |
490 | if (list_empty(&(fs_info)->tree_mod_seq_list)) | |
491 | return 1; | |
097b8a7c JS |
492 | if (eb && btrfs_header_level(eb) == 0) |
493 | return 1; | |
494 | ||
495 | tree_mod_log_write_lock(fs_info); | |
496 | if (list_empty(&fs_info->tree_mod_seq_list)) { | |
497 | /* | |
498 | * someone emptied the list while we were waiting for the lock. | |
499 | * we must not add to the list when no blocker exists. | |
500 | */ | |
501 | tree_mod_log_write_unlock(fs_info); | |
e9b7fd4d | 502 | return 1; |
097b8a7c JS |
503 | } |
504 | ||
e9b7fd4d JS |
505 | return 0; |
506 | } | |
507 | ||
3310c36e | 508 | /* |
097b8a7c | 509 | * This allocates memory and gets a tree modification sequence number. |
3310c36e | 510 | * |
097b8a7c JS |
511 | * Returns <0 on error. |
512 | * Returns >0 (the added sequence number) on success. | |
3310c36e | 513 | */ |
926dd8a6 JS |
514 | static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags, |
515 | struct tree_mod_elem **tm_ret) | |
bd989ba3 JS |
516 | { |
517 | struct tree_mod_elem *tm; | |
bd989ba3 | 518 | |
097b8a7c JS |
519 | /* |
520 | * once we switch from spin locks to something different, we should | |
521 | * honor the flags parameter here. | |
522 | */ | |
523 | tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC); | |
bd989ba3 JS |
524 | if (!tm) |
525 | return -ENOMEM; | |
526 | ||
097b8a7c JS |
527 | tm->seq = btrfs_inc_tree_mod_seq(fs_info); |
528 | return tm->seq; | |
bd989ba3 JS |
529 | } |
530 | ||
097b8a7c JS |
531 | static inline int |
532 | __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, | |
533 | struct extent_buffer *eb, int slot, | |
534 | enum mod_log_op op, gfp_t flags) | |
bd989ba3 | 535 | { |
bd989ba3 | 536 | int ret; |
097b8a7c | 537 | struct tree_mod_elem *tm; |
bd989ba3 JS |
538 | |
539 | ret = tree_mod_alloc(fs_info, flags, &tm); | |
097b8a7c | 540 | if (ret < 0) |
bd989ba3 JS |
541 | return ret; |
542 | ||
543 | tm->index = eb->start >> PAGE_CACHE_SHIFT; | |
544 | if (op != MOD_LOG_KEY_ADD) { | |
545 | btrfs_node_key(eb, &tm->key, slot); | |
546 | tm->blockptr = btrfs_node_blockptr(eb, slot); | |
547 | } | |
548 | tm->op = op; | |
549 | tm->slot = slot; | |
550 | tm->generation = btrfs_node_ptr_generation(eb, slot); | |
551 | ||
097b8a7c JS |
552 | return __tree_mod_log_insert(fs_info, tm); |
553 | } | |
554 | ||
555 | static noinline int | |
556 | tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info, | |
557 | struct extent_buffer *eb, int slot, | |
558 | enum mod_log_op op, gfp_t flags) | |
559 | { | |
560 | int ret; | |
561 | ||
562 | if (tree_mod_dont_log(fs_info, eb)) | |
563 | return 0; | |
564 | ||
565 | ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags); | |
566 | ||
567 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 568 | return ret; |
bd989ba3 JS |
569 | } |
570 | ||
571 | static noinline int | |
572 | tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
573 | int slot, enum mod_log_op op) | |
574 | { | |
575 | return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS); | |
576 | } | |
577 | ||
097b8a7c JS |
578 | static noinline int |
579 | tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info, | |
580 | struct extent_buffer *eb, int slot, | |
581 | enum mod_log_op op) | |
582 | { | |
583 | return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS); | |
584 | } | |
585 | ||
bd989ba3 JS |
586 | static noinline int |
587 | tree_mod_log_insert_move(struct btrfs_fs_info *fs_info, | |
588 | struct extent_buffer *eb, int dst_slot, int src_slot, | |
589 | int nr_items, gfp_t flags) | |
590 | { | |
591 | struct tree_mod_elem *tm; | |
592 | int ret; | |
593 | int i; | |
594 | ||
f395694c JS |
595 | if (tree_mod_dont_log(fs_info, eb)) |
596 | return 0; | |
bd989ba3 | 597 | |
01763a2e JS |
598 | /* |
599 | * When we override something during the move, we log these removals. | |
600 | * This can only happen when we move towards the beginning of the | |
601 | * buffer, i.e. dst_slot < src_slot. | |
602 | */ | |
bd989ba3 | 603 | for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) { |
097b8a7c | 604 | ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot, |
bd989ba3 JS |
605 | MOD_LOG_KEY_REMOVE_WHILE_MOVING); |
606 | BUG_ON(ret < 0); | |
607 | } | |
608 | ||
f395694c | 609 | ret = tree_mod_alloc(fs_info, flags, &tm); |
097b8a7c JS |
610 | if (ret < 0) |
611 | goto out; | |
f395694c | 612 | |
bd989ba3 JS |
613 | tm->index = eb->start >> PAGE_CACHE_SHIFT; |
614 | tm->slot = src_slot; | |
615 | tm->move.dst_slot = dst_slot; | |
616 | tm->move.nr_items = nr_items; | |
617 | tm->op = MOD_LOG_MOVE_KEYS; | |
618 | ||
3310c36e | 619 | ret = __tree_mod_log_insert(fs_info, tm); |
097b8a7c JS |
620 | out: |
621 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 622 | return ret; |
bd989ba3 JS |
623 | } |
624 | ||
097b8a7c JS |
625 | static inline void |
626 | __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
627 | { | |
628 | int i; | |
629 | u32 nritems; | |
630 | int ret; | |
631 | ||
b12a3b1e CM |
632 | if (btrfs_header_level(eb) == 0) |
633 | return; | |
634 | ||
097b8a7c JS |
635 | nritems = btrfs_header_nritems(eb); |
636 | for (i = nritems - 1; i >= 0; i--) { | |
637 | ret = tree_mod_log_insert_key_locked(fs_info, eb, i, | |
638 | MOD_LOG_KEY_REMOVE_WHILE_FREEING); | |
639 | BUG_ON(ret < 0); | |
640 | } | |
641 | } | |
642 | ||
bd989ba3 JS |
643 | static noinline int |
644 | tree_mod_log_insert_root(struct btrfs_fs_info *fs_info, | |
645 | struct extent_buffer *old_root, | |
646 | struct extent_buffer *new_root, gfp_t flags) | |
647 | { | |
648 | struct tree_mod_elem *tm; | |
649 | int ret; | |
650 | ||
097b8a7c JS |
651 | if (tree_mod_dont_log(fs_info, NULL)) |
652 | return 0; | |
653 | ||
bd989ba3 | 654 | ret = tree_mod_alloc(fs_info, flags, &tm); |
097b8a7c JS |
655 | if (ret < 0) |
656 | goto out; | |
bd989ba3 JS |
657 | |
658 | tm->index = new_root->start >> PAGE_CACHE_SHIFT; | |
659 | tm->old_root.logical = old_root->start; | |
660 | tm->old_root.level = btrfs_header_level(old_root); | |
661 | tm->generation = btrfs_header_generation(old_root); | |
662 | tm->op = MOD_LOG_ROOT_REPLACE; | |
663 | ||
3310c36e | 664 | ret = __tree_mod_log_insert(fs_info, tm); |
097b8a7c JS |
665 | out: |
666 | tree_mod_log_write_unlock(fs_info); | |
3310c36e | 667 | return ret; |
bd989ba3 JS |
668 | } |
669 | ||
670 | static struct tree_mod_elem * | |
671 | __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq, | |
672 | int smallest) | |
673 | { | |
674 | struct rb_root *tm_root; | |
675 | struct rb_node *node; | |
676 | struct tree_mod_elem *cur = NULL; | |
677 | struct tree_mod_elem *found = NULL; | |
678 | u64 index = start >> PAGE_CACHE_SHIFT; | |
679 | ||
097b8a7c | 680 | tree_mod_log_read_lock(fs_info); |
bd989ba3 JS |
681 | tm_root = &fs_info->tree_mod_log; |
682 | node = tm_root->rb_node; | |
683 | while (node) { | |
684 | cur = container_of(node, struct tree_mod_elem, node); | |
685 | if (cur->index < index) { | |
686 | node = node->rb_left; | |
687 | } else if (cur->index > index) { | |
688 | node = node->rb_right; | |
097b8a7c | 689 | } else if (cur->seq < min_seq) { |
bd989ba3 JS |
690 | node = node->rb_left; |
691 | } else if (!smallest) { | |
692 | /* we want the node with the highest seq */ | |
693 | if (found) | |
097b8a7c | 694 | BUG_ON(found->seq > cur->seq); |
bd989ba3 JS |
695 | found = cur; |
696 | node = node->rb_left; | |
097b8a7c | 697 | } else if (cur->seq > min_seq) { |
bd989ba3 JS |
698 | /* we want the node with the smallest seq */ |
699 | if (found) | |
097b8a7c | 700 | BUG_ON(found->seq < cur->seq); |
bd989ba3 JS |
701 | found = cur; |
702 | node = node->rb_right; | |
703 | } else { | |
704 | found = cur; | |
705 | break; | |
706 | } | |
707 | } | |
097b8a7c | 708 | tree_mod_log_read_unlock(fs_info); |
bd989ba3 JS |
709 | |
710 | return found; | |
711 | } | |
712 | ||
713 | /* | |
714 | * this returns the element from the log with the smallest time sequence | |
715 | * value that's in the log (the oldest log item). any element with a time | |
716 | * sequence lower than min_seq will be ignored. | |
717 | */ | |
718 | static struct tree_mod_elem * | |
719 | tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start, | |
720 | u64 min_seq) | |
721 | { | |
722 | return __tree_mod_log_search(fs_info, start, min_seq, 1); | |
723 | } | |
724 | ||
725 | /* | |
726 | * this returns the element from the log with the largest time sequence | |
727 | * value that's in the log (the most recent log item). any element with | |
728 | * a time sequence lower than min_seq will be ignored. | |
729 | */ | |
730 | static struct tree_mod_elem * | |
731 | tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq) | |
732 | { | |
733 | return __tree_mod_log_search(fs_info, start, min_seq, 0); | |
734 | } | |
735 | ||
097b8a7c | 736 | static noinline void |
bd989ba3 JS |
737 | tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, |
738 | struct extent_buffer *src, unsigned long dst_offset, | |
739 | unsigned long src_offset, int nr_items) | |
740 | { | |
741 | int ret; | |
742 | int i; | |
743 | ||
e9b7fd4d | 744 | if (tree_mod_dont_log(fs_info, NULL)) |
bd989ba3 JS |
745 | return; |
746 | ||
097b8a7c JS |
747 | if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) { |
748 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 | 749 | return; |
097b8a7c | 750 | } |
bd989ba3 | 751 | |
bd989ba3 | 752 | for (i = 0; i < nr_items; i++) { |
097b8a7c JS |
753 | ret = tree_mod_log_insert_key_locked(fs_info, src, |
754 | i + src_offset, | |
755 | MOD_LOG_KEY_REMOVE); | |
bd989ba3 | 756 | BUG_ON(ret < 0); |
097b8a7c JS |
757 | ret = tree_mod_log_insert_key_locked(fs_info, dst, |
758 | i + dst_offset, | |
759 | MOD_LOG_KEY_ADD); | |
bd989ba3 JS |
760 | BUG_ON(ret < 0); |
761 | } | |
097b8a7c JS |
762 | |
763 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 JS |
764 | } |
765 | ||
766 | static inline void | |
767 | tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst, | |
768 | int dst_offset, int src_offset, int nr_items) | |
769 | { | |
770 | int ret; | |
771 | ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset, | |
772 | nr_items, GFP_NOFS); | |
773 | BUG_ON(ret < 0); | |
774 | } | |
775 | ||
097b8a7c | 776 | static noinline void |
bd989ba3 | 777 | tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info, |
32adf090 | 778 | struct extent_buffer *eb, int slot, int atomic) |
bd989ba3 JS |
779 | { |
780 | int ret; | |
781 | ||
782 | ret = tree_mod_log_insert_key_mask(fs_info, eb, slot, | |
783 | MOD_LOG_KEY_REPLACE, | |
784 | atomic ? GFP_ATOMIC : GFP_NOFS); | |
785 | BUG_ON(ret < 0); | |
786 | } | |
787 | ||
097b8a7c JS |
788 | static noinline void |
789 | tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb) | |
bd989ba3 | 790 | { |
e9b7fd4d | 791 | if (tree_mod_dont_log(fs_info, eb)) |
bd989ba3 JS |
792 | return; |
793 | ||
097b8a7c JS |
794 | __tree_mod_log_free_eb(fs_info, eb); |
795 | ||
796 | tree_mod_log_write_unlock(fs_info); | |
bd989ba3 JS |
797 | } |
798 | ||
097b8a7c | 799 | static noinline void |
bd989ba3 JS |
800 | tree_mod_log_set_root_pointer(struct btrfs_root *root, |
801 | struct extent_buffer *new_root_node) | |
802 | { | |
803 | int ret; | |
bd989ba3 JS |
804 | ret = tree_mod_log_insert_root(root->fs_info, root->node, |
805 | new_root_node, GFP_NOFS); | |
806 | BUG_ON(ret < 0); | |
807 | } | |
808 | ||
5d4f98a2 YZ |
809 | /* |
810 | * check if the tree block can be shared by multiple trees | |
811 | */ | |
812 | int btrfs_block_can_be_shared(struct btrfs_root *root, | |
813 | struct extent_buffer *buf) | |
814 | { | |
815 | /* | |
816 | * Tree blocks not in refernece counted trees and tree roots | |
817 | * are never shared. If a block was allocated after the last | |
818 | * snapshot and the block was not allocated by tree relocation, | |
819 | * we know the block is not shared. | |
820 | */ | |
821 | if (root->ref_cows && | |
822 | buf != root->node && buf != root->commit_root && | |
823 | (btrfs_header_generation(buf) <= | |
824 | btrfs_root_last_snapshot(&root->root_item) || | |
825 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC))) | |
826 | return 1; | |
827 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
828 | if (root->ref_cows && | |
829 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
830 | return 1; | |
831 | #endif | |
832 | return 0; | |
833 | } | |
834 | ||
835 | static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans, | |
836 | struct btrfs_root *root, | |
837 | struct extent_buffer *buf, | |
f0486c68 YZ |
838 | struct extent_buffer *cow, |
839 | int *last_ref) | |
5d4f98a2 YZ |
840 | { |
841 | u64 refs; | |
842 | u64 owner; | |
843 | u64 flags; | |
844 | u64 new_flags = 0; | |
845 | int ret; | |
846 | ||
847 | /* | |
848 | * Backrefs update rules: | |
849 | * | |
850 | * Always use full backrefs for extent pointers in tree block | |
851 | * allocated by tree relocation. | |
852 | * | |
853 | * If a shared tree block is no longer referenced by its owner | |
854 | * tree (btrfs_header_owner(buf) == root->root_key.objectid), | |
855 | * use full backrefs for extent pointers in tree block. | |
856 | * | |
857 | * If a tree block is been relocating | |
858 | * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID), | |
859 | * use full backrefs for extent pointers in tree block. | |
860 | * The reason for this is some operations (such as drop tree) | |
861 | * are only allowed for blocks use full backrefs. | |
862 | */ | |
863 | ||
864 | if (btrfs_block_can_be_shared(root, buf)) { | |
865 | ret = btrfs_lookup_extent_info(trans, root, buf->start, | |
866 | buf->len, &refs, &flags); | |
be1a5564 MF |
867 | if (ret) |
868 | return ret; | |
e5df9573 MF |
869 | if (refs == 0) { |
870 | ret = -EROFS; | |
871 | btrfs_std_error(root->fs_info, ret); | |
872 | return ret; | |
873 | } | |
5d4f98a2 YZ |
874 | } else { |
875 | refs = 1; | |
876 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || | |
877 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
878 | flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
879 | else | |
880 | flags = 0; | |
881 | } | |
882 | ||
883 | owner = btrfs_header_owner(buf); | |
884 | BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID && | |
885 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
886 | ||
887 | if (refs > 1) { | |
888 | if ((owner == root->root_key.objectid || | |
889 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) && | |
890 | !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) { | |
66d7e7f0 | 891 | ret = btrfs_inc_ref(trans, root, buf, 1, 1); |
79787eaa | 892 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
893 | |
894 | if (root->root_key.objectid == | |
895 | BTRFS_TREE_RELOC_OBJECTID) { | |
66d7e7f0 | 896 | ret = btrfs_dec_ref(trans, root, buf, 0, 1); |
79787eaa | 897 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 898 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
79787eaa | 899 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
900 | } |
901 | new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
902 | } else { | |
903 | ||
904 | if (root->root_key.objectid == | |
905 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 906 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 907 | else |
66d7e7f0 | 908 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 909 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 YZ |
910 | } |
911 | if (new_flags != 0) { | |
912 | ret = btrfs_set_disk_extent_flags(trans, root, | |
913 | buf->start, | |
914 | buf->len, | |
915 | new_flags, 0); | |
be1a5564 MF |
916 | if (ret) |
917 | return ret; | |
5d4f98a2 YZ |
918 | } |
919 | } else { | |
920 | if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
921 | if (root->root_key.objectid == | |
922 | BTRFS_TREE_RELOC_OBJECTID) | |
66d7e7f0 | 923 | ret = btrfs_inc_ref(trans, root, cow, 1, 1); |
5d4f98a2 | 924 | else |
66d7e7f0 | 925 | ret = btrfs_inc_ref(trans, root, cow, 0, 1); |
79787eaa | 926 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 927 | ret = btrfs_dec_ref(trans, root, buf, 1, 1); |
79787eaa | 928 | BUG_ON(ret); /* -ENOMEM */ |
5d4f98a2 | 929 | } |
ba1bfbd5 | 930 | tree_mod_log_free_eb(root->fs_info, buf); |
5d4f98a2 | 931 | clean_tree_block(trans, root, buf); |
f0486c68 | 932 | *last_ref = 1; |
5d4f98a2 YZ |
933 | } |
934 | return 0; | |
935 | } | |
936 | ||
d352ac68 | 937 | /* |
d397712b CM |
938 | * does the dirty work in cow of a single block. The parent block (if |
939 | * supplied) is updated to point to the new cow copy. The new buffer is marked | |
940 | * dirty and returned locked. If you modify the block it needs to be marked | |
941 | * dirty again. | |
d352ac68 CM |
942 | * |
943 | * search_start -- an allocation hint for the new block | |
944 | * | |
d397712b CM |
945 | * empty_size -- a hint that you plan on doing more cow. This is the size in |
946 | * bytes the allocator should try to find free next to the block it returns. | |
947 | * This is just a hint and may be ignored by the allocator. | |
d352ac68 | 948 | */ |
d397712b | 949 | static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
950 | struct btrfs_root *root, |
951 | struct extent_buffer *buf, | |
952 | struct extent_buffer *parent, int parent_slot, | |
953 | struct extent_buffer **cow_ret, | |
9fa8cfe7 | 954 | u64 search_start, u64 empty_size) |
02217ed2 | 955 | { |
5d4f98a2 | 956 | struct btrfs_disk_key disk_key; |
5f39d397 | 957 | struct extent_buffer *cow; |
be1a5564 | 958 | int level, ret; |
f0486c68 | 959 | int last_ref = 0; |
925baedd | 960 | int unlock_orig = 0; |
5d4f98a2 | 961 | u64 parent_start; |
7bb86316 | 962 | |
925baedd CM |
963 | if (*cow_ret == buf) |
964 | unlock_orig = 1; | |
965 | ||
b9447ef8 | 966 | btrfs_assert_tree_locked(buf); |
925baedd | 967 | |
7bb86316 CM |
968 | WARN_ON(root->ref_cows && trans->transid != |
969 | root->fs_info->running_transaction->transid); | |
6702ed49 | 970 | WARN_ON(root->ref_cows && trans->transid != root->last_trans); |
5f39d397 | 971 | |
7bb86316 | 972 | level = btrfs_header_level(buf); |
31840ae1 | 973 | |
5d4f98a2 YZ |
974 | if (level == 0) |
975 | btrfs_item_key(buf, &disk_key, 0); | |
976 | else | |
977 | btrfs_node_key(buf, &disk_key, 0); | |
978 | ||
979 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
980 | if (parent) | |
981 | parent_start = parent->start; | |
982 | else | |
983 | parent_start = 0; | |
984 | } else | |
985 | parent_start = 0; | |
986 | ||
987 | cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start, | |
988 | root->root_key.objectid, &disk_key, | |
5581a51a | 989 | level, search_start, empty_size); |
54aa1f4d CM |
990 | if (IS_ERR(cow)) |
991 | return PTR_ERR(cow); | |
6702ed49 | 992 | |
b4ce94de CM |
993 | /* cow is set to blocking by btrfs_init_new_buffer */ |
994 | ||
5f39d397 | 995 | copy_extent_buffer(cow, buf, 0, 0, cow->len); |
db94535d | 996 | btrfs_set_header_bytenr(cow, cow->start); |
5f39d397 | 997 | btrfs_set_header_generation(cow, trans->transid); |
5d4f98a2 YZ |
998 | btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV); |
999 | btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN | | |
1000 | BTRFS_HEADER_FLAG_RELOC); | |
1001 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) | |
1002 | btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC); | |
1003 | else | |
1004 | btrfs_set_header_owner(cow, root->root_key.objectid); | |
6702ed49 | 1005 | |
2b82032c YZ |
1006 | write_extent_buffer(cow, root->fs_info->fsid, |
1007 | (unsigned long)btrfs_header_fsid(cow), | |
1008 | BTRFS_FSID_SIZE); | |
1009 | ||
be1a5564 | 1010 | ret = update_ref_for_cow(trans, root, buf, cow, &last_ref); |
b68dc2a9 | 1011 | if (ret) { |
79787eaa | 1012 | btrfs_abort_transaction(trans, root, ret); |
b68dc2a9 MF |
1013 | return ret; |
1014 | } | |
1a40e23b | 1015 | |
3fd0a558 YZ |
1016 | if (root->ref_cows) |
1017 | btrfs_reloc_cow_block(trans, root, buf, cow); | |
1018 | ||
02217ed2 | 1019 | if (buf == root->node) { |
925baedd | 1020 | WARN_ON(parent && parent != buf); |
5d4f98a2 YZ |
1021 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
1022 | btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV) | |
1023 | parent_start = buf->start; | |
1024 | else | |
1025 | parent_start = 0; | |
925baedd | 1026 | |
5f39d397 | 1027 | extent_buffer_get(cow); |
f230475e | 1028 | tree_mod_log_set_root_pointer(root, cow); |
240f62c8 | 1029 | rcu_assign_pointer(root->node, cow); |
925baedd | 1030 | |
f0486c68 | 1031 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1032 | last_ref); |
5f39d397 | 1033 | free_extent_buffer(buf); |
0b86a832 | 1034 | add_root_to_dirty_list(root); |
02217ed2 | 1035 | } else { |
5d4f98a2 YZ |
1036 | if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) |
1037 | parent_start = parent->start; | |
1038 | else | |
1039 | parent_start = 0; | |
1040 | ||
1041 | WARN_ON(trans->transid != btrfs_header_generation(parent)); | |
f230475e JS |
1042 | tree_mod_log_insert_key(root->fs_info, parent, parent_slot, |
1043 | MOD_LOG_KEY_REPLACE); | |
5f39d397 | 1044 | btrfs_set_node_blockptr(parent, parent_slot, |
db94535d | 1045 | cow->start); |
74493f7a CM |
1046 | btrfs_set_node_ptr_generation(parent, parent_slot, |
1047 | trans->transid); | |
d6025579 | 1048 | btrfs_mark_buffer_dirty(parent); |
f0486c68 | 1049 | btrfs_free_tree_block(trans, root, buf, parent_start, |
5581a51a | 1050 | last_ref); |
02217ed2 | 1051 | } |
925baedd CM |
1052 | if (unlock_orig) |
1053 | btrfs_tree_unlock(buf); | |
3083ee2e | 1054 | free_extent_buffer_stale(buf); |
ccd467d6 | 1055 | btrfs_mark_buffer_dirty(cow); |
2c90e5d6 | 1056 | *cow_ret = cow; |
02217ed2 CM |
1057 | return 0; |
1058 | } | |
1059 | ||
5d9e75c4 JS |
1060 | /* |
1061 | * returns the logical address of the oldest predecessor of the given root. | |
1062 | * entries older than time_seq are ignored. | |
1063 | */ | |
1064 | static struct tree_mod_elem * | |
1065 | __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info, | |
1066 | struct btrfs_root *root, u64 time_seq) | |
1067 | { | |
1068 | struct tree_mod_elem *tm; | |
1069 | struct tree_mod_elem *found = NULL; | |
1070 | u64 root_logical = root->node->start; | |
1071 | int looped = 0; | |
1072 | ||
1073 | if (!time_seq) | |
1074 | return 0; | |
1075 | ||
1076 | /* | |
1077 | * the very last operation that's logged for a root is the replacement | |
1078 | * operation (if it is replaced at all). this has the index of the *new* | |
1079 | * root, making it the very first operation that's logged for this root. | |
1080 | */ | |
1081 | while (1) { | |
1082 | tm = tree_mod_log_search_oldest(fs_info, root_logical, | |
1083 | time_seq); | |
1084 | if (!looped && !tm) | |
1085 | return 0; | |
1086 | /* | |
28da9fb4 JS |
1087 | * if there are no tree operation for the oldest root, we simply |
1088 | * return it. this should only happen if that (old) root is at | |
1089 | * level 0. | |
5d9e75c4 | 1090 | */ |
28da9fb4 JS |
1091 | if (!tm) |
1092 | break; | |
5d9e75c4 | 1093 | |
28da9fb4 JS |
1094 | /* |
1095 | * if there's an operation that's not a root replacement, we | |
1096 | * found the oldest version of our root. normally, we'll find a | |
1097 | * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here. | |
1098 | */ | |
5d9e75c4 JS |
1099 | if (tm->op != MOD_LOG_ROOT_REPLACE) |
1100 | break; | |
1101 | ||
1102 | found = tm; | |
1103 | root_logical = tm->old_root.logical; | |
1104 | BUG_ON(root_logical == root->node->start); | |
1105 | looped = 1; | |
1106 | } | |
1107 | ||
a95236d9 JS |
1108 | /* if there's no old root to return, return what we found instead */ |
1109 | if (!found) | |
1110 | found = tm; | |
1111 | ||
5d9e75c4 JS |
1112 | return found; |
1113 | } | |
1114 | ||
1115 | /* | |
1116 | * tm is a pointer to the first operation to rewind within eb. then, all | |
1117 | * previous operations will be rewinded (until we reach something older than | |
1118 | * time_seq). | |
1119 | */ | |
1120 | static void | |
1121 | __tree_mod_log_rewind(struct extent_buffer *eb, u64 time_seq, | |
1122 | struct tree_mod_elem *first_tm) | |
1123 | { | |
1124 | u32 n; | |
1125 | struct rb_node *next; | |
1126 | struct tree_mod_elem *tm = first_tm; | |
1127 | unsigned long o_dst; | |
1128 | unsigned long o_src; | |
1129 | unsigned long p_size = sizeof(struct btrfs_key_ptr); | |
1130 | ||
1131 | n = btrfs_header_nritems(eb); | |
097b8a7c | 1132 | while (tm && tm->seq >= time_seq) { |
5d9e75c4 JS |
1133 | /* |
1134 | * all the operations are recorded with the operator used for | |
1135 | * the modification. as we're going backwards, we do the | |
1136 | * opposite of each operation here. | |
1137 | */ | |
1138 | switch (tm->op) { | |
1139 | case MOD_LOG_KEY_REMOVE_WHILE_FREEING: | |
1140 | BUG_ON(tm->slot < n); | |
1c697d4a | 1141 | /* Fallthrough */ |
95c80bb1 | 1142 | case MOD_LOG_KEY_REMOVE_WHILE_MOVING: |
4c3e6969 | 1143 | case MOD_LOG_KEY_REMOVE: |
5d9e75c4 JS |
1144 | btrfs_set_node_key(eb, &tm->key, tm->slot); |
1145 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1146 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1147 | tm->generation); | |
4c3e6969 | 1148 | n++; |
5d9e75c4 JS |
1149 | break; |
1150 | case MOD_LOG_KEY_REPLACE: | |
1151 | BUG_ON(tm->slot >= n); | |
1152 | btrfs_set_node_key(eb, &tm->key, tm->slot); | |
1153 | btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr); | |
1154 | btrfs_set_node_ptr_generation(eb, tm->slot, | |
1155 | tm->generation); | |
1156 | break; | |
1157 | case MOD_LOG_KEY_ADD: | |
19956c7e | 1158 | /* if a move operation is needed it's in the log */ |
5d9e75c4 JS |
1159 | n--; |
1160 | break; | |
1161 | case MOD_LOG_MOVE_KEYS: | |
c3193108 JS |
1162 | o_dst = btrfs_node_key_ptr_offset(tm->slot); |
1163 | o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot); | |
1164 | memmove_extent_buffer(eb, o_dst, o_src, | |
5d9e75c4 JS |
1165 | tm->move.nr_items * p_size); |
1166 | break; | |
1167 | case MOD_LOG_ROOT_REPLACE: | |
1168 | /* | |
1169 | * this operation is special. for roots, this must be | |
1170 | * handled explicitly before rewinding. | |
1171 | * for non-roots, this operation may exist if the node | |
1172 | * was a root: root A -> child B; then A gets empty and | |
1173 | * B is promoted to the new root. in the mod log, we'll | |
1174 | * have a root-replace operation for B, a tree block | |
1175 | * that is no root. we simply ignore that operation. | |
1176 | */ | |
1177 | break; | |
1178 | } | |
1179 | next = rb_next(&tm->node); | |
1180 | if (!next) | |
1181 | break; | |
1182 | tm = container_of(next, struct tree_mod_elem, node); | |
1183 | if (tm->index != first_tm->index) | |
1184 | break; | |
1185 | } | |
1186 | btrfs_set_header_nritems(eb, n); | |
1187 | } | |
1188 | ||
1189 | static struct extent_buffer * | |
1190 | tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb, | |
1191 | u64 time_seq) | |
1192 | { | |
1193 | struct extent_buffer *eb_rewin; | |
1194 | struct tree_mod_elem *tm; | |
1195 | ||
1196 | if (!time_seq) | |
1197 | return eb; | |
1198 | ||
1199 | if (btrfs_header_level(eb) == 0) | |
1200 | return eb; | |
1201 | ||
1202 | tm = tree_mod_log_search(fs_info, eb->start, time_seq); | |
1203 | if (!tm) | |
1204 | return eb; | |
1205 | ||
1206 | if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) { | |
1207 | BUG_ON(tm->slot != 0); | |
1208 | eb_rewin = alloc_dummy_extent_buffer(eb->start, | |
1209 | fs_info->tree_root->nodesize); | |
1210 | BUG_ON(!eb_rewin); | |
1211 | btrfs_set_header_bytenr(eb_rewin, eb->start); | |
1212 | btrfs_set_header_backref_rev(eb_rewin, | |
1213 | btrfs_header_backref_rev(eb)); | |
1214 | btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb)); | |
c3193108 | 1215 | btrfs_set_header_level(eb_rewin, btrfs_header_level(eb)); |
5d9e75c4 JS |
1216 | } else { |
1217 | eb_rewin = btrfs_clone_extent_buffer(eb); | |
1218 | BUG_ON(!eb_rewin); | |
1219 | } | |
1220 | ||
1221 | extent_buffer_get(eb_rewin); | |
1222 | free_extent_buffer(eb); | |
1223 | ||
1224 | __tree_mod_log_rewind(eb_rewin, time_seq, tm); | |
57911b8b | 1225 | WARN_ON(btrfs_header_nritems(eb_rewin) > |
2a745b14 | 1226 | BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root)); |
5d9e75c4 JS |
1227 | |
1228 | return eb_rewin; | |
1229 | } | |
1230 | ||
8ba97a15 JS |
1231 | /* |
1232 | * get_old_root() rewinds the state of @root's root node to the given @time_seq | |
1233 | * value. If there are no changes, the current root->root_node is returned. If | |
1234 | * anything changed in between, there's a fresh buffer allocated on which the | |
1235 | * rewind operations are done. In any case, the returned buffer is read locked. | |
1236 | * Returns NULL on error (with no locks held). | |
1237 | */ | |
5d9e75c4 JS |
1238 | static inline struct extent_buffer * |
1239 | get_old_root(struct btrfs_root *root, u64 time_seq) | |
1240 | { | |
1241 | struct tree_mod_elem *tm; | |
1242 | struct extent_buffer *eb; | |
7bfdcf7f | 1243 | struct extent_buffer *old; |
a95236d9 | 1244 | struct tree_mod_root *old_root = NULL; |
4325edd0 | 1245 | u64 old_generation = 0; |
a95236d9 | 1246 | u64 logical; |
834328a8 | 1247 | u32 blocksize; |
5d9e75c4 | 1248 | |
8ba97a15 | 1249 | eb = btrfs_read_lock_root_node(root); |
5d9e75c4 JS |
1250 | tm = __tree_mod_log_oldest_root(root->fs_info, root, time_seq); |
1251 | if (!tm) | |
1252 | return root->node; | |
1253 | ||
a95236d9 JS |
1254 | if (tm->op == MOD_LOG_ROOT_REPLACE) { |
1255 | old_root = &tm->old_root; | |
1256 | old_generation = tm->generation; | |
1257 | logical = old_root->logical; | |
1258 | } else { | |
1259 | logical = root->node->start; | |
1260 | } | |
5d9e75c4 | 1261 | |
a95236d9 | 1262 | tm = tree_mod_log_search(root->fs_info, logical, time_seq); |
834328a8 JS |
1263 | if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) { |
1264 | btrfs_tree_read_unlock(root->node); | |
1265 | free_extent_buffer(root->node); | |
1266 | blocksize = btrfs_level_size(root, old_root->level); | |
7bfdcf7f LB |
1267 | old = read_tree_block(root, logical, blocksize, 0); |
1268 | if (!old) { | |
834328a8 JS |
1269 | pr_warn("btrfs: failed to read tree block %llu from get_old_root\n", |
1270 | logical); | |
1271 | WARN_ON(1); | |
1272 | } else { | |
7bfdcf7f LB |
1273 | eb = btrfs_clone_extent_buffer(old); |
1274 | free_extent_buffer(old); | |
834328a8 JS |
1275 | } |
1276 | } else if (old_root) { | |
1277 | btrfs_tree_read_unlock(root->node); | |
1278 | free_extent_buffer(root->node); | |
28da9fb4 | 1279 | eb = alloc_dummy_extent_buffer(logical, root->nodesize); |
834328a8 | 1280 | } else { |
a95236d9 | 1281 | eb = btrfs_clone_extent_buffer(root->node); |
834328a8 JS |
1282 | btrfs_tree_read_unlock(root->node); |
1283 | free_extent_buffer(root->node); | |
1284 | } | |
1285 | ||
8ba97a15 JS |
1286 | if (!eb) |
1287 | return NULL; | |
d6381084 | 1288 | extent_buffer_get(eb); |
8ba97a15 | 1289 | btrfs_tree_read_lock(eb); |
a95236d9 | 1290 | if (old_root) { |
5d9e75c4 JS |
1291 | btrfs_set_header_bytenr(eb, eb->start); |
1292 | btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV); | |
1293 | btrfs_set_header_owner(eb, root->root_key.objectid); | |
a95236d9 JS |
1294 | btrfs_set_header_level(eb, old_root->level); |
1295 | btrfs_set_header_generation(eb, old_generation); | |
5d9e75c4 | 1296 | } |
28da9fb4 JS |
1297 | if (tm) |
1298 | __tree_mod_log_rewind(eb, time_seq, tm); | |
1299 | else | |
1300 | WARN_ON(btrfs_header_level(eb) != 0); | |
57911b8b | 1301 | WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root)); |
5d9e75c4 JS |
1302 | |
1303 | return eb; | |
1304 | } | |
1305 | ||
5b6602e7 JS |
1306 | int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq) |
1307 | { | |
1308 | struct tree_mod_elem *tm; | |
1309 | int level; | |
1310 | ||
1311 | tm = __tree_mod_log_oldest_root(root->fs_info, root, time_seq); | |
1312 | if (tm && tm->op == MOD_LOG_ROOT_REPLACE) { | |
1313 | level = tm->old_root.level; | |
1314 | } else { | |
1315 | rcu_read_lock(); | |
1316 | level = btrfs_header_level(root->node); | |
1317 | rcu_read_unlock(); | |
1318 | } | |
1319 | ||
1320 | return level; | |
1321 | } | |
1322 | ||
5d4f98a2 YZ |
1323 | static inline int should_cow_block(struct btrfs_trans_handle *trans, |
1324 | struct btrfs_root *root, | |
1325 | struct extent_buffer *buf) | |
1326 | { | |
f1ebcc74 LB |
1327 | /* ensure we can see the force_cow */ |
1328 | smp_rmb(); | |
1329 | ||
1330 | /* | |
1331 | * We do not need to cow a block if | |
1332 | * 1) this block is not created or changed in this transaction; | |
1333 | * 2) this block does not belong to TREE_RELOC tree; | |
1334 | * 3) the root is not forced COW. | |
1335 | * | |
1336 | * What is forced COW: | |
1337 | * when we create snapshot during commiting the transaction, | |
1338 | * after we've finished coping src root, we must COW the shared | |
1339 | * block to ensure the metadata consistency. | |
1340 | */ | |
5d4f98a2 YZ |
1341 | if (btrfs_header_generation(buf) == trans->transid && |
1342 | !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) && | |
1343 | !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
f1ebcc74 LB |
1344 | btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) && |
1345 | !root->force_cow) | |
5d4f98a2 YZ |
1346 | return 0; |
1347 | return 1; | |
1348 | } | |
1349 | ||
d352ac68 CM |
1350 | /* |
1351 | * cows a single block, see __btrfs_cow_block for the real work. | |
1352 | * This version of it has extra checks so that a block isn't cow'd more than | |
1353 | * once per transaction, as long as it hasn't been written yet | |
1354 | */ | |
d397712b | 1355 | noinline int btrfs_cow_block(struct btrfs_trans_handle *trans, |
5f39d397 CM |
1356 | struct btrfs_root *root, struct extent_buffer *buf, |
1357 | struct extent_buffer *parent, int parent_slot, | |
9fa8cfe7 | 1358 | struct extent_buffer **cow_ret) |
6702ed49 CM |
1359 | { |
1360 | u64 search_start; | |
f510cfec | 1361 | int ret; |
dc17ff8f | 1362 | |
31b1a2bd JL |
1363 | if (trans->transaction != root->fs_info->running_transaction) |
1364 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
d397712b CM |
1365 | (unsigned long long)trans->transid, |
1366 | (unsigned long long) | |
6702ed49 | 1367 | root->fs_info->running_transaction->transid); |
31b1a2bd JL |
1368 | |
1369 | if (trans->transid != root->fs_info->generation) | |
1370 | WARN(1, KERN_CRIT "trans %llu running %llu\n", | |
d397712b CM |
1371 | (unsigned long long)trans->transid, |
1372 | (unsigned long long)root->fs_info->generation); | |
dc17ff8f | 1373 | |
5d4f98a2 | 1374 | if (!should_cow_block(trans, root, buf)) { |
6702ed49 CM |
1375 | *cow_ret = buf; |
1376 | return 0; | |
1377 | } | |
c487685d | 1378 | |
0b86a832 | 1379 | search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1); |
b4ce94de CM |
1380 | |
1381 | if (parent) | |
1382 | btrfs_set_lock_blocking(parent); | |
1383 | btrfs_set_lock_blocking(buf); | |
1384 | ||
f510cfec | 1385 | ret = __btrfs_cow_block(trans, root, buf, parent, |
9fa8cfe7 | 1386 | parent_slot, cow_ret, search_start, 0); |
1abe9b8a | 1387 | |
1388 | trace_btrfs_cow_block(root, buf, *cow_ret); | |
1389 | ||
f510cfec | 1390 | return ret; |
6702ed49 CM |
1391 | } |
1392 | ||
d352ac68 CM |
1393 | /* |
1394 | * helper function for defrag to decide if two blocks pointed to by a | |
1395 | * node are actually close by | |
1396 | */ | |
6b80053d | 1397 | static int close_blocks(u64 blocknr, u64 other, u32 blocksize) |
6702ed49 | 1398 | { |
6b80053d | 1399 | if (blocknr < other && other - (blocknr + blocksize) < 32768) |
6702ed49 | 1400 | return 1; |
6b80053d | 1401 | if (blocknr > other && blocknr - (other + blocksize) < 32768) |
6702ed49 CM |
1402 | return 1; |
1403 | return 0; | |
1404 | } | |
1405 | ||
081e9573 CM |
1406 | /* |
1407 | * compare two keys in a memcmp fashion | |
1408 | */ | |
1409 | static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2) | |
1410 | { | |
1411 | struct btrfs_key k1; | |
1412 | ||
1413 | btrfs_disk_key_to_cpu(&k1, disk); | |
1414 | ||
20736aba | 1415 | return btrfs_comp_cpu_keys(&k1, k2); |
081e9573 CM |
1416 | } |
1417 | ||
f3465ca4 JB |
1418 | /* |
1419 | * same as comp_keys only with two btrfs_key's | |
1420 | */ | |
5d4f98a2 | 1421 | int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2) |
f3465ca4 JB |
1422 | { |
1423 | if (k1->objectid > k2->objectid) | |
1424 | return 1; | |
1425 | if (k1->objectid < k2->objectid) | |
1426 | return -1; | |
1427 | if (k1->type > k2->type) | |
1428 | return 1; | |
1429 | if (k1->type < k2->type) | |
1430 | return -1; | |
1431 | if (k1->offset > k2->offset) | |
1432 | return 1; | |
1433 | if (k1->offset < k2->offset) | |
1434 | return -1; | |
1435 | return 0; | |
1436 | } | |
081e9573 | 1437 | |
d352ac68 CM |
1438 | /* |
1439 | * this is used by the defrag code to go through all the | |
1440 | * leaves pointed to by a node and reallocate them so that | |
1441 | * disk order is close to key order | |
1442 | */ | |
6702ed49 | 1443 | int btrfs_realloc_node(struct btrfs_trans_handle *trans, |
5f39d397 | 1444 | struct btrfs_root *root, struct extent_buffer *parent, |
de78b51a | 1445 | int start_slot, u64 *last_ret, |
a6b6e75e | 1446 | struct btrfs_key *progress) |
6702ed49 | 1447 | { |
6b80053d | 1448 | struct extent_buffer *cur; |
6702ed49 | 1449 | u64 blocknr; |
ca7a79ad | 1450 | u64 gen; |
e9d0b13b CM |
1451 | u64 search_start = *last_ret; |
1452 | u64 last_block = 0; | |
6702ed49 CM |
1453 | u64 other; |
1454 | u32 parent_nritems; | |
6702ed49 CM |
1455 | int end_slot; |
1456 | int i; | |
1457 | int err = 0; | |
f2183bde | 1458 | int parent_level; |
6b80053d CM |
1459 | int uptodate; |
1460 | u32 blocksize; | |
081e9573 CM |
1461 | int progress_passed = 0; |
1462 | struct btrfs_disk_key disk_key; | |
6702ed49 | 1463 | |
5708b959 | 1464 | parent_level = btrfs_header_level(parent); |
5708b959 | 1465 | |
6c1500f2 JL |
1466 | WARN_ON(trans->transaction != root->fs_info->running_transaction); |
1467 | WARN_ON(trans->transid != root->fs_info->generation); | |
86479a04 | 1468 | |
6b80053d | 1469 | parent_nritems = btrfs_header_nritems(parent); |
6b80053d | 1470 | blocksize = btrfs_level_size(root, parent_level - 1); |
6702ed49 CM |
1471 | end_slot = parent_nritems; |
1472 | ||
1473 | if (parent_nritems == 1) | |
1474 | return 0; | |
1475 | ||
b4ce94de CM |
1476 | btrfs_set_lock_blocking(parent); |
1477 | ||
6702ed49 CM |
1478 | for (i = start_slot; i < end_slot; i++) { |
1479 | int close = 1; | |
a6b6e75e | 1480 | |
081e9573 CM |
1481 | btrfs_node_key(parent, &disk_key, i); |
1482 | if (!progress_passed && comp_keys(&disk_key, progress) < 0) | |
1483 | continue; | |
1484 | ||
1485 | progress_passed = 1; | |
6b80053d | 1486 | blocknr = btrfs_node_blockptr(parent, i); |
ca7a79ad | 1487 | gen = btrfs_node_ptr_generation(parent, i); |
e9d0b13b CM |
1488 | if (last_block == 0) |
1489 | last_block = blocknr; | |
5708b959 | 1490 | |
6702ed49 | 1491 | if (i > 0) { |
6b80053d CM |
1492 | other = btrfs_node_blockptr(parent, i - 1); |
1493 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1494 | } |
0ef3e66b | 1495 | if (!close && i < end_slot - 2) { |
6b80053d CM |
1496 | other = btrfs_node_blockptr(parent, i + 1); |
1497 | close = close_blocks(blocknr, other, blocksize); | |
6702ed49 | 1498 | } |
e9d0b13b CM |
1499 | if (close) { |
1500 | last_block = blocknr; | |
6702ed49 | 1501 | continue; |
e9d0b13b | 1502 | } |
6702ed49 | 1503 | |
6b80053d CM |
1504 | cur = btrfs_find_tree_block(root, blocknr, blocksize); |
1505 | if (cur) | |
b9fab919 | 1506 | uptodate = btrfs_buffer_uptodate(cur, gen, 0); |
6b80053d CM |
1507 | else |
1508 | uptodate = 0; | |
5708b959 | 1509 | if (!cur || !uptodate) { |
6b80053d CM |
1510 | if (!cur) { |
1511 | cur = read_tree_block(root, blocknr, | |
ca7a79ad | 1512 | blocksize, gen); |
97d9a8a4 TI |
1513 | if (!cur) |
1514 | return -EIO; | |
6b80053d | 1515 | } else if (!uptodate) { |
018642a1 TI |
1516 | err = btrfs_read_buffer(cur, gen); |
1517 | if (err) { | |
1518 | free_extent_buffer(cur); | |
1519 | return err; | |
1520 | } | |
f2183bde | 1521 | } |
6702ed49 | 1522 | } |
e9d0b13b | 1523 | if (search_start == 0) |
6b80053d | 1524 | search_start = last_block; |
e9d0b13b | 1525 | |
e7a84565 | 1526 | btrfs_tree_lock(cur); |
b4ce94de | 1527 | btrfs_set_lock_blocking(cur); |
6b80053d | 1528 | err = __btrfs_cow_block(trans, root, cur, parent, i, |
e7a84565 | 1529 | &cur, search_start, |
6b80053d | 1530 | min(16 * blocksize, |
9fa8cfe7 | 1531 | (end_slot - i) * blocksize)); |
252c38f0 | 1532 | if (err) { |
e7a84565 | 1533 | btrfs_tree_unlock(cur); |
6b80053d | 1534 | free_extent_buffer(cur); |
6702ed49 | 1535 | break; |
252c38f0 | 1536 | } |
e7a84565 CM |
1537 | search_start = cur->start; |
1538 | last_block = cur->start; | |
f2183bde | 1539 | *last_ret = search_start; |
e7a84565 CM |
1540 | btrfs_tree_unlock(cur); |
1541 | free_extent_buffer(cur); | |
6702ed49 CM |
1542 | } |
1543 | return err; | |
1544 | } | |
1545 | ||
74123bd7 CM |
1546 | /* |
1547 | * The leaf data grows from end-to-front in the node. | |
1548 | * this returns the address of the start of the last item, | |
1549 | * which is the stop of the leaf data stack | |
1550 | */ | |
123abc88 | 1551 | static inline unsigned int leaf_data_end(struct btrfs_root *root, |
5f39d397 | 1552 | struct extent_buffer *leaf) |
be0e5c09 | 1553 | { |
5f39d397 | 1554 | u32 nr = btrfs_header_nritems(leaf); |
be0e5c09 | 1555 | if (nr == 0) |
123abc88 | 1556 | return BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 | 1557 | return btrfs_item_offset_nr(leaf, nr - 1); |
be0e5c09 CM |
1558 | } |
1559 | ||
aa5d6bed | 1560 | |
74123bd7 | 1561 | /* |
5f39d397 CM |
1562 | * search for key in the extent_buffer. The items start at offset p, |
1563 | * and they are item_size apart. There are 'max' items in p. | |
1564 | * | |
74123bd7 CM |
1565 | * the slot in the array is returned via slot, and it points to |
1566 | * the place where you would insert key if it is not found in | |
1567 | * the array. | |
1568 | * | |
1569 | * slot may point to max if the key is bigger than all of the keys | |
1570 | */ | |
e02119d5 CM |
1571 | static noinline int generic_bin_search(struct extent_buffer *eb, |
1572 | unsigned long p, | |
1573 | int item_size, struct btrfs_key *key, | |
1574 | int max, int *slot) | |
be0e5c09 CM |
1575 | { |
1576 | int low = 0; | |
1577 | int high = max; | |
1578 | int mid; | |
1579 | int ret; | |
479965d6 | 1580 | struct btrfs_disk_key *tmp = NULL; |
5f39d397 CM |
1581 | struct btrfs_disk_key unaligned; |
1582 | unsigned long offset; | |
5f39d397 CM |
1583 | char *kaddr = NULL; |
1584 | unsigned long map_start = 0; | |
1585 | unsigned long map_len = 0; | |
479965d6 | 1586 | int err; |
be0e5c09 | 1587 | |
d397712b | 1588 | while (low < high) { |
be0e5c09 | 1589 | mid = (low + high) / 2; |
5f39d397 CM |
1590 | offset = p + mid * item_size; |
1591 | ||
a6591715 | 1592 | if (!kaddr || offset < map_start || |
5f39d397 CM |
1593 | (offset + sizeof(struct btrfs_disk_key)) > |
1594 | map_start + map_len) { | |
934d375b CM |
1595 | |
1596 | err = map_private_extent_buffer(eb, offset, | |
479965d6 | 1597 | sizeof(struct btrfs_disk_key), |
a6591715 | 1598 | &kaddr, &map_start, &map_len); |
479965d6 CM |
1599 | |
1600 | if (!err) { | |
1601 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1602 | map_start); | |
1603 | } else { | |
1604 | read_extent_buffer(eb, &unaligned, | |
1605 | offset, sizeof(unaligned)); | |
1606 | tmp = &unaligned; | |
1607 | } | |
5f39d397 | 1608 | |
5f39d397 CM |
1609 | } else { |
1610 | tmp = (struct btrfs_disk_key *)(kaddr + offset - | |
1611 | map_start); | |
1612 | } | |
be0e5c09 CM |
1613 | ret = comp_keys(tmp, key); |
1614 | ||
1615 | if (ret < 0) | |
1616 | low = mid + 1; | |
1617 | else if (ret > 0) | |
1618 | high = mid; | |
1619 | else { | |
1620 | *slot = mid; | |
1621 | return 0; | |
1622 | } | |
1623 | } | |
1624 | *slot = low; | |
1625 | return 1; | |
1626 | } | |
1627 | ||
97571fd0 CM |
1628 | /* |
1629 | * simple bin_search frontend that does the right thing for | |
1630 | * leaves vs nodes | |
1631 | */ | |
5f39d397 CM |
1632 | static int bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1633 | int level, int *slot) | |
be0e5c09 | 1634 | { |
f775738f | 1635 | if (level == 0) |
5f39d397 CM |
1636 | return generic_bin_search(eb, |
1637 | offsetof(struct btrfs_leaf, items), | |
0783fcfc | 1638 | sizeof(struct btrfs_item), |
5f39d397 | 1639 | key, btrfs_header_nritems(eb), |
7518a238 | 1640 | slot); |
f775738f | 1641 | else |
5f39d397 CM |
1642 | return generic_bin_search(eb, |
1643 | offsetof(struct btrfs_node, ptrs), | |
123abc88 | 1644 | sizeof(struct btrfs_key_ptr), |
5f39d397 | 1645 | key, btrfs_header_nritems(eb), |
7518a238 | 1646 | slot); |
be0e5c09 CM |
1647 | } |
1648 | ||
5d4f98a2 YZ |
1649 | int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key, |
1650 | int level, int *slot) | |
1651 | { | |
1652 | return bin_search(eb, key, level, slot); | |
1653 | } | |
1654 | ||
f0486c68 YZ |
1655 | static void root_add_used(struct btrfs_root *root, u32 size) |
1656 | { | |
1657 | spin_lock(&root->accounting_lock); | |
1658 | btrfs_set_root_used(&root->root_item, | |
1659 | btrfs_root_used(&root->root_item) + size); | |
1660 | spin_unlock(&root->accounting_lock); | |
1661 | } | |
1662 | ||
1663 | static void root_sub_used(struct btrfs_root *root, u32 size) | |
1664 | { | |
1665 | spin_lock(&root->accounting_lock); | |
1666 | btrfs_set_root_used(&root->root_item, | |
1667 | btrfs_root_used(&root->root_item) - size); | |
1668 | spin_unlock(&root->accounting_lock); | |
1669 | } | |
1670 | ||
d352ac68 CM |
1671 | /* given a node and slot number, this reads the blocks it points to. The |
1672 | * extent buffer is returned with a reference taken (but unlocked). | |
1673 | * NULL is returned on error. | |
1674 | */ | |
e02119d5 | 1675 | static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root, |
5f39d397 | 1676 | struct extent_buffer *parent, int slot) |
bb803951 | 1677 | { |
ca7a79ad | 1678 | int level = btrfs_header_level(parent); |
bb803951 CM |
1679 | if (slot < 0) |
1680 | return NULL; | |
5f39d397 | 1681 | if (slot >= btrfs_header_nritems(parent)) |
bb803951 | 1682 | return NULL; |
ca7a79ad CM |
1683 | |
1684 | BUG_ON(level == 0); | |
1685 | ||
db94535d | 1686 | return read_tree_block(root, btrfs_node_blockptr(parent, slot), |
ca7a79ad CM |
1687 | btrfs_level_size(root, level - 1), |
1688 | btrfs_node_ptr_generation(parent, slot)); | |
bb803951 CM |
1689 | } |
1690 | ||
d352ac68 CM |
1691 | /* |
1692 | * node level balancing, used to make sure nodes are in proper order for | |
1693 | * item deletion. We balance from the top down, so we have to make sure | |
1694 | * that a deletion won't leave an node completely empty later on. | |
1695 | */ | |
e02119d5 | 1696 | static noinline int balance_level(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1697 | struct btrfs_root *root, |
1698 | struct btrfs_path *path, int level) | |
bb803951 | 1699 | { |
5f39d397 CM |
1700 | struct extent_buffer *right = NULL; |
1701 | struct extent_buffer *mid; | |
1702 | struct extent_buffer *left = NULL; | |
1703 | struct extent_buffer *parent = NULL; | |
bb803951 CM |
1704 | int ret = 0; |
1705 | int wret; | |
1706 | int pslot; | |
bb803951 | 1707 | int orig_slot = path->slots[level]; |
79f95c82 | 1708 | u64 orig_ptr; |
bb803951 CM |
1709 | |
1710 | if (level == 0) | |
1711 | return 0; | |
1712 | ||
5f39d397 | 1713 | mid = path->nodes[level]; |
b4ce94de | 1714 | |
bd681513 CM |
1715 | WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK && |
1716 | path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING); | |
7bb86316 CM |
1717 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
1718 | ||
1d4f8a0c | 1719 | orig_ptr = btrfs_node_blockptr(mid, orig_slot); |
79f95c82 | 1720 | |
a05a9bb1 | 1721 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1722 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1723 | pslot = path->slots[level + 1]; |
1724 | } | |
bb803951 | 1725 | |
40689478 CM |
1726 | /* |
1727 | * deal with the case where there is only one pointer in the root | |
1728 | * by promoting the node below to a root | |
1729 | */ | |
5f39d397 CM |
1730 | if (!parent) { |
1731 | struct extent_buffer *child; | |
bb803951 | 1732 | |
5f39d397 | 1733 | if (btrfs_header_nritems(mid) != 1) |
bb803951 CM |
1734 | return 0; |
1735 | ||
1736 | /* promote the child to a root */ | |
5f39d397 | 1737 | child = read_node_slot(root, mid, 0); |
305a26af MF |
1738 | if (!child) { |
1739 | ret = -EROFS; | |
1740 | btrfs_std_error(root->fs_info, ret); | |
1741 | goto enospc; | |
1742 | } | |
1743 | ||
925baedd | 1744 | btrfs_tree_lock(child); |
b4ce94de | 1745 | btrfs_set_lock_blocking(child); |
9fa8cfe7 | 1746 | ret = btrfs_cow_block(trans, root, child, mid, 0, &child); |
f0486c68 YZ |
1747 | if (ret) { |
1748 | btrfs_tree_unlock(child); | |
1749 | free_extent_buffer(child); | |
1750 | goto enospc; | |
1751 | } | |
2f375ab9 | 1752 | |
ba1bfbd5 | 1753 | tree_mod_log_free_eb(root->fs_info, root->node); |
f230475e | 1754 | tree_mod_log_set_root_pointer(root, child); |
240f62c8 | 1755 | rcu_assign_pointer(root->node, child); |
925baedd | 1756 | |
0b86a832 | 1757 | add_root_to_dirty_list(root); |
925baedd | 1758 | btrfs_tree_unlock(child); |
b4ce94de | 1759 | |
925baedd | 1760 | path->locks[level] = 0; |
bb803951 | 1761 | path->nodes[level] = NULL; |
5f39d397 | 1762 | clean_tree_block(trans, root, mid); |
925baedd | 1763 | btrfs_tree_unlock(mid); |
bb803951 | 1764 | /* once for the path */ |
5f39d397 | 1765 | free_extent_buffer(mid); |
f0486c68 YZ |
1766 | |
1767 | root_sub_used(root, mid->len); | |
5581a51a | 1768 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
bb803951 | 1769 | /* once for the root ptr */ |
3083ee2e | 1770 | free_extent_buffer_stale(mid); |
f0486c68 | 1771 | return 0; |
bb803951 | 1772 | } |
5f39d397 | 1773 | if (btrfs_header_nritems(mid) > |
123abc88 | 1774 | BTRFS_NODEPTRS_PER_BLOCK(root) / 4) |
bb803951 CM |
1775 | return 0; |
1776 | ||
5f39d397 CM |
1777 | left = read_node_slot(root, parent, pslot - 1); |
1778 | if (left) { | |
925baedd | 1779 | btrfs_tree_lock(left); |
b4ce94de | 1780 | btrfs_set_lock_blocking(left); |
5f39d397 | 1781 | wret = btrfs_cow_block(trans, root, left, |
9fa8cfe7 | 1782 | parent, pslot - 1, &left); |
54aa1f4d CM |
1783 | if (wret) { |
1784 | ret = wret; | |
1785 | goto enospc; | |
1786 | } | |
2cc58cf2 | 1787 | } |
5f39d397 CM |
1788 | right = read_node_slot(root, parent, pslot + 1); |
1789 | if (right) { | |
925baedd | 1790 | btrfs_tree_lock(right); |
b4ce94de | 1791 | btrfs_set_lock_blocking(right); |
5f39d397 | 1792 | wret = btrfs_cow_block(trans, root, right, |
9fa8cfe7 | 1793 | parent, pslot + 1, &right); |
2cc58cf2 CM |
1794 | if (wret) { |
1795 | ret = wret; | |
1796 | goto enospc; | |
1797 | } | |
1798 | } | |
1799 | ||
1800 | /* first, try to make some room in the middle buffer */ | |
5f39d397 CM |
1801 | if (left) { |
1802 | orig_slot += btrfs_header_nritems(left); | |
bce4eae9 | 1803 | wret = push_node_left(trans, root, left, mid, 1); |
79f95c82 CM |
1804 | if (wret < 0) |
1805 | ret = wret; | |
bb803951 | 1806 | } |
79f95c82 CM |
1807 | |
1808 | /* | |
1809 | * then try to empty the right most buffer into the middle | |
1810 | */ | |
5f39d397 | 1811 | if (right) { |
971a1f66 | 1812 | wret = push_node_left(trans, root, mid, right, 1); |
54aa1f4d | 1813 | if (wret < 0 && wret != -ENOSPC) |
79f95c82 | 1814 | ret = wret; |
5f39d397 | 1815 | if (btrfs_header_nritems(right) == 0) { |
5f39d397 | 1816 | clean_tree_block(trans, root, right); |
925baedd | 1817 | btrfs_tree_unlock(right); |
0e411ece | 1818 | del_ptr(trans, root, path, level + 1, pslot + 1); |
f0486c68 | 1819 | root_sub_used(root, right->len); |
5581a51a | 1820 | btrfs_free_tree_block(trans, root, right, 0, 1); |
3083ee2e | 1821 | free_extent_buffer_stale(right); |
f0486c68 | 1822 | right = NULL; |
bb803951 | 1823 | } else { |
5f39d397 CM |
1824 | struct btrfs_disk_key right_key; |
1825 | btrfs_node_key(right, &right_key, 0); | |
f230475e | 1826 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1827 | pslot + 1, 0); |
5f39d397 CM |
1828 | btrfs_set_node_key(parent, &right_key, pslot + 1); |
1829 | btrfs_mark_buffer_dirty(parent); | |
bb803951 CM |
1830 | } |
1831 | } | |
5f39d397 | 1832 | if (btrfs_header_nritems(mid) == 1) { |
79f95c82 CM |
1833 | /* |
1834 | * we're not allowed to leave a node with one item in the | |
1835 | * tree during a delete. A deletion from lower in the tree | |
1836 | * could try to delete the only pointer in this node. | |
1837 | * So, pull some keys from the left. | |
1838 | * There has to be a left pointer at this point because | |
1839 | * otherwise we would have pulled some pointers from the | |
1840 | * right | |
1841 | */ | |
305a26af MF |
1842 | if (!left) { |
1843 | ret = -EROFS; | |
1844 | btrfs_std_error(root->fs_info, ret); | |
1845 | goto enospc; | |
1846 | } | |
5f39d397 | 1847 | wret = balance_node_right(trans, root, mid, left); |
54aa1f4d | 1848 | if (wret < 0) { |
79f95c82 | 1849 | ret = wret; |
54aa1f4d CM |
1850 | goto enospc; |
1851 | } | |
bce4eae9 CM |
1852 | if (wret == 1) { |
1853 | wret = push_node_left(trans, root, left, mid, 1); | |
1854 | if (wret < 0) | |
1855 | ret = wret; | |
1856 | } | |
79f95c82 CM |
1857 | BUG_ON(wret == 1); |
1858 | } | |
5f39d397 | 1859 | if (btrfs_header_nritems(mid) == 0) { |
5f39d397 | 1860 | clean_tree_block(trans, root, mid); |
925baedd | 1861 | btrfs_tree_unlock(mid); |
0e411ece | 1862 | del_ptr(trans, root, path, level + 1, pslot); |
f0486c68 | 1863 | root_sub_used(root, mid->len); |
5581a51a | 1864 | btrfs_free_tree_block(trans, root, mid, 0, 1); |
3083ee2e | 1865 | free_extent_buffer_stale(mid); |
f0486c68 | 1866 | mid = NULL; |
79f95c82 CM |
1867 | } else { |
1868 | /* update the parent key to reflect our changes */ | |
5f39d397 CM |
1869 | struct btrfs_disk_key mid_key; |
1870 | btrfs_node_key(mid, &mid_key, 0); | |
32adf090 | 1871 | tree_mod_log_set_node_key(root->fs_info, parent, |
f230475e | 1872 | pslot, 0); |
5f39d397 CM |
1873 | btrfs_set_node_key(parent, &mid_key, pslot); |
1874 | btrfs_mark_buffer_dirty(parent); | |
79f95c82 | 1875 | } |
bb803951 | 1876 | |
79f95c82 | 1877 | /* update the path */ |
5f39d397 CM |
1878 | if (left) { |
1879 | if (btrfs_header_nritems(left) > orig_slot) { | |
1880 | extent_buffer_get(left); | |
925baedd | 1881 | /* left was locked after cow */ |
5f39d397 | 1882 | path->nodes[level] = left; |
bb803951 CM |
1883 | path->slots[level + 1] -= 1; |
1884 | path->slots[level] = orig_slot; | |
925baedd CM |
1885 | if (mid) { |
1886 | btrfs_tree_unlock(mid); | |
5f39d397 | 1887 | free_extent_buffer(mid); |
925baedd | 1888 | } |
bb803951 | 1889 | } else { |
5f39d397 | 1890 | orig_slot -= btrfs_header_nritems(left); |
bb803951 CM |
1891 | path->slots[level] = orig_slot; |
1892 | } | |
1893 | } | |
79f95c82 | 1894 | /* double check we haven't messed things up */ |
e20d96d6 | 1895 | if (orig_ptr != |
5f39d397 | 1896 | btrfs_node_blockptr(path->nodes[level], path->slots[level])) |
79f95c82 | 1897 | BUG(); |
54aa1f4d | 1898 | enospc: |
925baedd CM |
1899 | if (right) { |
1900 | btrfs_tree_unlock(right); | |
5f39d397 | 1901 | free_extent_buffer(right); |
925baedd CM |
1902 | } |
1903 | if (left) { | |
1904 | if (path->nodes[level] != left) | |
1905 | btrfs_tree_unlock(left); | |
5f39d397 | 1906 | free_extent_buffer(left); |
925baedd | 1907 | } |
bb803951 CM |
1908 | return ret; |
1909 | } | |
1910 | ||
d352ac68 CM |
1911 | /* Node balancing for insertion. Here we only split or push nodes around |
1912 | * when they are completely full. This is also done top down, so we | |
1913 | * have to be pessimistic. | |
1914 | */ | |
d397712b | 1915 | static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans, |
98ed5174 CM |
1916 | struct btrfs_root *root, |
1917 | struct btrfs_path *path, int level) | |
e66f709b | 1918 | { |
5f39d397 CM |
1919 | struct extent_buffer *right = NULL; |
1920 | struct extent_buffer *mid; | |
1921 | struct extent_buffer *left = NULL; | |
1922 | struct extent_buffer *parent = NULL; | |
e66f709b CM |
1923 | int ret = 0; |
1924 | int wret; | |
1925 | int pslot; | |
1926 | int orig_slot = path->slots[level]; | |
e66f709b CM |
1927 | |
1928 | if (level == 0) | |
1929 | return 1; | |
1930 | ||
5f39d397 | 1931 | mid = path->nodes[level]; |
7bb86316 | 1932 | WARN_ON(btrfs_header_generation(mid) != trans->transid); |
e66f709b | 1933 | |
a05a9bb1 | 1934 | if (level < BTRFS_MAX_LEVEL - 1) { |
5f39d397 | 1935 | parent = path->nodes[level + 1]; |
a05a9bb1 LZ |
1936 | pslot = path->slots[level + 1]; |
1937 | } | |
e66f709b | 1938 | |
5f39d397 | 1939 | if (!parent) |
e66f709b | 1940 | return 1; |
e66f709b | 1941 | |
5f39d397 | 1942 | left = read_node_slot(root, parent, pslot - 1); |
e66f709b CM |
1943 | |
1944 | /* first, try to make some room in the middle buffer */ | |
5f39d397 | 1945 | if (left) { |
e66f709b | 1946 | u32 left_nr; |
925baedd CM |
1947 | |
1948 | btrfs_tree_lock(left); | |
b4ce94de CM |
1949 | btrfs_set_lock_blocking(left); |
1950 | ||
5f39d397 | 1951 | left_nr = btrfs_header_nritems(left); |
33ade1f8 CM |
1952 | if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
1953 | wret = 1; | |
1954 | } else { | |
5f39d397 | 1955 | ret = btrfs_cow_block(trans, root, left, parent, |
9fa8cfe7 | 1956 | pslot - 1, &left); |
54aa1f4d CM |
1957 | if (ret) |
1958 | wret = 1; | |
1959 | else { | |
54aa1f4d | 1960 | wret = push_node_left(trans, root, |
971a1f66 | 1961 | left, mid, 0); |
54aa1f4d | 1962 | } |
33ade1f8 | 1963 | } |
e66f709b CM |
1964 | if (wret < 0) |
1965 | ret = wret; | |
1966 | if (wret == 0) { | |
5f39d397 | 1967 | struct btrfs_disk_key disk_key; |
e66f709b | 1968 | orig_slot += left_nr; |
5f39d397 | 1969 | btrfs_node_key(mid, &disk_key, 0); |
f230475e | 1970 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 1971 | pslot, 0); |
5f39d397 CM |
1972 | btrfs_set_node_key(parent, &disk_key, pslot); |
1973 | btrfs_mark_buffer_dirty(parent); | |
1974 | if (btrfs_header_nritems(left) > orig_slot) { | |
1975 | path->nodes[level] = left; | |
e66f709b CM |
1976 | path->slots[level + 1] -= 1; |
1977 | path->slots[level] = orig_slot; | |
925baedd | 1978 | btrfs_tree_unlock(mid); |
5f39d397 | 1979 | free_extent_buffer(mid); |
e66f709b CM |
1980 | } else { |
1981 | orig_slot -= | |
5f39d397 | 1982 | btrfs_header_nritems(left); |
e66f709b | 1983 | path->slots[level] = orig_slot; |
925baedd | 1984 | btrfs_tree_unlock(left); |
5f39d397 | 1985 | free_extent_buffer(left); |
e66f709b | 1986 | } |
e66f709b CM |
1987 | return 0; |
1988 | } | |
925baedd | 1989 | btrfs_tree_unlock(left); |
5f39d397 | 1990 | free_extent_buffer(left); |
e66f709b | 1991 | } |
925baedd | 1992 | right = read_node_slot(root, parent, pslot + 1); |
e66f709b CM |
1993 | |
1994 | /* | |
1995 | * then try to empty the right most buffer into the middle | |
1996 | */ | |
5f39d397 | 1997 | if (right) { |
33ade1f8 | 1998 | u32 right_nr; |
b4ce94de | 1999 | |
925baedd | 2000 | btrfs_tree_lock(right); |
b4ce94de CM |
2001 | btrfs_set_lock_blocking(right); |
2002 | ||
5f39d397 | 2003 | right_nr = btrfs_header_nritems(right); |
33ade1f8 CM |
2004 | if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) { |
2005 | wret = 1; | |
2006 | } else { | |
5f39d397 CM |
2007 | ret = btrfs_cow_block(trans, root, right, |
2008 | parent, pslot + 1, | |
9fa8cfe7 | 2009 | &right); |
54aa1f4d CM |
2010 | if (ret) |
2011 | wret = 1; | |
2012 | else { | |
54aa1f4d | 2013 | wret = balance_node_right(trans, root, |
5f39d397 | 2014 | right, mid); |
54aa1f4d | 2015 | } |
33ade1f8 | 2016 | } |
e66f709b CM |
2017 | if (wret < 0) |
2018 | ret = wret; | |
2019 | if (wret == 0) { | |
5f39d397 CM |
2020 | struct btrfs_disk_key disk_key; |
2021 | ||
2022 | btrfs_node_key(right, &disk_key, 0); | |
f230475e | 2023 | tree_mod_log_set_node_key(root->fs_info, parent, |
32adf090 | 2024 | pslot + 1, 0); |
5f39d397 CM |
2025 | btrfs_set_node_key(parent, &disk_key, pslot + 1); |
2026 | btrfs_mark_buffer_dirty(parent); | |
2027 | ||
2028 | if (btrfs_header_nritems(mid) <= orig_slot) { | |
2029 | path->nodes[level] = right; | |
e66f709b CM |
2030 | path->slots[level + 1] += 1; |
2031 | path->slots[level] = orig_slot - | |
5f39d397 | 2032 | btrfs_header_nritems(mid); |
925baedd | 2033 | btrfs_tree_unlock(mid); |
5f39d397 | 2034 | free_extent_buffer(mid); |
e66f709b | 2035 | } else { |
925baedd | 2036 | btrfs_tree_unlock(right); |
5f39d397 | 2037 | free_extent_buffer(right); |
e66f709b | 2038 | } |
e66f709b CM |
2039 | return 0; |
2040 | } | |
925baedd | 2041 | btrfs_tree_unlock(right); |
5f39d397 | 2042 | free_extent_buffer(right); |
e66f709b | 2043 | } |
e66f709b CM |
2044 | return 1; |
2045 | } | |
2046 | ||
3c69faec | 2047 | /* |
d352ac68 CM |
2048 | * readahead one full node of leaves, finding things that are close |
2049 | * to the block in 'slot', and triggering ra on them. | |
3c69faec | 2050 | */ |
c8c42864 CM |
2051 | static void reada_for_search(struct btrfs_root *root, |
2052 | struct btrfs_path *path, | |
2053 | int level, int slot, u64 objectid) | |
3c69faec | 2054 | { |
5f39d397 | 2055 | struct extent_buffer *node; |
01f46658 | 2056 | struct btrfs_disk_key disk_key; |
3c69faec | 2057 | u32 nritems; |
3c69faec | 2058 | u64 search; |
a7175319 | 2059 | u64 target; |
6b80053d | 2060 | u64 nread = 0; |
cb25c2ea | 2061 | u64 gen; |
3c69faec | 2062 | int direction = path->reada; |
5f39d397 | 2063 | struct extent_buffer *eb; |
6b80053d CM |
2064 | u32 nr; |
2065 | u32 blocksize; | |
2066 | u32 nscan = 0; | |
db94535d | 2067 | |
a6b6e75e | 2068 | if (level != 1) |
6702ed49 CM |
2069 | return; |
2070 | ||
2071 | if (!path->nodes[level]) | |
3c69faec CM |
2072 | return; |
2073 | ||
5f39d397 | 2074 | node = path->nodes[level]; |
925baedd | 2075 | |
3c69faec | 2076 | search = btrfs_node_blockptr(node, slot); |
6b80053d CM |
2077 | blocksize = btrfs_level_size(root, level - 1); |
2078 | eb = btrfs_find_tree_block(root, search, blocksize); | |
5f39d397 CM |
2079 | if (eb) { |
2080 | free_extent_buffer(eb); | |
3c69faec CM |
2081 | return; |
2082 | } | |
2083 | ||
a7175319 | 2084 | target = search; |
6b80053d | 2085 | |
5f39d397 | 2086 | nritems = btrfs_header_nritems(node); |
6b80053d | 2087 | nr = slot; |
25b8b936 | 2088 | |
d397712b | 2089 | while (1) { |
6b80053d CM |
2090 | if (direction < 0) { |
2091 | if (nr == 0) | |
2092 | break; | |
2093 | nr--; | |
2094 | } else if (direction > 0) { | |
2095 | nr++; | |
2096 | if (nr >= nritems) | |
2097 | break; | |
3c69faec | 2098 | } |
01f46658 CM |
2099 | if (path->reada < 0 && objectid) { |
2100 | btrfs_node_key(node, &disk_key, nr); | |
2101 | if (btrfs_disk_key_objectid(&disk_key) != objectid) | |
2102 | break; | |
2103 | } | |
6b80053d | 2104 | search = btrfs_node_blockptr(node, nr); |
a7175319 CM |
2105 | if ((search <= target && target - search <= 65536) || |
2106 | (search > target && search - target <= 65536)) { | |
cb25c2ea | 2107 | gen = btrfs_node_ptr_generation(node, nr); |
cb25c2ea | 2108 | readahead_tree_block(root, search, blocksize, gen); |
6b80053d CM |
2109 | nread += blocksize; |
2110 | } | |
2111 | nscan++; | |
a7175319 | 2112 | if ((nread > 65536 || nscan > 32)) |
6b80053d | 2113 | break; |
3c69faec CM |
2114 | } |
2115 | } | |
925baedd | 2116 | |
b4ce94de CM |
2117 | /* |
2118 | * returns -EAGAIN if it had to drop the path, or zero if everything was in | |
2119 | * cache | |
2120 | */ | |
2121 | static noinline int reada_for_balance(struct btrfs_root *root, | |
2122 | struct btrfs_path *path, int level) | |
2123 | { | |
2124 | int slot; | |
2125 | int nritems; | |
2126 | struct extent_buffer *parent; | |
2127 | struct extent_buffer *eb; | |
2128 | u64 gen; | |
2129 | u64 block1 = 0; | |
2130 | u64 block2 = 0; | |
2131 | int ret = 0; | |
2132 | int blocksize; | |
2133 | ||
8c594ea8 | 2134 | parent = path->nodes[level + 1]; |
b4ce94de CM |
2135 | if (!parent) |
2136 | return 0; | |
2137 | ||
2138 | nritems = btrfs_header_nritems(parent); | |
8c594ea8 | 2139 | slot = path->slots[level + 1]; |
b4ce94de CM |
2140 | blocksize = btrfs_level_size(root, level); |
2141 | ||
2142 | if (slot > 0) { | |
2143 | block1 = btrfs_node_blockptr(parent, slot - 1); | |
2144 | gen = btrfs_node_ptr_generation(parent, slot - 1); | |
2145 | eb = btrfs_find_tree_block(root, block1, blocksize); | |
b9fab919 CM |
2146 | /* |
2147 | * if we get -eagain from btrfs_buffer_uptodate, we | |
2148 | * don't want to return eagain here. That will loop | |
2149 | * forever | |
2150 | */ | |
2151 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) | |
b4ce94de CM |
2152 | block1 = 0; |
2153 | free_extent_buffer(eb); | |
2154 | } | |
8c594ea8 | 2155 | if (slot + 1 < nritems) { |
b4ce94de CM |
2156 | block2 = btrfs_node_blockptr(parent, slot + 1); |
2157 | gen = btrfs_node_ptr_generation(parent, slot + 1); | |
2158 | eb = btrfs_find_tree_block(root, block2, blocksize); | |
b9fab919 | 2159 | if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0) |
b4ce94de CM |
2160 | block2 = 0; |
2161 | free_extent_buffer(eb); | |
2162 | } | |
2163 | if (block1 || block2) { | |
2164 | ret = -EAGAIN; | |
8c594ea8 CM |
2165 | |
2166 | /* release the whole path */ | |
b3b4aa74 | 2167 | btrfs_release_path(path); |
8c594ea8 CM |
2168 | |
2169 | /* read the blocks */ | |
b4ce94de CM |
2170 | if (block1) |
2171 | readahead_tree_block(root, block1, blocksize, 0); | |
2172 | if (block2) | |
2173 | readahead_tree_block(root, block2, blocksize, 0); | |
2174 | ||
2175 | if (block1) { | |
2176 | eb = read_tree_block(root, block1, blocksize, 0); | |
2177 | free_extent_buffer(eb); | |
2178 | } | |
8c594ea8 | 2179 | if (block2) { |
b4ce94de CM |
2180 | eb = read_tree_block(root, block2, blocksize, 0); |
2181 | free_extent_buffer(eb); | |
2182 | } | |
2183 | } | |
2184 | return ret; | |
2185 | } | |
2186 | ||
2187 | ||
d352ac68 | 2188 | /* |
d397712b CM |
2189 | * when we walk down the tree, it is usually safe to unlock the higher layers |
2190 | * in the tree. The exceptions are when our path goes through slot 0, because | |
2191 | * operations on the tree might require changing key pointers higher up in the | |
2192 | * tree. | |
d352ac68 | 2193 | * |
d397712b CM |
2194 | * callers might also have set path->keep_locks, which tells this code to keep |
2195 | * the lock if the path points to the last slot in the block. This is part of | |
2196 | * walking through the tree, and selecting the next slot in the higher block. | |
d352ac68 | 2197 | * |
d397712b CM |
2198 | * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so |
2199 | * if lowest_unlock is 1, level 0 won't be unlocked | |
d352ac68 | 2200 | */ |
e02119d5 | 2201 | static noinline void unlock_up(struct btrfs_path *path, int level, |
f7c79f30 CM |
2202 | int lowest_unlock, int min_write_lock_level, |
2203 | int *write_lock_level) | |
925baedd CM |
2204 | { |
2205 | int i; | |
2206 | int skip_level = level; | |
051e1b9f | 2207 | int no_skips = 0; |
925baedd CM |
2208 | struct extent_buffer *t; |
2209 | ||
d6393786 JB |
2210 | if (path->really_keep_locks) |
2211 | return; | |
2212 | ||
925baedd CM |
2213 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
2214 | if (!path->nodes[i]) | |
2215 | break; | |
2216 | if (!path->locks[i]) | |
2217 | break; | |
051e1b9f | 2218 | if (!no_skips && path->slots[i] == 0) { |
925baedd CM |
2219 | skip_level = i + 1; |
2220 | continue; | |
2221 | } | |
051e1b9f | 2222 | if (!no_skips && path->keep_locks) { |
925baedd CM |
2223 | u32 nritems; |
2224 | t = path->nodes[i]; | |
2225 | nritems = btrfs_header_nritems(t); | |
051e1b9f | 2226 | if (nritems < 1 || path->slots[i] >= nritems - 1) { |
925baedd CM |
2227 | skip_level = i + 1; |
2228 | continue; | |
2229 | } | |
2230 | } | |
051e1b9f CM |
2231 | if (skip_level < i && i >= lowest_unlock) |
2232 | no_skips = 1; | |
2233 | ||
925baedd CM |
2234 | t = path->nodes[i]; |
2235 | if (i >= lowest_unlock && i > skip_level && path->locks[i]) { | |
bd681513 | 2236 | btrfs_tree_unlock_rw(t, path->locks[i]); |
925baedd | 2237 | path->locks[i] = 0; |
f7c79f30 CM |
2238 | if (write_lock_level && |
2239 | i > min_write_lock_level && | |
2240 | i <= *write_lock_level) { | |
2241 | *write_lock_level = i - 1; | |
2242 | } | |
925baedd CM |
2243 | } |
2244 | } | |
2245 | } | |
2246 | ||
b4ce94de CM |
2247 | /* |
2248 | * This releases any locks held in the path starting at level and | |
2249 | * going all the way up to the root. | |
2250 | * | |
2251 | * btrfs_search_slot will keep the lock held on higher nodes in a few | |
2252 | * corner cases, such as COW of the block at slot zero in the node. This | |
2253 | * ignores those rules, and it should only be called when there are no | |
2254 | * more updates to be done higher up in the tree. | |
2255 | */ | |
2256 | noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level) | |
2257 | { | |
2258 | int i; | |
2259 | ||
d6393786 | 2260 | if (path->keep_locks || path->really_keep_locks) |
b4ce94de CM |
2261 | return; |
2262 | ||
2263 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { | |
2264 | if (!path->nodes[i]) | |
12f4dacc | 2265 | continue; |
b4ce94de | 2266 | if (!path->locks[i]) |
12f4dacc | 2267 | continue; |
bd681513 | 2268 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); |
b4ce94de CM |
2269 | path->locks[i] = 0; |
2270 | } | |
2271 | } | |
2272 | ||
c8c42864 CM |
2273 | /* |
2274 | * helper function for btrfs_search_slot. The goal is to find a block | |
2275 | * in cache without setting the path to blocking. If we find the block | |
2276 | * we return zero and the path is unchanged. | |
2277 | * | |
2278 | * If we can't find the block, we set the path blocking and do some | |
2279 | * reada. -EAGAIN is returned and the search must be repeated. | |
2280 | */ | |
2281 | static int | |
2282 | read_block_for_search(struct btrfs_trans_handle *trans, | |
2283 | struct btrfs_root *root, struct btrfs_path *p, | |
2284 | struct extent_buffer **eb_ret, int level, int slot, | |
5d9e75c4 | 2285 | struct btrfs_key *key, u64 time_seq) |
c8c42864 CM |
2286 | { |
2287 | u64 blocknr; | |
2288 | u64 gen; | |
2289 | u32 blocksize; | |
2290 | struct extent_buffer *b = *eb_ret; | |
2291 | struct extent_buffer *tmp; | |
76a05b35 | 2292 | int ret; |
c8c42864 CM |
2293 | |
2294 | blocknr = btrfs_node_blockptr(b, slot); | |
2295 | gen = btrfs_node_ptr_generation(b, slot); | |
2296 | blocksize = btrfs_level_size(root, level - 1); | |
2297 | ||
2298 | tmp = btrfs_find_tree_block(root, blocknr, blocksize); | |
cb44921a | 2299 | if (tmp) { |
b9fab919 CM |
2300 | /* first we do an atomic uptodate check */ |
2301 | if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) { | |
2302 | if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) { | |
cb44921a CM |
2303 | /* |
2304 | * we found an up to date block without | |
2305 | * sleeping, return | |
2306 | * right away | |
2307 | */ | |
2308 | *eb_ret = tmp; | |
2309 | return 0; | |
2310 | } | |
2311 | /* the pages were up to date, but we failed | |
2312 | * the generation number check. Do a full | |
2313 | * read for the generation number that is correct. | |
2314 | * We must do this without dropping locks so | |
2315 | * we can trust our generation number | |
2316 | */ | |
2317 | free_extent_buffer(tmp); | |
bd681513 CM |
2318 | btrfs_set_path_blocking(p); |
2319 | ||
b9fab919 | 2320 | /* now we're allowed to do a blocking uptodate check */ |
cb44921a | 2321 | tmp = read_tree_block(root, blocknr, blocksize, gen); |
b9fab919 | 2322 | if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) { |
cb44921a CM |
2323 | *eb_ret = tmp; |
2324 | return 0; | |
2325 | } | |
2326 | free_extent_buffer(tmp); | |
b3b4aa74 | 2327 | btrfs_release_path(p); |
cb44921a CM |
2328 | return -EIO; |
2329 | } | |
c8c42864 CM |
2330 | } |
2331 | ||
2332 | /* | |
2333 | * reduce lock contention at high levels | |
2334 | * of the btree by dropping locks before | |
76a05b35 CM |
2335 | * we read. Don't release the lock on the current |
2336 | * level because we need to walk this node to figure | |
2337 | * out which blocks to read. | |
c8c42864 | 2338 | */ |
8c594ea8 CM |
2339 | btrfs_unlock_up_safe(p, level + 1); |
2340 | btrfs_set_path_blocking(p); | |
2341 | ||
cb44921a | 2342 | free_extent_buffer(tmp); |
c8c42864 CM |
2343 | if (p->reada) |
2344 | reada_for_search(root, p, level, slot, key->objectid); | |
2345 | ||
b3b4aa74 | 2346 | btrfs_release_path(p); |
76a05b35 CM |
2347 | |
2348 | ret = -EAGAIN; | |
5bdd3536 | 2349 | tmp = read_tree_block(root, blocknr, blocksize, 0); |
76a05b35 CM |
2350 | if (tmp) { |
2351 | /* | |
2352 | * If the read above didn't mark this buffer up to date, | |
2353 | * it will never end up being up to date. Set ret to EIO now | |
2354 | * and give up so that our caller doesn't loop forever | |
2355 | * on our EAGAINs. | |
2356 | */ | |
b9fab919 | 2357 | if (!btrfs_buffer_uptodate(tmp, 0, 0)) |
76a05b35 | 2358 | ret = -EIO; |
c8c42864 | 2359 | free_extent_buffer(tmp); |
76a05b35 CM |
2360 | } |
2361 | return ret; | |
c8c42864 CM |
2362 | } |
2363 | ||
2364 | /* | |
2365 | * helper function for btrfs_search_slot. This does all of the checks | |
2366 | * for node-level blocks and does any balancing required based on | |
2367 | * the ins_len. | |
2368 | * | |
2369 | * If no extra work was required, zero is returned. If we had to | |
2370 | * drop the path, -EAGAIN is returned and btrfs_search_slot must | |
2371 | * start over | |
2372 | */ | |
2373 | static int | |
2374 | setup_nodes_for_search(struct btrfs_trans_handle *trans, | |
2375 | struct btrfs_root *root, struct btrfs_path *p, | |
bd681513 CM |
2376 | struct extent_buffer *b, int level, int ins_len, |
2377 | int *write_lock_level) | |
c8c42864 CM |
2378 | { |
2379 | int ret; | |
2380 | if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >= | |
2381 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) { | |
2382 | int sret; | |
2383 | ||
bd681513 CM |
2384 | if (*write_lock_level < level + 1) { |
2385 | *write_lock_level = level + 1; | |
2386 | btrfs_release_path(p); | |
2387 | goto again; | |
2388 | } | |
2389 | ||
c8c42864 CM |
2390 | sret = reada_for_balance(root, p, level); |
2391 | if (sret) | |
2392 | goto again; | |
2393 | ||
2394 | btrfs_set_path_blocking(p); | |
2395 | sret = split_node(trans, root, p, level); | |
bd681513 | 2396 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2397 | |
2398 | BUG_ON(sret > 0); | |
2399 | if (sret) { | |
2400 | ret = sret; | |
2401 | goto done; | |
2402 | } | |
2403 | b = p->nodes[level]; | |
2404 | } else if (ins_len < 0 && btrfs_header_nritems(b) < | |
cfbb9308 | 2405 | BTRFS_NODEPTRS_PER_BLOCK(root) / 2) { |
c8c42864 CM |
2406 | int sret; |
2407 | ||
bd681513 CM |
2408 | if (*write_lock_level < level + 1) { |
2409 | *write_lock_level = level + 1; | |
2410 | btrfs_release_path(p); | |
2411 | goto again; | |
2412 | } | |
2413 | ||
c8c42864 CM |
2414 | sret = reada_for_balance(root, p, level); |
2415 | if (sret) | |
2416 | goto again; | |
2417 | ||
2418 | btrfs_set_path_blocking(p); | |
2419 | sret = balance_level(trans, root, p, level); | |
bd681513 | 2420 | btrfs_clear_path_blocking(p, NULL, 0); |
c8c42864 CM |
2421 | |
2422 | if (sret) { | |
2423 | ret = sret; | |
2424 | goto done; | |
2425 | } | |
2426 | b = p->nodes[level]; | |
2427 | if (!b) { | |
b3b4aa74 | 2428 | btrfs_release_path(p); |
c8c42864 CM |
2429 | goto again; |
2430 | } | |
2431 | BUG_ON(btrfs_header_nritems(b) == 1); | |
2432 | } | |
2433 | return 0; | |
2434 | ||
2435 | again: | |
2436 | ret = -EAGAIN; | |
2437 | done: | |
2438 | return ret; | |
2439 | } | |
2440 | ||
74123bd7 CM |
2441 | /* |
2442 | * look for key in the tree. path is filled in with nodes along the way | |
2443 | * if key is found, we return zero and you can find the item in the leaf | |
2444 | * level of the path (level 0) | |
2445 | * | |
2446 | * If the key isn't found, the path points to the slot where it should | |
aa5d6bed CM |
2447 | * be inserted, and 1 is returned. If there are other errors during the |
2448 | * search a negative error number is returned. | |
97571fd0 CM |
2449 | * |
2450 | * if ins_len > 0, nodes and leaves will be split as we walk down the | |
2451 | * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if | |
2452 | * possible) | |
74123bd7 | 2453 | */ |
e089f05c CM |
2454 | int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root |
2455 | *root, struct btrfs_key *key, struct btrfs_path *p, int | |
2456 | ins_len, int cow) | |
be0e5c09 | 2457 | { |
5f39d397 | 2458 | struct extent_buffer *b; |
be0e5c09 CM |
2459 | int slot; |
2460 | int ret; | |
33c66f43 | 2461 | int err; |
be0e5c09 | 2462 | int level; |
925baedd | 2463 | int lowest_unlock = 1; |
bd681513 CM |
2464 | int root_lock; |
2465 | /* everything at write_lock_level or lower must be write locked */ | |
2466 | int write_lock_level = 0; | |
9f3a7427 | 2467 | u8 lowest_level = 0; |
f7c79f30 | 2468 | int min_write_lock_level; |
9f3a7427 | 2469 | |
6702ed49 | 2470 | lowest_level = p->lowest_level; |
323ac95b | 2471 | WARN_ON(lowest_level && ins_len > 0); |
22b0ebda | 2472 | WARN_ON(p->nodes[0] != NULL); |
25179201 | 2473 | |
bd681513 | 2474 | if (ins_len < 0) { |
925baedd | 2475 | lowest_unlock = 2; |
65b51a00 | 2476 | |
bd681513 CM |
2477 | /* when we are removing items, we might have to go up to level |
2478 | * two as we update tree pointers Make sure we keep write | |
2479 | * for those levels as well | |
2480 | */ | |
2481 | write_lock_level = 2; | |
2482 | } else if (ins_len > 0) { | |
2483 | /* | |
2484 | * for inserting items, make sure we have a write lock on | |
2485 | * level 1 so we can update keys | |
2486 | */ | |
2487 | write_lock_level = 1; | |
2488 | } | |
2489 | ||
2490 | if (!cow) | |
2491 | write_lock_level = -1; | |
2492 | ||
d6393786 | 2493 | if (cow && (p->really_keep_locks || p->keep_locks || p->lowest_level)) |
bd681513 CM |
2494 | write_lock_level = BTRFS_MAX_LEVEL; |
2495 | ||
f7c79f30 CM |
2496 | min_write_lock_level = write_lock_level; |
2497 | ||
bb803951 | 2498 | again: |
bd681513 CM |
2499 | /* |
2500 | * we try very hard to do read locks on the root | |
2501 | */ | |
2502 | root_lock = BTRFS_READ_LOCK; | |
2503 | level = 0; | |
5d4f98a2 | 2504 | if (p->search_commit_root) { |
bd681513 CM |
2505 | /* |
2506 | * the commit roots are read only | |
2507 | * so we always do read locks | |
2508 | */ | |
5d4f98a2 YZ |
2509 | b = root->commit_root; |
2510 | extent_buffer_get(b); | |
bd681513 | 2511 | level = btrfs_header_level(b); |
5d4f98a2 | 2512 | if (!p->skip_locking) |
bd681513 | 2513 | btrfs_tree_read_lock(b); |
5d4f98a2 | 2514 | } else { |
bd681513 | 2515 | if (p->skip_locking) { |
5d4f98a2 | 2516 | b = btrfs_root_node(root); |
bd681513 CM |
2517 | level = btrfs_header_level(b); |
2518 | } else { | |
2519 | /* we don't know the level of the root node | |
2520 | * until we actually have it read locked | |
2521 | */ | |
2522 | b = btrfs_read_lock_root_node(root); | |
2523 | level = btrfs_header_level(b); | |
2524 | if (level <= write_lock_level) { | |
2525 | /* whoops, must trade for write lock */ | |
2526 | btrfs_tree_read_unlock(b); | |
2527 | free_extent_buffer(b); | |
2528 | b = btrfs_lock_root_node(root); | |
2529 | root_lock = BTRFS_WRITE_LOCK; | |
2530 | ||
2531 | /* the level might have changed, check again */ | |
2532 | level = btrfs_header_level(b); | |
2533 | } | |
2534 | } | |
5d4f98a2 | 2535 | } |
bd681513 CM |
2536 | p->nodes[level] = b; |
2537 | if (!p->skip_locking) | |
2538 | p->locks[level] = root_lock; | |
925baedd | 2539 | |
eb60ceac | 2540 | while (b) { |
5f39d397 | 2541 | level = btrfs_header_level(b); |
65b51a00 CM |
2542 | |
2543 | /* | |
2544 | * setup the path here so we can release it under lock | |
2545 | * contention with the cow code | |
2546 | */ | |
02217ed2 | 2547 | if (cow) { |
c8c42864 CM |
2548 | /* |
2549 | * if we don't really need to cow this block | |
2550 | * then we don't want to set the path blocking, | |
2551 | * so we test it here | |
2552 | */ | |
5d4f98a2 | 2553 | if (!should_cow_block(trans, root, b)) |
65b51a00 | 2554 | goto cow_done; |
5d4f98a2 | 2555 | |
b4ce94de CM |
2556 | btrfs_set_path_blocking(p); |
2557 | ||
bd681513 CM |
2558 | /* |
2559 | * must have write locks on this node and the | |
2560 | * parent | |
2561 | */ | |
5124e00e JB |
2562 | if (level > write_lock_level || |
2563 | (level + 1 > write_lock_level && | |
2564 | level + 1 < BTRFS_MAX_LEVEL && | |
2565 | p->nodes[level + 1])) { | |
bd681513 CM |
2566 | write_lock_level = level + 1; |
2567 | btrfs_release_path(p); | |
2568 | goto again; | |
2569 | } | |
2570 | ||
33c66f43 YZ |
2571 | err = btrfs_cow_block(trans, root, b, |
2572 | p->nodes[level + 1], | |
2573 | p->slots[level + 1], &b); | |
2574 | if (err) { | |
33c66f43 | 2575 | ret = err; |
65b51a00 | 2576 | goto done; |
54aa1f4d | 2577 | } |
02217ed2 | 2578 | } |
65b51a00 | 2579 | cow_done: |
02217ed2 | 2580 | BUG_ON(!cow && ins_len); |
65b51a00 | 2581 | |
eb60ceac | 2582 | p->nodes[level] = b; |
bd681513 | 2583 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de CM |
2584 | |
2585 | /* | |
2586 | * we have a lock on b and as long as we aren't changing | |
2587 | * the tree, there is no way to for the items in b to change. | |
2588 | * It is safe to drop the lock on our parent before we | |
2589 | * go through the expensive btree search on b. | |
2590 | * | |
2591 | * If cow is true, then we might be changing slot zero, | |
2592 | * which may require changing the parent. So, we can't | |
2593 | * drop the lock until after we know which slot we're | |
2594 | * operating on. | |
2595 | */ | |
2596 | if (!cow) | |
2597 | btrfs_unlock_up_safe(p, level + 1); | |
2598 | ||
5f39d397 | 2599 | ret = bin_search(b, key, level, &slot); |
b4ce94de | 2600 | |
5f39d397 | 2601 | if (level != 0) { |
33c66f43 YZ |
2602 | int dec = 0; |
2603 | if (ret && slot > 0) { | |
2604 | dec = 1; | |
be0e5c09 | 2605 | slot -= 1; |
33c66f43 | 2606 | } |
be0e5c09 | 2607 | p->slots[level] = slot; |
33c66f43 | 2608 | err = setup_nodes_for_search(trans, root, p, b, level, |
bd681513 | 2609 | ins_len, &write_lock_level); |
33c66f43 | 2610 | if (err == -EAGAIN) |
c8c42864 | 2611 | goto again; |
33c66f43 YZ |
2612 | if (err) { |
2613 | ret = err; | |
c8c42864 | 2614 | goto done; |
33c66f43 | 2615 | } |
c8c42864 CM |
2616 | b = p->nodes[level]; |
2617 | slot = p->slots[level]; | |
b4ce94de | 2618 | |
bd681513 CM |
2619 | /* |
2620 | * slot 0 is special, if we change the key | |
2621 | * we have to update the parent pointer | |
2622 | * which means we must have a write lock | |
2623 | * on the parent | |
2624 | */ | |
2625 | if (slot == 0 && cow && | |
2626 | write_lock_level < level + 1) { | |
2627 | write_lock_level = level + 1; | |
2628 | btrfs_release_path(p); | |
2629 | goto again; | |
2630 | } | |
2631 | ||
f7c79f30 CM |
2632 | unlock_up(p, level, lowest_unlock, |
2633 | min_write_lock_level, &write_lock_level); | |
f9efa9c7 | 2634 | |
925baedd | 2635 | if (level == lowest_level) { |
33c66f43 YZ |
2636 | if (dec) |
2637 | p->slots[level]++; | |
5b21f2ed | 2638 | goto done; |
925baedd | 2639 | } |
ca7a79ad | 2640 | |
33c66f43 | 2641 | err = read_block_for_search(trans, root, p, |
5d9e75c4 | 2642 | &b, level, slot, key, 0); |
33c66f43 | 2643 | if (err == -EAGAIN) |
c8c42864 | 2644 | goto again; |
33c66f43 YZ |
2645 | if (err) { |
2646 | ret = err; | |
76a05b35 | 2647 | goto done; |
33c66f43 | 2648 | } |
76a05b35 | 2649 | |
b4ce94de | 2650 | if (!p->skip_locking) { |
bd681513 CM |
2651 | level = btrfs_header_level(b); |
2652 | if (level <= write_lock_level) { | |
2653 | err = btrfs_try_tree_write_lock(b); | |
2654 | if (!err) { | |
2655 | btrfs_set_path_blocking(p); | |
2656 | btrfs_tree_lock(b); | |
2657 | btrfs_clear_path_blocking(p, b, | |
2658 | BTRFS_WRITE_LOCK); | |
2659 | } | |
2660 | p->locks[level] = BTRFS_WRITE_LOCK; | |
2661 | } else { | |
2662 | err = btrfs_try_tree_read_lock(b); | |
2663 | if (!err) { | |
2664 | btrfs_set_path_blocking(p); | |
2665 | btrfs_tree_read_lock(b); | |
2666 | btrfs_clear_path_blocking(p, b, | |
2667 | BTRFS_READ_LOCK); | |
2668 | } | |
2669 | p->locks[level] = BTRFS_READ_LOCK; | |
b4ce94de | 2670 | } |
bd681513 | 2671 | p->nodes[level] = b; |
b4ce94de | 2672 | } |
be0e5c09 CM |
2673 | } else { |
2674 | p->slots[level] = slot; | |
87b29b20 YZ |
2675 | if (ins_len > 0 && |
2676 | btrfs_leaf_free_space(root, b) < ins_len) { | |
bd681513 CM |
2677 | if (write_lock_level < 1) { |
2678 | write_lock_level = 1; | |
2679 | btrfs_release_path(p); | |
2680 | goto again; | |
2681 | } | |
2682 | ||
b4ce94de | 2683 | btrfs_set_path_blocking(p); |
33c66f43 YZ |
2684 | err = split_leaf(trans, root, key, |
2685 | p, ins_len, ret == 0); | |
bd681513 | 2686 | btrfs_clear_path_blocking(p, NULL, 0); |
b4ce94de | 2687 | |
33c66f43 YZ |
2688 | BUG_ON(err > 0); |
2689 | if (err) { | |
2690 | ret = err; | |
65b51a00 CM |
2691 | goto done; |
2692 | } | |
5c680ed6 | 2693 | } |
459931ec | 2694 | if (!p->search_for_split) |
f7c79f30 CM |
2695 | unlock_up(p, level, lowest_unlock, |
2696 | min_write_lock_level, &write_lock_level); | |
65b51a00 | 2697 | goto done; |
be0e5c09 CM |
2698 | } |
2699 | } | |
65b51a00 CM |
2700 | ret = 1; |
2701 | done: | |
b4ce94de CM |
2702 | /* |
2703 | * we don't really know what they plan on doing with the path | |
2704 | * from here on, so for now just mark it as blocking | |
2705 | */ | |
b9473439 CM |
2706 | if (!p->leave_spinning) |
2707 | btrfs_set_path_blocking(p); | |
76a05b35 | 2708 | if (ret < 0) |
b3b4aa74 | 2709 | btrfs_release_path(p); |
65b51a00 | 2710 | return ret; |
be0e5c09 CM |
2711 | } |
2712 | ||
5d9e75c4 JS |
2713 | /* |
2714 | * Like btrfs_search_slot, this looks for a key in the given tree. It uses the | |
2715 | * current state of the tree together with the operations recorded in the tree | |
2716 | * modification log to search for the key in a previous version of this tree, as | |
2717 | * denoted by the time_seq parameter. | |
2718 | * | |
2719 | * Naturally, there is no support for insert, delete or cow operations. | |
2720 | * | |
2721 | * The resulting path and return value will be set up as if we called | |
2722 | * btrfs_search_slot at that point in time with ins_len and cow both set to 0. | |
2723 | */ | |
2724 | int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key, | |
2725 | struct btrfs_path *p, u64 time_seq) | |
2726 | { | |
2727 | struct extent_buffer *b; | |
2728 | int slot; | |
2729 | int ret; | |
2730 | int err; | |
2731 | int level; | |
2732 | int lowest_unlock = 1; | |
2733 | u8 lowest_level = 0; | |
2734 | ||
2735 | lowest_level = p->lowest_level; | |
2736 | WARN_ON(p->nodes[0] != NULL); | |
2737 | ||
2738 | if (p->search_commit_root) { | |
2739 | BUG_ON(time_seq); | |
2740 | return btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2741 | } | |
2742 | ||
2743 | again: | |
5d9e75c4 | 2744 | b = get_old_root(root, time_seq); |
5d9e75c4 | 2745 | level = btrfs_header_level(b); |
5d9e75c4 JS |
2746 | p->locks[level] = BTRFS_READ_LOCK; |
2747 | ||
2748 | while (b) { | |
2749 | level = btrfs_header_level(b); | |
2750 | p->nodes[level] = b; | |
2751 | btrfs_clear_path_blocking(p, NULL, 0); | |
2752 | ||
2753 | /* | |
2754 | * we have a lock on b and as long as we aren't changing | |
2755 | * the tree, there is no way to for the items in b to change. | |
2756 | * It is safe to drop the lock on our parent before we | |
2757 | * go through the expensive btree search on b. | |
2758 | */ | |
2759 | btrfs_unlock_up_safe(p, level + 1); | |
2760 | ||
2761 | ret = bin_search(b, key, level, &slot); | |
2762 | ||
2763 | if (level != 0) { | |
2764 | int dec = 0; | |
2765 | if (ret && slot > 0) { | |
2766 | dec = 1; | |
2767 | slot -= 1; | |
2768 | } | |
2769 | p->slots[level] = slot; | |
2770 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2771 | ||
2772 | if (level == lowest_level) { | |
2773 | if (dec) | |
2774 | p->slots[level]++; | |
2775 | goto done; | |
2776 | } | |
2777 | ||
2778 | err = read_block_for_search(NULL, root, p, &b, level, | |
2779 | slot, key, time_seq); | |
2780 | if (err == -EAGAIN) | |
2781 | goto again; | |
2782 | if (err) { | |
2783 | ret = err; | |
2784 | goto done; | |
2785 | } | |
2786 | ||
2787 | level = btrfs_header_level(b); | |
2788 | err = btrfs_try_tree_read_lock(b); | |
2789 | if (!err) { | |
2790 | btrfs_set_path_blocking(p); | |
2791 | btrfs_tree_read_lock(b); | |
2792 | btrfs_clear_path_blocking(p, b, | |
2793 | BTRFS_READ_LOCK); | |
2794 | } | |
2795 | p->locks[level] = BTRFS_READ_LOCK; | |
2796 | p->nodes[level] = b; | |
2797 | b = tree_mod_log_rewind(root->fs_info, b, time_seq); | |
2798 | if (b != p->nodes[level]) { | |
2799 | btrfs_tree_unlock_rw(p->nodes[level], | |
2800 | p->locks[level]); | |
2801 | p->locks[level] = 0; | |
2802 | p->nodes[level] = b; | |
2803 | } | |
2804 | } else { | |
2805 | p->slots[level] = slot; | |
2806 | unlock_up(p, level, lowest_unlock, 0, NULL); | |
2807 | goto done; | |
2808 | } | |
2809 | } | |
2810 | ret = 1; | |
2811 | done: | |
2812 | if (!p->leave_spinning) | |
2813 | btrfs_set_path_blocking(p); | |
2814 | if (ret < 0) | |
2815 | btrfs_release_path(p); | |
2816 | ||
2817 | return ret; | |
2818 | } | |
2819 | ||
2f38b3e1 AJ |
2820 | /* |
2821 | * helper to use instead of search slot if no exact match is needed but | |
2822 | * instead the next or previous item should be returned. | |
2823 | * When find_higher is true, the next higher item is returned, the next lower | |
2824 | * otherwise. | |
2825 | * When return_any and find_higher are both true, and no higher item is found, | |
2826 | * return the next lower instead. | |
2827 | * When return_any is true and find_higher is false, and no lower item is found, | |
2828 | * return the next higher instead. | |
2829 | * It returns 0 if any item is found, 1 if none is found (tree empty), and | |
2830 | * < 0 on error | |
2831 | */ | |
2832 | int btrfs_search_slot_for_read(struct btrfs_root *root, | |
2833 | struct btrfs_key *key, struct btrfs_path *p, | |
2834 | int find_higher, int return_any) | |
2835 | { | |
2836 | int ret; | |
2837 | struct extent_buffer *leaf; | |
2838 | ||
2839 | again: | |
2840 | ret = btrfs_search_slot(NULL, root, key, p, 0, 0); | |
2841 | if (ret <= 0) | |
2842 | return ret; | |
2843 | /* | |
2844 | * a return value of 1 means the path is at the position where the | |
2845 | * item should be inserted. Normally this is the next bigger item, | |
2846 | * but in case the previous item is the last in a leaf, path points | |
2847 | * to the first free slot in the previous leaf, i.e. at an invalid | |
2848 | * item. | |
2849 | */ | |
2850 | leaf = p->nodes[0]; | |
2851 | ||
2852 | if (find_higher) { | |
2853 | if (p->slots[0] >= btrfs_header_nritems(leaf)) { | |
2854 | ret = btrfs_next_leaf(root, p); | |
2855 | if (ret <= 0) | |
2856 | return ret; | |
2857 | if (!return_any) | |
2858 | return 1; | |
2859 | /* | |
2860 | * no higher item found, return the next | |
2861 | * lower instead | |
2862 | */ | |
2863 | return_any = 0; | |
2864 | find_higher = 0; | |
2865 | btrfs_release_path(p); | |
2866 | goto again; | |
2867 | } | |
2868 | } else { | |
e6793769 AJ |
2869 | if (p->slots[0] == 0) { |
2870 | ret = btrfs_prev_leaf(root, p); | |
2871 | if (ret < 0) | |
2872 | return ret; | |
2873 | if (!ret) { | |
2874 | p->slots[0] = btrfs_header_nritems(leaf) - 1; | |
2875 | return 0; | |
2f38b3e1 | 2876 | } |
e6793769 AJ |
2877 | if (!return_any) |
2878 | return 1; | |
2879 | /* | |
2880 | * no lower item found, return the next | |
2881 | * higher instead | |
2882 | */ | |
2883 | return_any = 0; | |
2884 | find_higher = 1; | |
2885 | btrfs_release_path(p); | |
2886 | goto again; | |
2887 | } else { | |
2f38b3e1 AJ |
2888 | --p->slots[0]; |
2889 | } | |
2890 | } | |
2891 | return 0; | |
2892 | } | |
2893 | ||
74123bd7 CM |
2894 | /* |
2895 | * adjust the pointers going up the tree, starting at level | |
2896 | * making sure the right key of each node is points to 'key'. | |
2897 | * This is used after shifting pointers to the left, so it stops | |
2898 | * fixing up pointers when a given leaf/node is not in slot 0 of the | |
2899 | * higher levels | |
aa5d6bed | 2900 | * |
74123bd7 | 2901 | */ |
143bede5 JM |
2902 | static void fixup_low_keys(struct btrfs_trans_handle *trans, |
2903 | struct btrfs_root *root, struct btrfs_path *path, | |
2904 | struct btrfs_disk_key *key, int level) | |
be0e5c09 CM |
2905 | { |
2906 | int i; | |
5f39d397 CM |
2907 | struct extent_buffer *t; |
2908 | ||
234b63a0 | 2909 | for (i = level; i < BTRFS_MAX_LEVEL; i++) { |
be0e5c09 | 2910 | int tslot = path->slots[i]; |
eb60ceac | 2911 | if (!path->nodes[i]) |
be0e5c09 | 2912 | break; |
5f39d397 | 2913 | t = path->nodes[i]; |
32adf090 | 2914 | tree_mod_log_set_node_key(root->fs_info, t, tslot, 1); |
5f39d397 | 2915 | btrfs_set_node_key(t, key, tslot); |
d6025579 | 2916 | btrfs_mark_buffer_dirty(path->nodes[i]); |
be0e5c09 CM |
2917 | if (tslot != 0) |
2918 | break; | |
2919 | } | |
2920 | } | |
2921 | ||
31840ae1 ZY |
2922 | /* |
2923 | * update item key. | |
2924 | * | |
2925 | * This function isn't completely safe. It's the caller's responsibility | |
2926 | * that the new key won't break the order | |
2927 | */ | |
143bede5 JM |
2928 | void btrfs_set_item_key_safe(struct btrfs_trans_handle *trans, |
2929 | struct btrfs_root *root, struct btrfs_path *path, | |
2930 | struct btrfs_key *new_key) | |
31840ae1 ZY |
2931 | { |
2932 | struct btrfs_disk_key disk_key; | |
2933 | struct extent_buffer *eb; | |
2934 | int slot; | |
2935 | ||
2936 | eb = path->nodes[0]; | |
2937 | slot = path->slots[0]; | |
2938 | if (slot > 0) { | |
2939 | btrfs_item_key(eb, &disk_key, slot - 1); | |
143bede5 | 2940 | BUG_ON(comp_keys(&disk_key, new_key) >= 0); |
31840ae1 ZY |
2941 | } |
2942 | if (slot < btrfs_header_nritems(eb) - 1) { | |
2943 | btrfs_item_key(eb, &disk_key, slot + 1); | |
143bede5 | 2944 | BUG_ON(comp_keys(&disk_key, new_key) <= 0); |
31840ae1 ZY |
2945 | } |
2946 | ||
2947 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
2948 | btrfs_set_item_key(eb, &disk_key, slot); | |
2949 | btrfs_mark_buffer_dirty(eb); | |
2950 | if (slot == 0) | |
2951 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
31840ae1 ZY |
2952 | } |
2953 | ||
74123bd7 CM |
2954 | /* |
2955 | * try to push data from one node into the next node left in the | |
79f95c82 | 2956 | * tree. |
aa5d6bed CM |
2957 | * |
2958 | * returns 0 if some ptrs were pushed left, < 0 if there was some horrible | |
2959 | * error, and > 0 if there was no room in the left hand block. | |
74123bd7 | 2960 | */ |
98ed5174 CM |
2961 | static int push_node_left(struct btrfs_trans_handle *trans, |
2962 | struct btrfs_root *root, struct extent_buffer *dst, | |
971a1f66 | 2963 | struct extent_buffer *src, int empty) |
be0e5c09 | 2964 | { |
be0e5c09 | 2965 | int push_items = 0; |
bb803951 CM |
2966 | int src_nritems; |
2967 | int dst_nritems; | |
aa5d6bed | 2968 | int ret = 0; |
be0e5c09 | 2969 | |
5f39d397 CM |
2970 | src_nritems = btrfs_header_nritems(src); |
2971 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 2972 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
7bb86316 CM |
2973 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
2974 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
54aa1f4d | 2975 | |
bce4eae9 | 2976 | if (!empty && src_nritems <= 8) |
971a1f66 CM |
2977 | return 1; |
2978 | ||
d397712b | 2979 | if (push_items <= 0) |
be0e5c09 CM |
2980 | return 1; |
2981 | ||
bce4eae9 | 2982 | if (empty) { |
971a1f66 | 2983 | push_items = min(src_nritems, push_items); |
bce4eae9 CM |
2984 | if (push_items < src_nritems) { |
2985 | /* leave at least 8 pointers in the node if | |
2986 | * we aren't going to empty it | |
2987 | */ | |
2988 | if (src_nritems - push_items < 8) { | |
2989 | if (push_items <= 8) | |
2990 | return 1; | |
2991 | push_items -= 8; | |
2992 | } | |
2993 | } | |
2994 | } else | |
2995 | push_items = min(src_nritems - 8, push_items); | |
79f95c82 | 2996 | |
f230475e JS |
2997 | tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0, |
2998 | push_items); | |
5f39d397 CM |
2999 | copy_extent_buffer(dst, src, |
3000 | btrfs_node_key_ptr_offset(dst_nritems), | |
3001 | btrfs_node_key_ptr_offset(0), | |
d397712b | 3002 | push_items * sizeof(struct btrfs_key_ptr)); |
5f39d397 | 3003 | |
bb803951 | 3004 | if (push_items < src_nritems) { |
57911b8b JS |
3005 | /* |
3006 | * don't call tree_mod_log_eb_move here, key removal was already | |
3007 | * fully logged by tree_mod_log_eb_copy above. | |
3008 | */ | |
5f39d397 CM |
3009 | memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0), |
3010 | btrfs_node_key_ptr_offset(push_items), | |
3011 | (src_nritems - push_items) * | |
3012 | sizeof(struct btrfs_key_ptr)); | |
3013 | } | |
3014 | btrfs_set_header_nritems(src, src_nritems - push_items); | |
3015 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
3016 | btrfs_mark_buffer_dirty(src); | |
3017 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3018 | |
79f95c82 CM |
3019 | return ret; |
3020 | } | |
3021 | ||
3022 | /* | |
3023 | * try to push data from one node into the next node right in the | |
3024 | * tree. | |
3025 | * | |
3026 | * returns 0 if some ptrs were pushed, < 0 if there was some horrible | |
3027 | * error, and > 0 if there was no room in the right hand block. | |
3028 | * | |
3029 | * this will only push up to 1/2 the contents of the left node over | |
3030 | */ | |
5f39d397 CM |
3031 | static int balance_node_right(struct btrfs_trans_handle *trans, |
3032 | struct btrfs_root *root, | |
3033 | struct extent_buffer *dst, | |
3034 | struct extent_buffer *src) | |
79f95c82 | 3035 | { |
79f95c82 CM |
3036 | int push_items = 0; |
3037 | int max_push; | |
3038 | int src_nritems; | |
3039 | int dst_nritems; | |
3040 | int ret = 0; | |
79f95c82 | 3041 | |
7bb86316 CM |
3042 | WARN_ON(btrfs_header_generation(src) != trans->transid); |
3043 | WARN_ON(btrfs_header_generation(dst) != trans->transid); | |
3044 | ||
5f39d397 CM |
3045 | src_nritems = btrfs_header_nritems(src); |
3046 | dst_nritems = btrfs_header_nritems(dst); | |
123abc88 | 3047 | push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems; |
d397712b | 3048 | if (push_items <= 0) |
79f95c82 | 3049 | return 1; |
bce4eae9 | 3050 | |
d397712b | 3051 | if (src_nritems < 4) |
bce4eae9 | 3052 | return 1; |
79f95c82 CM |
3053 | |
3054 | max_push = src_nritems / 2 + 1; | |
3055 | /* don't try to empty the node */ | |
d397712b | 3056 | if (max_push >= src_nritems) |
79f95c82 | 3057 | return 1; |
252c38f0 | 3058 | |
79f95c82 CM |
3059 | if (max_push < push_items) |
3060 | push_items = max_push; | |
3061 | ||
f230475e | 3062 | tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems); |
5f39d397 CM |
3063 | memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items), |
3064 | btrfs_node_key_ptr_offset(0), | |
3065 | (dst_nritems) * | |
3066 | sizeof(struct btrfs_key_ptr)); | |
d6025579 | 3067 | |
f230475e JS |
3068 | tree_mod_log_eb_copy(root->fs_info, dst, src, 0, |
3069 | src_nritems - push_items, push_items); | |
5f39d397 CM |
3070 | copy_extent_buffer(dst, src, |
3071 | btrfs_node_key_ptr_offset(0), | |
3072 | btrfs_node_key_ptr_offset(src_nritems - push_items), | |
d397712b | 3073 | push_items * sizeof(struct btrfs_key_ptr)); |
79f95c82 | 3074 | |
5f39d397 CM |
3075 | btrfs_set_header_nritems(src, src_nritems - push_items); |
3076 | btrfs_set_header_nritems(dst, dst_nritems + push_items); | |
79f95c82 | 3077 | |
5f39d397 CM |
3078 | btrfs_mark_buffer_dirty(src); |
3079 | btrfs_mark_buffer_dirty(dst); | |
31840ae1 | 3080 | |
aa5d6bed | 3081 | return ret; |
be0e5c09 CM |
3082 | } |
3083 | ||
97571fd0 CM |
3084 | /* |
3085 | * helper function to insert a new root level in the tree. | |
3086 | * A new node is allocated, and a single item is inserted to | |
3087 | * point to the existing root | |
aa5d6bed CM |
3088 | * |
3089 | * returns zero on success or < 0 on failure. | |
97571fd0 | 3090 | */ |
d397712b | 3091 | static noinline int insert_new_root(struct btrfs_trans_handle *trans, |
5f39d397 CM |
3092 | struct btrfs_root *root, |
3093 | struct btrfs_path *path, int level) | |
5c680ed6 | 3094 | { |
7bb86316 | 3095 | u64 lower_gen; |
5f39d397 CM |
3096 | struct extent_buffer *lower; |
3097 | struct extent_buffer *c; | |
925baedd | 3098 | struct extent_buffer *old; |
5f39d397 | 3099 | struct btrfs_disk_key lower_key; |
5c680ed6 CM |
3100 | |
3101 | BUG_ON(path->nodes[level]); | |
3102 | BUG_ON(path->nodes[level-1] != root->node); | |
3103 | ||
7bb86316 CM |
3104 | lower = path->nodes[level-1]; |
3105 | if (level == 1) | |
3106 | btrfs_item_key(lower, &lower_key, 0); | |
3107 | else | |
3108 | btrfs_node_key(lower, &lower_key, 0); | |
3109 | ||
31840ae1 | 3110 | c = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
5d4f98a2 | 3111 | root->root_key.objectid, &lower_key, |
5581a51a | 3112 | level, root->node->start, 0); |
5f39d397 CM |
3113 | if (IS_ERR(c)) |
3114 | return PTR_ERR(c); | |
925baedd | 3115 | |
f0486c68 YZ |
3116 | root_add_used(root, root->nodesize); |
3117 | ||
5d4f98a2 | 3118 | memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 CM |
3119 | btrfs_set_header_nritems(c, 1); |
3120 | btrfs_set_header_level(c, level); | |
db94535d | 3121 | btrfs_set_header_bytenr(c, c->start); |
5f39d397 | 3122 | btrfs_set_header_generation(c, trans->transid); |
5d4f98a2 | 3123 | btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV); |
5f39d397 | 3124 | btrfs_set_header_owner(c, root->root_key.objectid); |
5f39d397 CM |
3125 | |
3126 | write_extent_buffer(c, root->fs_info->fsid, | |
3127 | (unsigned long)btrfs_header_fsid(c), | |
3128 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3129 | |
3130 | write_extent_buffer(c, root->fs_info->chunk_tree_uuid, | |
3131 | (unsigned long)btrfs_header_chunk_tree_uuid(c), | |
3132 | BTRFS_UUID_SIZE); | |
3133 | ||
5f39d397 | 3134 | btrfs_set_node_key(c, &lower_key, 0); |
db94535d | 3135 | btrfs_set_node_blockptr(c, 0, lower->start); |
7bb86316 | 3136 | lower_gen = btrfs_header_generation(lower); |
31840ae1 | 3137 | WARN_ON(lower_gen != trans->transid); |
7bb86316 CM |
3138 | |
3139 | btrfs_set_node_ptr_generation(c, 0, lower_gen); | |
d5719762 | 3140 | |
5f39d397 | 3141 | btrfs_mark_buffer_dirty(c); |
d5719762 | 3142 | |
925baedd | 3143 | old = root->node; |
f230475e | 3144 | tree_mod_log_set_root_pointer(root, c); |
240f62c8 | 3145 | rcu_assign_pointer(root->node, c); |
925baedd CM |
3146 | |
3147 | /* the super has an extra ref to root->node */ | |
3148 | free_extent_buffer(old); | |
3149 | ||
0b86a832 | 3150 | add_root_to_dirty_list(root); |
5f39d397 CM |
3151 | extent_buffer_get(c); |
3152 | path->nodes[level] = c; | |
bd681513 | 3153 | path->locks[level] = BTRFS_WRITE_LOCK; |
5c680ed6 CM |
3154 | path->slots[level] = 0; |
3155 | return 0; | |
3156 | } | |
3157 | ||
74123bd7 CM |
3158 | /* |
3159 | * worker function to insert a single pointer in a node. | |
3160 | * the node should have enough room for the pointer already | |
97571fd0 | 3161 | * |
74123bd7 CM |
3162 | * slot and level indicate where you want the key to go, and |
3163 | * blocknr is the block the key points to. | |
3164 | */ | |
143bede5 JM |
3165 | static void insert_ptr(struct btrfs_trans_handle *trans, |
3166 | struct btrfs_root *root, struct btrfs_path *path, | |
3167 | struct btrfs_disk_key *key, u64 bytenr, | |
c3e06965 | 3168 | int slot, int level) |
74123bd7 | 3169 | { |
5f39d397 | 3170 | struct extent_buffer *lower; |
74123bd7 | 3171 | int nritems; |
f3ea38da | 3172 | int ret; |
5c680ed6 CM |
3173 | |
3174 | BUG_ON(!path->nodes[level]); | |
f0486c68 | 3175 | btrfs_assert_tree_locked(path->nodes[level]); |
5f39d397 CM |
3176 | lower = path->nodes[level]; |
3177 | nritems = btrfs_header_nritems(lower); | |
c293498b | 3178 | BUG_ON(slot > nritems); |
143bede5 | 3179 | BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root)); |
74123bd7 | 3180 | if (slot != nritems) { |
c3e06965 | 3181 | if (level) |
f3ea38da JS |
3182 | tree_mod_log_eb_move(root->fs_info, lower, slot + 1, |
3183 | slot, nritems - slot); | |
5f39d397 CM |
3184 | memmove_extent_buffer(lower, |
3185 | btrfs_node_key_ptr_offset(slot + 1), | |
3186 | btrfs_node_key_ptr_offset(slot), | |
d6025579 | 3187 | (nritems - slot) * sizeof(struct btrfs_key_ptr)); |
74123bd7 | 3188 | } |
c3e06965 | 3189 | if (level) { |
f3ea38da JS |
3190 | ret = tree_mod_log_insert_key(root->fs_info, lower, slot, |
3191 | MOD_LOG_KEY_ADD); | |
3192 | BUG_ON(ret < 0); | |
3193 | } | |
5f39d397 | 3194 | btrfs_set_node_key(lower, key, slot); |
db94535d | 3195 | btrfs_set_node_blockptr(lower, slot, bytenr); |
74493f7a CM |
3196 | WARN_ON(trans->transid == 0); |
3197 | btrfs_set_node_ptr_generation(lower, slot, trans->transid); | |
5f39d397 CM |
3198 | btrfs_set_header_nritems(lower, nritems + 1); |
3199 | btrfs_mark_buffer_dirty(lower); | |
74123bd7 CM |
3200 | } |
3201 | ||
97571fd0 CM |
3202 | /* |
3203 | * split the node at the specified level in path in two. | |
3204 | * The path is corrected to point to the appropriate node after the split | |
3205 | * | |
3206 | * Before splitting this tries to make some room in the node by pushing | |
3207 | * left and right, if either one works, it returns right away. | |
aa5d6bed CM |
3208 | * |
3209 | * returns 0 on success and < 0 on failure | |
97571fd0 | 3210 | */ |
e02119d5 CM |
3211 | static noinline int split_node(struct btrfs_trans_handle *trans, |
3212 | struct btrfs_root *root, | |
3213 | struct btrfs_path *path, int level) | |
be0e5c09 | 3214 | { |
5f39d397 CM |
3215 | struct extent_buffer *c; |
3216 | struct extent_buffer *split; | |
3217 | struct btrfs_disk_key disk_key; | |
be0e5c09 | 3218 | int mid; |
5c680ed6 | 3219 | int ret; |
7518a238 | 3220 | u32 c_nritems; |
eb60ceac | 3221 | |
5f39d397 | 3222 | c = path->nodes[level]; |
7bb86316 | 3223 | WARN_ON(btrfs_header_generation(c) != trans->transid); |
5f39d397 | 3224 | if (c == root->node) { |
5c680ed6 | 3225 | /* trying to split the root, lets make a new one */ |
e089f05c | 3226 | ret = insert_new_root(trans, root, path, level + 1); |
5c680ed6 CM |
3227 | if (ret) |
3228 | return ret; | |
b3612421 | 3229 | } else { |
e66f709b | 3230 | ret = push_nodes_for_insert(trans, root, path, level); |
5f39d397 CM |
3231 | c = path->nodes[level]; |
3232 | if (!ret && btrfs_header_nritems(c) < | |
c448acf0 | 3233 | BTRFS_NODEPTRS_PER_BLOCK(root) - 3) |
e66f709b | 3234 | return 0; |
54aa1f4d CM |
3235 | if (ret < 0) |
3236 | return ret; | |
be0e5c09 | 3237 | } |
e66f709b | 3238 | |
5f39d397 | 3239 | c_nritems = btrfs_header_nritems(c); |
5d4f98a2 YZ |
3240 | mid = (c_nritems + 1) / 2; |
3241 | btrfs_node_key(c, &disk_key, mid); | |
7bb86316 | 3242 | |
5d4f98a2 | 3243 | split = btrfs_alloc_free_block(trans, root, root->nodesize, 0, |
31840ae1 | 3244 | root->root_key.objectid, |
5581a51a | 3245 | &disk_key, level, c->start, 0); |
5f39d397 CM |
3246 | if (IS_ERR(split)) |
3247 | return PTR_ERR(split); | |
3248 | ||
f0486c68 YZ |
3249 | root_add_used(root, root->nodesize); |
3250 | ||
5d4f98a2 | 3251 | memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header)); |
5f39d397 | 3252 | btrfs_set_header_level(split, btrfs_header_level(c)); |
db94535d | 3253 | btrfs_set_header_bytenr(split, split->start); |
5f39d397 | 3254 | btrfs_set_header_generation(split, trans->transid); |
5d4f98a2 | 3255 | btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
3256 | btrfs_set_header_owner(split, root->root_key.objectid); |
3257 | write_extent_buffer(split, root->fs_info->fsid, | |
3258 | (unsigned long)btrfs_header_fsid(split), | |
3259 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
3260 | write_extent_buffer(split, root->fs_info->chunk_tree_uuid, |
3261 | (unsigned long)btrfs_header_chunk_tree_uuid(split), | |
3262 | BTRFS_UUID_SIZE); | |
54aa1f4d | 3263 | |
f230475e | 3264 | tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid); |
5f39d397 CM |
3265 | copy_extent_buffer(split, c, |
3266 | btrfs_node_key_ptr_offset(0), | |
3267 | btrfs_node_key_ptr_offset(mid), | |
3268 | (c_nritems - mid) * sizeof(struct btrfs_key_ptr)); | |
3269 | btrfs_set_header_nritems(split, c_nritems - mid); | |
3270 | btrfs_set_header_nritems(c, mid); | |
aa5d6bed CM |
3271 | ret = 0; |
3272 | ||
5f39d397 CM |
3273 | btrfs_mark_buffer_dirty(c); |
3274 | btrfs_mark_buffer_dirty(split); | |
3275 | ||
143bede5 | 3276 | insert_ptr(trans, root, path, &disk_key, split->start, |
c3e06965 | 3277 | path->slots[level + 1] + 1, level + 1); |
aa5d6bed | 3278 | |
5de08d7d | 3279 | if (path->slots[level] >= mid) { |
5c680ed6 | 3280 | path->slots[level] -= mid; |
925baedd | 3281 | btrfs_tree_unlock(c); |
5f39d397 CM |
3282 | free_extent_buffer(c); |
3283 | path->nodes[level] = split; | |
5c680ed6 CM |
3284 | path->slots[level + 1] += 1; |
3285 | } else { | |
925baedd | 3286 | btrfs_tree_unlock(split); |
5f39d397 | 3287 | free_extent_buffer(split); |
be0e5c09 | 3288 | } |
aa5d6bed | 3289 | return ret; |
be0e5c09 CM |
3290 | } |
3291 | ||
74123bd7 CM |
3292 | /* |
3293 | * how many bytes are required to store the items in a leaf. start | |
3294 | * and nr indicate which items in the leaf to check. This totals up the | |
3295 | * space used both by the item structs and the item data | |
3296 | */ | |
5f39d397 | 3297 | static int leaf_space_used(struct extent_buffer *l, int start, int nr) |
be0e5c09 | 3298 | { |
41be1f3b JB |
3299 | struct btrfs_item *start_item; |
3300 | struct btrfs_item *end_item; | |
3301 | struct btrfs_map_token token; | |
be0e5c09 | 3302 | int data_len; |
5f39d397 | 3303 | int nritems = btrfs_header_nritems(l); |
d4dbff95 | 3304 | int end = min(nritems, start + nr) - 1; |
be0e5c09 CM |
3305 | |
3306 | if (!nr) | |
3307 | return 0; | |
41be1f3b JB |
3308 | btrfs_init_map_token(&token); |
3309 | start_item = btrfs_item_nr(l, start); | |
3310 | end_item = btrfs_item_nr(l, end); | |
3311 | data_len = btrfs_token_item_offset(l, start_item, &token) + | |
3312 | btrfs_token_item_size(l, start_item, &token); | |
3313 | data_len = data_len - btrfs_token_item_offset(l, end_item, &token); | |
0783fcfc | 3314 | data_len += sizeof(struct btrfs_item) * nr; |
d4dbff95 | 3315 | WARN_ON(data_len < 0); |
be0e5c09 CM |
3316 | return data_len; |
3317 | } | |
3318 | ||
d4dbff95 CM |
3319 | /* |
3320 | * The space between the end of the leaf items and | |
3321 | * the start of the leaf data. IOW, how much room | |
3322 | * the leaf has left for both items and data | |
3323 | */ | |
d397712b | 3324 | noinline int btrfs_leaf_free_space(struct btrfs_root *root, |
e02119d5 | 3325 | struct extent_buffer *leaf) |
d4dbff95 | 3326 | { |
5f39d397 CM |
3327 | int nritems = btrfs_header_nritems(leaf); |
3328 | int ret; | |
3329 | ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems); | |
3330 | if (ret < 0) { | |
d397712b CM |
3331 | printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, " |
3332 | "used %d nritems %d\n", | |
ae2f5411 | 3333 | ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root), |
5f39d397 CM |
3334 | leaf_space_used(leaf, 0, nritems), nritems); |
3335 | } | |
3336 | return ret; | |
d4dbff95 CM |
3337 | } |
3338 | ||
99d8f83c CM |
3339 | /* |
3340 | * min slot controls the lowest index we're willing to push to the | |
3341 | * right. We'll push up to and including min_slot, but no lower | |
3342 | */ | |
44871b1b CM |
3343 | static noinline int __push_leaf_right(struct btrfs_trans_handle *trans, |
3344 | struct btrfs_root *root, | |
3345 | struct btrfs_path *path, | |
3346 | int data_size, int empty, | |
3347 | struct extent_buffer *right, | |
99d8f83c CM |
3348 | int free_space, u32 left_nritems, |
3349 | u32 min_slot) | |
00ec4c51 | 3350 | { |
5f39d397 | 3351 | struct extent_buffer *left = path->nodes[0]; |
44871b1b | 3352 | struct extent_buffer *upper = path->nodes[1]; |
cfed81a0 | 3353 | struct btrfs_map_token token; |
5f39d397 | 3354 | struct btrfs_disk_key disk_key; |
00ec4c51 | 3355 | int slot; |
34a38218 | 3356 | u32 i; |
00ec4c51 CM |
3357 | int push_space = 0; |
3358 | int push_items = 0; | |
0783fcfc | 3359 | struct btrfs_item *item; |
34a38218 | 3360 | u32 nr; |
7518a238 | 3361 | u32 right_nritems; |
5f39d397 | 3362 | u32 data_end; |
db94535d | 3363 | u32 this_item_size; |
00ec4c51 | 3364 | |
cfed81a0 CM |
3365 | btrfs_init_map_token(&token); |
3366 | ||
34a38218 CM |
3367 | if (empty) |
3368 | nr = 0; | |
3369 | else | |
99d8f83c | 3370 | nr = max_t(u32, 1, min_slot); |
34a38218 | 3371 | |
31840ae1 | 3372 | if (path->slots[0] >= left_nritems) |
87b29b20 | 3373 | push_space += data_size; |
31840ae1 | 3374 | |
44871b1b | 3375 | slot = path->slots[1]; |
34a38218 CM |
3376 | i = left_nritems - 1; |
3377 | while (i >= nr) { | |
5f39d397 | 3378 | item = btrfs_item_nr(left, i); |
db94535d | 3379 | |
31840ae1 ZY |
3380 | if (!empty && push_items > 0) { |
3381 | if (path->slots[0] > i) | |
3382 | break; | |
3383 | if (path->slots[0] == i) { | |
3384 | int space = btrfs_leaf_free_space(root, left); | |
3385 | if (space + push_space * 2 > free_space) | |
3386 | break; | |
3387 | } | |
3388 | } | |
3389 | ||
00ec4c51 | 3390 | if (path->slots[0] == i) |
87b29b20 | 3391 | push_space += data_size; |
db94535d | 3392 | |
db94535d CM |
3393 | this_item_size = btrfs_item_size(left, item); |
3394 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
00ec4c51 | 3395 | break; |
31840ae1 | 3396 | |
00ec4c51 | 3397 | push_items++; |
db94535d | 3398 | push_space += this_item_size + sizeof(*item); |
34a38218 CM |
3399 | if (i == 0) |
3400 | break; | |
3401 | i--; | |
db94535d | 3402 | } |
5f39d397 | 3403 | |
925baedd CM |
3404 | if (push_items == 0) |
3405 | goto out_unlock; | |
5f39d397 | 3406 | |
6c1500f2 | 3407 | WARN_ON(!empty && push_items == left_nritems); |
5f39d397 | 3408 | |
00ec4c51 | 3409 | /* push left to right */ |
5f39d397 | 3410 | right_nritems = btrfs_header_nritems(right); |
34a38218 | 3411 | |
5f39d397 | 3412 | push_space = btrfs_item_end_nr(left, left_nritems - push_items); |
123abc88 | 3413 | push_space -= leaf_data_end(root, left); |
5f39d397 | 3414 | |
00ec4c51 | 3415 | /* make room in the right data area */ |
5f39d397 CM |
3416 | data_end = leaf_data_end(root, right); |
3417 | memmove_extent_buffer(right, | |
3418 | btrfs_leaf_data(right) + data_end - push_space, | |
3419 | btrfs_leaf_data(right) + data_end, | |
3420 | BTRFS_LEAF_DATA_SIZE(root) - data_end); | |
3421 | ||
00ec4c51 | 3422 | /* copy from the left data area */ |
5f39d397 | 3423 | copy_extent_buffer(right, left, btrfs_leaf_data(right) + |
d6025579 CM |
3424 | BTRFS_LEAF_DATA_SIZE(root) - push_space, |
3425 | btrfs_leaf_data(left) + leaf_data_end(root, left), | |
3426 | push_space); | |
5f39d397 CM |
3427 | |
3428 | memmove_extent_buffer(right, btrfs_item_nr_offset(push_items), | |
3429 | btrfs_item_nr_offset(0), | |
3430 | right_nritems * sizeof(struct btrfs_item)); | |
3431 | ||
00ec4c51 | 3432 | /* copy the items from left to right */ |
5f39d397 CM |
3433 | copy_extent_buffer(right, left, btrfs_item_nr_offset(0), |
3434 | btrfs_item_nr_offset(left_nritems - push_items), | |
3435 | push_items * sizeof(struct btrfs_item)); | |
00ec4c51 CM |
3436 | |
3437 | /* update the item pointers */ | |
7518a238 | 3438 | right_nritems += push_items; |
5f39d397 | 3439 | btrfs_set_header_nritems(right, right_nritems); |
123abc88 | 3440 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
7518a238 | 3441 | for (i = 0; i < right_nritems; i++) { |
5f39d397 | 3442 | item = btrfs_item_nr(right, i); |
cfed81a0 CM |
3443 | push_space -= btrfs_token_item_size(right, item, &token); |
3444 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d CM |
3445 | } |
3446 | ||
7518a238 | 3447 | left_nritems -= push_items; |
5f39d397 | 3448 | btrfs_set_header_nritems(left, left_nritems); |
00ec4c51 | 3449 | |
34a38218 CM |
3450 | if (left_nritems) |
3451 | btrfs_mark_buffer_dirty(left); | |
f0486c68 YZ |
3452 | else |
3453 | clean_tree_block(trans, root, left); | |
3454 | ||
5f39d397 | 3455 | btrfs_mark_buffer_dirty(right); |
a429e513 | 3456 | |
5f39d397 CM |
3457 | btrfs_item_key(right, &disk_key, 0); |
3458 | btrfs_set_node_key(upper, &disk_key, slot + 1); | |
d6025579 | 3459 | btrfs_mark_buffer_dirty(upper); |
02217ed2 | 3460 | |
00ec4c51 | 3461 | /* then fixup the leaf pointer in the path */ |
7518a238 CM |
3462 | if (path->slots[0] >= left_nritems) { |
3463 | path->slots[0] -= left_nritems; | |
925baedd CM |
3464 | if (btrfs_header_nritems(path->nodes[0]) == 0) |
3465 | clean_tree_block(trans, root, path->nodes[0]); | |
3466 | btrfs_tree_unlock(path->nodes[0]); | |
5f39d397 CM |
3467 | free_extent_buffer(path->nodes[0]); |
3468 | path->nodes[0] = right; | |
00ec4c51 CM |
3469 | path->slots[1] += 1; |
3470 | } else { | |
925baedd | 3471 | btrfs_tree_unlock(right); |
5f39d397 | 3472 | free_extent_buffer(right); |
00ec4c51 CM |
3473 | } |
3474 | return 0; | |
925baedd CM |
3475 | |
3476 | out_unlock: | |
3477 | btrfs_tree_unlock(right); | |
3478 | free_extent_buffer(right); | |
3479 | return 1; | |
00ec4c51 | 3480 | } |
925baedd | 3481 | |
44871b1b CM |
3482 | /* |
3483 | * push some data in the path leaf to the right, trying to free up at | |
3484 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
3485 | * | |
3486 | * returns 1 if the push failed because the other node didn't have enough | |
3487 | * room, 0 if everything worked out and < 0 if there were major errors. | |
99d8f83c CM |
3488 | * |
3489 | * this will push starting from min_slot to the end of the leaf. It won't | |
3490 | * push any slot lower than min_slot | |
44871b1b CM |
3491 | */ |
3492 | static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3493 | *root, struct btrfs_path *path, |
3494 | int min_data_size, int data_size, | |
3495 | int empty, u32 min_slot) | |
44871b1b CM |
3496 | { |
3497 | struct extent_buffer *left = path->nodes[0]; | |
3498 | struct extent_buffer *right; | |
3499 | struct extent_buffer *upper; | |
3500 | int slot; | |
3501 | int free_space; | |
3502 | u32 left_nritems; | |
3503 | int ret; | |
3504 | ||
3505 | if (!path->nodes[1]) | |
3506 | return 1; | |
3507 | ||
3508 | slot = path->slots[1]; | |
3509 | upper = path->nodes[1]; | |
3510 | if (slot >= btrfs_header_nritems(upper) - 1) | |
3511 | return 1; | |
3512 | ||
3513 | btrfs_assert_tree_locked(path->nodes[1]); | |
3514 | ||
3515 | right = read_node_slot(root, upper, slot + 1); | |
91ca338d TI |
3516 | if (right == NULL) |
3517 | return 1; | |
3518 | ||
44871b1b CM |
3519 | btrfs_tree_lock(right); |
3520 | btrfs_set_lock_blocking(right); | |
3521 | ||
3522 | free_space = btrfs_leaf_free_space(root, right); | |
3523 | if (free_space < data_size) | |
3524 | goto out_unlock; | |
3525 | ||
3526 | /* cow and double check */ | |
3527 | ret = btrfs_cow_block(trans, root, right, upper, | |
3528 | slot + 1, &right); | |
3529 | if (ret) | |
3530 | goto out_unlock; | |
3531 | ||
3532 | free_space = btrfs_leaf_free_space(root, right); | |
3533 | if (free_space < data_size) | |
3534 | goto out_unlock; | |
3535 | ||
3536 | left_nritems = btrfs_header_nritems(left); | |
3537 | if (left_nritems == 0) | |
3538 | goto out_unlock; | |
3539 | ||
99d8f83c CM |
3540 | return __push_leaf_right(trans, root, path, min_data_size, empty, |
3541 | right, free_space, left_nritems, min_slot); | |
44871b1b CM |
3542 | out_unlock: |
3543 | btrfs_tree_unlock(right); | |
3544 | free_extent_buffer(right); | |
3545 | return 1; | |
3546 | } | |
3547 | ||
74123bd7 CM |
3548 | /* |
3549 | * push some data in the path leaf to the left, trying to free up at | |
3550 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3551 | * |
3552 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3553 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the | |
3554 | * items | |
74123bd7 | 3555 | */ |
44871b1b CM |
3556 | static noinline int __push_leaf_left(struct btrfs_trans_handle *trans, |
3557 | struct btrfs_root *root, | |
3558 | struct btrfs_path *path, int data_size, | |
3559 | int empty, struct extent_buffer *left, | |
99d8f83c CM |
3560 | int free_space, u32 right_nritems, |
3561 | u32 max_slot) | |
be0e5c09 | 3562 | { |
5f39d397 CM |
3563 | struct btrfs_disk_key disk_key; |
3564 | struct extent_buffer *right = path->nodes[0]; | |
be0e5c09 | 3565 | int i; |
be0e5c09 CM |
3566 | int push_space = 0; |
3567 | int push_items = 0; | |
0783fcfc | 3568 | struct btrfs_item *item; |
7518a238 | 3569 | u32 old_left_nritems; |
34a38218 | 3570 | u32 nr; |
aa5d6bed | 3571 | int ret = 0; |
db94535d CM |
3572 | u32 this_item_size; |
3573 | u32 old_left_item_size; | |
cfed81a0 CM |
3574 | struct btrfs_map_token token; |
3575 | ||
3576 | btrfs_init_map_token(&token); | |
be0e5c09 | 3577 | |
34a38218 | 3578 | if (empty) |
99d8f83c | 3579 | nr = min(right_nritems, max_slot); |
34a38218 | 3580 | else |
99d8f83c | 3581 | nr = min(right_nritems - 1, max_slot); |
34a38218 CM |
3582 | |
3583 | for (i = 0; i < nr; i++) { | |
5f39d397 | 3584 | item = btrfs_item_nr(right, i); |
db94535d | 3585 | |
31840ae1 ZY |
3586 | if (!empty && push_items > 0) { |
3587 | if (path->slots[0] < i) | |
3588 | break; | |
3589 | if (path->slots[0] == i) { | |
3590 | int space = btrfs_leaf_free_space(root, right); | |
3591 | if (space + push_space * 2 > free_space) | |
3592 | break; | |
3593 | } | |
3594 | } | |
3595 | ||
be0e5c09 | 3596 | if (path->slots[0] == i) |
87b29b20 | 3597 | push_space += data_size; |
db94535d CM |
3598 | |
3599 | this_item_size = btrfs_item_size(right, item); | |
3600 | if (this_item_size + sizeof(*item) + push_space > free_space) | |
be0e5c09 | 3601 | break; |
db94535d | 3602 | |
be0e5c09 | 3603 | push_items++; |
db94535d CM |
3604 | push_space += this_item_size + sizeof(*item); |
3605 | } | |
3606 | ||
be0e5c09 | 3607 | if (push_items == 0) { |
925baedd CM |
3608 | ret = 1; |
3609 | goto out; | |
be0e5c09 | 3610 | } |
34a38218 | 3611 | if (!empty && push_items == btrfs_header_nritems(right)) |
a429e513 | 3612 | WARN_ON(1); |
5f39d397 | 3613 | |
be0e5c09 | 3614 | /* push data from right to left */ |
5f39d397 CM |
3615 | copy_extent_buffer(left, right, |
3616 | btrfs_item_nr_offset(btrfs_header_nritems(left)), | |
3617 | btrfs_item_nr_offset(0), | |
3618 | push_items * sizeof(struct btrfs_item)); | |
3619 | ||
123abc88 | 3620 | push_space = BTRFS_LEAF_DATA_SIZE(root) - |
d397712b | 3621 | btrfs_item_offset_nr(right, push_items - 1); |
5f39d397 CM |
3622 | |
3623 | copy_extent_buffer(left, right, btrfs_leaf_data(left) + | |
d6025579 CM |
3624 | leaf_data_end(root, left) - push_space, |
3625 | btrfs_leaf_data(right) + | |
5f39d397 | 3626 | btrfs_item_offset_nr(right, push_items - 1), |
d6025579 | 3627 | push_space); |
5f39d397 | 3628 | old_left_nritems = btrfs_header_nritems(left); |
87b29b20 | 3629 | BUG_ON(old_left_nritems <= 0); |
eb60ceac | 3630 | |
db94535d | 3631 | old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1); |
0783fcfc | 3632 | for (i = old_left_nritems; i < old_left_nritems + push_items; i++) { |
5f39d397 | 3633 | u32 ioff; |
db94535d | 3634 | |
5f39d397 | 3635 | item = btrfs_item_nr(left, i); |
db94535d | 3636 | |
cfed81a0 CM |
3637 | ioff = btrfs_token_item_offset(left, item, &token); |
3638 | btrfs_set_token_item_offset(left, item, | |
3639 | ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size), | |
3640 | &token); | |
be0e5c09 | 3641 | } |
5f39d397 | 3642 | btrfs_set_header_nritems(left, old_left_nritems + push_items); |
be0e5c09 CM |
3643 | |
3644 | /* fixup right node */ | |
31b1a2bd JL |
3645 | if (push_items > right_nritems) |
3646 | WARN(1, KERN_CRIT "push items %d nr %u\n", push_items, | |
d397712b | 3647 | right_nritems); |
34a38218 CM |
3648 | |
3649 | if (push_items < right_nritems) { | |
3650 | push_space = btrfs_item_offset_nr(right, push_items - 1) - | |
3651 | leaf_data_end(root, right); | |
3652 | memmove_extent_buffer(right, btrfs_leaf_data(right) + | |
3653 | BTRFS_LEAF_DATA_SIZE(root) - push_space, | |
3654 | btrfs_leaf_data(right) + | |
3655 | leaf_data_end(root, right), push_space); | |
3656 | ||
3657 | memmove_extent_buffer(right, btrfs_item_nr_offset(0), | |
5f39d397 CM |
3658 | btrfs_item_nr_offset(push_items), |
3659 | (btrfs_header_nritems(right) - push_items) * | |
3660 | sizeof(struct btrfs_item)); | |
34a38218 | 3661 | } |
eef1c494 Y |
3662 | right_nritems -= push_items; |
3663 | btrfs_set_header_nritems(right, right_nritems); | |
123abc88 | 3664 | push_space = BTRFS_LEAF_DATA_SIZE(root); |
5f39d397 CM |
3665 | for (i = 0; i < right_nritems; i++) { |
3666 | item = btrfs_item_nr(right, i); | |
db94535d | 3667 | |
cfed81a0 CM |
3668 | push_space = push_space - btrfs_token_item_size(right, |
3669 | item, &token); | |
3670 | btrfs_set_token_item_offset(right, item, push_space, &token); | |
db94535d | 3671 | } |
eb60ceac | 3672 | |
5f39d397 | 3673 | btrfs_mark_buffer_dirty(left); |
34a38218 CM |
3674 | if (right_nritems) |
3675 | btrfs_mark_buffer_dirty(right); | |
f0486c68 YZ |
3676 | else |
3677 | clean_tree_block(trans, root, right); | |
098f59c2 | 3678 | |
5f39d397 | 3679 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3680 | fixup_low_keys(trans, root, path, &disk_key, 1); |
be0e5c09 CM |
3681 | |
3682 | /* then fixup the leaf pointer in the path */ | |
3683 | if (path->slots[0] < push_items) { | |
3684 | path->slots[0] += old_left_nritems; | |
925baedd | 3685 | btrfs_tree_unlock(path->nodes[0]); |
5f39d397 CM |
3686 | free_extent_buffer(path->nodes[0]); |
3687 | path->nodes[0] = left; | |
be0e5c09 CM |
3688 | path->slots[1] -= 1; |
3689 | } else { | |
925baedd | 3690 | btrfs_tree_unlock(left); |
5f39d397 | 3691 | free_extent_buffer(left); |
be0e5c09 CM |
3692 | path->slots[0] -= push_items; |
3693 | } | |
eb60ceac | 3694 | BUG_ON(path->slots[0] < 0); |
aa5d6bed | 3695 | return ret; |
925baedd CM |
3696 | out: |
3697 | btrfs_tree_unlock(left); | |
3698 | free_extent_buffer(left); | |
3699 | return ret; | |
be0e5c09 CM |
3700 | } |
3701 | ||
44871b1b CM |
3702 | /* |
3703 | * push some data in the path leaf to the left, trying to free up at | |
3704 | * least data_size bytes. returns zero if the push worked, nonzero otherwise | |
99d8f83c CM |
3705 | * |
3706 | * max_slot can put a limit on how far into the leaf we'll push items. The | |
3707 | * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the | |
3708 | * items | |
44871b1b CM |
3709 | */ |
3710 | static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root | |
99d8f83c CM |
3711 | *root, struct btrfs_path *path, int min_data_size, |
3712 | int data_size, int empty, u32 max_slot) | |
44871b1b CM |
3713 | { |
3714 | struct extent_buffer *right = path->nodes[0]; | |
3715 | struct extent_buffer *left; | |
3716 | int slot; | |
3717 | int free_space; | |
3718 | u32 right_nritems; | |
3719 | int ret = 0; | |
3720 | ||
3721 | slot = path->slots[1]; | |
3722 | if (slot == 0) | |
3723 | return 1; | |
3724 | if (!path->nodes[1]) | |
3725 | return 1; | |
3726 | ||
3727 | right_nritems = btrfs_header_nritems(right); | |
3728 | if (right_nritems == 0) | |
3729 | return 1; | |
3730 | ||
3731 | btrfs_assert_tree_locked(path->nodes[1]); | |
3732 | ||
3733 | left = read_node_slot(root, path->nodes[1], slot - 1); | |
91ca338d TI |
3734 | if (left == NULL) |
3735 | return 1; | |
3736 | ||
44871b1b CM |
3737 | btrfs_tree_lock(left); |
3738 | btrfs_set_lock_blocking(left); | |
3739 | ||
3740 | free_space = btrfs_leaf_free_space(root, left); | |
3741 | if (free_space < data_size) { | |
3742 | ret = 1; | |
3743 | goto out; | |
3744 | } | |
3745 | ||
3746 | /* cow and double check */ | |
3747 | ret = btrfs_cow_block(trans, root, left, | |
3748 | path->nodes[1], slot - 1, &left); | |
3749 | if (ret) { | |
3750 | /* we hit -ENOSPC, but it isn't fatal here */ | |
79787eaa JM |
3751 | if (ret == -ENOSPC) |
3752 | ret = 1; | |
44871b1b CM |
3753 | goto out; |
3754 | } | |
3755 | ||
3756 | free_space = btrfs_leaf_free_space(root, left); | |
3757 | if (free_space < data_size) { | |
3758 | ret = 1; | |
3759 | goto out; | |
3760 | } | |
3761 | ||
99d8f83c CM |
3762 | return __push_leaf_left(trans, root, path, min_data_size, |
3763 | empty, left, free_space, right_nritems, | |
3764 | max_slot); | |
44871b1b CM |
3765 | out: |
3766 | btrfs_tree_unlock(left); | |
3767 | free_extent_buffer(left); | |
3768 | return ret; | |
3769 | } | |
3770 | ||
3771 | /* | |
3772 | * split the path's leaf in two, making sure there is at least data_size | |
3773 | * available for the resulting leaf level of the path. | |
44871b1b | 3774 | */ |
143bede5 JM |
3775 | static noinline void copy_for_split(struct btrfs_trans_handle *trans, |
3776 | struct btrfs_root *root, | |
3777 | struct btrfs_path *path, | |
3778 | struct extent_buffer *l, | |
3779 | struct extent_buffer *right, | |
3780 | int slot, int mid, int nritems) | |
44871b1b CM |
3781 | { |
3782 | int data_copy_size; | |
3783 | int rt_data_off; | |
3784 | int i; | |
44871b1b | 3785 | struct btrfs_disk_key disk_key; |
cfed81a0 CM |
3786 | struct btrfs_map_token token; |
3787 | ||
3788 | btrfs_init_map_token(&token); | |
44871b1b CM |
3789 | |
3790 | nritems = nritems - mid; | |
3791 | btrfs_set_header_nritems(right, nritems); | |
3792 | data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l); | |
3793 | ||
3794 | copy_extent_buffer(right, l, btrfs_item_nr_offset(0), | |
3795 | btrfs_item_nr_offset(mid), | |
3796 | nritems * sizeof(struct btrfs_item)); | |
3797 | ||
3798 | copy_extent_buffer(right, l, | |
3799 | btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - | |
3800 | data_copy_size, btrfs_leaf_data(l) + | |
3801 | leaf_data_end(root, l), data_copy_size); | |
3802 | ||
3803 | rt_data_off = BTRFS_LEAF_DATA_SIZE(root) - | |
3804 | btrfs_item_end_nr(l, mid); | |
3805 | ||
3806 | for (i = 0; i < nritems; i++) { | |
3807 | struct btrfs_item *item = btrfs_item_nr(right, i); | |
3808 | u32 ioff; | |
3809 | ||
cfed81a0 CM |
3810 | ioff = btrfs_token_item_offset(right, item, &token); |
3811 | btrfs_set_token_item_offset(right, item, | |
3812 | ioff + rt_data_off, &token); | |
44871b1b CM |
3813 | } |
3814 | ||
44871b1b | 3815 | btrfs_set_header_nritems(l, mid); |
44871b1b | 3816 | btrfs_item_key(right, &disk_key, 0); |
143bede5 | 3817 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 3818 | path->slots[1] + 1, 1); |
44871b1b CM |
3819 | |
3820 | btrfs_mark_buffer_dirty(right); | |
3821 | btrfs_mark_buffer_dirty(l); | |
3822 | BUG_ON(path->slots[0] != slot); | |
3823 | ||
44871b1b CM |
3824 | if (mid <= slot) { |
3825 | btrfs_tree_unlock(path->nodes[0]); | |
3826 | free_extent_buffer(path->nodes[0]); | |
3827 | path->nodes[0] = right; | |
3828 | path->slots[0] -= mid; | |
3829 | path->slots[1] += 1; | |
3830 | } else { | |
3831 | btrfs_tree_unlock(right); | |
3832 | free_extent_buffer(right); | |
3833 | } | |
3834 | ||
3835 | BUG_ON(path->slots[0] < 0); | |
44871b1b CM |
3836 | } |
3837 | ||
99d8f83c CM |
3838 | /* |
3839 | * double splits happen when we need to insert a big item in the middle | |
3840 | * of a leaf. A double split can leave us with 3 mostly empty leaves: | |
3841 | * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ] | |
3842 | * A B C | |
3843 | * | |
3844 | * We avoid this by trying to push the items on either side of our target | |
3845 | * into the adjacent leaves. If all goes well we can avoid the double split | |
3846 | * completely. | |
3847 | */ | |
3848 | static noinline int push_for_double_split(struct btrfs_trans_handle *trans, | |
3849 | struct btrfs_root *root, | |
3850 | struct btrfs_path *path, | |
3851 | int data_size) | |
3852 | { | |
3853 | int ret; | |
3854 | int progress = 0; | |
3855 | int slot; | |
3856 | u32 nritems; | |
3857 | ||
3858 | slot = path->slots[0]; | |
3859 | ||
3860 | /* | |
3861 | * try to push all the items after our slot into the | |
3862 | * right leaf | |
3863 | */ | |
3864 | ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot); | |
3865 | if (ret < 0) | |
3866 | return ret; | |
3867 | ||
3868 | if (ret == 0) | |
3869 | progress++; | |
3870 | ||
3871 | nritems = btrfs_header_nritems(path->nodes[0]); | |
3872 | /* | |
3873 | * our goal is to get our slot at the start or end of a leaf. If | |
3874 | * we've done so we're done | |
3875 | */ | |
3876 | if (path->slots[0] == 0 || path->slots[0] == nritems) | |
3877 | return 0; | |
3878 | ||
3879 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
3880 | return 0; | |
3881 | ||
3882 | /* try to push all the items before our slot into the next leaf */ | |
3883 | slot = path->slots[0]; | |
3884 | ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot); | |
3885 | if (ret < 0) | |
3886 | return ret; | |
3887 | ||
3888 | if (ret == 0) | |
3889 | progress++; | |
3890 | ||
3891 | if (progress) | |
3892 | return 0; | |
3893 | return 1; | |
3894 | } | |
3895 | ||
74123bd7 CM |
3896 | /* |
3897 | * split the path's leaf in two, making sure there is at least data_size | |
3898 | * available for the resulting leaf level of the path. | |
aa5d6bed CM |
3899 | * |
3900 | * returns 0 if all went well and < 0 on failure. | |
74123bd7 | 3901 | */ |
e02119d5 CM |
3902 | static noinline int split_leaf(struct btrfs_trans_handle *trans, |
3903 | struct btrfs_root *root, | |
3904 | struct btrfs_key *ins_key, | |
3905 | struct btrfs_path *path, int data_size, | |
3906 | int extend) | |
be0e5c09 | 3907 | { |
5d4f98a2 | 3908 | struct btrfs_disk_key disk_key; |
5f39d397 | 3909 | struct extent_buffer *l; |
7518a238 | 3910 | u32 nritems; |
eb60ceac CM |
3911 | int mid; |
3912 | int slot; | |
5f39d397 | 3913 | struct extent_buffer *right; |
d4dbff95 | 3914 | int ret = 0; |
aa5d6bed | 3915 | int wret; |
5d4f98a2 | 3916 | int split; |
cc0c5538 | 3917 | int num_doubles = 0; |
99d8f83c | 3918 | int tried_avoid_double = 0; |
aa5d6bed | 3919 | |
a5719521 YZ |
3920 | l = path->nodes[0]; |
3921 | slot = path->slots[0]; | |
3922 | if (extend && data_size + btrfs_item_size_nr(l, slot) + | |
3923 | sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root)) | |
3924 | return -EOVERFLOW; | |
3925 | ||
40689478 | 3926 | /* first try to make some room by pushing left and right */ |
99d8f83c CM |
3927 | if (data_size) { |
3928 | wret = push_leaf_right(trans, root, path, data_size, | |
3929 | data_size, 0, 0); | |
d397712b | 3930 | if (wret < 0) |
eaee50e8 | 3931 | return wret; |
3685f791 | 3932 | if (wret) { |
99d8f83c CM |
3933 | wret = push_leaf_left(trans, root, path, data_size, |
3934 | data_size, 0, (u32)-1); | |
3685f791 CM |
3935 | if (wret < 0) |
3936 | return wret; | |
3937 | } | |
3938 | l = path->nodes[0]; | |
aa5d6bed | 3939 | |
3685f791 | 3940 | /* did the pushes work? */ |
87b29b20 | 3941 | if (btrfs_leaf_free_space(root, l) >= data_size) |
3685f791 | 3942 | return 0; |
3326d1b0 | 3943 | } |
aa5d6bed | 3944 | |
5c680ed6 | 3945 | if (!path->nodes[1]) { |
e089f05c | 3946 | ret = insert_new_root(trans, root, path, 1); |
5c680ed6 CM |
3947 | if (ret) |
3948 | return ret; | |
3949 | } | |
cc0c5538 | 3950 | again: |
5d4f98a2 | 3951 | split = 1; |
cc0c5538 | 3952 | l = path->nodes[0]; |
eb60ceac | 3953 | slot = path->slots[0]; |
5f39d397 | 3954 | nritems = btrfs_header_nritems(l); |
d397712b | 3955 | mid = (nritems + 1) / 2; |
54aa1f4d | 3956 | |
5d4f98a2 YZ |
3957 | if (mid <= slot) { |
3958 | if (nritems == 1 || | |
3959 | leaf_space_used(l, mid, nritems - mid) + data_size > | |
3960 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3961 | if (slot >= nritems) { | |
3962 | split = 0; | |
3963 | } else { | |
3964 | mid = slot; | |
3965 | if (mid != nritems && | |
3966 | leaf_space_used(l, mid, nritems - mid) + | |
3967 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3968 | if (data_size && !tried_avoid_double) |
3969 | goto push_for_double; | |
5d4f98a2 YZ |
3970 | split = 2; |
3971 | } | |
3972 | } | |
3973 | } | |
3974 | } else { | |
3975 | if (leaf_space_used(l, 0, mid) + data_size > | |
3976 | BTRFS_LEAF_DATA_SIZE(root)) { | |
3977 | if (!extend && data_size && slot == 0) { | |
3978 | split = 0; | |
3979 | } else if ((extend || !data_size) && slot == 0) { | |
3980 | mid = 1; | |
3981 | } else { | |
3982 | mid = slot; | |
3983 | if (mid != nritems && | |
3984 | leaf_space_used(l, mid, nritems - mid) + | |
3985 | data_size > BTRFS_LEAF_DATA_SIZE(root)) { | |
99d8f83c CM |
3986 | if (data_size && !tried_avoid_double) |
3987 | goto push_for_double; | |
5d4f98a2 YZ |
3988 | split = 2 ; |
3989 | } | |
3990 | } | |
3991 | } | |
3992 | } | |
3993 | ||
3994 | if (split == 0) | |
3995 | btrfs_cpu_key_to_disk(&disk_key, ins_key); | |
3996 | else | |
3997 | btrfs_item_key(l, &disk_key, mid); | |
3998 | ||
3999 | right = btrfs_alloc_free_block(trans, root, root->leafsize, 0, | |
31840ae1 | 4000 | root->root_key.objectid, |
5581a51a | 4001 | &disk_key, 0, l->start, 0); |
f0486c68 | 4002 | if (IS_ERR(right)) |
5f39d397 | 4003 | return PTR_ERR(right); |
f0486c68 YZ |
4004 | |
4005 | root_add_used(root, root->leafsize); | |
5f39d397 CM |
4006 | |
4007 | memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header)); | |
db94535d | 4008 | btrfs_set_header_bytenr(right, right->start); |
5f39d397 | 4009 | btrfs_set_header_generation(right, trans->transid); |
5d4f98a2 | 4010 | btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV); |
5f39d397 CM |
4011 | btrfs_set_header_owner(right, root->root_key.objectid); |
4012 | btrfs_set_header_level(right, 0); | |
4013 | write_extent_buffer(right, root->fs_info->fsid, | |
4014 | (unsigned long)btrfs_header_fsid(right), | |
4015 | BTRFS_FSID_SIZE); | |
e17cade2 CM |
4016 | |
4017 | write_extent_buffer(right, root->fs_info->chunk_tree_uuid, | |
4018 | (unsigned long)btrfs_header_chunk_tree_uuid(right), | |
4019 | BTRFS_UUID_SIZE); | |
44871b1b | 4020 | |
5d4f98a2 YZ |
4021 | if (split == 0) { |
4022 | if (mid <= slot) { | |
4023 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4024 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4025 | path->slots[1] + 1, 1); |
5d4f98a2 YZ |
4026 | btrfs_tree_unlock(path->nodes[0]); |
4027 | free_extent_buffer(path->nodes[0]); | |
4028 | path->nodes[0] = right; | |
4029 | path->slots[0] = 0; | |
4030 | path->slots[1] += 1; | |
4031 | } else { | |
4032 | btrfs_set_header_nritems(right, 0); | |
143bede5 | 4033 | insert_ptr(trans, root, path, &disk_key, right->start, |
c3e06965 | 4034 | path->slots[1], 1); |
5d4f98a2 YZ |
4035 | btrfs_tree_unlock(path->nodes[0]); |
4036 | free_extent_buffer(path->nodes[0]); | |
4037 | path->nodes[0] = right; | |
4038 | path->slots[0] = 0; | |
143bede5 JM |
4039 | if (path->slots[1] == 0) |
4040 | fixup_low_keys(trans, root, path, | |
4041 | &disk_key, 1); | |
d4dbff95 | 4042 | } |
5d4f98a2 YZ |
4043 | btrfs_mark_buffer_dirty(right); |
4044 | return ret; | |
d4dbff95 | 4045 | } |
74123bd7 | 4046 | |
143bede5 | 4047 | copy_for_split(trans, root, path, l, right, slot, mid, nritems); |
31840ae1 | 4048 | |
5d4f98a2 | 4049 | if (split == 2) { |
cc0c5538 CM |
4050 | BUG_ON(num_doubles != 0); |
4051 | num_doubles++; | |
4052 | goto again; | |
a429e513 | 4053 | } |
44871b1b | 4054 | |
143bede5 | 4055 | return 0; |
99d8f83c CM |
4056 | |
4057 | push_for_double: | |
4058 | push_for_double_split(trans, root, path, data_size); | |
4059 | tried_avoid_double = 1; | |
4060 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size) | |
4061 | return 0; | |
4062 | goto again; | |
be0e5c09 CM |
4063 | } |
4064 | ||
ad48fd75 YZ |
4065 | static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans, |
4066 | struct btrfs_root *root, | |
4067 | struct btrfs_path *path, int ins_len) | |
459931ec | 4068 | { |
ad48fd75 | 4069 | struct btrfs_key key; |
459931ec | 4070 | struct extent_buffer *leaf; |
ad48fd75 YZ |
4071 | struct btrfs_file_extent_item *fi; |
4072 | u64 extent_len = 0; | |
4073 | u32 item_size; | |
4074 | int ret; | |
459931ec CM |
4075 | |
4076 | leaf = path->nodes[0]; | |
ad48fd75 YZ |
4077 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
4078 | ||
4079 | BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY && | |
4080 | key.type != BTRFS_EXTENT_CSUM_KEY); | |
4081 | ||
4082 | if (btrfs_leaf_free_space(root, leaf) >= ins_len) | |
4083 | return 0; | |
459931ec CM |
4084 | |
4085 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ad48fd75 YZ |
4086 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4087 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4088 | struct btrfs_file_extent_item); | |
4089 | extent_len = btrfs_file_extent_num_bytes(leaf, fi); | |
4090 | } | |
b3b4aa74 | 4091 | btrfs_release_path(path); |
459931ec | 4092 | |
459931ec | 4093 | path->keep_locks = 1; |
ad48fd75 YZ |
4094 | path->search_for_split = 1; |
4095 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
459931ec | 4096 | path->search_for_split = 0; |
ad48fd75 YZ |
4097 | if (ret < 0) |
4098 | goto err; | |
459931ec | 4099 | |
ad48fd75 YZ |
4100 | ret = -EAGAIN; |
4101 | leaf = path->nodes[0]; | |
459931ec | 4102 | /* if our item isn't there or got smaller, return now */ |
ad48fd75 YZ |
4103 | if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0])) |
4104 | goto err; | |
4105 | ||
109f6aef CM |
4106 | /* the leaf has changed, it now has room. return now */ |
4107 | if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len) | |
4108 | goto err; | |
4109 | ||
ad48fd75 YZ |
4110 | if (key.type == BTRFS_EXTENT_DATA_KEY) { |
4111 | fi = btrfs_item_ptr(leaf, path->slots[0], | |
4112 | struct btrfs_file_extent_item); | |
4113 | if (extent_len != btrfs_file_extent_num_bytes(leaf, fi)) | |
4114 | goto err; | |
459931ec CM |
4115 | } |
4116 | ||
b9473439 | 4117 | btrfs_set_path_blocking(path); |
ad48fd75 | 4118 | ret = split_leaf(trans, root, &key, path, ins_len, 1); |
f0486c68 YZ |
4119 | if (ret) |
4120 | goto err; | |
459931ec | 4121 | |
ad48fd75 | 4122 | path->keep_locks = 0; |
b9473439 | 4123 | btrfs_unlock_up_safe(path, 1); |
ad48fd75 YZ |
4124 | return 0; |
4125 | err: | |
4126 | path->keep_locks = 0; | |
4127 | return ret; | |
4128 | } | |
4129 | ||
4130 | static noinline int split_item(struct btrfs_trans_handle *trans, | |
4131 | struct btrfs_root *root, | |
4132 | struct btrfs_path *path, | |
4133 | struct btrfs_key *new_key, | |
4134 | unsigned long split_offset) | |
4135 | { | |
4136 | struct extent_buffer *leaf; | |
4137 | struct btrfs_item *item; | |
4138 | struct btrfs_item *new_item; | |
4139 | int slot; | |
4140 | char *buf; | |
4141 | u32 nritems; | |
4142 | u32 item_size; | |
4143 | u32 orig_offset; | |
4144 | struct btrfs_disk_key disk_key; | |
4145 | ||
b9473439 CM |
4146 | leaf = path->nodes[0]; |
4147 | BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item)); | |
4148 | ||
b4ce94de CM |
4149 | btrfs_set_path_blocking(path); |
4150 | ||
459931ec CM |
4151 | item = btrfs_item_nr(leaf, path->slots[0]); |
4152 | orig_offset = btrfs_item_offset(leaf, item); | |
4153 | item_size = btrfs_item_size(leaf, item); | |
4154 | ||
459931ec | 4155 | buf = kmalloc(item_size, GFP_NOFS); |
ad48fd75 YZ |
4156 | if (!buf) |
4157 | return -ENOMEM; | |
4158 | ||
459931ec CM |
4159 | read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, |
4160 | path->slots[0]), item_size); | |
459931ec | 4161 | |
ad48fd75 | 4162 | slot = path->slots[0] + 1; |
459931ec | 4163 | nritems = btrfs_header_nritems(leaf); |
459931ec CM |
4164 | if (slot != nritems) { |
4165 | /* shift the items */ | |
4166 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1), | |
ad48fd75 YZ |
4167 | btrfs_item_nr_offset(slot), |
4168 | (nritems - slot) * sizeof(struct btrfs_item)); | |
459931ec CM |
4169 | } |
4170 | ||
4171 | btrfs_cpu_key_to_disk(&disk_key, new_key); | |
4172 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4173 | ||
4174 | new_item = btrfs_item_nr(leaf, slot); | |
4175 | ||
4176 | btrfs_set_item_offset(leaf, new_item, orig_offset); | |
4177 | btrfs_set_item_size(leaf, new_item, item_size - split_offset); | |
4178 | ||
4179 | btrfs_set_item_offset(leaf, item, | |
4180 | orig_offset + item_size - split_offset); | |
4181 | btrfs_set_item_size(leaf, item, split_offset); | |
4182 | ||
4183 | btrfs_set_header_nritems(leaf, nritems + 1); | |
4184 | ||
4185 | /* write the data for the start of the original item */ | |
4186 | write_extent_buffer(leaf, buf, | |
4187 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4188 | split_offset); | |
4189 | ||
4190 | /* write the data for the new item */ | |
4191 | write_extent_buffer(leaf, buf + split_offset, | |
4192 | btrfs_item_ptr_offset(leaf, slot), | |
4193 | item_size - split_offset); | |
4194 | btrfs_mark_buffer_dirty(leaf); | |
4195 | ||
ad48fd75 | 4196 | BUG_ON(btrfs_leaf_free_space(root, leaf) < 0); |
459931ec | 4197 | kfree(buf); |
ad48fd75 YZ |
4198 | return 0; |
4199 | } | |
4200 | ||
4201 | /* | |
4202 | * This function splits a single item into two items, | |
4203 | * giving 'new_key' to the new item and splitting the | |
4204 | * old one at split_offset (from the start of the item). | |
4205 | * | |
4206 | * The path may be released by this operation. After | |
4207 | * the split, the path is pointing to the old item. The | |
4208 | * new item is going to be in the same node as the old one. | |
4209 | * | |
4210 | * Note, the item being split must be smaller enough to live alone on | |
4211 | * a tree block with room for one extra struct btrfs_item | |
4212 | * | |
4213 | * This allows us to split the item in place, keeping a lock on the | |
4214 | * leaf the entire time. | |
4215 | */ | |
4216 | int btrfs_split_item(struct btrfs_trans_handle *trans, | |
4217 | struct btrfs_root *root, | |
4218 | struct btrfs_path *path, | |
4219 | struct btrfs_key *new_key, | |
4220 | unsigned long split_offset) | |
4221 | { | |
4222 | int ret; | |
4223 | ret = setup_leaf_for_split(trans, root, path, | |
4224 | sizeof(struct btrfs_item)); | |
4225 | if (ret) | |
4226 | return ret; | |
4227 | ||
4228 | ret = split_item(trans, root, path, new_key, split_offset); | |
459931ec CM |
4229 | return ret; |
4230 | } | |
4231 | ||
ad48fd75 YZ |
4232 | /* |
4233 | * This function duplicate a item, giving 'new_key' to the new item. | |
4234 | * It guarantees both items live in the same tree leaf and the new item | |
4235 | * is contiguous with the original item. | |
4236 | * | |
4237 | * This allows us to split file extent in place, keeping a lock on the | |
4238 | * leaf the entire time. | |
4239 | */ | |
4240 | int btrfs_duplicate_item(struct btrfs_trans_handle *trans, | |
4241 | struct btrfs_root *root, | |
4242 | struct btrfs_path *path, | |
4243 | struct btrfs_key *new_key) | |
4244 | { | |
4245 | struct extent_buffer *leaf; | |
4246 | int ret; | |
4247 | u32 item_size; | |
4248 | ||
4249 | leaf = path->nodes[0]; | |
4250 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
4251 | ret = setup_leaf_for_split(trans, root, path, | |
4252 | item_size + sizeof(struct btrfs_item)); | |
4253 | if (ret) | |
4254 | return ret; | |
4255 | ||
4256 | path->slots[0]++; | |
143bede5 JM |
4257 | setup_items_for_insert(trans, root, path, new_key, &item_size, |
4258 | item_size, item_size + | |
4259 | sizeof(struct btrfs_item), 1); | |
ad48fd75 YZ |
4260 | leaf = path->nodes[0]; |
4261 | memcpy_extent_buffer(leaf, | |
4262 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
4263 | btrfs_item_ptr_offset(leaf, path->slots[0] - 1), | |
4264 | item_size); | |
4265 | return 0; | |
4266 | } | |
4267 | ||
d352ac68 CM |
4268 | /* |
4269 | * make the item pointed to by the path smaller. new_size indicates | |
4270 | * how small to make it, and from_end tells us if we just chop bytes | |
4271 | * off the end of the item or if we shift the item to chop bytes off | |
4272 | * the front. | |
4273 | */ | |
143bede5 JM |
4274 | void btrfs_truncate_item(struct btrfs_trans_handle *trans, |
4275 | struct btrfs_root *root, | |
4276 | struct btrfs_path *path, | |
4277 | u32 new_size, int from_end) | |
b18c6685 | 4278 | { |
b18c6685 | 4279 | int slot; |
5f39d397 CM |
4280 | struct extent_buffer *leaf; |
4281 | struct btrfs_item *item; | |
b18c6685 CM |
4282 | u32 nritems; |
4283 | unsigned int data_end; | |
4284 | unsigned int old_data_start; | |
4285 | unsigned int old_size; | |
4286 | unsigned int size_diff; | |
4287 | int i; | |
cfed81a0 CM |
4288 | struct btrfs_map_token token; |
4289 | ||
4290 | btrfs_init_map_token(&token); | |
b18c6685 | 4291 | |
5f39d397 | 4292 | leaf = path->nodes[0]; |
179e29e4 CM |
4293 | slot = path->slots[0]; |
4294 | ||
4295 | old_size = btrfs_item_size_nr(leaf, slot); | |
4296 | if (old_size == new_size) | |
143bede5 | 4297 | return; |
b18c6685 | 4298 | |
5f39d397 | 4299 | nritems = btrfs_header_nritems(leaf); |
b18c6685 CM |
4300 | data_end = leaf_data_end(root, leaf); |
4301 | ||
5f39d397 | 4302 | old_data_start = btrfs_item_offset_nr(leaf, slot); |
179e29e4 | 4303 | |
b18c6685 CM |
4304 | size_diff = old_size - new_size; |
4305 | ||
4306 | BUG_ON(slot < 0); | |
4307 | BUG_ON(slot >= nritems); | |
4308 | ||
4309 | /* | |
4310 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4311 | */ | |
4312 | /* first correct the data pointers */ | |
4313 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4314 | u32 ioff; |
4315 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4316 | |
cfed81a0 CM |
4317 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4318 | btrfs_set_token_item_offset(leaf, item, | |
4319 | ioff + size_diff, &token); | |
b18c6685 | 4320 | } |
db94535d | 4321 | |
b18c6685 | 4322 | /* shift the data */ |
179e29e4 CM |
4323 | if (from_end) { |
4324 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4325 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4326 | data_end, old_data_start + new_size - data_end); | |
4327 | } else { | |
4328 | struct btrfs_disk_key disk_key; | |
4329 | u64 offset; | |
4330 | ||
4331 | btrfs_item_key(leaf, &disk_key, slot); | |
4332 | ||
4333 | if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) { | |
4334 | unsigned long ptr; | |
4335 | struct btrfs_file_extent_item *fi; | |
4336 | ||
4337 | fi = btrfs_item_ptr(leaf, slot, | |
4338 | struct btrfs_file_extent_item); | |
4339 | fi = (struct btrfs_file_extent_item *)( | |
4340 | (unsigned long)fi - size_diff); | |
4341 | ||
4342 | if (btrfs_file_extent_type(leaf, fi) == | |
4343 | BTRFS_FILE_EXTENT_INLINE) { | |
4344 | ptr = btrfs_item_ptr_offset(leaf, slot); | |
4345 | memmove_extent_buffer(leaf, ptr, | |
d397712b CM |
4346 | (unsigned long)fi, |
4347 | offsetof(struct btrfs_file_extent_item, | |
179e29e4 CM |
4348 | disk_bytenr)); |
4349 | } | |
4350 | } | |
4351 | ||
4352 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
4353 | data_end + size_diff, btrfs_leaf_data(leaf) + | |
4354 | data_end, old_data_start - data_end); | |
4355 | ||
4356 | offset = btrfs_disk_key_offset(&disk_key); | |
4357 | btrfs_set_disk_key_offset(&disk_key, offset + size_diff); | |
4358 | btrfs_set_item_key(leaf, &disk_key, slot); | |
4359 | if (slot == 0) | |
4360 | fixup_low_keys(trans, root, path, &disk_key, 1); | |
4361 | } | |
5f39d397 CM |
4362 | |
4363 | item = btrfs_item_nr(leaf, slot); | |
4364 | btrfs_set_item_size(leaf, item, new_size); | |
4365 | btrfs_mark_buffer_dirty(leaf); | |
b18c6685 | 4366 | |
5f39d397 CM |
4367 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4368 | btrfs_print_leaf(root, leaf); | |
b18c6685 | 4369 | BUG(); |
5f39d397 | 4370 | } |
b18c6685 CM |
4371 | } |
4372 | ||
d352ac68 CM |
4373 | /* |
4374 | * make the item pointed to by the path bigger, data_size is the new size. | |
4375 | */ | |
143bede5 JM |
4376 | void btrfs_extend_item(struct btrfs_trans_handle *trans, |
4377 | struct btrfs_root *root, struct btrfs_path *path, | |
4378 | u32 data_size) | |
6567e837 | 4379 | { |
6567e837 | 4380 | int slot; |
5f39d397 CM |
4381 | struct extent_buffer *leaf; |
4382 | struct btrfs_item *item; | |
6567e837 CM |
4383 | u32 nritems; |
4384 | unsigned int data_end; | |
4385 | unsigned int old_data; | |
4386 | unsigned int old_size; | |
4387 | int i; | |
cfed81a0 CM |
4388 | struct btrfs_map_token token; |
4389 | ||
4390 | btrfs_init_map_token(&token); | |
6567e837 | 4391 | |
5f39d397 | 4392 | leaf = path->nodes[0]; |
6567e837 | 4393 | |
5f39d397 | 4394 | nritems = btrfs_header_nritems(leaf); |
6567e837 CM |
4395 | data_end = leaf_data_end(root, leaf); |
4396 | ||
5f39d397 CM |
4397 | if (btrfs_leaf_free_space(root, leaf) < data_size) { |
4398 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4399 | BUG(); |
5f39d397 | 4400 | } |
6567e837 | 4401 | slot = path->slots[0]; |
5f39d397 | 4402 | old_data = btrfs_item_end_nr(leaf, slot); |
6567e837 CM |
4403 | |
4404 | BUG_ON(slot < 0); | |
3326d1b0 CM |
4405 | if (slot >= nritems) { |
4406 | btrfs_print_leaf(root, leaf); | |
d397712b CM |
4407 | printk(KERN_CRIT "slot %d too large, nritems %d\n", |
4408 | slot, nritems); | |
3326d1b0 CM |
4409 | BUG_ON(1); |
4410 | } | |
6567e837 CM |
4411 | |
4412 | /* | |
4413 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4414 | */ | |
4415 | /* first correct the data pointers */ | |
4416 | for (i = slot; i < nritems; i++) { | |
5f39d397 CM |
4417 | u32 ioff; |
4418 | item = btrfs_item_nr(leaf, i); | |
db94535d | 4419 | |
cfed81a0 CM |
4420 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4421 | btrfs_set_token_item_offset(leaf, item, | |
4422 | ioff - data_size, &token); | |
6567e837 | 4423 | } |
5f39d397 | 4424 | |
6567e837 | 4425 | /* shift the data */ |
5f39d397 | 4426 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
6567e837 CM |
4427 | data_end - data_size, btrfs_leaf_data(leaf) + |
4428 | data_end, old_data - data_end); | |
5f39d397 | 4429 | |
6567e837 | 4430 | data_end = old_data; |
5f39d397 CM |
4431 | old_size = btrfs_item_size_nr(leaf, slot); |
4432 | item = btrfs_item_nr(leaf, slot); | |
4433 | btrfs_set_item_size(leaf, item, old_size + data_size); | |
4434 | btrfs_mark_buffer_dirty(leaf); | |
6567e837 | 4435 | |
5f39d397 CM |
4436 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4437 | btrfs_print_leaf(root, leaf); | |
6567e837 | 4438 | BUG(); |
5f39d397 | 4439 | } |
6567e837 CM |
4440 | } |
4441 | ||
74123bd7 | 4442 | /* |
44871b1b CM |
4443 | * this is a helper for btrfs_insert_empty_items, the main goal here is |
4444 | * to save stack depth by doing the bulk of the work in a function | |
4445 | * that doesn't call btrfs_search_slot | |
74123bd7 | 4446 | */ |
143bede5 JM |
4447 | void setup_items_for_insert(struct btrfs_trans_handle *trans, |
4448 | struct btrfs_root *root, struct btrfs_path *path, | |
4449 | struct btrfs_key *cpu_key, u32 *data_size, | |
4450 | u32 total_data, u32 total_size, int nr) | |
be0e5c09 | 4451 | { |
5f39d397 | 4452 | struct btrfs_item *item; |
9c58309d | 4453 | int i; |
7518a238 | 4454 | u32 nritems; |
be0e5c09 | 4455 | unsigned int data_end; |
e2fa7227 | 4456 | struct btrfs_disk_key disk_key; |
44871b1b CM |
4457 | struct extent_buffer *leaf; |
4458 | int slot; | |
cfed81a0 CM |
4459 | struct btrfs_map_token token; |
4460 | ||
4461 | btrfs_init_map_token(&token); | |
e2fa7227 | 4462 | |
5f39d397 | 4463 | leaf = path->nodes[0]; |
44871b1b | 4464 | slot = path->slots[0]; |
74123bd7 | 4465 | |
5f39d397 | 4466 | nritems = btrfs_header_nritems(leaf); |
123abc88 | 4467 | data_end = leaf_data_end(root, leaf); |
eb60ceac | 4468 | |
f25956cc | 4469 | if (btrfs_leaf_free_space(root, leaf) < total_size) { |
3326d1b0 | 4470 | btrfs_print_leaf(root, leaf); |
d397712b | 4471 | printk(KERN_CRIT "not enough freespace need %u have %d\n", |
9c58309d | 4472 | total_size, btrfs_leaf_free_space(root, leaf)); |
be0e5c09 | 4473 | BUG(); |
d4dbff95 | 4474 | } |
5f39d397 | 4475 | |
be0e5c09 | 4476 | if (slot != nritems) { |
5f39d397 | 4477 | unsigned int old_data = btrfs_item_end_nr(leaf, slot); |
be0e5c09 | 4478 | |
5f39d397 CM |
4479 | if (old_data < data_end) { |
4480 | btrfs_print_leaf(root, leaf); | |
d397712b | 4481 | printk(KERN_CRIT "slot %d old_data %d data_end %d\n", |
5f39d397 CM |
4482 | slot, old_data, data_end); |
4483 | BUG_ON(1); | |
4484 | } | |
be0e5c09 CM |
4485 | /* |
4486 | * item0..itemN ... dataN.offset..dataN.size .. data0.size | |
4487 | */ | |
4488 | /* first correct the data pointers */ | |
0783fcfc | 4489 | for (i = slot; i < nritems; i++) { |
5f39d397 | 4490 | u32 ioff; |
db94535d | 4491 | |
5f39d397 | 4492 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4493 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4494 | btrfs_set_token_item_offset(leaf, item, | |
4495 | ioff - total_data, &token); | |
0783fcfc | 4496 | } |
be0e5c09 | 4497 | /* shift the items */ |
9c58309d | 4498 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr), |
5f39d397 | 4499 | btrfs_item_nr_offset(slot), |
d6025579 | 4500 | (nritems - slot) * sizeof(struct btrfs_item)); |
be0e5c09 CM |
4501 | |
4502 | /* shift the data */ | |
5f39d397 | 4503 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + |
9c58309d | 4504 | data_end - total_data, btrfs_leaf_data(leaf) + |
d6025579 | 4505 | data_end, old_data - data_end); |
be0e5c09 CM |
4506 | data_end = old_data; |
4507 | } | |
5f39d397 | 4508 | |
62e2749e | 4509 | /* setup the item for the new data */ |
9c58309d CM |
4510 | for (i = 0; i < nr; i++) { |
4511 | btrfs_cpu_key_to_disk(&disk_key, cpu_key + i); | |
4512 | btrfs_set_item_key(leaf, &disk_key, slot + i); | |
4513 | item = btrfs_item_nr(leaf, slot + i); | |
cfed81a0 CM |
4514 | btrfs_set_token_item_offset(leaf, item, |
4515 | data_end - data_size[i], &token); | |
9c58309d | 4516 | data_end -= data_size[i]; |
cfed81a0 | 4517 | btrfs_set_token_item_size(leaf, item, data_size[i], &token); |
9c58309d | 4518 | } |
44871b1b | 4519 | |
9c58309d | 4520 | btrfs_set_header_nritems(leaf, nritems + nr); |
aa5d6bed | 4521 | |
5a01a2e3 CM |
4522 | if (slot == 0) { |
4523 | btrfs_cpu_key_to_disk(&disk_key, cpu_key); | |
143bede5 | 4524 | fixup_low_keys(trans, root, path, &disk_key, 1); |
5a01a2e3 | 4525 | } |
b9473439 CM |
4526 | btrfs_unlock_up_safe(path, 1); |
4527 | btrfs_mark_buffer_dirty(leaf); | |
aa5d6bed | 4528 | |
5f39d397 CM |
4529 | if (btrfs_leaf_free_space(root, leaf) < 0) { |
4530 | btrfs_print_leaf(root, leaf); | |
be0e5c09 | 4531 | BUG(); |
5f39d397 | 4532 | } |
44871b1b CM |
4533 | } |
4534 | ||
4535 | /* | |
4536 | * Given a key and some data, insert items into the tree. | |
4537 | * This does all the path init required, making room in the tree if needed. | |
4538 | */ | |
4539 | int btrfs_insert_empty_items(struct btrfs_trans_handle *trans, | |
4540 | struct btrfs_root *root, | |
4541 | struct btrfs_path *path, | |
4542 | struct btrfs_key *cpu_key, u32 *data_size, | |
4543 | int nr) | |
4544 | { | |
44871b1b CM |
4545 | int ret = 0; |
4546 | int slot; | |
4547 | int i; | |
4548 | u32 total_size = 0; | |
4549 | u32 total_data = 0; | |
4550 | ||
4551 | for (i = 0; i < nr; i++) | |
4552 | total_data += data_size[i]; | |
4553 | ||
4554 | total_size = total_data + (nr * sizeof(struct btrfs_item)); | |
4555 | ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1); | |
4556 | if (ret == 0) | |
4557 | return -EEXIST; | |
4558 | if (ret < 0) | |
143bede5 | 4559 | return ret; |
44871b1b | 4560 | |
44871b1b CM |
4561 | slot = path->slots[0]; |
4562 | BUG_ON(slot < 0); | |
4563 | ||
143bede5 | 4564 | setup_items_for_insert(trans, root, path, cpu_key, data_size, |
44871b1b | 4565 | total_data, total_size, nr); |
143bede5 | 4566 | return 0; |
62e2749e CM |
4567 | } |
4568 | ||
4569 | /* | |
4570 | * Given a key and some data, insert an item into the tree. | |
4571 | * This does all the path init required, making room in the tree if needed. | |
4572 | */ | |
e089f05c CM |
4573 | int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root |
4574 | *root, struct btrfs_key *cpu_key, void *data, u32 | |
4575 | data_size) | |
62e2749e CM |
4576 | { |
4577 | int ret = 0; | |
2c90e5d6 | 4578 | struct btrfs_path *path; |
5f39d397 CM |
4579 | struct extent_buffer *leaf; |
4580 | unsigned long ptr; | |
62e2749e | 4581 | |
2c90e5d6 | 4582 | path = btrfs_alloc_path(); |
db5b493a TI |
4583 | if (!path) |
4584 | return -ENOMEM; | |
2c90e5d6 | 4585 | ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size); |
62e2749e | 4586 | if (!ret) { |
5f39d397 CM |
4587 | leaf = path->nodes[0]; |
4588 | ptr = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
4589 | write_extent_buffer(leaf, data, ptr, data_size); | |
4590 | btrfs_mark_buffer_dirty(leaf); | |
62e2749e | 4591 | } |
2c90e5d6 | 4592 | btrfs_free_path(path); |
aa5d6bed | 4593 | return ret; |
be0e5c09 CM |
4594 | } |
4595 | ||
74123bd7 | 4596 | /* |
5de08d7d | 4597 | * delete the pointer from a given node. |
74123bd7 | 4598 | * |
d352ac68 CM |
4599 | * the tree should have been previously balanced so the deletion does not |
4600 | * empty a node. | |
74123bd7 | 4601 | */ |
143bede5 | 4602 | static void del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
0e411ece | 4603 | struct btrfs_path *path, int level, int slot) |
be0e5c09 | 4604 | { |
5f39d397 | 4605 | struct extent_buffer *parent = path->nodes[level]; |
7518a238 | 4606 | u32 nritems; |
f3ea38da | 4607 | int ret; |
be0e5c09 | 4608 | |
5f39d397 | 4609 | nritems = btrfs_header_nritems(parent); |
d397712b | 4610 | if (slot != nritems - 1) { |
0e411ece | 4611 | if (level) |
f3ea38da JS |
4612 | tree_mod_log_eb_move(root->fs_info, parent, slot, |
4613 | slot + 1, nritems - slot - 1); | |
5f39d397 CM |
4614 | memmove_extent_buffer(parent, |
4615 | btrfs_node_key_ptr_offset(slot), | |
4616 | btrfs_node_key_ptr_offset(slot + 1), | |
d6025579 CM |
4617 | sizeof(struct btrfs_key_ptr) * |
4618 | (nritems - slot - 1)); | |
57ba86c0 CM |
4619 | } else if (level) { |
4620 | ret = tree_mod_log_insert_key(root->fs_info, parent, slot, | |
4621 | MOD_LOG_KEY_REMOVE); | |
4622 | BUG_ON(ret < 0); | |
bb803951 | 4623 | } |
f3ea38da | 4624 | |
7518a238 | 4625 | nritems--; |
5f39d397 | 4626 | btrfs_set_header_nritems(parent, nritems); |
7518a238 | 4627 | if (nritems == 0 && parent == root->node) { |
5f39d397 | 4628 | BUG_ON(btrfs_header_level(root->node) != 1); |
bb803951 | 4629 | /* just turn the root into a leaf and break */ |
5f39d397 | 4630 | btrfs_set_header_level(root->node, 0); |
bb803951 | 4631 | } else if (slot == 0) { |
5f39d397 CM |
4632 | struct btrfs_disk_key disk_key; |
4633 | ||
4634 | btrfs_node_key(parent, &disk_key, 0); | |
143bede5 | 4635 | fixup_low_keys(trans, root, path, &disk_key, level + 1); |
be0e5c09 | 4636 | } |
d6025579 | 4637 | btrfs_mark_buffer_dirty(parent); |
be0e5c09 CM |
4638 | } |
4639 | ||
323ac95b CM |
4640 | /* |
4641 | * a helper function to delete the leaf pointed to by path->slots[1] and | |
5d4f98a2 | 4642 | * path->nodes[1]. |
323ac95b CM |
4643 | * |
4644 | * This deletes the pointer in path->nodes[1] and frees the leaf | |
4645 | * block extent. zero is returned if it all worked out, < 0 otherwise. | |
4646 | * | |
4647 | * The path must have already been setup for deleting the leaf, including | |
4648 | * all the proper balancing. path->nodes[1] must be locked. | |
4649 | */ | |
143bede5 JM |
4650 | static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans, |
4651 | struct btrfs_root *root, | |
4652 | struct btrfs_path *path, | |
4653 | struct extent_buffer *leaf) | |
323ac95b | 4654 | { |
5d4f98a2 | 4655 | WARN_ON(btrfs_header_generation(leaf) != trans->transid); |
0e411ece | 4656 | del_ptr(trans, root, path, 1, path->slots[1]); |
323ac95b | 4657 | |
4d081c41 CM |
4658 | /* |
4659 | * btrfs_free_extent is expensive, we want to make sure we | |
4660 | * aren't holding any locks when we call it | |
4661 | */ | |
4662 | btrfs_unlock_up_safe(path, 0); | |
4663 | ||
f0486c68 YZ |
4664 | root_sub_used(root, leaf->len); |
4665 | ||
3083ee2e | 4666 | extent_buffer_get(leaf); |
5581a51a | 4667 | btrfs_free_tree_block(trans, root, leaf, 0, 1); |
3083ee2e | 4668 | free_extent_buffer_stale(leaf); |
323ac95b | 4669 | } |
74123bd7 CM |
4670 | /* |
4671 | * delete the item at the leaf level in path. If that empties | |
4672 | * the leaf, remove it from the tree | |
4673 | */ | |
85e21bac CM |
4674 | int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
4675 | struct btrfs_path *path, int slot, int nr) | |
be0e5c09 | 4676 | { |
5f39d397 CM |
4677 | struct extent_buffer *leaf; |
4678 | struct btrfs_item *item; | |
85e21bac CM |
4679 | int last_off; |
4680 | int dsize = 0; | |
aa5d6bed CM |
4681 | int ret = 0; |
4682 | int wret; | |
85e21bac | 4683 | int i; |
7518a238 | 4684 | u32 nritems; |
cfed81a0 CM |
4685 | struct btrfs_map_token token; |
4686 | ||
4687 | btrfs_init_map_token(&token); | |
be0e5c09 | 4688 | |
5f39d397 | 4689 | leaf = path->nodes[0]; |
85e21bac CM |
4690 | last_off = btrfs_item_offset_nr(leaf, slot + nr - 1); |
4691 | ||
4692 | for (i = 0; i < nr; i++) | |
4693 | dsize += btrfs_item_size_nr(leaf, slot + i); | |
4694 | ||
5f39d397 | 4695 | nritems = btrfs_header_nritems(leaf); |
be0e5c09 | 4696 | |
85e21bac | 4697 | if (slot + nr != nritems) { |
123abc88 | 4698 | int data_end = leaf_data_end(root, leaf); |
5f39d397 CM |
4699 | |
4700 | memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) + | |
d6025579 CM |
4701 | data_end + dsize, |
4702 | btrfs_leaf_data(leaf) + data_end, | |
85e21bac | 4703 | last_off - data_end); |
5f39d397 | 4704 | |
85e21bac | 4705 | for (i = slot + nr; i < nritems; i++) { |
5f39d397 | 4706 | u32 ioff; |
db94535d | 4707 | |
5f39d397 | 4708 | item = btrfs_item_nr(leaf, i); |
cfed81a0 CM |
4709 | ioff = btrfs_token_item_offset(leaf, item, &token); |
4710 | btrfs_set_token_item_offset(leaf, item, | |
4711 | ioff + dsize, &token); | |
0783fcfc | 4712 | } |
db94535d | 4713 | |
5f39d397 | 4714 | memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot), |
85e21bac | 4715 | btrfs_item_nr_offset(slot + nr), |
d6025579 | 4716 | sizeof(struct btrfs_item) * |
85e21bac | 4717 | (nritems - slot - nr)); |
be0e5c09 | 4718 | } |
85e21bac CM |
4719 | btrfs_set_header_nritems(leaf, nritems - nr); |
4720 | nritems -= nr; | |
5f39d397 | 4721 | |
74123bd7 | 4722 | /* delete the leaf if we've emptied it */ |
7518a238 | 4723 | if (nritems == 0) { |
5f39d397 CM |
4724 | if (leaf == root->node) { |
4725 | btrfs_set_header_level(leaf, 0); | |
9a8dd150 | 4726 | } else { |
f0486c68 YZ |
4727 | btrfs_set_path_blocking(path); |
4728 | clean_tree_block(trans, root, leaf); | |
143bede5 | 4729 | btrfs_del_leaf(trans, root, path, leaf); |
9a8dd150 | 4730 | } |
be0e5c09 | 4731 | } else { |
7518a238 | 4732 | int used = leaf_space_used(leaf, 0, nritems); |
aa5d6bed | 4733 | if (slot == 0) { |
5f39d397 CM |
4734 | struct btrfs_disk_key disk_key; |
4735 | ||
4736 | btrfs_item_key(leaf, &disk_key, 0); | |
143bede5 | 4737 | fixup_low_keys(trans, root, path, &disk_key, 1); |
aa5d6bed | 4738 | } |
aa5d6bed | 4739 | |
74123bd7 | 4740 | /* delete the leaf if it is mostly empty */ |
d717aa1d | 4741 | if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) { |
be0e5c09 CM |
4742 | /* push_leaf_left fixes the path. |
4743 | * make sure the path still points to our leaf | |
4744 | * for possible call to del_ptr below | |
4745 | */ | |
4920c9ac | 4746 | slot = path->slots[1]; |
5f39d397 CM |
4747 | extent_buffer_get(leaf); |
4748 | ||
b9473439 | 4749 | btrfs_set_path_blocking(path); |
99d8f83c CM |
4750 | wret = push_leaf_left(trans, root, path, 1, 1, |
4751 | 1, (u32)-1); | |
54aa1f4d | 4752 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed | 4753 | ret = wret; |
5f39d397 CM |
4754 | |
4755 | if (path->nodes[0] == leaf && | |
4756 | btrfs_header_nritems(leaf)) { | |
99d8f83c CM |
4757 | wret = push_leaf_right(trans, root, path, 1, |
4758 | 1, 1, 0); | |
54aa1f4d | 4759 | if (wret < 0 && wret != -ENOSPC) |
aa5d6bed CM |
4760 | ret = wret; |
4761 | } | |
5f39d397 CM |
4762 | |
4763 | if (btrfs_header_nritems(leaf) == 0) { | |
323ac95b | 4764 | path->slots[1] = slot; |
143bede5 | 4765 | btrfs_del_leaf(trans, root, path, leaf); |
5f39d397 | 4766 | free_extent_buffer(leaf); |
143bede5 | 4767 | ret = 0; |
5de08d7d | 4768 | } else { |
925baedd CM |
4769 | /* if we're still in the path, make sure |
4770 | * we're dirty. Otherwise, one of the | |
4771 | * push_leaf functions must have already | |
4772 | * dirtied this buffer | |
4773 | */ | |
4774 | if (path->nodes[0] == leaf) | |
4775 | btrfs_mark_buffer_dirty(leaf); | |
5f39d397 | 4776 | free_extent_buffer(leaf); |
be0e5c09 | 4777 | } |
d5719762 | 4778 | } else { |
5f39d397 | 4779 | btrfs_mark_buffer_dirty(leaf); |
be0e5c09 CM |
4780 | } |
4781 | } | |
aa5d6bed | 4782 | return ret; |
be0e5c09 CM |
4783 | } |
4784 | ||
7bb86316 | 4785 | /* |
925baedd | 4786 | * search the tree again to find a leaf with lesser keys |
7bb86316 CM |
4787 | * returns 0 if it found something or 1 if there are no lesser leaves. |
4788 | * returns < 0 on io errors. | |
d352ac68 CM |
4789 | * |
4790 | * This may release the path, and so you may lose any locks held at the | |
4791 | * time you call it. | |
7bb86316 CM |
4792 | */ |
4793 | int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path) | |
4794 | { | |
925baedd CM |
4795 | struct btrfs_key key; |
4796 | struct btrfs_disk_key found_key; | |
4797 | int ret; | |
7bb86316 | 4798 | |
925baedd | 4799 | btrfs_item_key_to_cpu(path->nodes[0], &key, 0); |
7bb86316 | 4800 | |
925baedd CM |
4801 | if (key.offset > 0) |
4802 | key.offset--; | |
4803 | else if (key.type > 0) | |
4804 | key.type--; | |
4805 | else if (key.objectid > 0) | |
4806 | key.objectid--; | |
4807 | else | |
4808 | return 1; | |
7bb86316 | 4809 | |
b3b4aa74 | 4810 | btrfs_release_path(path); |
925baedd CM |
4811 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4812 | if (ret < 0) | |
4813 | return ret; | |
4814 | btrfs_item_key(path->nodes[0], &found_key, 0); | |
4815 | ret = comp_keys(&found_key, &key); | |
4816 | if (ret < 0) | |
4817 | return 0; | |
4818 | return 1; | |
7bb86316 CM |
4819 | } |
4820 | ||
3f157a2f CM |
4821 | /* |
4822 | * A helper function to walk down the tree starting at min_key, and looking | |
de78b51a ES |
4823 | * for nodes or leaves that are have a minimum transaction id. |
4824 | * This is used by the btree defrag code, and tree logging | |
3f157a2f CM |
4825 | * |
4826 | * This does not cow, but it does stuff the starting key it finds back | |
4827 | * into min_key, so you can call btrfs_search_slot with cow=1 on the | |
4828 | * key and get a writable path. | |
4829 | * | |
4830 | * This does lock as it descends, and path->keep_locks should be set | |
4831 | * to 1 by the caller. | |
4832 | * | |
4833 | * This honors path->lowest_level to prevent descent past a given level | |
4834 | * of the tree. | |
4835 | * | |
d352ac68 CM |
4836 | * min_trans indicates the oldest transaction that you are interested |
4837 | * in walking through. Any nodes or leaves older than min_trans are | |
4838 | * skipped over (without reading them). | |
4839 | * | |
3f157a2f CM |
4840 | * returns zero if something useful was found, < 0 on error and 1 if there |
4841 | * was nothing in the tree that matched the search criteria. | |
4842 | */ | |
4843 | int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key, | |
e02119d5 | 4844 | struct btrfs_key *max_key, |
de78b51a | 4845 | struct btrfs_path *path, |
3f157a2f CM |
4846 | u64 min_trans) |
4847 | { | |
4848 | struct extent_buffer *cur; | |
4849 | struct btrfs_key found_key; | |
4850 | int slot; | |
9652480b | 4851 | int sret; |
3f157a2f CM |
4852 | u32 nritems; |
4853 | int level; | |
4854 | int ret = 1; | |
4855 | ||
934d375b | 4856 | WARN_ON(!path->keep_locks); |
3f157a2f | 4857 | again: |
bd681513 | 4858 | cur = btrfs_read_lock_root_node(root); |
3f157a2f | 4859 | level = btrfs_header_level(cur); |
e02119d5 | 4860 | WARN_ON(path->nodes[level]); |
3f157a2f | 4861 | path->nodes[level] = cur; |
bd681513 | 4862 | path->locks[level] = BTRFS_READ_LOCK; |
3f157a2f CM |
4863 | |
4864 | if (btrfs_header_generation(cur) < min_trans) { | |
4865 | ret = 1; | |
4866 | goto out; | |
4867 | } | |
d397712b | 4868 | while (1) { |
3f157a2f CM |
4869 | nritems = btrfs_header_nritems(cur); |
4870 | level = btrfs_header_level(cur); | |
9652480b | 4871 | sret = bin_search(cur, min_key, level, &slot); |
3f157a2f | 4872 | |
323ac95b CM |
4873 | /* at the lowest level, we're done, setup the path and exit */ |
4874 | if (level == path->lowest_level) { | |
e02119d5 CM |
4875 | if (slot >= nritems) |
4876 | goto find_next_key; | |
3f157a2f CM |
4877 | ret = 0; |
4878 | path->slots[level] = slot; | |
4879 | btrfs_item_key_to_cpu(cur, &found_key, slot); | |
4880 | goto out; | |
4881 | } | |
9652480b Y |
4882 | if (sret && slot > 0) |
4883 | slot--; | |
3f157a2f | 4884 | /* |
de78b51a ES |
4885 | * check this node pointer against the min_trans parameters. |
4886 | * If it is too old, old, skip to the next one. | |
3f157a2f | 4887 | */ |
d397712b | 4888 | while (slot < nritems) { |
3f157a2f CM |
4889 | u64 blockptr; |
4890 | u64 gen; | |
e02119d5 | 4891 | |
3f157a2f CM |
4892 | blockptr = btrfs_node_blockptr(cur, slot); |
4893 | gen = btrfs_node_ptr_generation(cur, slot); | |
4894 | if (gen < min_trans) { | |
4895 | slot++; | |
4896 | continue; | |
4897 | } | |
de78b51a | 4898 | break; |
3f157a2f | 4899 | } |
e02119d5 | 4900 | find_next_key: |
3f157a2f CM |
4901 | /* |
4902 | * we didn't find a candidate key in this node, walk forward | |
4903 | * and find another one | |
4904 | */ | |
4905 | if (slot >= nritems) { | |
e02119d5 | 4906 | path->slots[level] = slot; |
b4ce94de | 4907 | btrfs_set_path_blocking(path); |
e02119d5 | 4908 | sret = btrfs_find_next_key(root, path, min_key, level, |
de78b51a | 4909 | min_trans); |
e02119d5 | 4910 | if (sret == 0) { |
b3b4aa74 | 4911 | btrfs_release_path(path); |
3f157a2f CM |
4912 | goto again; |
4913 | } else { | |
4914 | goto out; | |
4915 | } | |
4916 | } | |
4917 | /* save our key for returning back */ | |
4918 | btrfs_node_key_to_cpu(cur, &found_key, slot); | |
4919 | path->slots[level] = slot; | |
4920 | if (level == path->lowest_level) { | |
4921 | ret = 0; | |
f7c79f30 | 4922 | unlock_up(path, level, 1, 0, NULL); |
3f157a2f CM |
4923 | goto out; |
4924 | } | |
b4ce94de | 4925 | btrfs_set_path_blocking(path); |
3f157a2f | 4926 | cur = read_node_slot(root, cur, slot); |
79787eaa | 4927 | BUG_ON(!cur); /* -ENOMEM */ |
3f157a2f | 4928 | |
bd681513 | 4929 | btrfs_tree_read_lock(cur); |
b4ce94de | 4930 | |
bd681513 | 4931 | path->locks[level - 1] = BTRFS_READ_LOCK; |
3f157a2f | 4932 | path->nodes[level - 1] = cur; |
f7c79f30 | 4933 | unlock_up(path, level, 1, 0, NULL); |
bd681513 | 4934 | btrfs_clear_path_blocking(path, NULL, 0); |
3f157a2f CM |
4935 | } |
4936 | out: | |
4937 | if (ret == 0) | |
4938 | memcpy(min_key, &found_key, sizeof(found_key)); | |
b4ce94de | 4939 | btrfs_set_path_blocking(path); |
3f157a2f CM |
4940 | return ret; |
4941 | } | |
4942 | ||
7069830a AB |
4943 | static void tree_move_down(struct btrfs_root *root, |
4944 | struct btrfs_path *path, | |
4945 | int *level, int root_level) | |
4946 | { | |
74dd17fb | 4947 | BUG_ON(*level == 0); |
7069830a AB |
4948 | path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level], |
4949 | path->slots[*level]); | |
4950 | path->slots[*level - 1] = 0; | |
4951 | (*level)--; | |
4952 | } | |
4953 | ||
4954 | static int tree_move_next_or_upnext(struct btrfs_root *root, | |
4955 | struct btrfs_path *path, | |
4956 | int *level, int root_level) | |
4957 | { | |
4958 | int ret = 0; | |
4959 | int nritems; | |
4960 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4961 | ||
4962 | path->slots[*level]++; | |
4963 | ||
74dd17fb | 4964 | while (path->slots[*level] >= nritems) { |
7069830a AB |
4965 | if (*level == root_level) |
4966 | return -1; | |
4967 | ||
4968 | /* move upnext */ | |
4969 | path->slots[*level] = 0; | |
4970 | free_extent_buffer(path->nodes[*level]); | |
4971 | path->nodes[*level] = NULL; | |
4972 | (*level)++; | |
4973 | path->slots[*level]++; | |
4974 | ||
4975 | nritems = btrfs_header_nritems(path->nodes[*level]); | |
4976 | ret = 1; | |
4977 | } | |
4978 | return ret; | |
4979 | } | |
4980 | ||
4981 | /* | |
4982 | * Returns 1 if it had to move up and next. 0 is returned if it moved only next | |
4983 | * or down. | |
4984 | */ | |
4985 | static int tree_advance(struct btrfs_root *root, | |
4986 | struct btrfs_path *path, | |
4987 | int *level, int root_level, | |
4988 | int allow_down, | |
4989 | struct btrfs_key *key) | |
4990 | { | |
4991 | int ret; | |
4992 | ||
4993 | if (*level == 0 || !allow_down) { | |
4994 | ret = tree_move_next_or_upnext(root, path, level, root_level); | |
4995 | } else { | |
4996 | tree_move_down(root, path, level, root_level); | |
4997 | ret = 0; | |
4998 | } | |
4999 | if (ret >= 0) { | |
5000 | if (*level == 0) | |
5001 | btrfs_item_key_to_cpu(path->nodes[*level], key, | |
5002 | path->slots[*level]); | |
5003 | else | |
5004 | btrfs_node_key_to_cpu(path->nodes[*level], key, | |
5005 | path->slots[*level]); | |
5006 | } | |
5007 | return ret; | |
5008 | } | |
5009 | ||
5010 | static int tree_compare_item(struct btrfs_root *left_root, | |
5011 | struct btrfs_path *left_path, | |
5012 | struct btrfs_path *right_path, | |
5013 | char *tmp_buf) | |
5014 | { | |
5015 | int cmp; | |
5016 | int len1, len2; | |
5017 | unsigned long off1, off2; | |
5018 | ||
5019 | len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]); | |
5020 | len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]); | |
5021 | if (len1 != len2) | |
5022 | return 1; | |
5023 | ||
5024 | off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]); | |
5025 | off2 = btrfs_item_ptr_offset(right_path->nodes[0], | |
5026 | right_path->slots[0]); | |
5027 | ||
5028 | read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1); | |
5029 | ||
5030 | cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1); | |
5031 | if (cmp) | |
5032 | return 1; | |
5033 | return 0; | |
5034 | } | |
5035 | ||
5036 | #define ADVANCE 1 | |
5037 | #define ADVANCE_ONLY_NEXT -1 | |
5038 | ||
5039 | /* | |
5040 | * This function compares two trees and calls the provided callback for | |
5041 | * every changed/new/deleted item it finds. | |
5042 | * If shared tree blocks are encountered, whole subtrees are skipped, making | |
5043 | * the compare pretty fast on snapshotted subvolumes. | |
5044 | * | |
5045 | * This currently works on commit roots only. As commit roots are read only, | |
5046 | * we don't do any locking. The commit roots are protected with transactions. | |
5047 | * Transactions are ended and rejoined when a commit is tried in between. | |
5048 | * | |
5049 | * This function checks for modifications done to the trees while comparing. | |
5050 | * If it detects a change, it aborts immediately. | |
5051 | */ | |
5052 | int btrfs_compare_trees(struct btrfs_root *left_root, | |
5053 | struct btrfs_root *right_root, | |
5054 | btrfs_changed_cb_t changed_cb, void *ctx) | |
5055 | { | |
5056 | int ret; | |
5057 | int cmp; | |
5058 | struct btrfs_trans_handle *trans = NULL; | |
5059 | struct btrfs_path *left_path = NULL; | |
5060 | struct btrfs_path *right_path = NULL; | |
5061 | struct btrfs_key left_key; | |
5062 | struct btrfs_key right_key; | |
5063 | char *tmp_buf = NULL; | |
5064 | int left_root_level; | |
5065 | int right_root_level; | |
5066 | int left_level; | |
5067 | int right_level; | |
5068 | int left_end_reached; | |
5069 | int right_end_reached; | |
5070 | int advance_left; | |
5071 | int advance_right; | |
5072 | u64 left_blockptr; | |
5073 | u64 right_blockptr; | |
5074 | u64 left_start_ctransid; | |
5075 | u64 right_start_ctransid; | |
5076 | u64 ctransid; | |
5077 | ||
5078 | left_path = btrfs_alloc_path(); | |
5079 | if (!left_path) { | |
5080 | ret = -ENOMEM; | |
5081 | goto out; | |
5082 | } | |
5083 | right_path = btrfs_alloc_path(); | |
5084 | if (!right_path) { | |
5085 | ret = -ENOMEM; | |
5086 | goto out; | |
5087 | } | |
5088 | ||
5089 | tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS); | |
5090 | if (!tmp_buf) { | |
5091 | ret = -ENOMEM; | |
5092 | goto out; | |
5093 | } | |
5094 | ||
5095 | left_path->search_commit_root = 1; | |
5096 | left_path->skip_locking = 1; | |
5097 | right_path->search_commit_root = 1; | |
5098 | right_path->skip_locking = 1; | |
5099 | ||
5f3ab90a | 5100 | spin_lock(&left_root->root_item_lock); |
7069830a | 5101 | left_start_ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5102 | spin_unlock(&left_root->root_item_lock); |
7069830a | 5103 | |
5f3ab90a | 5104 | spin_lock(&right_root->root_item_lock); |
7069830a | 5105 | right_start_ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5106 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5107 | |
5108 | trans = btrfs_join_transaction(left_root); | |
5109 | if (IS_ERR(trans)) { | |
5110 | ret = PTR_ERR(trans); | |
5111 | trans = NULL; | |
5112 | goto out; | |
5113 | } | |
5114 | ||
5115 | /* | |
5116 | * Strategy: Go to the first items of both trees. Then do | |
5117 | * | |
5118 | * If both trees are at level 0 | |
5119 | * Compare keys of current items | |
5120 | * If left < right treat left item as new, advance left tree | |
5121 | * and repeat | |
5122 | * If left > right treat right item as deleted, advance right tree | |
5123 | * and repeat | |
5124 | * If left == right do deep compare of items, treat as changed if | |
5125 | * needed, advance both trees and repeat | |
5126 | * If both trees are at the same level but not at level 0 | |
5127 | * Compare keys of current nodes/leafs | |
5128 | * If left < right advance left tree and repeat | |
5129 | * If left > right advance right tree and repeat | |
5130 | * If left == right compare blockptrs of the next nodes/leafs | |
5131 | * If they match advance both trees but stay at the same level | |
5132 | * and repeat | |
5133 | * If they don't match advance both trees while allowing to go | |
5134 | * deeper and repeat | |
5135 | * If tree levels are different | |
5136 | * Advance the tree that needs it and repeat | |
5137 | * | |
5138 | * Advancing a tree means: | |
5139 | * If we are at level 0, try to go to the next slot. If that's not | |
5140 | * possible, go one level up and repeat. Stop when we found a level | |
5141 | * where we could go to the next slot. We may at this point be on a | |
5142 | * node or a leaf. | |
5143 | * | |
5144 | * If we are not at level 0 and not on shared tree blocks, go one | |
5145 | * level deeper. | |
5146 | * | |
5147 | * If we are not at level 0 and on shared tree blocks, go one slot to | |
5148 | * the right if possible or go up and right. | |
5149 | */ | |
5150 | ||
5151 | left_level = btrfs_header_level(left_root->commit_root); | |
5152 | left_root_level = left_level; | |
5153 | left_path->nodes[left_level] = left_root->commit_root; | |
5154 | extent_buffer_get(left_path->nodes[left_level]); | |
5155 | ||
5156 | right_level = btrfs_header_level(right_root->commit_root); | |
5157 | right_root_level = right_level; | |
5158 | right_path->nodes[right_level] = right_root->commit_root; | |
5159 | extent_buffer_get(right_path->nodes[right_level]); | |
5160 | ||
5161 | if (left_level == 0) | |
5162 | btrfs_item_key_to_cpu(left_path->nodes[left_level], | |
5163 | &left_key, left_path->slots[left_level]); | |
5164 | else | |
5165 | btrfs_node_key_to_cpu(left_path->nodes[left_level], | |
5166 | &left_key, left_path->slots[left_level]); | |
5167 | if (right_level == 0) | |
5168 | btrfs_item_key_to_cpu(right_path->nodes[right_level], | |
5169 | &right_key, right_path->slots[right_level]); | |
5170 | else | |
5171 | btrfs_node_key_to_cpu(right_path->nodes[right_level], | |
5172 | &right_key, right_path->slots[right_level]); | |
5173 | ||
5174 | left_end_reached = right_end_reached = 0; | |
5175 | advance_left = advance_right = 0; | |
5176 | ||
5177 | while (1) { | |
5178 | /* | |
5179 | * We need to make sure the transaction does not get committed | |
5180 | * while we do anything on commit roots. This means, we need to | |
5181 | * join and leave transactions for every item that we process. | |
5182 | */ | |
5183 | if (trans && btrfs_should_end_transaction(trans, left_root)) { | |
5184 | btrfs_release_path(left_path); | |
5185 | btrfs_release_path(right_path); | |
5186 | ||
5187 | ret = btrfs_end_transaction(trans, left_root); | |
5188 | trans = NULL; | |
5189 | if (ret < 0) | |
5190 | goto out; | |
5191 | } | |
5192 | /* now rejoin the transaction */ | |
5193 | if (!trans) { | |
5194 | trans = btrfs_join_transaction(left_root); | |
5195 | if (IS_ERR(trans)) { | |
5196 | ret = PTR_ERR(trans); | |
5197 | trans = NULL; | |
5198 | goto out; | |
5199 | } | |
5200 | ||
5f3ab90a | 5201 | spin_lock(&left_root->root_item_lock); |
7069830a | 5202 | ctransid = btrfs_root_ctransid(&left_root->root_item); |
5f3ab90a | 5203 | spin_unlock(&left_root->root_item_lock); |
7069830a AB |
5204 | if (ctransid != left_start_ctransid) |
5205 | left_start_ctransid = 0; | |
5206 | ||
5f3ab90a | 5207 | spin_lock(&right_root->root_item_lock); |
7069830a | 5208 | ctransid = btrfs_root_ctransid(&right_root->root_item); |
5f3ab90a | 5209 | spin_unlock(&right_root->root_item_lock); |
7069830a AB |
5210 | if (ctransid != right_start_ctransid) |
5211 | right_start_ctransid = 0; | |
5212 | ||
5213 | if (!left_start_ctransid || !right_start_ctransid) { | |
5214 | WARN(1, KERN_WARNING | |
5215 | "btrfs: btrfs_compare_tree detected " | |
5216 | "a change in one of the trees while " | |
5217 | "iterating. This is probably a " | |
5218 | "bug.\n"); | |
5219 | ret = -EIO; | |
5220 | goto out; | |
5221 | } | |
5222 | ||
5223 | /* | |
5224 | * the commit root may have changed, so start again | |
5225 | * where we stopped | |
5226 | */ | |
5227 | left_path->lowest_level = left_level; | |
5228 | right_path->lowest_level = right_level; | |
5229 | ret = btrfs_search_slot(NULL, left_root, | |
5230 | &left_key, left_path, 0, 0); | |
5231 | if (ret < 0) | |
5232 | goto out; | |
5233 | ret = btrfs_search_slot(NULL, right_root, | |
5234 | &right_key, right_path, 0, 0); | |
5235 | if (ret < 0) | |
5236 | goto out; | |
5237 | } | |
5238 | ||
5239 | if (advance_left && !left_end_reached) { | |
5240 | ret = tree_advance(left_root, left_path, &left_level, | |
5241 | left_root_level, | |
5242 | advance_left != ADVANCE_ONLY_NEXT, | |
5243 | &left_key); | |
5244 | if (ret < 0) | |
5245 | left_end_reached = ADVANCE; | |
5246 | advance_left = 0; | |
5247 | } | |
5248 | if (advance_right && !right_end_reached) { | |
5249 | ret = tree_advance(right_root, right_path, &right_level, | |
5250 | right_root_level, | |
5251 | advance_right != ADVANCE_ONLY_NEXT, | |
5252 | &right_key); | |
5253 | if (ret < 0) | |
5254 | right_end_reached = ADVANCE; | |
5255 | advance_right = 0; | |
5256 | } | |
5257 | ||
5258 | if (left_end_reached && right_end_reached) { | |
5259 | ret = 0; | |
5260 | goto out; | |
5261 | } else if (left_end_reached) { | |
5262 | if (right_level == 0) { | |
5263 | ret = changed_cb(left_root, right_root, | |
5264 | left_path, right_path, | |
5265 | &right_key, | |
5266 | BTRFS_COMPARE_TREE_DELETED, | |
5267 | ctx); | |
5268 | if (ret < 0) | |
5269 | goto out; | |
5270 | } | |
5271 | advance_right = ADVANCE; | |
5272 | continue; | |
5273 | } else if (right_end_reached) { | |
5274 | if (left_level == 0) { | |
5275 | ret = changed_cb(left_root, right_root, | |
5276 | left_path, right_path, | |
5277 | &left_key, | |
5278 | BTRFS_COMPARE_TREE_NEW, | |
5279 | ctx); | |
5280 | if (ret < 0) | |
5281 | goto out; | |
5282 | } | |
5283 | advance_left = ADVANCE; | |
5284 | continue; | |
5285 | } | |
5286 | ||
5287 | if (left_level == 0 && right_level == 0) { | |
5288 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5289 | if (cmp < 0) { | |
5290 | ret = changed_cb(left_root, right_root, | |
5291 | left_path, right_path, | |
5292 | &left_key, | |
5293 | BTRFS_COMPARE_TREE_NEW, | |
5294 | ctx); | |
5295 | if (ret < 0) | |
5296 | goto out; | |
5297 | advance_left = ADVANCE; | |
5298 | } else if (cmp > 0) { | |
5299 | ret = changed_cb(left_root, right_root, | |
5300 | left_path, right_path, | |
5301 | &right_key, | |
5302 | BTRFS_COMPARE_TREE_DELETED, | |
5303 | ctx); | |
5304 | if (ret < 0) | |
5305 | goto out; | |
5306 | advance_right = ADVANCE; | |
5307 | } else { | |
74dd17fb | 5308 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5309 | ret = tree_compare_item(left_root, left_path, |
5310 | right_path, tmp_buf); | |
5311 | if (ret) { | |
74dd17fb | 5312 | WARN_ON(!extent_buffer_uptodate(left_path->nodes[0])); |
7069830a AB |
5313 | ret = changed_cb(left_root, right_root, |
5314 | left_path, right_path, | |
5315 | &left_key, | |
5316 | BTRFS_COMPARE_TREE_CHANGED, | |
5317 | ctx); | |
5318 | if (ret < 0) | |
5319 | goto out; | |
5320 | } | |
5321 | advance_left = ADVANCE; | |
5322 | advance_right = ADVANCE; | |
5323 | } | |
5324 | } else if (left_level == right_level) { | |
5325 | cmp = btrfs_comp_cpu_keys(&left_key, &right_key); | |
5326 | if (cmp < 0) { | |
5327 | advance_left = ADVANCE; | |
5328 | } else if (cmp > 0) { | |
5329 | advance_right = ADVANCE; | |
5330 | } else { | |
5331 | left_blockptr = btrfs_node_blockptr( | |
5332 | left_path->nodes[left_level], | |
5333 | left_path->slots[left_level]); | |
5334 | right_blockptr = btrfs_node_blockptr( | |
5335 | right_path->nodes[right_level], | |
5336 | right_path->slots[right_level]); | |
5337 | if (left_blockptr == right_blockptr) { | |
5338 | /* | |
5339 | * As we're on a shared block, don't | |
5340 | * allow to go deeper. | |
5341 | */ | |
5342 | advance_left = ADVANCE_ONLY_NEXT; | |
5343 | advance_right = ADVANCE_ONLY_NEXT; | |
5344 | } else { | |
5345 | advance_left = ADVANCE; | |
5346 | advance_right = ADVANCE; | |
5347 | } | |
5348 | } | |
5349 | } else if (left_level < right_level) { | |
5350 | advance_right = ADVANCE; | |
5351 | } else { | |
5352 | advance_left = ADVANCE; | |
5353 | } | |
5354 | } | |
5355 | ||
5356 | out: | |
5357 | btrfs_free_path(left_path); | |
5358 | btrfs_free_path(right_path); | |
5359 | kfree(tmp_buf); | |
5360 | ||
5361 | if (trans) { | |
5362 | if (!ret) | |
5363 | ret = btrfs_end_transaction(trans, left_root); | |
5364 | else | |
5365 | btrfs_end_transaction(trans, left_root); | |
5366 | } | |
5367 | ||
5368 | return ret; | |
5369 | } | |
5370 | ||
3f157a2f CM |
5371 | /* |
5372 | * this is similar to btrfs_next_leaf, but does not try to preserve | |
5373 | * and fixup the path. It looks for and returns the next key in the | |
de78b51a | 5374 | * tree based on the current path and the min_trans parameters. |
3f157a2f CM |
5375 | * |
5376 | * 0 is returned if another key is found, < 0 if there are any errors | |
5377 | * and 1 is returned if there are no higher keys in the tree | |
5378 | * | |
5379 | * path->keep_locks should be set to 1 on the search made before | |
5380 | * calling this function. | |
5381 | */ | |
e7a84565 | 5382 | int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path, |
de78b51a | 5383 | struct btrfs_key *key, int level, u64 min_trans) |
e7a84565 | 5384 | { |
e7a84565 CM |
5385 | int slot; |
5386 | struct extent_buffer *c; | |
5387 | ||
934d375b | 5388 | WARN_ON(!path->keep_locks); |
d397712b | 5389 | while (level < BTRFS_MAX_LEVEL) { |
e7a84565 CM |
5390 | if (!path->nodes[level]) |
5391 | return 1; | |
5392 | ||
5393 | slot = path->slots[level] + 1; | |
5394 | c = path->nodes[level]; | |
3f157a2f | 5395 | next: |
e7a84565 | 5396 | if (slot >= btrfs_header_nritems(c)) { |
33c66f43 YZ |
5397 | int ret; |
5398 | int orig_lowest; | |
5399 | struct btrfs_key cur_key; | |
5400 | if (level + 1 >= BTRFS_MAX_LEVEL || | |
5401 | !path->nodes[level + 1]) | |
e7a84565 | 5402 | return 1; |
33c66f43 YZ |
5403 | |
5404 | if (path->locks[level + 1]) { | |
5405 | level++; | |
5406 | continue; | |
5407 | } | |
5408 | ||
5409 | slot = btrfs_header_nritems(c) - 1; | |
5410 | if (level == 0) | |
5411 | btrfs_item_key_to_cpu(c, &cur_key, slot); | |
5412 | else | |
5413 | btrfs_node_key_to_cpu(c, &cur_key, slot); | |
5414 | ||
5415 | orig_lowest = path->lowest_level; | |
b3b4aa74 | 5416 | btrfs_release_path(path); |
33c66f43 YZ |
5417 | path->lowest_level = level; |
5418 | ret = btrfs_search_slot(NULL, root, &cur_key, path, | |
5419 | 0, 0); | |
5420 | path->lowest_level = orig_lowest; | |
5421 | if (ret < 0) | |
5422 | return ret; | |
5423 | ||
5424 | c = path->nodes[level]; | |
5425 | slot = path->slots[level]; | |
5426 | if (ret == 0) | |
5427 | slot++; | |
5428 | goto next; | |
e7a84565 | 5429 | } |
33c66f43 | 5430 | |
e7a84565 CM |
5431 | if (level == 0) |
5432 | btrfs_item_key_to_cpu(c, key, slot); | |
3f157a2f | 5433 | else { |
3f157a2f CM |
5434 | u64 gen = btrfs_node_ptr_generation(c, slot); |
5435 | ||
3f157a2f CM |
5436 | if (gen < min_trans) { |
5437 | slot++; | |
5438 | goto next; | |
5439 | } | |
e7a84565 | 5440 | btrfs_node_key_to_cpu(c, key, slot); |
3f157a2f | 5441 | } |
e7a84565 CM |
5442 | return 0; |
5443 | } | |
5444 | return 1; | |
5445 | } | |
5446 | ||
97571fd0 | 5447 | /* |
925baedd | 5448 | * search the tree again to find a leaf with greater keys |
0f70abe2 CM |
5449 | * returns 0 if it found something or 1 if there are no greater leaves. |
5450 | * returns < 0 on io errors. | |
97571fd0 | 5451 | */ |
234b63a0 | 5452 | int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path) |
3d7806ec JS |
5453 | { |
5454 | return btrfs_next_old_leaf(root, path, 0); | |
5455 | } | |
5456 | ||
70c8a91c JB |
5457 | /* Release the path up to but not including the given level */ |
5458 | static void btrfs_release_level(struct btrfs_path *path, int level) | |
5459 | { | |
5460 | int i; | |
5461 | ||
5462 | for (i = 0; i < level; i++) { | |
5463 | path->slots[i] = 0; | |
5464 | if (!path->nodes[i]) | |
5465 | continue; | |
5466 | if (path->locks[i]) { | |
5467 | btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]); | |
5468 | path->locks[i] = 0; | |
5469 | } | |
5470 | free_extent_buffer(path->nodes[i]); | |
5471 | path->nodes[i] = NULL; | |
5472 | } | |
5473 | } | |
5474 | ||
5475 | /* | |
5476 | * This function assumes 2 things | |
5477 | * | |
5478 | * 1) You are using path->keep_locks | |
5479 | * 2) You are not inserting items. | |
5480 | * | |
5481 | * If either of these are not true do not use this function. If you need a next | |
5482 | * leaf with either of these not being true then this function can be easily | |
5483 | * adapted to do that, but at the moment these are the limitations. | |
5484 | */ | |
5485 | int btrfs_next_leaf_write(struct btrfs_trans_handle *trans, | |
5486 | struct btrfs_root *root, struct btrfs_path *path, | |
5487 | int del) | |
5488 | { | |
5489 | struct extent_buffer *b; | |
5490 | struct btrfs_key key; | |
5491 | u32 nritems; | |
5492 | int level = 1; | |
5493 | int slot; | |
5494 | int ret = 1; | |
5495 | int write_lock_level = BTRFS_MAX_LEVEL; | |
5496 | int ins_len = del ? -1 : 0; | |
5497 | ||
5498 | WARN_ON(!(path->keep_locks || path->really_keep_locks)); | |
5499 | ||
5500 | nritems = btrfs_header_nritems(path->nodes[0]); | |
5501 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); | |
5502 | ||
5503 | while (path->nodes[level]) { | |
5504 | nritems = btrfs_header_nritems(path->nodes[level]); | |
5505 | if (!(path->locks[level] & BTRFS_WRITE_LOCK)) { | |
5506 | search: | |
5507 | btrfs_release_path(path); | |
5508 | ret = btrfs_search_slot(trans, root, &key, path, | |
5509 | ins_len, 1); | |
5510 | if (ret < 0) | |
5511 | goto out; | |
5512 | level = 1; | |
5513 | continue; | |
5514 | } | |
5515 | ||
5516 | if (path->slots[level] >= nritems - 1) { | |
5517 | level++; | |
5518 | continue; | |
5519 | } | |
5520 | ||
5521 | btrfs_release_level(path, level); | |
5522 | break; | |
5523 | } | |
5524 | ||
5525 | if (!path->nodes[level]) { | |
5526 | ret = 1; | |
5527 | goto out; | |
5528 | } | |
5529 | ||
5530 | path->slots[level]++; | |
5531 | b = path->nodes[level]; | |
5532 | ||
5533 | while (b) { | |
5534 | level = btrfs_header_level(b); | |
5535 | ||
5536 | if (!should_cow_block(trans, root, b)) | |
5537 | goto cow_done; | |
5538 | ||
5539 | btrfs_set_path_blocking(path); | |
5540 | ret = btrfs_cow_block(trans, root, b, | |
5541 | path->nodes[level + 1], | |
5542 | path->slots[level + 1], &b); | |
5543 | if (ret) | |
5544 | goto out; | |
5545 | cow_done: | |
5546 | path->nodes[level] = b; | |
5547 | btrfs_clear_path_blocking(path, NULL, 0); | |
5548 | if (level != 0) { | |
5549 | ret = setup_nodes_for_search(trans, root, path, b, | |
5550 | level, ins_len, | |
5551 | &write_lock_level); | |
5552 | if (ret == -EAGAIN) | |
5553 | goto search; | |
5554 | if (ret) | |
5555 | goto out; | |
5556 | ||
5557 | b = path->nodes[level]; | |
5558 | slot = path->slots[level]; | |
5559 | ||
5560 | ret = read_block_for_search(trans, root, path, | |
5561 | &b, level, slot, &key, 0); | |
5562 | if (ret == -EAGAIN) | |
5563 | goto search; | |
5564 | if (ret) | |
5565 | goto out; | |
5566 | level = btrfs_header_level(b); | |
5567 | if (!btrfs_try_tree_write_lock(b)) { | |
5568 | btrfs_set_path_blocking(path); | |
5569 | btrfs_tree_lock(b); | |
5570 | btrfs_clear_path_blocking(path, b, | |
5571 | BTRFS_WRITE_LOCK); | |
5572 | } | |
5573 | path->locks[level] = BTRFS_WRITE_LOCK; | |
5574 | path->nodes[level] = b; | |
5575 | path->slots[level] = 0; | |
5576 | } else { | |
5577 | path->slots[level] = 0; | |
5578 | ret = 0; | |
5579 | break; | |
5580 | } | |
5581 | } | |
5582 | ||
5583 | out: | |
5584 | if (ret) | |
5585 | btrfs_release_path(path); | |
5586 | ||
5587 | return ret; | |
5588 | } | |
5589 | ||
3d7806ec JS |
5590 | int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path, |
5591 | u64 time_seq) | |
d97e63b6 CM |
5592 | { |
5593 | int slot; | |
8e73f275 | 5594 | int level; |
5f39d397 | 5595 | struct extent_buffer *c; |
8e73f275 | 5596 | struct extent_buffer *next; |
925baedd CM |
5597 | struct btrfs_key key; |
5598 | u32 nritems; | |
5599 | int ret; | |
8e73f275 | 5600 | int old_spinning = path->leave_spinning; |
bd681513 | 5601 | int next_rw_lock = 0; |
925baedd CM |
5602 | |
5603 | nritems = btrfs_header_nritems(path->nodes[0]); | |
d397712b | 5604 | if (nritems == 0) |
925baedd | 5605 | return 1; |
925baedd | 5606 | |
8e73f275 CM |
5607 | btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1); |
5608 | again: | |
5609 | level = 1; | |
5610 | next = NULL; | |
bd681513 | 5611 | next_rw_lock = 0; |
b3b4aa74 | 5612 | btrfs_release_path(path); |
8e73f275 | 5613 | |
a2135011 | 5614 | path->keep_locks = 1; |
31533fb2 | 5615 | path->leave_spinning = 1; |
8e73f275 | 5616 | |
3d7806ec JS |
5617 | if (time_seq) |
5618 | ret = btrfs_search_old_slot(root, &key, path, time_seq); | |
5619 | else | |
5620 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
925baedd CM |
5621 | path->keep_locks = 0; |
5622 | ||
5623 | if (ret < 0) | |
5624 | return ret; | |
5625 | ||
a2135011 | 5626 | nritems = btrfs_header_nritems(path->nodes[0]); |
168fd7d2 CM |
5627 | /* |
5628 | * by releasing the path above we dropped all our locks. A balance | |
5629 | * could have added more items next to the key that used to be | |
5630 | * at the very end of the block. So, check again here and | |
5631 | * advance the path if there are now more items available. | |
5632 | */ | |
a2135011 | 5633 | if (nritems > 0 && path->slots[0] < nritems - 1) { |
e457afec YZ |
5634 | if (ret == 0) |
5635 | path->slots[0]++; | |
8e73f275 | 5636 | ret = 0; |
925baedd CM |
5637 | goto done; |
5638 | } | |
d97e63b6 | 5639 | |
d397712b | 5640 | while (level < BTRFS_MAX_LEVEL) { |
8e73f275 CM |
5641 | if (!path->nodes[level]) { |
5642 | ret = 1; | |
5643 | goto done; | |
5644 | } | |
5f39d397 | 5645 | |
d97e63b6 CM |
5646 | slot = path->slots[level] + 1; |
5647 | c = path->nodes[level]; | |
5f39d397 | 5648 | if (slot >= btrfs_header_nritems(c)) { |
d97e63b6 | 5649 | level++; |
8e73f275 CM |
5650 | if (level == BTRFS_MAX_LEVEL) { |
5651 | ret = 1; | |
5652 | goto done; | |
5653 | } | |
d97e63b6 CM |
5654 | continue; |
5655 | } | |
5f39d397 | 5656 | |
925baedd | 5657 | if (next) { |
bd681513 | 5658 | btrfs_tree_unlock_rw(next, next_rw_lock); |
5f39d397 | 5659 | free_extent_buffer(next); |
925baedd | 5660 | } |
5f39d397 | 5661 | |
8e73f275 | 5662 | next = c; |
bd681513 | 5663 | next_rw_lock = path->locks[level]; |
8e73f275 | 5664 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5665 | slot, &key, 0); |
8e73f275 CM |
5666 | if (ret == -EAGAIN) |
5667 | goto again; | |
5f39d397 | 5668 | |
76a05b35 | 5669 | if (ret < 0) { |
b3b4aa74 | 5670 | btrfs_release_path(path); |
76a05b35 CM |
5671 | goto done; |
5672 | } | |
5673 | ||
5cd57b2c | 5674 | if (!path->skip_locking) { |
bd681513 | 5675 | ret = btrfs_try_tree_read_lock(next); |
d42244a0 JS |
5676 | if (!ret && time_seq) { |
5677 | /* | |
5678 | * If we don't get the lock, we may be racing | |
5679 | * with push_leaf_left, holding that lock while | |
5680 | * itself waiting for the leaf we've currently | |
5681 | * locked. To solve this situation, we give up | |
5682 | * on our lock and cycle. | |
5683 | */ | |
cf538830 | 5684 | free_extent_buffer(next); |
d42244a0 JS |
5685 | btrfs_release_path(path); |
5686 | cond_resched(); | |
5687 | goto again; | |
5688 | } | |
8e73f275 CM |
5689 | if (!ret) { |
5690 | btrfs_set_path_blocking(path); | |
bd681513 | 5691 | btrfs_tree_read_lock(next); |
31533fb2 | 5692 | btrfs_clear_path_blocking(path, next, |
bd681513 | 5693 | BTRFS_READ_LOCK); |
8e73f275 | 5694 | } |
31533fb2 | 5695 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5696 | } |
d97e63b6 CM |
5697 | break; |
5698 | } | |
5699 | path->slots[level] = slot; | |
d397712b | 5700 | while (1) { |
d97e63b6 CM |
5701 | level--; |
5702 | c = path->nodes[level]; | |
925baedd | 5703 | if (path->locks[level]) |
bd681513 | 5704 | btrfs_tree_unlock_rw(c, path->locks[level]); |
8e73f275 | 5705 | |
5f39d397 | 5706 | free_extent_buffer(c); |
d97e63b6 CM |
5707 | path->nodes[level] = next; |
5708 | path->slots[level] = 0; | |
a74a4b97 | 5709 | if (!path->skip_locking) |
bd681513 | 5710 | path->locks[level] = next_rw_lock; |
d97e63b6 CM |
5711 | if (!level) |
5712 | break; | |
b4ce94de | 5713 | |
8e73f275 | 5714 | ret = read_block_for_search(NULL, root, path, &next, level, |
5d9e75c4 | 5715 | 0, &key, 0); |
8e73f275 CM |
5716 | if (ret == -EAGAIN) |
5717 | goto again; | |
5718 | ||
76a05b35 | 5719 | if (ret < 0) { |
b3b4aa74 | 5720 | btrfs_release_path(path); |
76a05b35 CM |
5721 | goto done; |
5722 | } | |
5723 | ||
5cd57b2c | 5724 | if (!path->skip_locking) { |
bd681513 | 5725 | ret = btrfs_try_tree_read_lock(next); |
8e73f275 CM |
5726 | if (!ret) { |
5727 | btrfs_set_path_blocking(path); | |
bd681513 | 5728 | btrfs_tree_read_lock(next); |
31533fb2 | 5729 | btrfs_clear_path_blocking(path, next, |
bd681513 CM |
5730 | BTRFS_READ_LOCK); |
5731 | } | |
31533fb2 | 5732 | next_rw_lock = BTRFS_READ_LOCK; |
5cd57b2c | 5733 | } |
d97e63b6 | 5734 | } |
8e73f275 | 5735 | ret = 0; |
925baedd | 5736 | done: |
f7c79f30 | 5737 | unlock_up(path, 0, 1, 0, NULL); |
8e73f275 CM |
5738 | path->leave_spinning = old_spinning; |
5739 | if (!old_spinning) | |
5740 | btrfs_set_path_blocking(path); | |
5741 | ||
5742 | return ret; | |
d97e63b6 | 5743 | } |
0b86a832 | 5744 | |
3f157a2f CM |
5745 | /* |
5746 | * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps | |
5747 | * searching until it gets past min_objectid or finds an item of 'type' | |
5748 | * | |
5749 | * returns 0 if something is found, 1 if nothing was found and < 0 on error | |
5750 | */ | |
0b86a832 CM |
5751 | int btrfs_previous_item(struct btrfs_root *root, |
5752 | struct btrfs_path *path, u64 min_objectid, | |
5753 | int type) | |
5754 | { | |
5755 | struct btrfs_key found_key; | |
5756 | struct extent_buffer *leaf; | |
e02119d5 | 5757 | u32 nritems; |
0b86a832 CM |
5758 | int ret; |
5759 | ||
d397712b | 5760 | while (1) { |
0b86a832 | 5761 | if (path->slots[0] == 0) { |
b4ce94de | 5762 | btrfs_set_path_blocking(path); |
0b86a832 CM |
5763 | ret = btrfs_prev_leaf(root, path); |
5764 | if (ret != 0) | |
5765 | return ret; | |
5766 | } else { | |
5767 | path->slots[0]--; | |
5768 | } | |
5769 | leaf = path->nodes[0]; | |
e02119d5 CM |
5770 | nritems = btrfs_header_nritems(leaf); |
5771 | if (nritems == 0) | |
5772 | return 1; | |
5773 | if (path->slots[0] == nritems) | |
5774 | path->slots[0]--; | |
5775 | ||
0b86a832 | 5776 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
e02119d5 CM |
5777 | if (found_key.objectid < min_objectid) |
5778 | break; | |
0a4eefbb YZ |
5779 | if (found_key.type == type) |
5780 | return 0; | |
e02119d5 CM |
5781 | if (found_key.objectid == min_objectid && |
5782 | found_key.type < type) | |
5783 | break; | |
0b86a832 CM |
5784 | } |
5785 | return 1; | |
5786 | } |